]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
core/manager: split out creation of serialization fd out to a helper
[thirdparty/systemd.git] / src / core / manager.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <errno.h>
21 #include <fcntl.h>
22 #include <linux/kd.h>
23 #include <signal.h>
24 #include <string.h>
25 #include <sys/epoll.h>
26 #include <sys/inotify.h>
27 #include <sys/ioctl.h>
28 #include <sys/reboot.h>
29 #include <sys/timerfd.h>
30 #include <sys/wait.h>
31 #include <unistd.h>
32
33 #ifdef HAVE_AUDIT
34 #include <libaudit.h>
35 #endif
36
37 #include "sd-daemon.h"
38 #include "sd-messages.h"
39
40 #include "alloc-util.h"
41 #include "audit-fd.h"
42 #include "boot-timestamps.h"
43 #include "bus-common-errors.h"
44 #include "bus-error.h"
45 #include "bus-kernel.h"
46 #include "bus-util.h"
47 #include "clean-ipc.h"
48 #include "dbus-job.h"
49 #include "dbus-manager.h"
50 #include "dbus-unit.h"
51 #include "dbus.h"
52 #include "dirent-util.h"
53 #include "env-util.h"
54 #include "escape.h"
55 #include "exec-util.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 /* We do not want to show the cylon animation if the user
115 * needs to confirm service executions otherwise confirmation
116 * messages will be screwed by the cylon animation. */
117 if (!manager_is_confirm_spawn_disabled(m))
118 return;
119
120 if (m->jobs_in_progress_event_source)
121 return;
122
123 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
124 r = sd_event_add_time(
125 m->event,
126 &m->jobs_in_progress_event_source,
127 CLOCK_MONOTONIC,
128 next, 0,
129 manager_dispatch_jobs_in_progress, m);
130 if (r < 0)
131 return;
132
133 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
134 }
135
136 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
137
138 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
139 char *p = buffer;
140
141 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
142 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
143
144 if (pos > 1) {
145 if (pos > 2)
146 p = mempset(p, ' ', pos-2);
147 if (log_get_show_color())
148 p = stpcpy(p, ANSI_RED);
149 *p++ = '*';
150 }
151
152 if (pos > 0 && pos <= width) {
153 if (log_get_show_color())
154 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
155 *p++ = '*';
156 }
157
158 if (log_get_show_color())
159 p = stpcpy(p, ANSI_NORMAL);
160
161 if (pos < width) {
162 if (log_get_show_color())
163 p = stpcpy(p, ANSI_RED);
164 *p++ = '*';
165 if (pos < width-1)
166 p = mempset(p, ' ', width-1-pos);
167 if (log_get_show_color())
168 strcpy(p, ANSI_NORMAL);
169 }
170 }
171
172 void manager_flip_auto_status(Manager *m, bool enable) {
173 assert(m);
174
175 if (enable) {
176 if (m->show_status == SHOW_STATUS_AUTO)
177 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
178 } else {
179 if (m->show_status == SHOW_STATUS_TEMPORARY)
180 manager_set_show_status(m, SHOW_STATUS_AUTO);
181 }
182 }
183
184 static void manager_print_jobs_in_progress(Manager *m) {
185 _cleanup_free_ char *job_of_n = NULL;
186 Iterator i;
187 Job *j;
188 unsigned counter = 0, print_nr;
189 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
190 unsigned cylon_pos;
191 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
192 uint64_t x;
193
194 assert(m);
195 assert(m->n_running_jobs > 0);
196
197 manager_flip_auto_status(m, true);
198
199 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
200
201 HASHMAP_FOREACH(j, m->jobs, i)
202 if (j->state == JOB_RUNNING && counter++ == print_nr)
203 break;
204
205 /* m->n_running_jobs must be consistent with the contents of m->jobs,
206 * so the above loop must have succeeded in finding j. */
207 assert(counter == print_nr + 1);
208 assert(j);
209
210 cylon_pos = m->jobs_in_progress_iteration % 14;
211 if (cylon_pos >= 8)
212 cylon_pos = 14 - cylon_pos;
213 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
214
215 m->jobs_in_progress_iteration++;
216
217 if (m->n_running_jobs > 1) {
218 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
219 job_of_n = NULL;
220 }
221
222 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
223 if (job_get_timeout(j, &x) > 0)
224 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
225
226 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
227 "%sA %s job is running for %s (%s / %s)",
228 strempty(job_of_n),
229 job_type_to_string(j->type),
230 unit_description(j->unit),
231 time, limit);
232 }
233
234 static int have_ask_password(void) {
235 _cleanup_closedir_ DIR *dir;
236 struct dirent *de;
237
238 dir = opendir("/run/systemd/ask-password");
239 if (!dir) {
240 if (errno == ENOENT)
241 return false;
242 else
243 return -errno;
244 }
245
246 FOREACH_DIRENT_ALL(de, dir, return -errno) {
247 if (startswith(de->d_name, "ask."))
248 return true;
249 }
250 return false;
251 }
252
253 static int manager_dispatch_ask_password_fd(sd_event_source *source,
254 int fd, uint32_t revents, void *userdata) {
255 Manager *m = userdata;
256
257 assert(m);
258
259 flush_fd(fd);
260
261 m->have_ask_password = have_ask_password();
262 if (m->have_ask_password < 0)
263 /* Log error but continue. Negative have_ask_password
264 * is treated as unknown status. */
265 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
266
267 return 0;
268 }
269
270 static void manager_close_ask_password(Manager *m) {
271 assert(m);
272
273 m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
274 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
275 m->have_ask_password = -EINVAL;
276 }
277
278 static int manager_check_ask_password(Manager *m) {
279 int r;
280
281 assert(m);
282
283 if (!m->ask_password_event_source) {
284 assert(m->ask_password_inotify_fd < 0);
285
286 mkdir_p_label("/run/systemd/ask-password", 0755);
287
288 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
289 if (m->ask_password_inotify_fd < 0)
290 return log_error_errno(errno, "inotify_init1() failed: %m");
291
292 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
293 log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
294 manager_close_ask_password(m);
295 return -errno;
296 }
297
298 r = sd_event_add_io(m->event, &m->ask_password_event_source,
299 m->ask_password_inotify_fd, EPOLLIN,
300 manager_dispatch_ask_password_fd, m);
301 if (r < 0) {
302 log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
303 manager_close_ask_password(m);
304 return -errno;
305 }
306
307 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
308
309 /* Queries might have been added meanwhile... */
310 manager_dispatch_ask_password_fd(m->ask_password_event_source,
311 m->ask_password_inotify_fd, EPOLLIN, m);
312 }
313
314 return m->have_ask_password;
315 }
316
317 static int manager_watch_idle_pipe(Manager *m) {
318 int r;
319
320 assert(m);
321
322 if (m->idle_pipe_event_source)
323 return 0;
324
325 if (m->idle_pipe[2] < 0)
326 return 0;
327
328 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
329 if (r < 0)
330 return log_error_errno(r, "Failed to watch idle pipe: %m");
331
332 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
333
334 return 0;
335 }
336
337 static void manager_close_idle_pipe(Manager *m) {
338 assert(m);
339
340 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
341
342 safe_close_pair(m->idle_pipe);
343 safe_close_pair(m->idle_pipe + 2);
344 }
345
346 static int manager_setup_time_change(Manager *m) {
347 int r;
348
349 /* We only care for the cancellation event, hence we set the
350 * timeout to the latest possible value. */
351 struct itimerspec its = {
352 .it_value.tv_sec = TIME_T_MAX,
353 };
354
355 assert(m);
356 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
357
358 if (m->test_run)
359 return 0;
360
361 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
362 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
363
364 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
365 if (m->time_change_fd < 0)
366 return log_error_errno(errno, "Failed to create timerfd: %m");
367
368 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
369 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
370 m->time_change_fd = safe_close(m->time_change_fd);
371 return 0;
372 }
373
374 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
375 if (r < 0)
376 return log_error_errno(r, "Failed to create time change event source: %m");
377
378 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
379
380 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
381
382 return 0;
383 }
384
385 static int enable_special_signals(Manager *m) {
386 _cleanup_close_ int fd = -1;
387
388 assert(m);
389
390 if (m->test_run)
391 return 0;
392
393 /* Enable that we get SIGINT on control-alt-del. In containers
394 * this will fail with EPERM (older) or EINVAL (newer), so
395 * ignore that. */
396 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
397 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
398
399 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
400 if (fd < 0) {
401 /* Support systems without virtual console */
402 if (fd != -ENOENT)
403 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
404 } else {
405 /* Enable that we get SIGWINCH on kbrequest */
406 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
407 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
408 }
409
410 return 0;
411 }
412
413 static int manager_setup_signals(Manager *m) {
414 struct sigaction sa = {
415 .sa_handler = SIG_DFL,
416 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
417 };
418 sigset_t mask;
419 int r;
420
421 assert(m);
422
423 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
424
425 /* We make liberal use of realtime signals here. On
426 * Linux/glibc we have 30 of them (with the exception of Linux
427 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
428 * (aka SIGRTMAX). */
429
430 assert_se(sigemptyset(&mask) == 0);
431 sigset_add_many(&mask,
432 SIGCHLD, /* Child died */
433 SIGTERM, /* Reexecute daemon */
434 SIGHUP, /* Reload configuration */
435 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
436 SIGUSR2, /* systemd: dump status */
437 SIGINT, /* Kernel sends us this on control-alt-del */
438 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
439 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
440
441 SIGRTMIN+0, /* systemd: start default.target */
442 SIGRTMIN+1, /* systemd: isolate rescue.target */
443 SIGRTMIN+2, /* systemd: isolate emergency.target */
444 SIGRTMIN+3, /* systemd: start halt.target */
445 SIGRTMIN+4, /* systemd: start poweroff.target */
446 SIGRTMIN+5, /* systemd: start reboot.target */
447 SIGRTMIN+6, /* systemd: start kexec.target */
448
449 /* ... space for more special targets ... */
450
451 SIGRTMIN+13, /* systemd: Immediate halt */
452 SIGRTMIN+14, /* systemd: Immediate poweroff */
453 SIGRTMIN+15, /* systemd: Immediate reboot */
454 SIGRTMIN+16, /* systemd: Immediate kexec */
455
456 /* ... space for more immediate system state changes ... */
457
458 SIGRTMIN+20, /* systemd: enable status messages */
459 SIGRTMIN+21, /* systemd: disable status messages */
460 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
461 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
462 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
463
464 /* .. one free signal here ... */
465
466 #if !defined(__hppa64__) && !defined(__hppa__)
467 /* Apparently Linux on hppa has fewer RT
468 * signals (SIGRTMAX is SIGRTMIN+25 there),
469 * hence let's not try to make use of them
470 * here. Since these commands are accessible
471 * by different means and only really a safety
472 * net, the missing functionality on hppa
473 * shouldn't matter. */
474
475 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
476 SIGRTMIN+27, /* systemd: set log target to console */
477 SIGRTMIN+28, /* systemd: set log target to kmsg */
478 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
479
480 /* ... one free signal here SIGRTMIN+30 ... */
481 #endif
482 -1);
483 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
484
485 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
486 if (m->signal_fd < 0)
487 return -errno;
488
489 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
490 if (r < 0)
491 return r;
492
493 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
494
495 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
496 * notify processing can still figure out to which process/service a message belongs, before we reap the
497 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
498 * status information before detecting that there's no process in a cgroup. */
499 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
500 if (r < 0)
501 return r;
502
503 if (MANAGER_IS_SYSTEM(m))
504 return enable_special_signals(m);
505
506 return 0;
507 }
508
509 static void manager_clean_environment(Manager *m) {
510 assert(m);
511
512 /* Let's remove some environment variables that we
513 * need ourselves to communicate with our clients */
514 strv_env_unset_many(
515 m->environment,
516 "NOTIFY_SOCKET",
517 "MAINPID",
518 "MANAGERPID",
519 "LISTEN_PID",
520 "LISTEN_FDS",
521 "LISTEN_FDNAMES",
522 "WATCHDOG_PID",
523 "WATCHDOG_USEC",
524 "INVOCATION_ID",
525 NULL);
526 }
527
528 static int manager_default_environment(Manager *m) {
529 assert(m);
530
531 if (MANAGER_IS_SYSTEM(m)) {
532 /* The system manager always starts with a clean
533 * environment for its children. It does not import
534 * the kernel or the parents exported variables.
535 *
536 * The initial passed environ is untouched to keep
537 * /proc/self/environ valid; it is used for tagging
538 * the init process inside containers. */
539 m->environment = strv_new("PATH=" DEFAULT_PATH,
540 NULL);
541
542 /* Import locale variables LC_*= from configuration */
543 locale_setup(&m->environment);
544 } else {
545 /* The user manager passes its own environment
546 * along to its children. */
547 m->environment = strv_copy(environ);
548 }
549
550 if (!m->environment)
551 return -ENOMEM;
552
553 manager_clean_environment(m);
554 strv_sort(m->environment);
555
556 return 0;
557 }
558
559 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
560 Manager *m;
561 int r;
562
563 assert(_m);
564 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
565
566 m = new0(Manager, 1);
567 if (!m)
568 return -ENOMEM;
569
570 m->unit_file_scope = scope;
571 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
572 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
573 m->default_tasks_accounting = true;
574 m->default_tasks_max = UINT64_MAX;
575
576 #ifdef ENABLE_EFI
577 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
578 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
579 #endif
580
581 /* Prepare log fields we can use for structured logging */
582 if (MANAGER_IS_SYSTEM(m)) {
583 m->unit_log_field = "UNIT=";
584 m->unit_log_format_string = "UNIT=%s";
585
586 m->invocation_log_field = "INVOCATION_ID=";
587 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
588 } else {
589 m->unit_log_field = "USER_UNIT=";
590 m->unit_log_format_string = "USER_UNIT=%s";
591
592 m->invocation_log_field = "USER_INVOCATION_ID=";
593 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
594 }
595
596 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
597
598 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
599 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
600 m->ask_password_inotify_fd = -1;
601
602 m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
603
604 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
605
606 m->have_ask_password = -EINVAL; /* we don't know */
607 m->first_boot = -1;
608
609 m->test_run = test_run;
610
611 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
612 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
613
614 r = manager_default_environment(m);
615 if (r < 0)
616 goto fail;
617
618 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
619 if (r < 0)
620 goto fail;
621
622 r = hashmap_ensure_allocated(&m->jobs, NULL);
623 if (r < 0)
624 goto fail;
625
626 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
627 if (r < 0)
628 goto fail;
629
630 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
631 if (r < 0)
632 goto fail;
633
634 r = sd_event_default(&m->event);
635 if (r < 0)
636 goto fail;
637
638 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
639 if (r < 0)
640 goto fail;
641
642 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
643 if (r < 0)
644 goto fail;
645
646 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
647 if (r < 0)
648 goto fail;
649
650 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
651
652 r = manager_setup_signals(m);
653 if (r < 0)
654 goto fail;
655
656 r = manager_setup_cgroup(m);
657 if (r < 0)
658 goto fail;
659
660 r = manager_setup_time_change(m);
661 if (r < 0)
662 goto fail;
663
664 m->udev = udev_new();
665 if (!m->udev) {
666 r = -ENOMEM;
667 goto fail;
668 }
669
670 /* Note that we do not set up the notify fd here. We do that after deserialization,
671 * since they might have gotten serialized across the reexec. */
672
673 m->taint_usr = dir_is_empty("/usr") > 0;
674
675 *_m = m;
676 return 0;
677
678 fail:
679 manager_free(m);
680 return r;
681 }
682
683 static int manager_setup_notify(Manager *m) {
684 int r;
685
686 if (m->test_run)
687 return 0;
688
689 if (m->notify_fd < 0) {
690 _cleanup_close_ int fd = -1;
691 union sockaddr_union sa = {
692 .sa.sa_family = AF_UNIX,
693 };
694 static const int one = 1;
695 const char *e;
696
697 /* First free all secondary fields */
698 m->notify_socket = mfree(m->notify_socket);
699 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
700
701 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
702 if (fd < 0)
703 return log_error_errno(errno, "Failed to allocate notification socket: %m");
704
705 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
706
707 e = manager_get_runtime_prefix(m);
708 if (!e) {
709 log_error("Failed to determine runtime prefix.");
710 return -EINVAL;
711 }
712
713 m->notify_socket = strappend(e, "/systemd/notify");
714 if (!m->notify_socket)
715 return log_oom();
716
717 (void) mkdir_parents_label(m->notify_socket, 0755);
718 (void) unlink(m->notify_socket);
719
720 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
721 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
722 if (r < 0)
723 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
724
725 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
726 if (r < 0)
727 return log_error_errno(errno, "SO_PASSCRED failed: %m");
728
729 m->notify_fd = fd;
730 fd = -1;
731
732 log_debug("Using notification socket %s", m->notify_socket);
733 }
734
735 if (!m->notify_event_source) {
736 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
737 if (r < 0)
738 return log_error_errno(r, "Failed to allocate notify event source: %m");
739
740 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
741 * service an exit message belongs. */
742 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
743 if (r < 0)
744 return log_error_errno(r, "Failed to set priority of notify event source: %m");
745
746 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
747 }
748
749 return 0;
750 }
751
752 static int manager_setup_cgroups_agent(Manager *m) {
753
754 static const union sockaddr_union sa = {
755 .un.sun_family = AF_UNIX,
756 .un.sun_path = "/run/systemd/cgroups-agent",
757 };
758 int r;
759
760 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
761 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
762 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
763 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
764 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
765 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
766 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
767 * we thus won't lose messages.
768 *
769 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
770 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
771 * bus for these messages. */
772
773 if (m->test_run)
774 return 0;
775
776 if (!MANAGER_IS_SYSTEM(m))
777 return 0;
778
779 if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* We don't need this anymore on the unified hierarchy */
780 return 0;
781
782 if (m->cgroups_agent_fd < 0) {
783 _cleanup_close_ int fd = -1;
784
785 /* First free all secondary fields */
786 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
787
788 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
789 if (fd < 0)
790 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
791
792 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
793
794 (void) unlink(sa.un.sun_path);
795
796 /* Only allow root to connect to this socket */
797 RUN_WITH_UMASK(0077)
798 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
799 if (r < 0)
800 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
801
802 m->cgroups_agent_fd = fd;
803 fd = -1;
804 }
805
806 if (!m->cgroups_agent_event_source) {
807 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
808 if (r < 0)
809 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
810
811 /* Process cgroups notifications early, but after having processed service notification messages or
812 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
813 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
814 * cgroup inotify for the unified cgroup stuff. */
815 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
816 if (r < 0)
817 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
818
819 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
820 }
821
822 return 0;
823 }
824
825 static int manager_setup_user_lookup_fd(Manager *m) {
826 int r;
827
828 assert(m);
829
830 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
831 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
832 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
833 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
834 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
835 * hence we establish this communication channel so that forked off processes can pass their UID/GID
836 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
837 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
838 * this purpose.
839 *
840 * You might wonder why we need a communication channel for this that is independent of the usual notification
841 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
842 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
843 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
844 *
845 * Note that this function is called under two circumstances: when we first initialize (in which case we
846 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
847 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
848
849 if (m->user_lookup_fds[0] < 0) {
850
851 /* Free all secondary fields */
852 safe_close_pair(m->user_lookup_fds);
853 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
854
855 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
856 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
857
858 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
859 }
860
861 if (!m->user_lookup_event_source) {
862 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
863 if (r < 0)
864 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
865
866 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
867 * resolutions */
868 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
869 if (r < 0)
870 return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
871
872 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
873 }
874
875 return 0;
876 }
877
878 static int manager_connect_bus(Manager *m, bool reexecuting) {
879 bool try_bus_connect;
880
881 assert(m);
882
883 if (m->test_run)
884 return 0;
885
886 try_bus_connect =
887 reexecuting ||
888 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
889
890 /* Try to connect to the buses, if possible. */
891 return bus_init(m, try_bus_connect);
892 }
893
894 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
895 Unit *u;
896 unsigned n = 0;
897
898 assert(m);
899
900 while ((u = m->cleanup_queue)) {
901 assert(u->in_cleanup_queue);
902
903 unit_free(u);
904 n++;
905 }
906
907 return n;
908 }
909
910 enum {
911 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
912 GC_OFFSET_UNSURE, /* No clue */
913 GC_OFFSET_GOOD, /* We still need this unit */
914 GC_OFFSET_BAD, /* We don't need this unit anymore */
915 _GC_OFFSET_MAX
916 };
917
918 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
919 Iterator i;
920 Unit *other;
921
922 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
923
924 /* Recursively mark referenced units as GOOD as well */
925 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
926 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
927 unit_gc_mark_good(other, gc_marker);
928 }
929
930 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
931 Iterator i;
932 Unit *other;
933 bool is_bad;
934
935 assert(u);
936
937 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
938 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
939 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
940 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
941 return;
942
943 if (u->in_cleanup_queue)
944 goto bad;
945
946 if (unit_check_gc(u))
947 goto good;
948
949 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
950
951 is_bad = true;
952
953 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
954 unit_gc_sweep(other, gc_marker);
955
956 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
957 goto good;
958
959 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
960 is_bad = false;
961 }
962
963 if (is_bad)
964 goto bad;
965
966 /* We were unable to find anything out about this entry, so
967 * let's investigate it later */
968 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
969 unit_add_to_gc_queue(u);
970 return;
971
972 bad:
973 /* We definitely know that this one is not useful anymore, so
974 * let's mark it for deletion */
975 u->gc_marker = gc_marker + GC_OFFSET_BAD;
976 unit_add_to_cleanup_queue(u);
977 return;
978
979 good:
980 unit_gc_mark_good(u, gc_marker);
981 }
982
983 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
984 unsigned n = 0, gc_marker;
985 Unit *u;
986
987 assert(m);
988
989 /* log_debug("Running GC..."); */
990
991 m->gc_marker += _GC_OFFSET_MAX;
992 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
993 m->gc_marker = 1;
994
995 gc_marker = m->gc_marker;
996
997 while ((u = m->gc_unit_queue)) {
998 assert(u->in_gc_queue);
999
1000 unit_gc_sweep(u, gc_marker);
1001
1002 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1003 u->in_gc_queue = false;
1004
1005 n++;
1006
1007 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1008 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1009 if (u->id)
1010 log_unit_debug(u, "Collecting.");
1011 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1012 unit_add_to_cleanup_queue(u);
1013 }
1014 }
1015
1016 return n;
1017 }
1018
1019 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1020 unsigned n = 0;
1021 Job *j;
1022
1023 assert(m);
1024
1025 while ((j = m->gc_job_queue)) {
1026 assert(j->in_gc_queue);
1027
1028 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1029 j->in_gc_queue = false;
1030
1031 n++;
1032
1033 if (job_check_gc(j))
1034 continue;
1035
1036 log_unit_debug(j->unit, "Collecting job.");
1037 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1038 }
1039
1040 return n;
1041 }
1042
1043 static void manager_clear_jobs_and_units(Manager *m) {
1044 Unit *u;
1045
1046 assert(m);
1047
1048 while ((u = hashmap_first(m->units)))
1049 unit_free(u);
1050
1051 manager_dispatch_cleanup_queue(m);
1052
1053 assert(!m->load_queue);
1054 assert(!m->run_queue);
1055 assert(!m->dbus_unit_queue);
1056 assert(!m->dbus_job_queue);
1057 assert(!m->cleanup_queue);
1058 assert(!m->gc_unit_queue);
1059 assert(!m->gc_job_queue);
1060
1061 assert(hashmap_isempty(m->jobs));
1062 assert(hashmap_isempty(m->units));
1063
1064 m->n_on_console = 0;
1065 m->n_running_jobs = 0;
1066 }
1067
1068 Manager* manager_free(Manager *m) {
1069 UnitType c;
1070 int i;
1071
1072 if (!m)
1073 return NULL;
1074
1075 manager_clear_jobs_and_units(m);
1076
1077 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1078 if (unit_vtable[c]->shutdown)
1079 unit_vtable[c]->shutdown(m);
1080
1081 /* If we reexecute ourselves, we keep the root cgroup
1082 * around */
1083 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1084
1085 lookup_paths_flush_generator(&m->lookup_paths);
1086
1087 bus_done(m);
1088
1089 dynamic_user_vacuum(m, false);
1090 hashmap_free(m->dynamic_users);
1091
1092 hashmap_free(m->units);
1093 hashmap_free(m->units_by_invocation_id);
1094 hashmap_free(m->jobs);
1095 hashmap_free(m->watch_pids1);
1096 hashmap_free(m->watch_pids2);
1097 hashmap_free(m->watch_bus);
1098
1099 set_free(m->startup_units);
1100 set_free(m->failed_units);
1101
1102 sd_event_source_unref(m->signal_event_source);
1103 sd_event_source_unref(m->notify_event_source);
1104 sd_event_source_unref(m->cgroups_agent_event_source);
1105 sd_event_source_unref(m->time_change_event_source);
1106 sd_event_source_unref(m->jobs_in_progress_event_source);
1107 sd_event_source_unref(m->run_queue_event_source);
1108 sd_event_source_unref(m->user_lookup_event_source);
1109
1110 safe_close(m->signal_fd);
1111 safe_close(m->notify_fd);
1112 safe_close(m->cgroups_agent_fd);
1113 safe_close(m->time_change_fd);
1114 safe_close_pair(m->user_lookup_fds);
1115
1116 manager_close_ask_password(m);
1117
1118 manager_close_idle_pipe(m);
1119
1120 udev_unref(m->udev);
1121 sd_event_unref(m->event);
1122
1123 free(m->notify_socket);
1124
1125 lookup_paths_free(&m->lookup_paths);
1126 strv_free(m->environment);
1127
1128 hashmap_free(m->cgroup_unit);
1129 set_free_free(m->unit_path_cache);
1130
1131 free(m->switch_root);
1132 free(m->switch_root_init);
1133
1134 for (i = 0; i < _RLIMIT_MAX; i++)
1135 m->rlimit[i] = mfree(m->rlimit[i]);
1136
1137 assert(hashmap_isempty(m->units_requiring_mounts_for));
1138 hashmap_free(m->units_requiring_mounts_for);
1139
1140 hashmap_free(m->uid_refs);
1141 hashmap_free(m->gid_refs);
1142
1143 return mfree(m);
1144 }
1145
1146 void manager_enumerate(Manager *m) {
1147 UnitType c;
1148
1149 assert(m);
1150
1151 /* Let's ask every type to load all units from disk/kernel
1152 * that it might know */
1153 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1154 if (!unit_type_supported(c)) {
1155 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1156 continue;
1157 }
1158
1159 if (!unit_vtable[c]->enumerate)
1160 continue;
1161
1162 unit_vtable[c]->enumerate(m);
1163 }
1164
1165 manager_dispatch_load_queue(m);
1166 }
1167
1168 static void manager_coldplug(Manager *m) {
1169 Iterator i;
1170 Unit *u;
1171 char *k;
1172 int r;
1173
1174 assert(m);
1175
1176 /* Then, let's set up their initial state. */
1177 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1178
1179 /* ignore aliases */
1180 if (u->id != k)
1181 continue;
1182
1183 r = unit_coldplug(u);
1184 if (r < 0)
1185 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1186 }
1187 }
1188
1189 static void manager_build_unit_path_cache(Manager *m) {
1190 char **i;
1191 int r;
1192
1193 assert(m);
1194
1195 set_free_free(m->unit_path_cache);
1196
1197 m->unit_path_cache = set_new(&string_hash_ops);
1198 if (!m->unit_path_cache) {
1199 r = -ENOMEM;
1200 goto fail;
1201 }
1202
1203 /* This simply builds a list of files we know exist, so that
1204 * we don't always have to go to disk */
1205
1206 STRV_FOREACH(i, m->lookup_paths.search_path) {
1207 _cleanup_closedir_ DIR *d = NULL;
1208 struct dirent *de;
1209
1210 d = opendir(*i);
1211 if (!d) {
1212 if (errno != ENOENT)
1213 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1214 continue;
1215 }
1216
1217 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1218 char *p;
1219
1220 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1221 if (!p) {
1222 r = -ENOMEM;
1223 goto fail;
1224 }
1225
1226 r = set_consume(m->unit_path_cache, p);
1227 if (r < 0)
1228 goto fail;
1229 }
1230 }
1231
1232 return;
1233
1234 fail:
1235 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1236 m->unit_path_cache = set_free_free(m->unit_path_cache);
1237 }
1238
1239 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1240 Iterator i;
1241 Unit *u;
1242
1243 assert(m);
1244
1245 HASHMAP_FOREACH(u, m->units, i) {
1246
1247 if (fdset_size(fds) <= 0)
1248 break;
1249
1250 if (!UNIT_VTABLE(u)->distribute_fds)
1251 continue;
1252
1253 UNIT_VTABLE(u)->distribute_fds(u, fds);
1254 }
1255 }
1256
1257 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1258 int r, q;
1259
1260 assert(m);
1261
1262 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1263 if (r < 0)
1264 return r;
1265
1266 /* Make sure the transient directory always exists, so that it remains in the search path */
1267 if (!m->test_run) {
1268 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1269 if (r < 0)
1270 return r;
1271 }
1272
1273 dual_timestamp_get(&m->generators_start_timestamp);
1274 r = manager_run_generators(m);
1275 dual_timestamp_get(&m->generators_finish_timestamp);
1276 if (r < 0)
1277 return r;
1278
1279 lookup_paths_reduce(&m->lookup_paths);
1280 manager_build_unit_path_cache(m);
1281
1282 /* If we will deserialize make sure that during enumeration
1283 * this is already known, so we increase the counter here
1284 * already */
1285 if (serialization)
1286 m->n_reloading++;
1287
1288 /* First, enumerate what we can from all config files */
1289 dual_timestamp_get(&m->units_load_start_timestamp);
1290 manager_enumerate(m);
1291 dual_timestamp_get(&m->units_load_finish_timestamp);
1292
1293 /* Second, deserialize if there is something to deserialize */
1294 if (serialization)
1295 r = manager_deserialize(m, serialization, fds);
1296
1297 /* Any fds left? Find some unit which wants them. This is
1298 * useful to allow container managers to pass some file
1299 * descriptors to us pre-initialized. This enables
1300 * socket-based activation of entire containers. */
1301 manager_distribute_fds(m, fds);
1302
1303 /* We might have deserialized the notify fd, but if we didn't
1304 * then let's create the bus now */
1305 q = manager_setup_notify(m);
1306 if (q < 0 && r == 0)
1307 r = q;
1308
1309 q = manager_setup_cgroups_agent(m);
1310 if (q < 0 && r == 0)
1311 r = q;
1312
1313 q = manager_setup_user_lookup_fd(m);
1314 if (q < 0 && r == 0)
1315 r = q;
1316
1317 /* Let's connect to the bus now. */
1318 (void) manager_connect_bus(m, !!serialization);
1319
1320 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1321 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1322
1323 /* Third, fire things up! */
1324 manager_coldplug(m);
1325
1326 /* Release any dynamic users no longer referenced */
1327 dynamic_user_vacuum(m, true);
1328
1329 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1330 manager_vacuum_uid_refs(m);
1331 manager_vacuum_gid_refs(m);
1332
1333 if (serialization) {
1334 assert(m->n_reloading > 0);
1335 m->n_reloading--;
1336
1337 /* Let's wait for the UnitNew/JobNew messages being
1338 * sent, before we notify that the reload is
1339 * finished */
1340 m->send_reloading_done = true;
1341 }
1342
1343 return r;
1344 }
1345
1346 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1347 int r;
1348 Transaction *tr;
1349
1350 assert(m);
1351 assert(type < _JOB_TYPE_MAX);
1352 assert(unit);
1353 assert(mode < _JOB_MODE_MAX);
1354
1355 if (mode == JOB_ISOLATE && type != JOB_START)
1356 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1357
1358 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1359 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1360
1361 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1362
1363 type = job_type_collapse(type, unit);
1364
1365 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1366 if (!tr)
1367 return -ENOMEM;
1368
1369 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1370 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1371 mode == JOB_IGNORE_DEPENDENCIES, e);
1372 if (r < 0)
1373 goto tr_abort;
1374
1375 if (mode == JOB_ISOLATE) {
1376 r = transaction_add_isolate_jobs(tr, m);
1377 if (r < 0)
1378 goto tr_abort;
1379 }
1380
1381 r = transaction_activate(tr, m, mode, e);
1382 if (r < 0)
1383 goto tr_abort;
1384
1385 log_unit_debug(unit,
1386 "Enqueued job %s/%s as %u", unit->id,
1387 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1388
1389 if (_ret)
1390 *_ret = tr->anchor_job;
1391
1392 transaction_free(tr);
1393 return 0;
1394
1395 tr_abort:
1396 transaction_abort(tr);
1397 transaction_free(tr);
1398 return r;
1399 }
1400
1401 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1402 Unit *unit;
1403 int r;
1404
1405 assert(m);
1406 assert(type < _JOB_TYPE_MAX);
1407 assert(name);
1408 assert(mode < _JOB_MODE_MAX);
1409
1410 r = manager_load_unit(m, name, NULL, NULL, &unit);
1411 if (r < 0)
1412 return r;
1413
1414 return manager_add_job(m, type, unit, mode, e, ret);
1415 }
1416
1417 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1418 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1419 int r;
1420
1421 assert(m);
1422 assert(type < _JOB_TYPE_MAX);
1423 assert(name);
1424 assert(mode < _JOB_MODE_MAX);
1425
1426 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1427 if (r < 0)
1428 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1429
1430 return r;
1431 }
1432
1433 Job *manager_get_job(Manager *m, uint32_t id) {
1434 assert(m);
1435
1436 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1437 }
1438
1439 Unit *manager_get_unit(Manager *m, const char *name) {
1440 assert(m);
1441 assert(name);
1442
1443 return hashmap_get(m->units, name);
1444 }
1445
1446 unsigned manager_dispatch_load_queue(Manager *m) {
1447 Unit *u;
1448 unsigned n = 0;
1449
1450 assert(m);
1451
1452 /* Make sure we are not run recursively */
1453 if (m->dispatching_load_queue)
1454 return 0;
1455
1456 m->dispatching_load_queue = true;
1457
1458 /* Dispatches the load queue. Takes a unit from the queue and
1459 * tries to load its data until the queue is empty */
1460
1461 while ((u = m->load_queue)) {
1462 assert(u->in_load_queue);
1463
1464 unit_load(u);
1465 n++;
1466 }
1467
1468 m->dispatching_load_queue = false;
1469 return n;
1470 }
1471
1472 int manager_load_unit_prepare(
1473 Manager *m,
1474 const char *name,
1475 const char *path,
1476 sd_bus_error *e,
1477 Unit **_ret) {
1478
1479 Unit *ret;
1480 UnitType t;
1481 int r;
1482
1483 assert(m);
1484 assert(name || path);
1485
1486 /* This will prepare the unit for loading, but not actually
1487 * load anything from disk. */
1488
1489 if (path && !is_path(path))
1490 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1491
1492 if (!name)
1493 name = basename(path);
1494
1495 t = unit_name_to_type(name);
1496
1497 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1498 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1499 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1500
1501 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1502 }
1503
1504 ret = manager_get_unit(m, name);
1505 if (ret) {
1506 *_ret = ret;
1507 return 1;
1508 }
1509
1510 ret = unit_new(m, unit_vtable[t]->object_size);
1511 if (!ret)
1512 return -ENOMEM;
1513
1514 if (path) {
1515 ret->fragment_path = strdup(path);
1516 if (!ret->fragment_path) {
1517 unit_free(ret);
1518 return -ENOMEM;
1519 }
1520 }
1521
1522 r = unit_add_name(ret, name);
1523 if (r < 0) {
1524 unit_free(ret);
1525 return r;
1526 }
1527
1528 unit_add_to_load_queue(ret);
1529 unit_add_to_dbus_queue(ret);
1530 unit_add_to_gc_queue(ret);
1531
1532 if (_ret)
1533 *_ret = ret;
1534
1535 return 0;
1536 }
1537
1538 int manager_load_unit(
1539 Manager *m,
1540 const char *name,
1541 const char *path,
1542 sd_bus_error *e,
1543 Unit **_ret) {
1544
1545 int r;
1546
1547 assert(m);
1548
1549 /* This will load the service information files, but not actually
1550 * start any services or anything. */
1551
1552 r = manager_load_unit_prepare(m, name, path, e, _ret);
1553 if (r != 0)
1554 return r;
1555
1556 manager_dispatch_load_queue(m);
1557
1558 if (_ret)
1559 *_ret = unit_follow_merge(*_ret);
1560
1561 return 0;
1562 }
1563
1564 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1565 Iterator i;
1566 Job *j;
1567
1568 assert(s);
1569 assert(f);
1570
1571 HASHMAP_FOREACH(j, s->jobs, i)
1572 job_dump(j, f, prefix);
1573 }
1574
1575 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1576 Iterator i;
1577 Unit *u;
1578 const char *t;
1579
1580 assert(s);
1581 assert(f);
1582
1583 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1584 if (u->id == t)
1585 unit_dump(u, f, prefix);
1586 }
1587
1588 void manager_clear_jobs(Manager *m) {
1589 Job *j;
1590
1591 assert(m);
1592
1593 while ((j = hashmap_first(m->jobs)))
1594 /* No need to recurse. We're cancelling all jobs. */
1595 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1596 }
1597
1598 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1599 Manager *m = userdata;
1600 Job *j;
1601
1602 assert(source);
1603 assert(m);
1604
1605 while ((j = m->run_queue)) {
1606 assert(j->installed);
1607 assert(j->in_run_queue);
1608
1609 job_run_and_invalidate(j);
1610 }
1611
1612 if (m->n_running_jobs > 0)
1613 manager_watch_jobs_in_progress(m);
1614
1615 if (m->n_on_console > 0)
1616 manager_watch_idle_pipe(m);
1617
1618 return 1;
1619 }
1620
1621 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1622 Job *j;
1623 Unit *u;
1624 unsigned n = 0;
1625
1626 assert(m);
1627
1628 if (m->dispatching_dbus_queue)
1629 return 0;
1630
1631 m->dispatching_dbus_queue = true;
1632
1633 while ((u = m->dbus_unit_queue)) {
1634 assert(u->in_dbus_queue);
1635
1636 bus_unit_send_change_signal(u);
1637 n++;
1638 }
1639
1640 while ((j = m->dbus_job_queue)) {
1641 assert(j->in_dbus_queue);
1642
1643 bus_job_send_change_signal(j);
1644 n++;
1645 }
1646
1647 m->dispatching_dbus_queue = false;
1648
1649 if (m->send_reloading_done) {
1650 m->send_reloading_done = false;
1651
1652 bus_manager_send_reloading(m, false);
1653 }
1654
1655 if (m->queued_message)
1656 bus_send_queued_message(m);
1657
1658 return n;
1659 }
1660
1661 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1662 Manager *m = userdata;
1663 char buf[PATH_MAX+1];
1664 ssize_t n;
1665
1666 n = recv(fd, buf, sizeof(buf), 0);
1667 if (n < 0)
1668 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1669 if (n == 0) {
1670 log_error("Got zero-length cgroups agent message, ignoring.");
1671 return 0;
1672 }
1673 if ((size_t) n >= sizeof(buf)) {
1674 log_error("Got overly long cgroups agent message, ignoring.");
1675 return 0;
1676 }
1677
1678 if (memchr(buf, 0, n)) {
1679 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1680 return 0;
1681 }
1682 buf[n] = 0;
1683
1684 manager_notify_cgroup_empty(m, buf);
1685 bus_forward_agent_released(m, buf);
1686
1687 return 0;
1688 }
1689
1690 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1691 _cleanup_strv_free_ char **tags = NULL;
1692
1693 assert(m);
1694 assert(u);
1695 assert(buf);
1696
1697 tags = strv_split(buf, "\n\r");
1698 if (!tags) {
1699 log_oom();
1700 return;
1701 }
1702
1703 if (UNIT_VTABLE(u)->notify_message)
1704 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1705 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1706 _cleanup_free_ char *x = NULL, *y = NULL;
1707
1708 x = cescape(buf);
1709 if (x)
1710 y = ellipsize(x, 20, 90);
1711 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1712 }
1713 }
1714
1715 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1716
1717 _cleanup_fdset_free_ FDSet *fds = NULL;
1718 Manager *m = userdata;
1719 char buf[NOTIFY_BUFFER_MAX+1];
1720 struct iovec iovec = {
1721 .iov_base = buf,
1722 .iov_len = sizeof(buf)-1,
1723 };
1724 union {
1725 struct cmsghdr cmsghdr;
1726 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1727 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1728 } control = {};
1729 struct msghdr msghdr = {
1730 .msg_iov = &iovec,
1731 .msg_iovlen = 1,
1732 .msg_control = &control,
1733 .msg_controllen = sizeof(control),
1734 };
1735
1736 struct cmsghdr *cmsg;
1737 struct ucred *ucred = NULL;
1738 Unit *u1, *u2, *u3;
1739 int r, *fd_array = NULL;
1740 unsigned n_fds = 0;
1741 ssize_t n;
1742
1743 assert(m);
1744 assert(m->notify_fd == fd);
1745
1746 if (revents != EPOLLIN) {
1747 log_warning("Got unexpected poll event for notify fd.");
1748 return 0;
1749 }
1750
1751 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1752 if (n < 0) {
1753 if (IN_SET(errno, EAGAIN, EINTR))
1754 return 0; /* Spurious wakeup, try again */
1755
1756 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1757 * won't take notification messages anymore, but that's still better than busy looping around this:
1758 * being woken up over and over again but being unable to actually read the message off the socket. */
1759 return log_error_errno(errno, "Failed to receive notification message: %m");
1760 }
1761
1762 CMSG_FOREACH(cmsg, &msghdr) {
1763 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1764
1765 fd_array = (int*) CMSG_DATA(cmsg);
1766 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1767
1768 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1769 cmsg->cmsg_type == SCM_CREDENTIALS &&
1770 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1771
1772 ucred = (struct ucred*) CMSG_DATA(cmsg);
1773 }
1774 }
1775
1776 if (n_fds > 0) {
1777 assert(fd_array);
1778
1779 r = fdset_new_array(&fds, fd_array, n_fds);
1780 if (r < 0) {
1781 close_many(fd_array, n_fds);
1782 log_oom();
1783 return 0;
1784 }
1785 }
1786
1787 if (!ucred || ucred->pid <= 0) {
1788 log_warning("Received notify message without valid credentials. Ignoring.");
1789 return 0;
1790 }
1791
1792 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1793 log_warning("Received notify message exceeded maximum size. Ignoring.");
1794 return 0;
1795 }
1796
1797 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1798 * trailing NUL byte in the message, but don't expect it. */
1799 if (n > 1 && memchr(buf, 0, n-1)) {
1800 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1801 return 0;
1802 }
1803
1804 /* Make sure it's NUL-terminated. */
1805 buf[n] = 0;
1806
1807 /* Notify every unit that might be interested, but try
1808 * to avoid notifying the same one multiple times. */
1809 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1810 if (u1)
1811 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1812
1813 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1814 if (u2 && u2 != u1)
1815 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1816
1817 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1818 if (u3 && u3 != u2 && u3 != u1)
1819 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1820
1821 if (!u1 && !u2 && !u3)
1822 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1823
1824 if (fdset_size(fds) > 0)
1825 log_warning("Got extra auxiliary fds with notification message, closing them.");
1826
1827 return 0;
1828 }
1829
1830 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1831 uint64_t iteration;
1832
1833 assert(m);
1834 assert(u);
1835 assert(si);
1836
1837 sd_event_get_iteration(m->event, &iteration);
1838
1839 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1840
1841 unit_unwatch_pid(u, si->si_pid);
1842
1843 if (UNIT_VTABLE(u)->sigchld_event) {
1844 if (set_size(u->pids) <= 1 ||
1845 iteration != u->sigchldgen ||
1846 unit_main_pid(u) == si->si_pid ||
1847 unit_control_pid(u) == si->si_pid) {
1848 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1849 u->sigchldgen = iteration;
1850 } else
1851 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1852 }
1853 }
1854
1855 static int manager_dispatch_sigchld(Manager *m) {
1856 assert(m);
1857
1858 for (;;) {
1859 siginfo_t si = {};
1860
1861 /* First we call waitd() for a PID and do not reap the
1862 * zombie. That way we can still access /proc/$PID for
1863 * it while it is a zombie. */
1864 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1865
1866 if (errno == ECHILD)
1867 break;
1868
1869 if (errno == EINTR)
1870 continue;
1871
1872 return -errno;
1873 }
1874
1875 if (si.si_pid <= 0)
1876 break;
1877
1878 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1879 _cleanup_free_ char *name = NULL;
1880 Unit *u1, *u2, *u3;
1881
1882 get_process_comm(si.si_pid, &name);
1883
1884 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1885 si.si_pid, strna(name),
1886 sigchld_code_to_string(si.si_code),
1887 si.si_status,
1888 strna(si.si_code == CLD_EXITED
1889 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1890 : signal_to_string(si.si_status)));
1891
1892 /* And now figure out the unit this belongs
1893 * to, it might be multiple... */
1894 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1895 if (u1)
1896 invoke_sigchld_event(m, u1, &si);
1897 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1898 if (u2 && u2 != u1)
1899 invoke_sigchld_event(m, u2, &si);
1900 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1901 if (u3 && u3 != u2 && u3 != u1)
1902 invoke_sigchld_event(m, u3, &si);
1903 }
1904
1905 /* And now, we actually reap the zombie. */
1906 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1907 if (errno == EINTR)
1908 continue;
1909
1910 return -errno;
1911 }
1912 }
1913
1914 return 0;
1915 }
1916
1917 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1918 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1919 int r;
1920
1921 log_debug("Activating special unit %s", name);
1922
1923 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1924 if (r < 0)
1925 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1926
1927 return r;
1928 }
1929
1930 static void manager_handle_ctrl_alt_del(Manager *m) {
1931 /* If the user presses C-A-D more than
1932 * 7 times within 2s, we reboot/shutdown immediately,
1933 * unless it was disabled in system.conf */
1934
1935 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
1936 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1937 else
1938 emergency_action(m, m->cad_burst_action, NULL,
1939 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1940 }
1941
1942 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1943 Manager *m = userdata;
1944 ssize_t n;
1945 struct signalfd_siginfo sfsi;
1946 bool sigchld = false;
1947 int r;
1948
1949 assert(m);
1950 assert(m->signal_fd == fd);
1951
1952 if (revents != EPOLLIN) {
1953 log_warning("Got unexpected events from signal file descriptor.");
1954 return 0;
1955 }
1956
1957 for (;;) {
1958 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1959 if (n != sizeof(sfsi)) {
1960 if (n >= 0) {
1961 log_warning("Truncated read from signal fd (%zu bytes)!", n);
1962 return 0;
1963 }
1964
1965 if (IN_SET(errno, EINTR, EAGAIN))
1966 break;
1967
1968 /* We return an error here, which will kill this handler,
1969 * to avoid a busy loop on read error. */
1970 return log_error_errno(errno, "Reading from signal fd failed: %m");
1971 }
1972
1973 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1974 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1975 ? LOG_DEBUG : LOG_INFO,
1976 &sfsi);
1977
1978 switch (sfsi.ssi_signo) {
1979
1980 case SIGCHLD:
1981 sigchld = true;
1982 break;
1983
1984 case SIGTERM:
1985 if (MANAGER_IS_SYSTEM(m)) {
1986 /* This is for compatibility with the
1987 * original sysvinit */
1988 r = verify_run_space_and_log("Refusing to reexecute");
1989 if (r >= 0)
1990 m->exit_code = MANAGER_REEXECUTE;
1991 break;
1992 }
1993
1994 /* Fall through */
1995
1996 case SIGINT:
1997 if (MANAGER_IS_SYSTEM(m)) {
1998 manager_handle_ctrl_alt_del(m);
1999 break;
2000 }
2001
2002 /* Run the exit target if there is one, if not, just exit. */
2003 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
2004 m->exit_code = MANAGER_EXIT;
2005 return 0;
2006 }
2007
2008 break;
2009
2010 case SIGWINCH:
2011 if (MANAGER_IS_SYSTEM(m))
2012 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2013
2014 /* This is a nop on non-init */
2015 break;
2016
2017 case SIGPWR:
2018 if (MANAGER_IS_SYSTEM(m))
2019 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2020
2021 /* This is a nop on non-init */
2022 break;
2023
2024 case SIGUSR1: {
2025 Unit *u;
2026
2027 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2028
2029 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2030 log_info("Trying to reconnect to bus...");
2031 bus_init(m, true);
2032 }
2033
2034 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2035 log_info("Loading D-Bus service...");
2036 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2037 }
2038
2039 break;
2040 }
2041
2042 case SIGUSR2: {
2043 _cleanup_free_ char *dump = NULL;
2044 _cleanup_fclose_ FILE *f = NULL;
2045 size_t size;
2046
2047 f = open_memstream(&dump, &size);
2048 if (!f) {
2049 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2050 break;
2051 }
2052
2053 manager_dump_units(m, f, "\t");
2054 manager_dump_jobs(m, f, "\t");
2055
2056 r = fflush_and_check(f);
2057 if (r < 0) {
2058 log_warning_errno(r, "Failed to write status stream: %m");
2059 break;
2060 }
2061
2062 log_dump(LOG_INFO, dump);
2063 break;
2064 }
2065
2066 case SIGHUP:
2067 r = verify_run_space_and_log("Refusing to reload");
2068 if (r >= 0)
2069 m->exit_code = MANAGER_RELOAD;
2070 break;
2071
2072 default: {
2073
2074 /* Starting SIGRTMIN+0 */
2075 static const char * const target_table[] = {
2076 [0] = SPECIAL_DEFAULT_TARGET,
2077 [1] = SPECIAL_RESCUE_TARGET,
2078 [2] = SPECIAL_EMERGENCY_TARGET,
2079 [3] = SPECIAL_HALT_TARGET,
2080 [4] = SPECIAL_POWEROFF_TARGET,
2081 [5] = SPECIAL_REBOOT_TARGET,
2082 [6] = SPECIAL_KEXEC_TARGET
2083 };
2084
2085 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2086 static const ManagerExitCode code_table[] = {
2087 [0] = MANAGER_HALT,
2088 [1] = MANAGER_POWEROFF,
2089 [2] = MANAGER_REBOOT,
2090 [3] = MANAGER_KEXEC
2091 };
2092
2093 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2094 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2095 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2096 manager_start_target(m, target_table[idx],
2097 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2098 break;
2099 }
2100
2101 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2102 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2103 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2104 break;
2105 }
2106
2107 switch (sfsi.ssi_signo - SIGRTMIN) {
2108
2109 case 20:
2110 manager_set_show_status(m, SHOW_STATUS_YES);
2111 break;
2112
2113 case 21:
2114 manager_set_show_status(m, SHOW_STATUS_NO);
2115 break;
2116
2117 case 22:
2118 log_set_max_level(LOG_DEBUG);
2119 log_info("Setting log level to debug.");
2120 break;
2121
2122 case 23:
2123 log_set_max_level(LOG_INFO);
2124 log_info("Setting log level to info.");
2125 break;
2126
2127 case 24:
2128 if (MANAGER_IS_USER(m)) {
2129 m->exit_code = MANAGER_EXIT;
2130 return 0;
2131 }
2132
2133 /* This is a nop on init */
2134 break;
2135
2136 case 26:
2137 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2138 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2139 log_notice("Setting log target to journal-or-kmsg.");
2140 break;
2141
2142 case 27:
2143 log_set_target(LOG_TARGET_CONSOLE);
2144 log_notice("Setting log target to console.");
2145 break;
2146
2147 case 28:
2148 log_set_target(LOG_TARGET_KMSG);
2149 log_notice("Setting log target to kmsg.");
2150 break;
2151
2152 default:
2153 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2154 }
2155 }
2156 }
2157 }
2158
2159 if (sigchld)
2160 manager_dispatch_sigchld(m);
2161
2162 return 0;
2163 }
2164
2165 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2166 Manager *m = userdata;
2167 Iterator i;
2168 Unit *u;
2169
2170 assert(m);
2171 assert(m->time_change_fd == fd);
2172
2173 log_struct(LOG_DEBUG,
2174 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
2175 LOG_MESSAGE("Time has been changed"),
2176 NULL);
2177
2178 /* Restart the watch */
2179 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2180 m->time_change_fd = safe_close(m->time_change_fd);
2181
2182 manager_setup_time_change(m);
2183
2184 HASHMAP_FOREACH(u, m->units, i)
2185 if (UNIT_VTABLE(u)->time_change)
2186 UNIT_VTABLE(u)->time_change(u);
2187
2188 return 0;
2189 }
2190
2191 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2192 Manager *m = userdata;
2193
2194 assert(m);
2195 assert(m->idle_pipe[2] == fd);
2196
2197 m->no_console_output = m->n_on_console > 0;
2198
2199 manager_close_idle_pipe(m);
2200
2201 return 0;
2202 }
2203
2204 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2205 Manager *m = userdata;
2206 int r;
2207 uint64_t next;
2208
2209 assert(m);
2210 assert(source);
2211
2212 manager_print_jobs_in_progress(m);
2213
2214 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2215 r = sd_event_source_set_time(source, next);
2216 if (r < 0)
2217 return r;
2218
2219 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2220 }
2221
2222 int manager_loop(Manager *m) {
2223 int r;
2224
2225 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2226
2227 assert(m);
2228 m->exit_code = MANAGER_OK;
2229
2230 /* Release the path cache */
2231 m->unit_path_cache = set_free_free(m->unit_path_cache);
2232
2233 manager_check_finished(m);
2234
2235 /* There might still be some zombies hanging around from
2236 * before we were exec()'ed. Let's reap them. */
2237 r = manager_dispatch_sigchld(m);
2238 if (r < 0)
2239 return r;
2240
2241 while (m->exit_code == MANAGER_OK) {
2242 usec_t wait_usec;
2243
2244 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2245 watchdog_ping();
2246
2247 if (!ratelimit_test(&rl)) {
2248 /* Yay, something is going seriously wrong, pause a little */
2249 log_warning("Looping too fast. Throttling execution a little.");
2250 sleep(1);
2251 }
2252
2253 if (manager_dispatch_load_queue(m) > 0)
2254 continue;
2255
2256 if (manager_dispatch_gc_job_queue(m) > 0)
2257 continue;
2258
2259 if (manager_dispatch_gc_unit_queue(m) > 0)
2260 continue;
2261
2262 if (manager_dispatch_cleanup_queue(m) > 0)
2263 continue;
2264
2265 if (manager_dispatch_cgroup_queue(m) > 0)
2266 continue;
2267
2268 if (manager_dispatch_dbus_queue(m) > 0)
2269 continue;
2270
2271 /* Sleep for half the watchdog time */
2272 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2273 wait_usec = m->runtime_watchdog / 2;
2274 if (wait_usec <= 0)
2275 wait_usec = 1;
2276 } else
2277 wait_usec = USEC_INFINITY;
2278
2279 r = sd_event_run(m->event, wait_usec);
2280 if (r < 0)
2281 return log_error_errno(r, "Failed to run event loop: %m");
2282 }
2283
2284 return m->exit_code;
2285 }
2286
2287 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2288 _cleanup_free_ char *n = NULL;
2289 sd_id128_t invocation_id;
2290 Unit *u;
2291 int r;
2292
2293 assert(m);
2294 assert(s);
2295 assert(_u);
2296
2297 r = unit_name_from_dbus_path(s, &n);
2298 if (r < 0)
2299 return r;
2300
2301 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2302 * as invocation ID. */
2303 r = sd_id128_from_string(n, &invocation_id);
2304 if (r >= 0) {
2305 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2306 if (u) {
2307 *_u = u;
2308 return 0;
2309 }
2310
2311 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));
2312 }
2313
2314 /* If this didn't work, we use the suffix as unit name. */
2315 r = manager_load_unit(m, n, NULL, e, &u);
2316 if (r < 0)
2317 return r;
2318
2319 *_u = u;
2320 return 0;
2321 }
2322
2323 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2324 const char *p;
2325 unsigned id;
2326 Job *j;
2327 int r;
2328
2329 assert(m);
2330 assert(s);
2331 assert(_j);
2332
2333 p = startswith(s, "/org/freedesktop/systemd1/job/");
2334 if (!p)
2335 return -EINVAL;
2336
2337 r = safe_atou(p, &id);
2338 if (r < 0)
2339 return r;
2340
2341 j = manager_get_job(m, id);
2342 if (!j)
2343 return -ENOENT;
2344
2345 *_j = j;
2346
2347 return 0;
2348 }
2349
2350 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2351
2352 #ifdef HAVE_AUDIT
2353 _cleanup_free_ char *p = NULL;
2354 const char *msg;
2355 int audit_fd, r;
2356
2357 if (!MANAGER_IS_SYSTEM(m))
2358 return;
2359
2360 audit_fd = get_audit_fd();
2361 if (audit_fd < 0)
2362 return;
2363
2364 /* Don't generate audit events if the service was already
2365 * started and we're just deserializing */
2366 if (MANAGER_IS_RELOADING(m))
2367 return;
2368
2369 if (u->type != UNIT_SERVICE)
2370 return;
2371
2372 r = unit_name_to_prefix_and_instance(u->id, &p);
2373 if (r < 0) {
2374 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2375 return;
2376 }
2377
2378 msg = strjoina("unit=", p);
2379 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2380 if (errno == EPERM)
2381 /* We aren't allowed to send audit messages?
2382 * Then let's not retry again. */
2383 close_audit_fd();
2384 else
2385 log_warning_errno(errno, "Failed to send audit message: %m");
2386 }
2387 #endif
2388
2389 }
2390
2391 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2392 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2393 _cleanup_free_ char *message = NULL;
2394 _cleanup_close_ int fd = -1;
2395 int n = 0;
2396
2397 /* Don't generate plymouth events if the service was already
2398 * started and we're just deserializing */
2399 if (MANAGER_IS_RELOADING(m))
2400 return;
2401
2402 if (!MANAGER_IS_SYSTEM(m))
2403 return;
2404
2405 if (detect_container() > 0)
2406 return;
2407
2408 if (u->type != UNIT_SERVICE &&
2409 u->type != UNIT_MOUNT &&
2410 u->type != UNIT_SWAP)
2411 return;
2412
2413 /* We set SOCK_NONBLOCK here so that we rather drop the
2414 * message then wait for plymouth */
2415 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2416 if (fd < 0) {
2417 log_error_errno(errno, "socket() failed: %m");
2418 return;
2419 }
2420
2421 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2422
2423 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2424 log_error_errno(errno, "connect() failed: %m");
2425 return;
2426 }
2427
2428 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2429 log_oom();
2430 return;
2431 }
2432
2433 errno = 0;
2434 if (write(fd, message, n + 1) != n + 1)
2435 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2436 log_error_errno(errno, "Failed to write Plymouth message: %m");
2437 }
2438
2439 int manager_open_serialization(Manager *m, FILE **_f) {
2440 int fd;
2441 FILE *f;
2442
2443 assert(_f);
2444
2445 fd = open_serialization_fd("systemd-state");
2446 if (fd < 0)
2447 return fd;
2448
2449 f = fdopen(fd, "w+");
2450 if (!f) {
2451 safe_close(fd);
2452 return -errno;
2453 }
2454
2455 *_f = f;
2456 return 0;
2457 }
2458
2459 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2460 Iterator i;
2461 Unit *u;
2462 const char *t;
2463 char **e;
2464 int r;
2465
2466 assert(m);
2467 assert(f);
2468 assert(fds);
2469
2470 m->n_reloading++;
2471
2472 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2473 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2474 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2475 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2476
2477 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2478 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2479 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2480 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2481
2482 if (!in_initrd()) {
2483 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2484 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2485 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2486 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2487 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2488 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2489 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2490 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2491 }
2492
2493 if (!switching_root) {
2494 STRV_FOREACH(e, m->environment) {
2495 _cleanup_free_ char *ce;
2496
2497 ce = cescape(*e);
2498 if (!ce)
2499 return -ENOMEM;
2500
2501 fprintf(f, "env=%s\n", *e);
2502 }
2503 }
2504
2505 if (m->notify_fd >= 0) {
2506 int copy;
2507
2508 copy = fdset_put_dup(fds, m->notify_fd);
2509 if (copy < 0)
2510 return copy;
2511
2512 fprintf(f, "notify-fd=%i\n", copy);
2513 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2514 }
2515
2516 if (m->cgroups_agent_fd >= 0) {
2517 int copy;
2518
2519 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2520 if (copy < 0)
2521 return copy;
2522
2523 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2524 }
2525
2526 if (m->user_lookup_fds[0] >= 0) {
2527 int copy0, copy1;
2528
2529 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2530 if (copy0 < 0)
2531 return copy0;
2532
2533 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2534 if (copy1 < 0)
2535 return copy1;
2536
2537 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2538 }
2539
2540 bus_track_serialize(m->subscribed, f, "subscribed");
2541
2542 r = dynamic_user_serialize(m, f, fds);
2543 if (r < 0)
2544 return r;
2545
2546 manager_serialize_uid_refs(m, f);
2547 manager_serialize_gid_refs(m, f);
2548
2549 fputc('\n', f);
2550
2551 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2552 if (u->id != t)
2553 continue;
2554
2555 /* Start marker */
2556 fputs(u->id, f);
2557 fputc('\n', f);
2558
2559 r = unit_serialize(u, f, fds, !switching_root);
2560 if (r < 0) {
2561 m->n_reloading--;
2562 return r;
2563 }
2564 }
2565
2566 assert(m->n_reloading > 0);
2567 m->n_reloading--;
2568
2569 if (ferror(f))
2570 return -EIO;
2571
2572 r = bus_fdset_add_all(m, fds);
2573 if (r < 0)
2574 return r;
2575
2576 return 0;
2577 }
2578
2579 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2580 int r = 0;
2581
2582 assert(m);
2583 assert(f);
2584
2585 log_debug("Deserializing state...");
2586
2587 m->n_reloading++;
2588
2589 for (;;) {
2590 char line[LINE_MAX], *l;
2591 const char *val;
2592
2593 if (!fgets(line, sizeof(line), f)) {
2594 if (feof(f))
2595 r = 0;
2596 else
2597 r = -errno;
2598
2599 goto finish;
2600 }
2601
2602 char_array_0(line);
2603 l = strstrip(line);
2604
2605 if (l[0] == 0)
2606 break;
2607
2608 if ((val = startswith(l, "current-job-id="))) {
2609 uint32_t id;
2610
2611 if (safe_atou32(val, &id) < 0)
2612 log_debug("Failed to parse current job id value %s", val);
2613 else
2614 m->current_job_id = MAX(m->current_job_id, id);
2615
2616 } else if ((val = startswith(l, "n-installed-jobs="))) {
2617 uint32_t n;
2618
2619 if (safe_atou32(val, &n) < 0)
2620 log_debug("Failed to parse installed jobs counter %s", val);
2621 else
2622 m->n_installed_jobs += n;
2623
2624 } else if ((val = startswith(l, "n-failed-jobs="))) {
2625 uint32_t n;
2626
2627 if (safe_atou32(val, &n) < 0)
2628 log_debug("Failed to parse failed jobs counter %s", val);
2629 else
2630 m->n_failed_jobs += n;
2631
2632 } else if ((val = startswith(l, "taint-usr="))) {
2633 int b;
2634
2635 b = parse_boolean(val);
2636 if (b < 0)
2637 log_debug("Failed to parse taint /usr flag %s", val);
2638 else
2639 m->taint_usr = m->taint_usr || b;
2640
2641 } else if ((val = startswith(l, "firmware-timestamp=")))
2642 dual_timestamp_deserialize(val, &m->firmware_timestamp);
2643 else if ((val = startswith(l, "loader-timestamp=")))
2644 dual_timestamp_deserialize(val, &m->loader_timestamp);
2645 else if ((val = startswith(l, "kernel-timestamp=")))
2646 dual_timestamp_deserialize(val, &m->kernel_timestamp);
2647 else if ((val = startswith(l, "initrd-timestamp=")))
2648 dual_timestamp_deserialize(val, &m->initrd_timestamp);
2649 else if ((val = startswith(l, "userspace-timestamp=")))
2650 dual_timestamp_deserialize(val, &m->userspace_timestamp);
2651 else if ((val = startswith(l, "finish-timestamp=")))
2652 dual_timestamp_deserialize(val, &m->finish_timestamp);
2653 else if ((val = startswith(l, "security-start-timestamp=")))
2654 dual_timestamp_deserialize(val, &m->security_start_timestamp);
2655 else if ((val = startswith(l, "security-finish-timestamp=")))
2656 dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2657 else if ((val = startswith(l, "generators-start-timestamp=")))
2658 dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2659 else if ((val = startswith(l, "generators-finish-timestamp=")))
2660 dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2661 else if ((val = startswith(l, "units-load-start-timestamp=")))
2662 dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2663 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2664 dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2665 else if (startswith(l, "env=")) {
2666 _cleanup_free_ char *uce = NULL;
2667 char **e;
2668
2669 r = cunescape(l + 4, UNESCAPE_RELAX, &uce);
2670 if (r < 0)
2671 goto finish;
2672
2673 e = strv_env_set(m->environment, uce);
2674 if (!e) {
2675 r = -ENOMEM;
2676 goto finish;
2677 }
2678
2679 strv_free(m->environment);
2680 m->environment = e;
2681
2682 } else if ((val = startswith(l, "notify-fd="))) {
2683 int fd;
2684
2685 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2686 log_debug("Failed to parse notify fd: %s", val);
2687 else {
2688 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2689 safe_close(m->notify_fd);
2690 m->notify_fd = fdset_remove(fds, fd);
2691 }
2692
2693 } else if ((val = startswith(l, "notify-socket="))) {
2694 char *n;
2695
2696 n = strdup(val);
2697 if (!n) {
2698 r = -ENOMEM;
2699 goto finish;
2700 }
2701
2702 free(m->notify_socket);
2703 m->notify_socket = n;
2704
2705 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2706 int fd;
2707
2708 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2709 log_debug("Failed to parse cgroups agent fd: %s", val);
2710 else {
2711 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2712 safe_close(m->cgroups_agent_fd);
2713 m->cgroups_agent_fd = fdset_remove(fds, fd);
2714 }
2715
2716 } else if ((val = startswith(l, "user-lookup="))) {
2717 int fd0, fd1;
2718
2719 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2720 log_debug("Failed to parse user lookup fd: %s", val);
2721 else {
2722 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2723 safe_close_pair(m->user_lookup_fds);
2724 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2725 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2726 }
2727
2728 } else if ((val = startswith(l, "dynamic-user=")))
2729 dynamic_user_deserialize_one(m, val, fds);
2730 else if ((val = startswith(l, "destroy-ipc-uid=")))
2731 manager_deserialize_uid_refs_one(m, val);
2732 else if ((val = startswith(l, "destroy-ipc-gid=")))
2733 manager_deserialize_gid_refs_one(m, val);
2734 else if ((val = startswith(l, "subscribed="))) {
2735
2736 if (strv_extend(&m->deserialized_subscribed, val) < 0)
2737 log_oom();
2738
2739 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2740 log_debug("Unknown serialization item '%s'", l);
2741 }
2742
2743 for (;;) {
2744 Unit *u;
2745 char name[UNIT_NAME_MAX+2];
2746
2747 /* Start marker */
2748 if (!fgets(name, sizeof(name), f)) {
2749 if (feof(f))
2750 r = 0;
2751 else
2752 r = -errno;
2753
2754 goto finish;
2755 }
2756
2757 char_array_0(name);
2758
2759 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2760 if (r < 0)
2761 goto finish;
2762
2763 r = unit_deserialize(u, f, fds);
2764 if (r < 0)
2765 goto finish;
2766 }
2767
2768 finish:
2769 if (ferror(f))
2770 r = -EIO;
2771
2772 assert(m->n_reloading > 0);
2773 m->n_reloading--;
2774
2775 return r;
2776 }
2777
2778 int manager_reload(Manager *m) {
2779 int r, q;
2780 _cleanup_fclose_ FILE *f = NULL;
2781 _cleanup_fdset_free_ FDSet *fds = NULL;
2782
2783 assert(m);
2784
2785 r = manager_open_serialization(m, &f);
2786 if (r < 0)
2787 return r;
2788
2789 m->n_reloading++;
2790 bus_manager_send_reloading(m, true);
2791
2792 fds = fdset_new();
2793 if (!fds) {
2794 m->n_reloading--;
2795 return -ENOMEM;
2796 }
2797
2798 r = manager_serialize(m, f, fds, false);
2799 if (r < 0) {
2800 m->n_reloading--;
2801 return r;
2802 }
2803
2804 if (fseeko(f, 0, SEEK_SET) < 0) {
2805 m->n_reloading--;
2806 return -errno;
2807 }
2808
2809 /* From here on there is no way back. */
2810 manager_clear_jobs_and_units(m);
2811 lookup_paths_flush_generator(&m->lookup_paths);
2812 lookup_paths_free(&m->lookup_paths);
2813 dynamic_user_vacuum(m, false);
2814 m->uid_refs = hashmap_free(m->uid_refs);
2815 m->gid_refs = hashmap_free(m->gid_refs);
2816
2817 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2818 if (q < 0 && r >= 0)
2819 r = q;
2820
2821 /* Find new unit paths */
2822 q = manager_run_generators(m);
2823 if (q < 0 && r >= 0)
2824 r = q;
2825
2826 lookup_paths_reduce(&m->lookup_paths);
2827 manager_build_unit_path_cache(m);
2828
2829 /* First, enumerate what we can from all config files */
2830 manager_enumerate(m);
2831
2832 /* Second, deserialize our stored data */
2833 q = manager_deserialize(m, f, fds);
2834 if (q < 0 && r >= 0)
2835 r = q;
2836
2837 fclose(f);
2838 f = NULL;
2839
2840 /* Re-register notify_fd as event source */
2841 q = manager_setup_notify(m);
2842 if (q < 0 && r >= 0)
2843 r = q;
2844
2845 q = manager_setup_cgroups_agent(m);
2846 if (q < 0 && r >= 0)
2847 r = q;
2848
2849 q = manager_setup_user_lookup_fd(m);
2850 if (q < 0 && r >= 0)
2851 r = q;
2852
2853 /* Third, fire things up! */
2854 manager_coldplug(m);
2855
2856 /* Release any dynamic users no longer referenced */
2857 dynamic_user_vacuum(m, true);
2858
2859 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2860 manager_vacuum_uid_refs(m);
2861 manager_vacuum_gid_refs(m);
2862
2863 /* Sync current state of bus names with our set of listening units */
2864 if (m->api_bus)
2865 manager_sync_bus_names(m, m->api_bus);
2866
2867 assert(m->n_reloading > 0);
2868 m->n_reloading--;
2869
2870 m->send_reloading_done = true;
2871
2872 return r;
2873 }
2874
2875 void manager_reset_failed(Manager *m) {
2876 Unit *u;
2877 Iterator i;
2878
2879 assert(m);
2880
2881 HASHMAP_FOREACH(u, m->units, i)
2882 unit_reset_failed(u);
2883 }
2884
2885 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2886 Unit *u;
2887
2888 assert(m);
2889 assert(name);
2890
2891 /* Returns true if the unit is inactive or going down */
2892 u = manager_get_unit(m, name);
2893 if (!u)
2894 return true;
2895
2896 return unit_inactive_or_pending(u);
2897 }
2898
2899 static void manager_notify_finished(Manager *m) {
2900 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2901 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2902
2903 if (m->test_run)
2904 return;
2905
2906 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2907
2908 /* Note that m->kernel_usec.monotonic is always at 0,
2909 * and m->firmware_usec.monotonic and
2910 * m->loader_usec.monotonic should be considered
2911 * negative values. */
2912
2913 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2914 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2915 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2916 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2917
2918 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2919
2920 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2921 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2922
2923 log_struct(LOG_INFO,
2924 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2925 "KERNEL_USEC="USEC_FMT, kernel_usec,
2926 "INITRD_USEC="USEC_FMT, initrd_usec,
2927 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2928 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2929 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2930 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2931 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2932 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2933 NULL);
2934 } else {
2935 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2936 initrd_usec = 0;
2937
2938 log_struct(LOG_INFO,
2939 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2940 "KERNEL_USEC="USEC_FMT, kernel_usec,
2941 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2942 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2943 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2944 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2945 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2946 NULL);
2947 }
2948 } else {
2949 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2950 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2951
2952 log_struct(LOG_INFO,
2953 LOG_MESSAGE_ID(SD_MESSAGE_USER_STARTUP_FINISHED),
2954 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2955 LOG_MESSAGE("Startup finished in %s.",
2956 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2957 NULL);
2958 }
2959
2960 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2961
2962 sd_notifyf(false,
2963 "READY=1\n"
2964 "STATUS=Startup finished in %s.",
2965 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2966 }
2967
2968 void manager_check_finished(Manager *m) {
2969 assert(m);
2970
2971 if (MANAGER_IS_RELOADING(m))
2972 return;
2973
2974 /* Verify that we are actually running currently. Initially
2975 * the exit code is set to invalid, and during operation it is
2976 * then set to MANAGER_OK */
2977 if (m->exit_code != MANAGER_OK)
2978 return;
2979
2980 if (hashmap_size(m->jobs) > 0) {
2981 if (m->jobs_in_progress_event_source)
2982 /* Ignore any failure, this is only for feedback */
2983 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2984
2985 return;
2986 }
2987
2988 manager_flip_auto_status(m, false);
2989
2990 /* Notify Type=idle units that we are done now */
2991 manager_close_idle_pipe(m);
2992
2993 /* Turn off confirm spawn now */
2994 m->confirm_spawn = NULL;
2995
2996 /* No need to update ask password status when we're going non-interactive */
2997 manager_close_ask_password(m);
2998
2999 /* This is no longer the first boot */
3000 manager_set_first_boot(m, false);
3001
3002 if (dual_timestamp_is_set(&m->finish_timestamp))
3003 return;
3004
3005 dual_timestamp_get(&m->finish_timestamp);
3006
3007 manager_notify_finished(m);
3008
3009 manager_invalidate_startup_units(m);
3010 }
3011
3012 static int manager_run_generators(Manager *m) {
3013 _cleanup_strv_free_ char **paths = NULL;
3014 const char *argv[5];
3015 char **path;
3016 int r;
3017
3018 assert(m);
3019
3020 if (m->test_run)
3021 return 0;
3022
3023 paths = generator_binary_paths(m->unit_file_scope);
3024 if (!paths)
3025 return log_oom();
3026
3027 /* Optimize by skipping the whole process by not creating output directories
3028 * if no generators are found. */
3029 STRV_FOREACH(path, paths) {
3030 if (access(*path, F_OK) >= 0)
3031 goto found;
3032 if (errno != ENOENT)
3033 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3034 }
3035
3036 return 0;
3037
3038 found:
3039 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3040 if (r < 0)
3041 goto finish;
3042
3043 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3044 argv[1] = m->lookup_paths.generator;
3045 argv[2] = m->lookup_paths.generator_early;
3046 argv[3] = m->lookup_paths.generator_late;
3047 argv[4] = NULL;
3048
3049 RUN_WITH_UMASK(0022)
3050 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, (char**) argv);
3051
3052 finish:
3053 lookup_paths_trim_generator(&m->lookup_paths);
3054 return r;
3055 }
3056
3057 int manager_environment_add(Manager *m, char **minus, char **plus) {
3058 char **a = NULL, **b = NULL, **l;
3059 assert(m);
3060
3061 l = m->environment;
3062
3063 if (!strv_isempty(minus)) {
3064 a = strv_env_delete(l, 1, minus);
3065 if (!a)
3066 return -ENOMEM;
3067
3068 l = a;
3069 }
3070
3071 if (!strv_isempty(plus)) {
3072 b = strv_env_merge(2, l, plus);
3073 if (!b) {
3074 strv_free(a);
3075 return -ENOMEM;
3076 }
3077
3078 l = b;
3079 }
3080
3081 if (m->environment != l)
3082 strv_free(m->environment);
3083 if (a != l)
3084 strv_free(a);
3085 if (b != l)
3086 strv_free(b);
3087
3088 m->environment = l;
3089 manager_clean_environment(m);
3090 strv_sort(m->environment);
3091
3092 return 0;
3093 }
3094
3095 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3096 int i;
3097
3098 assert(m);
3099
3100 for (i = 0; i < _RLIMIT_MAX; i++) {
3101 m->rlimit[i] = mfree(m->rlimit[i]);
3102
3103 if (!default_rlimit[i])
3104 continue;
3105
3106 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3107 if (!m->rlimit[i])
3108 return log_oom();
3109 }
3110
3111 return 0;
3112 }
3113
3114 void manager_recheck_journal(Manager *m) {
3115 Unit *u;
3116
3117 assert(m);
3118
3119 if (!MANAGER_IS_SYSTEM(m))
3120 return;
3121
3122 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3123 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3124 log_close_journal();
3125 return;
3126 }
3127
3128 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3129 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3130 log_close_journal();
3131 return;
3132 }
3133
3134 /* Hmm, OK, so the socket is fully up and the service is up
3135 * too, then let's make use of the thing. */
3136 log_open();
3137 }
3138
3139 void manager_set_show_status(Manager *m, ShowStatus mode) {
3140 assert(m);
3141 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3142
3143 if (!MANAGER_IS_SYSTEM(m))
3144 return;
3145
3146 if (m->show_status != mode)
3147 log_debug("%s showing of status.",
3148 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3149 m->show_status = mode;
3150
3151 if (mode > 0)
3152 (void) touch("/run/systemd/show-status");
3153 else
3154 (void) unlink("/run/systemd/show-status");
3155 }
3156
3157 static bool manager_get_show_status(Manager *m, StatusType type) {
3158 assert(m);
3159
3160 if (!MANAGER_IS_SYSTEM(m))
3161 return false;
3162
3163 if (m->no_console_output)
3164 return false;
3165
3166 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3167 return false;
3168
3169 /* If we cannot find out the status properly, just proceed. */
3170 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3171 return false;
3172
3173 if (m->show_status > 0)
3174 return true;
3175
3176 return false;
3177 }
3178
3179 const char *manager_get_confirm_spawn(Manager *m) {
3180 static int last_errno = 0;
3181 const char *vc = m->confirm_spawn;
3182 struct stat st;
3183 int r;
3184
3185 /* Here's the deal: we want to test the validity of the console but don't want
3186 * PID1 to go through the whole console process which might block. But we also
3187 * want to warn the user only once if something is wrong with the console so we
3188 * cannot do the sanity checks after spawning our children. So here we simply do
3189 * really basic tests to hopefully trap common errors.
3190 *
3191 * If the console suddenly disappear at the time our children will really it
3192 * then they will simply fail to acquire it and a positive answer will be
3193 * assumed. New children will fallback to /dev/console though.
3194 *
3195 * Note: TTYs are devices that can come and go any time, and frequently aren't
3196 * available yet during early boot (consider a USB rs232 dongle...). If for any
3197 * reason the configured console is not ready, we fallback to the default
3198 * console. */
3199
3200 if (!vc || path_equal(vc, "/dev/console"))
3201 return vc;
3202
3203 r = stat(vc, &st);
3204 if (r < 0)
3205 goto fail;
3206
3207 if (!S_ISCHR(st.st_mode)) {
3208 errno = ENOTTY;
3209 goto fail;
3210 }
3211
3212 last_errno = 0;
3213 return vc;
3214 fail:
3215 if (last_errno != errno) {
3216 last_errno = errno;
3217 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3218 }
3219 return "/dev/console";
3220 }
3221
3222 void manager_set_first_boot(Manager *m, bool b) {
3223 assert(m);
3224
3225 if (!MANAGER_IS_SYSTEM(m))
3226 return;
3227
3228 if (m->first_boot != (int) b) {
3229 if (b)
3230 (void) touch("/run/systemd/first-boot");
3231 else
3232 (void) unlink("/run/systemd/first-boot");
3233 }
3234
3235 m->first_boot = b;
3236 }
3237
3238 void manager_disable_confirm_spawn(void) {
3239 (void) touch("/run/systemd/confirm_spawn_disabled");
3240 }
3241
3242 bool manager_is_confirm_spawn_disabled(Manager *m) {
3243 if (!m->confirm_spawn)
3244 return true;
3245
3246 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3247 }
3248
3249 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3250 va_list ap;
3251
3252 /* If m is NULL, assume we're after shutdown and let the messages through. */
3253
3254 if (m && !manager_get_show_status(m, type))
3255 return;
3256
3257 /* XXX We should totally drop the check for ephemeral here
3258 * and thus effectively make 'Type=idle' pointless. */
3259 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3260 return;
3261
3262 va_start(ap, format);
3263 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3264 va_end(ap);
3265 }
3266
3267 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3268 char p[strlen(path)+1];
3269
3270 assert(m);
3271 assert(path);
3272
3273 strcpy(p, path);
3274 path_kill_slashes(p);
3275
3276 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3277 }
3278
3279 const char *manager_get_runtime_prefix(Manager *m) {
3280 assert(m);
3281
3282 return MANAGER_IS_SYSTEM(m) ?
3283 "/run" :
3284 getenv("XDG_RUNTIME_DIR");
3285 }
3286
3287 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3288 unsigned size;
3289 int r;
3290
3291 assert(m);
3292 assert(u->manager == m);
3293
3294 size = set_size(m->failed_units);
3295
3296 if (failed) {
3297 r = set_ensure_allocated(&m->failed_units, NULL);
3298 if (r < 0)
3299 return log_oom();
3300
3301 if (set_put(m->failed_units, u) < 0)
3302 return log_oom();
3303 } else
3304 (void) set_remove(m->failed_units, u);
3305
3306 if (set_size(m->failed_units) != size)
3307 bus_manager_send_change_signal(m);
3308
3309 return 0;
3310 }
3311
3312 ManagerState manager_state(Manager *m) {
3313 Unit *u;
3314
3315 assert(m);
3316
3317 /* Did we ever finish booting? If not then we are still starting up */
3318 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3319
3320 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3321 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3322 return MANAGER_INITIALIZING;
3323
3324 return MANAGER_STARTING;
3325 }
3326
3327 /* Is the special shutdown target queued? If so, we are in shutdown state */
3328 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3329 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3330 return MANAGER_STOPPING;
3331
3332 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3333 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3334 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3335 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3336 return MANAGER_MAINTENANCE;
3337
3338 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3339 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3340 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3341 return MANAGER_MAINTENANCE;
3342
3343 /* Are there any failed units? If so, we are in degraded mode */
3344 if (set_size(m->failed_units) > 0)
3345 return MANAGER_DEGRADED;
3346
3347 return MANAGER_RUNNING;
3348 }
3349
3350 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3351
3352 static void manager_unref_uid_internal(
3353 Manager *m,
3354 Hashmap **uid_refs,
3355 uid_t uid,
3356 bool destroy_now,
3357 int (*_clean_ipc)(uid_t uid)) {
3358
3359 uint32_t c, n;
3360
3361 assert(m);
3362 assert(uid_refs);
3363 assert(uid_is_valid(uid));
3364 assert(_clean_ipc);
3365
3366 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3367 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3368 *
3369 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3370 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3371 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3372 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3373
3374 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3375 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3376
3377 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3378 return;
3379
3380 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3381
3382 n = c & ~DESTROY_IPC_FLAG;
3383 assert(n > 0);
3384 n--;
3385
3386 if (destroy_now && n == 0) {
3387 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3388
3389 if (c & DESTROY_IPC_FLAG) {
3390 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3391 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3392 uid);
3393 (void) _clean_ipc(uid);
3394 }
3395 } else {
3396 c = n | (c & DESTROY_IPC_FLAG);
3397 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3398 }
3399 }
3400
3401 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3402 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3403 }
3404
3405 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3406 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3407 }
3408
3409 static int manager_ref_uid_internal(
3410 Manager *m,
3411 Hashmap **uid_refs,
3412 uid_t uid,
3413 bool clean_ipc) {
3414
3415 uint32_t c, n;
3416 int r;
3417
3418 assert(m);
3419 assert(uid_refs);
3420 assert(uid_is_valid(uid));
3421
3422 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3423 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3424
3425 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3426 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3427
3428 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3429 return 0;
3430
3431 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3432 if (r < 0)
3433 return r;
3434
3435 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3436
3437 n = c & ~DESTROY_IPC_FLAG;
3438 n++;
3439
3440 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3441 return -EOVERFLOW;
3442
3443 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3444
3445 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3446 }
3447
3448 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3449 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3450 }
3451
3452 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3453 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3454 }
3455
3456 static void manager_vacuum_uid_refs_internal(
3457 Manager *m,
3458 Hashmap **uid_refs,
3459 int (*_clean_ipc)(uid_t uid)) {
3460
3461 Iterator i;
3462 void *p, *k;
3463
3464 assert(m);
3465 assert(uid_refs);
3466 assert(_clean_ipc);
3467
3468 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3469 uint32_t c, n;
3470 uid_t uid;
3471
3472 uid = PTR_TO_UID(k);
3473 c = PTR_TO_UINT32(p);
3474
3475 n = c & ~DESTROY_IPC_FLAG;
3476 if (n > 0)
3477 continue;
3478
3479 if (c & DESTROY_IPC_FLAG) {
3480 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3481 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3482 uid);
3483 (void) _clean_ipc(uid);
3484 }
3485
3486 assert_se(hashmap_remove(*uid_refs, k) == p);
3487 }
3488 }
3489
3490 void manager_vacuum_uid_refs(Manager *m) {
3491 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3492 }
3493
3494 void manager_vacuum_gid_refs(Manager *m) {
3495 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3496 }
3497
3498 static void manager_serialize_uid_refs_internal(
3499 Manager *m,
3500 FILE *f,
3501 Hashmap **uid_refs,
3502 const char *field_name) {
3503
3504 Iterator i;
3505 void *p, *k;
3506
3507 assert(m);
3508 assert(f);
3509 assert(uid_refs);
3510 assert(field_name);
3511
3512 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3513 * of it is better rebuild after a reload/reexec. */
3514
3515 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3516 uint32_t c;
3517 uid_t uid;
3518
3519 uid = PTR_TO_UID(k);
3520 c = PTR_TO_UINT32(p);
3521
3522 if (!(c & DESTROY_IPC_FLAG))
3523 continue;
3524
3525 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3526 }
3527 }
3528
3529 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3530 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3531 }
3532
3533 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3534 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3535 }
3536
3537 static void manager_deserialize_uid_refs_one_internal(
3538 Manager *m,
3539 Hashmap** uid_refs,
3540 const char *value) {
3541
3542 uid_t uid;
3543 uint32_t c;
3544 int r;
3545
3546 assert(m);
3547 assert(uid_refs);
3548 assert(value);
3549
3550 r = parse_uid(value, &uid);
3551 if (r < 0 || uid == 0) {
3552 log_debug("Unable to parse UID reference serialization");
3553 return;
3554 }
3555
3556 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3557 if (r < 0) {
3558 log_oom();
3559 return;
3560 }
3561
3562 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3563 if (c & DESTROY_IPC_FLAG)
3564 return;
3565
3566 c |= DESTROY_IPC_FLAG;
3567
3568 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3569 if (r < 0) {
3570 log_debug("Failed to add UID reference entry");
3571 return;
3572 }
3573 }
3574
3575 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3576 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3577 }
3578
3579 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3580 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3581 }
3582
3583 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3584 struct buffer {
3585 uid_t uid;
3586 gid_t gid;
3587 char unit_name[UNIT_NAME_MAX+1];
3588 } _packed_ buffer;
3589
3590 Manager *m = userdata;
3591 ssize_t l;
3592 size_t n;
3593 Unit *u;
3594
3595 assert_se(source);
3596 assert_se(m);
3597
3598 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3599 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3600 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3601
3602 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3603 if (l < 0) {
3604 if (errno == EINTR || errno == EAGAIN)
3605 return 0;
3606
3607 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3608 }
3609
3610 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3611 log_warning("Received too short user lookup message, ignoring.");
3612 return 0;
3613 }
3614
3615 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3616 log_warning("Received too long user lookup message, ignoring.");
3617 return 0;
3618 }
3619
3620 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3621 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3622 return 0;
3623 }
3624
3625 n = (size_t) l - offsetof(struct buffer, unit_name);
3626 if (memchr(buffer.unit_name, 0, n)) {
3627 log_warning("Received lookup message with embedded NUL character, ignoring.");
3628 return 0;
3629 }
3630
3631 buffer.unit_name[n] = 0;
3632 u = manager_get_unit(m, buffer.unit_name);
3633 if (!u) {
3634 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3635 return 0;
3636 }
3637
3638 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3639
3640 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3641 return 0;
3642 }
3643
3644 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3645 [MANAGER_INITIALIZING] = "initializing",
3646 [MANAGER_STARTING] = "starting",
3647 [MANAGER_RUNNING] = "running",
3648 [MANAGER_DEGRADED] = "degraded",
3649 [MANAGER_MAINTENANCE] = "maintenance",
3650 [MANAGER_STOPPING] = "stopping",
3651 };
3652
3653 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
Unnamed repository; edit this file 'description' to name the repository.