]> 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: add "invocation ID" concept to service manager
[thirdparty/systemd.git] / src / core / manager.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <dirent.h>
21 #include <errno.h>
22 #include <fcntl.h>
23 #include <linux/kd.h>
24 #include <signal.h>
25 #include <string.h>
26 #include <sys/epoll.h>
27 #include <sys/inotify.h>
28 #include <sys/ioctl.h>
29 #include <sys/reboot.h>
30 #include <sys/timerfd.h>
31 #include <sys/wait.h>
32 #include <unistd.h>
33
34 #ifdef HAVE_AUDIT
35 #include <libaudit.h>
36 #endif
37
38 #include "sd-daemon.h"
39 #include "sd-messages.h"
40
41 #include "alloc-util.h"
42 #include "audit-fd.h"
43 #include "boot-timestamps.h"
44 #include "bus-common-errors.h"
45 #include "bus-error.h"
46 #include "bus-kernel.h"
47 #include "bus-util.h"
48 #include "clean-ipc.h"
49 #include "dbus-job.h"
50 #include "dbus-manager.h"
51 #include "dbus-unit.h"
52 #include "dbus.h"
53 #include "dirent-util.h"
54 #include "env-util.h"
55 #include "escape.h"
56 #include "exit-status.h"
57 #include "fd-util.h"
58 #include "fileio.h"
59 #include "fs-util.h"
60 #include "hashmap.h"
61 #include "io-util.h"
62 #include "locale-setup.h"
63 #include "log.h"
64 #include "macro.h"
65 #include "manager.h"
66 #include "missing.h"
67 #include "mkdir.h"
68 #include "parse-util.h"
69 #include "path-lookup.h"
70 #include "path-util.h"
71 #include "process-util.h"
72 #include "ratelimit.h"
73 #include "rm-rf.h"
74 #include "signal-util.h"
75 #include "special.h"
76 #include "stat-util.h"
77 #include "string-table.h"
78 #include "string-util.h"
79 #include "strv.h"
80 #include "terminal-util.h"
81 #include "time-util.h"
82 #include "transaction.h"
83 #include "umask-util.h"
84 #include "unit-name.h"
85 #include "user-util.h"
86 #include "util.h"
87 #include "virt.h"
88 #include "watchdog.h"
89
90 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
91 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
92
93 /* Initial delay and the interval for printing status messages about running jobs */
94 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
97
98 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
99 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
100 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
101 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
102 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
103 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
104 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
105 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
106 static int manager_run_generators(Manager *m);
107
108 static void manager_watch_jobs_in_progress(Manager *m) {
109 usec_t next;
110 int r;
111
112 assert(m);
113
114 if (m->jobs_in_progress_event_source)
115 return;
116
117 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
118 r = sd_event_add_time(
119 m->event,
120 &m->jobs_in_progress_event_source,
121 CLOCK_MONOTONIC,
122 next, 0,
123 manager_dispatch_jobs_in_progress, m);
124 if (r < 0)
125 return;
126
127 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
128 }
129
130 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
131
132 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
133 char *p = buffer;
134
135 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
136 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
137
138 if (pos > 1) {
139 if (pos > 2)
140 p = mempset(p, ' ', pos-2);
141 if (log_get_show_color())
142 p = stpcpy(p, ANSI_RED);
143 *p++ = '*';
144 }
145
146 if (pos > 0 && pos <= width) {
147 if (log_get_show_color())
148 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
149 *p++ = '*';
150 }
151
152 if (log_get_show_color())
153 p = stpcpy(p, ANSI_NORMAL);
154
155 if (pos < width) {
156 if (log_get_show_color())
157 p = stpcpy(p, ANSI_RED);
158 *p++ = '*';
159 if (pos < width-1)
160 p = mempset(p, ' ', width-1-pos);
161 if (log_get_show_color())
162 strcpy(p, ANSI_NORMAL);
163 }
164 }
165
166 void manager_flip_auto_status(Manager *m, bool enable) {
167 assert(m);
168
169 if (enable) {
170 if (m->show_status == SHOW_STATUS_AUTO)
171 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
172 } else {
173 if (m->show_status == SHOW_STATUS_TEMPORARY)
174 manager_set_show_status(m, SHOW_STATUS_AUTO);
175 }
176 }
177
178 static void manager_print_jobs_in_progress(Manager *m) {
179 _cleanup_free_ char *job_of_n = NULL;
180 Iterator i;
181 Job *j;
182 unsigned counter = 0, print_nr;
183 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
184 unsigned cylon_pos;
185 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
186 uint64_t x;
187
188 assert(m);
189 assert(m->n_running_jobs > 0);
190
191 manager_flip_auto_status(m, true);
192
193 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
194
195 HASHMAP_FOREACH(j, m->jobs, i)
196 if (j->state == JOB_RUNNING && counter++ == print_nr)
197 break;
198
199 /* m->n_running_jobs must be consistent with the contents of m->jobs,
200 * so the above loop must have succeeded in finding j. */
201 assert(counter == print_nr + 1);
202 assert(j);
203
204 cylon_pos = m->jobs_in_progress_iteration % 14;
205 if (cylon_pos >= 8)
206 cylon_pos = 14 - cylon_pos;
207 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
208
209 m->jobs_in_progress_iteration++;
210
211 if (m->n_running_jobs > 1) {
212 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
213 job_of_n = NULL;
214 }
215
216 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
217 if (job_get_timeout(j, &x) > 0)
218 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
219
220 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
221 "%sA %s job is running for %s (%s / %s)",
222 strempty(job_of_n),
223 job_type_to_string(j->type),
224 unit_description(j->unit),
225 time, limit);
226 }
227
228 static int have_ask_password(void) {
229 _cleanup_closedir_ DIR *dir;
230
231 dir = opendir("/run/systemd/ask-password");
232 if (!dir) {
233 if (errno == ENOENT)
234 return false;
235 else
236 return -errno;
237 }
238
239 for (;;) {
240 struct dirent *de;
241
242 errno = 0;
243 de = readdir(dir);
244 if (!de && errno > 0)
245 return -errno;
246 if (!de)
247 return false;
248
249 if (startswith(de->d_name, "ask."))
250 return true;
251 }
252 }
253
254 static int manager_dispatch_ask_password_fd(sd_event_source *source,
255 int fd, uint32_t revents, void *userdata) {
256 Manager *m = userdata;
257
258 assert(m);
259
260 flush_fd(fd);
261
262 m->have_ask_password = have_ask_password();
263 if (m->have_ask_password < 0)
264 /* Log error but continue. Negative have_ask_password
265 * is treated as unknown status. */
266 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
267
268 return 0;
269 }
270
271 static void manager_close_ask_password(Manager *m) {
272 assert(m);
273
274 m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
275 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
276 m->have_ask_password = -EINVAL;
277 }
278
279 static int manager_check_ask_password(Manager *m) {
280 int r;
281
282 assert(m);
283
284 if (!m->ask_password_event_source) {
285 assert(m->ask_password_inotify_fd < 0);
286
287 mkdir_p_label("/run/systemd/ask-password", 0755);
288
289 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
290 if (m->ask_password_inotify_fd < 0)
291 return log_error_errno(errno, "inotify_init1() failed: %m");
292
293 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
294 log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
295 manager_close_ask_password(m);
296 return -errno;
297 }
298
299 r = sd_event_add_io(m->event, &m->ask_password_event_source,
300 m->ask_password_inotify_fd, EPOLLIN,
301 manager_dispatch_ask_password_fd, m);
302 if (r < 0) {
303 log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
304 manager_close_ask_password(m);
305 return -errno;
306 }
307
308 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
309
310 /* Queries might have been added meanwhile... */
311 manager_dispatch_ask_password_fd(m->ask_password_event_source,
312 m->ask_password_inotify_fd, EPOLLIN, m);
313 }
314
315 return m->have_ask_password;
316 }
317
318 static int manager_watch_idle_pipe(Manager *m) {
319 int r;
320
321 assert(m);
322
323 if (m->idle_pipe_event_source)
324 return 0;
325
326 if (m->idle_pipe[2] < 0)
327 return 0;
328
329 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
330 if (r < 0)
331 return log_error_errno(r, "Failed to watch idle pipe: %m");
332
333 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
334
335 return 0;
336 }
337
338 static void manager_close_idle_pipe(Manager *m) {
339 assert(m);
340
341 m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
342
343 safe_close_pair(m->idle_pipe);
344 safe_close_pair(m->idle_pipe + 2);
345 }
346
347 static int manager_setup_time_change(Manager *m) {
348 int r;
349
350 /* We only care for the cancellation event, hence we set the
351 * timeout to the latest possible value. */
352 struct itimerspec its = {
353 .it_value.tv_sec = TIME_T_MAX,
354 };
355
356 assert(m);
357 assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
358
359 if (m->test_run)
360 return 0;
361
362 /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
363 * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
364
365 m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
366 if (m->time_change_fd < 0)
367 return log_error_errno(errno, "Failed to create timerfd: %m");
368
369 if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
370 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
371 m->time_change_fd = safe_close(m->time_change_fd);
372 return 0;
373 }
374
375 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
376 if (r < 0)
377 return log_error_errno(r, "Failed to create time change event source: %m");
378
379 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
380
381 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
382
383 return 0;
384 }
385
386 static int enable_special_signals(Manager *m) {
387 _cleanup_close_ int fd = -1;
388
389 assert(m);
390
391 if (m->test_run)
392 return 0;
393
394 /* Enable that we get SIGINT on control-alt-del. In containers
395 * this will fail with EPERM (older) or EINVAL (newer), so
396 * ignore that. */
397 if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
398 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
399
400 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
401 if (fd < 0) {
402 /* Support systems without virtual console */
403 if (fd != -ENOENT)
404 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
405 } else {
406 /* Enable that we get SIGWINCH on kbrequest */
407 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
408 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
409 }
410
411 return 0;
412 }
413
414 static int manager_setup_signals(Manager *m) {
415 struct sigaction sa = {
416 .sa_handler = SIG_DFL,
417 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
418 };
419 sigset_t mask;
420 int r;
421
422 assert(m);
423
424 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
425
426 /* We make liberal use of realtime signals here. On
427 * Linux/glibc we have 30 of them (with the exception of Linux
428 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
429 * (aka SIGRTMAX). */
430
431 assert_se(sigemptyset(&mask) == 0);
432 sigset_add_many(&mask,
433 SIGCHLD, /* Child died */
434 SIGTERM, /* Reexecute daemon */
435 SIGHUP, /* Reload configuration */
436 SIGUSR1, /* systemd/upstart: reconnect to D-Bus */
437 SIGUSR2, /* systemd: dump status */
438 SIGINT, /* Kernel sends us this on control-alt-del */
439 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
440 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
441
442 SIGRTMIN+0, /* systemd: start default.target */
443 SIGRTMIN+1, /* systemd: isolate rescue.target */
444 SIGRTMIN+2, /* systemd: isolate emergency.target */
445 SIGRTMIN+3, /* systemd: start halt.target */
446 SIGRTMIN+4, /* systemd: start poweroff.target */
447 SIGRTMIN+5, /* systemd: start reboot.target */
448 SIGRTMIN+6, /* systemd: start kexec.target */
449
450 /* ... space for more special targets ... */
451
452 SIGRTMIN+13, /* systemd: Immediate halt */
453 SIGRTMIN+14, /* systemd: Immediate poweroff */
454 SIGRTMIN+15, /* systemd: Immediate reboot */
455 SIGRTMIN+16, /* systemd: Immediate kexec */
456
457 /* ... space for more immediate system state changes ... */
458
459 SIGRTMIN+20, /* systemd: enable status messages */
460 SIGRTMIN+21, /* systemd: disable status messages */
461 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
462 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
463 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
464
465 /* .. one free signal here ... */
466
467 #if !defined(__hppa64__) && !defined(__hppa__)
468 /* Apparently Linux on hppa has fewer RT
469 * signals (SIGRTMAX is SIGRTMIN+25 there),
470 * hence let's not try to make use of them
471 * here. Since these commands are accessible
472 * by different means and only really a safety
473 * net, the missing functionality on hppa
474 * shouldn't matter. */
475
476 SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
477 SIGRTMIN+27, /* systemd: set log target to console */
478 SIGRTMIN+28, /* systemd: set log target to kmsg */
479 SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
480
481 /* ... one free signal here SIGRTMIN+30 ... */
482 #endif
483 -1);
484 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
485
486 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
487 if (m->signal_fd < 0)
488 return -errno;
489
490 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
491 if (r < 0)
492 return r;
493
494 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
495
496 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
497 * notify processing can still figure out to which process/service a message belongs, before we reap the
498 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
499 * status information before detecting that there's no process in a cgroup. */
500 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
501 if (r < 0)
502 return r;
503
504 if (MANAGER_IS_SYSTEM(m))
505 return enable_special_signals(m);
506
507 return 0;
508 }
509
510 static void manager_clean_environment(Manager *m) {
511 assert(m);
512
513 /* Let's remove some environment variables that we
514 * need ourselves to communicate with our clients */
515 strv_env_unset_many(
516 m->environment,
517 "NOTIFY_SOCKET",
518 "MAINPID",
519 "MANAGERPID",
520 "LISTEN_PID",
521 "LISTEN_FDS",
522 "LISTEN_FDNAMES",
523 "WATCHDOG_PID",
524 "WATCHDOG_USEC",
525 "INVOCATION_ID",
526 NULL);
527 }
528
529 static int manager_default_environment(Manager *m) {
530 assert(m);
531
532 if (MANAGER_IS_SYSTEM(m)) {
533 /* The system manager always starts with a clean
534 * environment for its children. It does not import
535 * the kernel or the parents exported variables.
536 *
537 * The initial passed environ is untouched to keep
538 * /proc/self/environ valid; it is used for tagging
539 * the init process inside containers. */
540 m->environment = strv_new("PATH=" DEFAULT_PATH,
541 NULL);
542
543 /* Import locale variables LC_*= from configuration */
544 locale_setup(&m->environment);
545 } else {
546 /* The user manager passes its own environment
547 * along to its children. */
548 m->environment = strv_copy(environ);
549 }
550
551 if (!m->environment)
552 return -ENOMEM;
553
554 manager_clean_environment(m);
555 strv_sort(m->environment);
556
557 return 0;
558 }
559
560 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
561 Manager *m;
562 int r;
563
564 assert(_m);
565 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
566
567 m = new0(Manager, 1);
568 if (!m)
569 return -ENOMEM;
570
571 m->unit_file_scope = scope;
572 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
573 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
574 m->default_tasks_accounting = true;
575 m->default_tasks_max = UINT64_MAX;
576
577 #ifdef ENABLE_EFI
578 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
579 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
580 #endif
581
582 /* Prepare log fields we can use for structured logging */
583 if (MANAGER_IS_SYSTEM(m)) {
584 m->unit_log_field = "UNIT=";
585 m->unit_log_format_string = "UNIT=%s";
586
587 m->invocation_log_field = "INVOCATION_ID=";
588 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
589 } else {
590 m->unit_log_field = "USER_UNIT=";
591 m->unit_log_format_string = "USER_UNIT=%s";
592
593 m->invocation_log_field = "USER_INVOCATION_ID=";
594 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
595 }
596
597 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
598
599 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
600 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
601 m->ask_password_inotify_fd = -1;
602
603 m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
604
605 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
606
607 m->have_ask_password = -EINVAL; /* we don't know */
608 m->first_boot = -1;
609
610 m->test_run = test_run;
611
612 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
613 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
614
615 r = manager_default_environment(m);
616 if (r < 0)
617 goto fail;
618
619 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
620 if (r < 0)
621 goto fail;
622
623 r = hashmap_ensure_allocated(&m->jobs, NULL);
624 if (r < 0)
625 goto fail;
626
627 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
628 if (r < 0)
629 goto fail;
630
631 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
632 if (r < 0)
633 goto fail;
634
635 r = sd_event_default(&m->event);
636 if (r < 0)
637 goto fail;
638
639 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
640 if (r < 0)
641 goto fail;
642
643 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
644 if (r < 0)
645 goto fail;
646
647 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
648 if (r < 0)
649 goto fail;
650
651 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
652
653 r = manager_setup_signals(m);
654 if (r < 0)
655 goto fail;
656
657 r = manager_setup_cgroup(m);
658 if (r < 0)
659 goto fail;
660
661 r = manager_setup_time_change(m);
662 if (r < 0)
663 goto fail;
664
665 m->udev = udev_new();
666 if (!m->udev) {
667 r = -ENOMEM;
668 goto fail;
669 }
670
671 /* Note that we do not set up the notify fd here. We do that after deserialization,
672 * since they might have gotten serialized across the reexec. */
673
674 m->taint_usr = dir_is_empty("/usr") > 0;
675
676 *_m = m;
677 return 0;
678
679 fail:
680 manager_free(m);
681 return r;
682 }
683
684 static int manager_setup_notify(Manager *m) {
685 int r;
686
687 if (m->test_run)
688 return 0;
689
690 if (m->notify_fd < 0) {
691 _cleanup_close_ int fd = -1;
692 union sockaddr_union sa = {
693 .sa.sa_family = AF_UNIX,
694 };
695 static const int one = 1;
696 const char *e;
697
698 /* First free all secondary fields */
699 m->notify_socket = mfree(m->notify_socket);
700 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
701
702 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
703 if (fd < 0)
704 return log_error_errno(errno, "Failed to allocate notification socket: %m");
705
706 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
707
708 e = manager_get_runtime_prefix(m);
709 if (!e) {
710 log_error("Failed to determine runtime prefix.");
711 return -EINVAL;
712 }
713
714 m->notify_socket = strappend(e, "/systemd/notify");
715 if (!m->notify_socket)
716 return log_oom();
717
718 (void) mkdir_parents_label(m->notify_socket, 0755);
719 (void) unlink(m->notify_socket);
720
721 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
722 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
723 if (r < 0)
724 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
725
726 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
727 if (r < 0)
728 return log_error_errno(errno, "SO_PASSCRED failed: %m");
729
730 m->notify_fd = fd;
731 fd = -1;
732
733 log_debug("Using notification socket %s", m->notify_socket);
734 }
735
736 if (!m->notify_event_source) {
737 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
738 if (r < 0)
739 return log_error_errno(r, "Failed to allocate notify event source: %m");
740
741 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
742 * service an exit message belongs. */
743 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
744 if (r < 0)
745 return log_error_errno(r, "Failed to set priority of notify event source: %m");
746
747 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
748 }
749
750 return 0;
751 }
752
753 static int manager_setup_cgroups_agent(Manager *m) {
754
755 static const union sockaddr_union sa = {
756 .un.sun_family = AF_UNIX,
757 .un.sun_path = "/run/systemd/cgroups-agent",
758 };
759 int r;
760
761 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
762 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
763 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
764 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
765 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
766 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
767 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
768 * we thus won't lose messages.
769 *
770 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
771 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
772 * bus for these messages. */
773
774 if (m->test_run)
775 return 0;
776
777 if (!MANAGER_IS_SYSTEM(m))
778 return 0;
779
780 if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* We don't need this anymore on the unified hierarchy */
781 return 0;
782
783 if (m->cgroups_agent_fd < 0) {
784 _cleanup_close_ int fd = -1;
785
786 /* First free all secondary fields */
787 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
788
789 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
790 if (fd < 0)
791 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
792
793 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
794
795 (void) unlink(sa.un.sun_path);
796
797 /* Only allow root to connect to this socket */
798 RUN_WITH_UMASK(0077)
799 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
800 if (r < 0)
801 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
802
803 m->cgroups_agent_fd = fd;
804 fd = -1;
805 }
806
807 if (!m->cgroups_agent_event_source) {
808 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
809 if (r < 0)
810 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
811
812 /* Process cgroups notifications early, but after having processed service notification messages or
813 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
814 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
815 * cgroup inotify for the unified cgroup stuff. */
816 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
817 if (r < 0)
818 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
819
820 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
821 }
822
823 return 0;
824 }
825
826 static int manager_setup_user_lookup_fd(Manager *m) {
827 int r;
828
829 assert(m);
830
831 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
832 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
833 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
834 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
835 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
836 * hence we establish this communication channel so that forked off processes can pass their UID/GID
837 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
838 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
839 * this purpose.
840 *
841 * You might wonder why we need a communication channel for this that is independent of the usual notification
842 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
843 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
844 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
845 *
846 * Note that this function is called under two circumstances: when we first initialize (in which case we
847 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
848 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
849
850 if (m->user_lookup_fds[0] < 0) {
851
852 /* Free all secondary fields */
853 safe_close_pair(m->user_lookup_fds);
854 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
855
856 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
857 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
858
859 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
860 }
861
862 if (!m->user_lookup_event_source) {
863 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
864 if (r < 0)
865 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
866
867 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
868 * resolutions */
869 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
870 if (r < 0)
871 return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
872
873 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
874 }
875
876 return 0;
877 }
878
879 static int manager_connect_bus(Manager *m, bool reexecuting) {
880 bool try_bus_connect;
881
882 assert(m);
883
884 if (m->test_run)
885 return 0;
886
887 try_bus_connect =
888 reexecuting ||
889 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
890
891 /* Try to connect to the buses, if possible. */
892 return bus_init(m, try_bus_connect);
893 }
894
895 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
896 Unit *u;
897 unsigned n = 0;
898
899 assert(m);
900
901 while ((u = m->cleanup_queue)) {
902 assert(u->in_cleanup_queue);
903
904 unit_free(u);
905 n++;
906 }
907
908 return n;
909 }
910
911 enum {
912 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
913 GC_OFFSET_UNSURE, /* No clue */
914 GC_OFFSET_GOOD, /* We still need this unit */
915 GC_OFFSET_BAD, /* We don't need this unit anymore */
916 _GC_OFFSET_MAX
917 };
918
919 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
920 Iterator i;
921 Unit *other;
922
923 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
924
925 /* Recursively mark referenced units as GOOD as well */
926 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
927 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
928 unit_gc_mark_good(other, gc_marker);
929 }
930
931 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
932 Iterator i;
933 Unit *other;
934 bool is_bad;
935
936 assert(u);
937
938 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
939 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
940 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
941 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
942 return;
943
944 if (u->in_cleanup_queue)
945 goto bad;
946
947 if (unit_check_gc(u))
948 goto good;
949
950 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
951
952 is_bad = true;
953
954 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
955 unit_gc_sweep(other, gc_marker);
956
957 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
958 goto good;
959
960 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
961 is_bad = false;
962 }
963
964 if (is_bad)
965 goto bad;
966
967 /* We were unable to find anything out about this entry, so
968 * let's investigate it later */
969 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
970 unit_add_to_gc_queue(u);
971 return;
972
973 bad:
974 /* We definitely know that this one is not useful anymore, so
975 * let's mark it for deletion */
976 u->gc_marker = gc_marker + GC_OFFSET_BAD;
977 unit_add_to_cleanup_queue(u);
978 return;
979
980 good:
981 unit_gc_mark_good(u, gc_marker);
982 }
983
984 static unsigned manager_dispatch_gc_queue(Manager *m) {
985 Unit *u;
986 unsigned n = 0;
987 unsigned gc_marker;
988
989 assert(m);
990
991 /* log_debug("Running GC..."); */
992
993 m->gc_marker += _GC_OFFSET_MAX;
994 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
995 m->gc_marker = 1;
996
997 gc_marker = m->gc_marker;
998
999 while ((u = m->gc_queue)) {
1000 assert(u->in_gc_queue);
1001
1002 unit_gc_sweep(u, gc_marker);
1003
1004 LIST_REMOVE(gc_queue, m->gc_queue, u);
1005 u->in_gc_queue = false;
1006
1007 n++;
1008
1009 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1010 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1011 if (u->id)
1012 log_unit_debug(u, "Collecting.");
1013 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1014 unit_add_to_cleanup_queue(u);
1015 }
1016 }
1017
1018 m->n_in_gc_queue = 0;
1019
1020 return n;
1021 }
1022
1023 static void manager_clear_jobs_and_units(Manager *m) {
1024 Unit *u;
1025
1026 assert(m);
1027
1028 while ((u = hashmap_first(m->units)))
1029 unit_free(u);
1030
1031 manager_dispatch_cleanup_queue(m);
1032
1033 assert(!m->load_queue);
1034 assert(!m->run_queue);
1035 assert(!m->dbus_unit_queue);
1036 assert(!m->dbus_job_queue);
1037 assert(!m->cleanup_queue);
1038 assert(!m->gc_queue);
1039
1040 assert(hashmap_isempty(m->jobs));
1041 assert(hashmap_isempty(m->units));
1042
1043 m->n_on_console = 0;
1044 m->n_running_jobs = 0;
1045 }
1046
1047 Manager* manager_free(Manager *m) {
1048 UnitType c;
1049 int i;
1050
1051 if (!m)
1052 return NULL;
1053
1054 manager_clear_jobs_and_units(m);
1055
1056 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1057 if (unit_vtable[c]->shutdown)
1058 unit_vtable[c]->shutdown(m);
1059
1060 /* If we reexecute ourselves, we keep the root cgroup
1061 * around */
1062 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1063
1064 lookup_paths_flush_generator(&m->lookup_paths);
1065
1066 bus_done(m);
1067
1068 dynamic_user_vacuum(m, false);
1069 hashmap_free(m->dynamic_users);
1070
1071 hashmap_free(m->units);
1072 hashmap_free(m->units_by_invocation_id);
1073 hashmap_free(m->jobs);
1074 hashmap_free(m->watch_pids1);
1075 hashmap_free(m->watch_pids2);
1076 hashmap_free(m->watch_bus);
1077
1078 set_free(m->startup_units);
1079 set_free(m->failed_units);
1080
1081 sd_event_source_unref(m->signal_event_source);
1082 sd_event_source_unref(m->notify_event_source);
1083 sd_event_source_unref(m->cgroups_agent_event_source);
1084 sd_event_source_unref(m->time_change_event_source);
1085 sd_event_source_unref(m->jobs_in_progress_event_source);
1086 sd_event_source_unref(m->run_queue_event_source);
1087 sd_event_source_unref(m->user_lookup_event_source);
1088
1089 safe_close(m->signal_fd);
1090 safe_close(m->notify_fd);
1091 safe_close(m->cgroups_agent_fd);
1092 safe_close(m->time_change_fd);
1093 safe_close_pair(m->user_lookup_fds);
1094
1095 manager_close_ask_password(m);
1096
1097 manager_close_idle_pipe(m);
1098
1099 udev_unref(m->udev);
1100 sd_event_unref(m->event);
1101
1102 free(m->notify_socket);
1103
1104 lookup_paths_free(&m->lookup_paths);
1105 strv_free(m->environment);
1106
1107 hashmap_free(m->cgroup_unit);
1108 set_free_free(m->unit_path_cache);
1109
1110 free(m->switch_root);
1111 free(m->switch_root_init);
1112
1113 for (i = 0; i < _RLIMIT_MAX; i++)
1114 m->rlimit[i] = mfree(m->rlimit[i]);
1115
1116 assert(hashmap_isempty(m->units_requiring_mounts_for));
1117 hashmap_free(m->units_requiring_mounts_for);
1118
1119 hashmap_free(m->uid_refs);
1120 hashmap_free(m->gid_refs);
1121
1122 free(m);
1123 return NULL;
1124 }
1125
1126 void manager_enumerate(Manager *m) {
1127 UnitType c;
1128
1129 assert(m);
1130
1131 /* Let's ask every type to load all units from disk/kernel
1132 * that it might know */
1133 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1134 if (!unit_type_supported(c)) {
1135 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1136 continue;
1137 }
1138
1139 if (!unit_vtable[c]->enumerate)
1140 continue;
1141
1142 unit_vtable[c]->enumerate(m);
1143 }
1144
1145 manager_dispatch_load_queue(m);
1146 }
1147
1148 static void manager_coldplug(Manager *m) {
1149 Iterator i;
1150 Unit *u;
1151 char *k;
1152 int r;
1153
1154 assert(m);
1155
1156 /* Then, let's set up their initial state. */
1157 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1158
1159 /* ignore aliases */
1160 if (u->id != k)
1161 continue;
1162
1163 r = unit_coldplug(u);
1164 if (r < 0)
1165 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1166 }
1167 }
1168
1169 static void manager_build_unit_path_cache(Manager *m) {
1170 char **i;
1171 int r;
1172
1173 assert(m);
1174
1175 set_free_free(m->unit_path_cache);
1176
1177 m->unit_path_cache = set_new(&string_hash_ops);
1178 if (!m->unit_path_cache) {
1179 r = -ENOMEM;
1180 goto fail;
1181 }
1182
1183 /* This simply builds a list of files we know exist, so that
1184 * we don't always have to go to disk */
1185
1186 STRV_FOREACH(i, m->lookup_paths.search_path) {
1187 _cleanup_closedir_ DIR *d = NULL;
1188 struct dirent *de;
1189
1190 d = opendir(*i);
1191 if (!d) {
1192 if (errno != ENOENT)
1193 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1194 continue;
1195 }
1196
1197 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1198 char *p;
1199
1200 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name, NULL);
1201 if (!p) {
1202 r = -ENOMEM;
1203 goto fail;
1204 }
1205
1206 r = set_consume(m->unit_path_cache, p);
1207 if (r < 0)
1208 goto fail;
1209 }
1210 }
1211
1212 return;
1213
1214 fail:
1215 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1216 m->unit_path_cache = set_free_free(m->unit_path_cache);
1217 }
1218
1219 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1220 Iterator i;
1221 Unit *u;
1222
1223 assert(m);
1224
1225 HASHMAP_FOREACH(u, m->units, i) {
1226
1227 if (fdset_size(fds) <= 0)
1228 break;
1229
1230 if (!UNIT_VTABLE(u)->distribute_fds)
1231 continue;
1232
1233 UNIT_VTABLE(u)->distribute_fds(u, fds);
1234 }
1235 }
1236
1237 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1238 int r, q;
1239
1240 assert(m);
1241
1242 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1243 if (r < 0)
1244 return r;
1245
1246 /* Make sure the transient directory always exists, so that it remains in the search path */
1247 if (!m->test_run) {
1248 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1249 if (r < 0)
1250 return r;
1251 }
1252
1253 dual_timestamp_get(&m->generators_start_timestamp);
1254 r = manager_run_generators(m);
1255 dual_timestamp_get(&m->generators_finish_timestamp);
1256 if (r < 0)
1257 return r;
1258
1259 lookup_paths_reduce(&m->lookup_paths);
1260 manager_build_unit_path_cache(m);
1261
1262 /* If we will deserialize make sure that during enumeration
1263 * this is already known, so we increase the counter here
1264 * already */
1265 if (serialization)
1266 m->n_reloading++;
1267
1268 /* First, enumerate what we can from all config files */
1269 dual_timestamp_get(&m->units_load_start_timestamp);
1270 manager_enumerate(m);
1271 dual_timestamp_get(&m->units_load_finish_timestamp);
1272
1273 /* Second, deserialize if there is something to deserialize */
1274 if (serialization)
1275 r = manager_deserialize(m, serialization, fds);
1276
1277 /* Any fds left? Find some unit which wants them. This is
1278 * useful to allow container managers to pass some file
1279 * descriptors to us pre-initialized. This enables
1280 * socket-based activation of entire containers. */
1281 manager_distribute_fds(m, fds);
1282
1283 /* We might have deserialized the notify fd, but if we didn't
1284 * then let's create the bus now */
1285 q = manager_setup_notify(m);
1286 if (q < 0 && r == 0)
1287 r = q;
1288
1289 q = manager_setup_cgroups_agent(m);
1290 if (q < 0 && r == 0)
1291 r = q;
1292
1293 q = manager_setup_user_lookup_fd(m);
1294 if (q < 0 && r == 0)
1295 r = q;
1296
1297 /* Let's connect to the bus now. */
1298 (void) manager_connect_bus(m, !!serialization);
1299
1300 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1301 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1302
1303 /* Third, fire things up! */
1304 manager_coldplug(m);
1305
1306 /* Release any dynamic users no longer referenced */
1307 dynamic_user_vacuum(m, true);
1308
1309 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1310 manager_vacuum_uid_refs(m);
1311 manager_vacuum_gid_refs(m);
1312
1313 if (serialization) {
1314 assert(m->n_reloading > 0);
1315 m->n_reloading--;
1316
1317 /* Let's wait for the UnitNew/JobNew messages being
1318 * sent, before we notify that the reload is
1319 * finished */
1320 m->send_reloading_done = true;
1321 }
1322
1323 return r;
1324 }
1325
1326 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1327 int r;
1328 Transaction *tr;
1329
1330 assert(m);
1331 assert(type < _JOB_TYPE_MAX);
1332 assert(unit);
1333 assert(mode < _JOB_MODE_MAX);
1334
1335 if (mode == JOB_ISOLATE && type != JOB_START)
1336 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1337
1338 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1339 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1340
1341 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1342
1343 type = job_type_collapse(type, unit);
1344
1345 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1346 if (!tr)
1347 return -ENOMEM;
1348
1349 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1350 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1351 mode == JOB_IGNORE_DEPENDENCIES, e);
1352 if (r < 0)
1353 goto tr_abort;
1354
1355 if (mode == JOB_ISOLATE) {
1356 r = transaction_add_isolate_jobs(tr, m);
1357 if (r < 0)
1358 goto tr_abort;
1359 }
1360
1361 r = transaction_activate(tr, m, mode, e);
1362 if (r < 0)
1363 goto tr_abort;
1364
1365 log_unit_debug(unit,
1366 "Enqueued job %s/%s as %u", unit->id,
1367 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1368
1369 if (_ret)
1370 *_ret = tr->anchor_job;
1371
1372 transaction_free(tr);
1373 return 0;
1374
1375 tr_abort:
1376 transaction_abort(tr);
1377 transaction_free(tr);
1378 return r;
1379 }
1380
1381 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1382 Unit *unit;
1383 int r;
1384
1385 assert(m);
1386 assert(type < _JOB_TYPE_MAX);
1387 assert(name);
1388 assert(mode < _JOB_MODE_MAX);
1389
1390 r = manager_load_unit(m, name, NULL, NULL, &unit);
1391 if (r < 0)
1392 return r;
1393
1394 return manager_add_job(m, type, unit, mode, e, ret);
1395 }
1396
1397 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1398 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1399 int r;
1400
1401 assert(m);
1402 assert(type < _JOB_TYPE_MAX);
1403 assert(name);
1404 assert(mode < _JOB_MODE_MAX);
1405
1406 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1407 if (r < 0)
1408 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1409
1410 return r;
1411 }
1412
1413 Job *manager_get_job(Manager *m, uint32_t id) {
1414 assert(m);
1415
1416 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1417 }
1418
1419 Unit *manager_get_unit(Manager *m, const char *name) {
1420 assert(m);
1421 assert(name);
1422
1423 return hashmap_get(m->units, name);
1424 }
1425
1426 unsigned manager_dispatch_load_queue(Manager *m) {
1427 Unit *u;
1428 unsigned n = 0;
1429
1430 assert(m);
1431
1432 /* Make sure we are not run recursively */
1433 if (m->dispatching_load_queue)
1434 return 0;
1435
1436 m->dispatching_load_queue = true;
1437
1438 /* Dispatches the load queue. Takes a unit from the queue and
1439 * tries to load its data until the queue is empty */
1440
1441 while ((u = m->load_queue)) {
1442 assert(u->in_load_queue);
1443
1444 unit_load(u);
1445 n++;
1446 }
1447
1448 m->dispatching_load_queue = false;
1449 return n;
1450 }
1451
1452 int manager_load_unit_prepare(
1453 Manager *m,
1454 const char *name,
1455 const char *path,
1456 sd_bus_error *e,
1457 Unit **_ret) {
1458
1459 Unit *ret;
1460 UnitType t;
1461 int r;
1462
1463 assert(m);
1464 assert(name || path);
1465
1466 /* This will prepare the unit for loading, but not actually
1467 * load anything from disk. */
1468
1469 if (path && !is_path(path))
1470 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1471
1472 if (!name)
1473 name = basename(path);
1474
1475 t = unit_name_to_type(name);
1476
1477 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1478 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1479 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1480
1481 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1482 }
1483
1484 ret = manager_get_unit(m, name);
1485 if (ret) {
1486 *_ret = ret;
1487 return 1;
1488 }
1489
1490 ret = unit_new(m, unit_vtable[t]->object_size);
1491 if (!ret)
1492 return -ENOMEM;
1493
1494 if (path) {
1495 ret->fragment_path = strdup(path);
1496 if (!ret->fragment_path) {
1497 unit_free(ret);
1498 return -ENOMEM;
1499 }
1500 }
1501
1502 r = unit_add_name(ret, name);
1503 if (r < 0) {
1504 unit_free(ret);
1505 return r;
1506 }
1507
1508 unit_add_to_load_queue(ret);
1509 unit_add_to_dbus_queue(ret);
1510 unit_add_to_gc_queue(ret);
1511
1512 if (_ret)
1513 *_ret = ret;
1514
1515 return 0;
1516 }
1517
1518 int manager_load_unit(
1519 Manager *m,
1520 const char *name,
1521 const char *path,
1522 sd_bus_error *e,
1523 Unit **_ret) {
1524
1525 int r;
1526
1527 assert(m);
1528
1529 /* This will load the service information files, but not actually
1530 * start any services or anything. */
1531
1532 r = manager_load_unit_prepare(m, name, path, e, _ret);
1533 if (r != 0)
1534 return r;
1535
1536 manager_dispatch_load_queue(m);
1537
1538 if (_ret)
1539 *_ret = unit_follow_merge(*_ret);
1540
1541 return 0;
1542 }
1543
1544 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1545 Iterator i;
1546 Job *j;
1547
1548 assert(s);
1549 assert(f);
1550
1551 HASHMAP_FOREACH(j, s->jobs, i)
1552 job_dump(j, f, prefix);
1553 }
1554
1555 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1556 Iterator i;
1557 Unit *u;
1558 const char *t;
1559
1560 assert(s);
1561 assert(f);
1562
1563 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1564 if (u->id == t)
1565 unit_dump(u, f, prefix);
1566 }
1567
1568 void manager_clear_jobs(Manager *m) {
1569 Job *j;
1570
1571 assert(m);
1572
1573 while ((j = hashmap_first(m->jobs)))
1574 /* No need to recurse. We're cancelling all jobs. */
1575 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1576 }
1577
1578 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1579 Manager *m = userdata;
1580 Job *j;
1581
1582 assert(source);
1583 assert(m);
1584
1585 while ((j = m->run_queue)) {
1586 assert(j->installed);
1587 assert(j->in_run_queue);
1588
1589 job_run_and_invalidate(j);
1590 }
1591
1592 if (m->n_running_jobs > 0)
1593 manager_watch_jobs_in_progress(m);
1594
1595 if (m->n_on_console > 0)
1596 manager_watch_idle_pipe(m);
1597
1598 return 1;
1599 }
1600
1601 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1602 Job *j;
1603 Unit *u;
1604 unsigned n = 0;
1605
1606 assert(m);
1607
1608 if (m->dispatching_dbus_queue)
1609 return 0;
1610
1611 m->dispatching_dbus_queue = true;
1612
1613 while ((u = m->dbus_unit_queue)) {
1614 assert(u->in_dbus_queue);
1615
1616 bus_unit_send_change_signal(u);
1617 n++;
1618 }
1619
1620 while ((j = m->dbus_job_queue)) {
1621 assert(j->in_dbus_queue);
1622
1623 bus_job_send_change_signal(j);
1624 n++;
1625 }
1626
1627 m->dispatching_dbus_queue = false;
1628
1629 if (m->send_reloading_done) {
1630 m->send_reloading_done = false;
1631
1632 bus_manager_send_reloading(m, false);
1633 }
1634
1635 if (m->queued_message)
1636 bus_send_queued_message(m);
1637
1638 return n;
1639 }
1640
1641 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1642 Manager *m = userdata;
1643 char buf[PATH_MAX+1];
1644 ssize_t n;
1645
1646 n = recv(fd, buf, sizeof(buf), 0);
1647 if (n < 0)
1648 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1649 if (n == 0) {
1650 log_error("Got zero-length cgroups agent message, ignoring.");
1651 return 0;
1652 }
1653 if ((size_t) n >= sizeof(buf)) {
1654 log_error("Got overly long cgroups agent message, ignoring.");
1655 return 0;
1656 }
1657
1658 if (memchr(buf, 0, n)) {
1659 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1660 return 0;
1661 }
1662 buf[n] = 0;
1663
1664 manager_notify_cgroup_empty(m, buf);
1665 bus_forward_agent_released(m, buf);
1666
1667 return 0;
1668 }
1669
1670 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1671 _cleanup_strv_free_ char **tags = NULL;
1672
1673 assert(m);
1674 assert(u);
1675 assert(buf);
1676
1677 tags = strv_split(buf, "\n\r");
1678 if (!tags) {
1679 log_oom();
1680 return;
1681 }
1682
1683 if (UNIT_VTABLE(u)->notify_message)
1684 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1685 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1686 _cleanup_free_ char *x = NULL, *y = NULL;
1687
1688 x = cescape(buf);
1689 if (x)
1690 y = ellipsize(x, 20, 90);
1691 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1692 }
1693 }
1694
1695 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1696
1697 _cleanup_fdset_free_ FDSet *fds = NULL;
1698 Manager *m = userdata;
1699 char buf[NOTIFY_BUFFER_MAX+1];
1700 struct iovec iovec = {
1701 .iov_base = buf,
1702 .iov_len = sizeof(buf)-1,
1703 };
1704 union {
1705 struct cmsghdr cmsghdr;
1706 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1707 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1708 } control = {};
1709 struct msghdr msghdr = {
1710 .msg_iov = &iovec,
1711 .msg_iovlen = 1,
1712 .msg_control = &control,
1713 .msg_controllen = sizeof(control),
1714 };
1715
1716 struct cmsghdr *cmsg;
1717 struct ucred *ucred = NULL;
1718 Unit *u1, *u2, *u3;
1719 int r, *fd_array = NULL;
1720 unsigned n_fds = 0;
1721 ssize_t n;
1722
1723 assert(m);
1724 assert(m->notify_fd == fd);
1725
1726 if (revents != EPOLLIN) {
1727 log_warning("Got unexpected poll event for notify fd.");
1728 return 0;
1729 }
1730
1731 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1732 if (n < 0) {
1733 if (IN_SET(errno, EAGAIN, EINTR))
1734 return 0; /* Spurious wakeup, try again */
1735
1736 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1737 * won't take notification messages anymore, but that's still better than busy looping around this:
1738 * being woken up over and over again but being unable to actually read the message off the socket. */
1739 return log_error_errno(errno, "Failed to receive notification message: %m");
1740 }
1741
1742 CMSG_FOREACH(cmsg, &msghdr) {
1743 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1744
1745 fd_array = (int*) CMSG_DATA(cmsg);
1746 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1747
1748 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1749 cmsg->cmsg_type == SCM_CREDENTIALS &&
1750 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1751
1752 ucred = (struct ucred*) CMSG_DATA(cmsg);
1753 }
1754 }
1755
1756 if (n_fds > 0) {
1757 assert(fd_array);
1758
1759 r = fdset_new_array(&fds, fd_array, n_fds);
1760 if (r < 0) {
1761 close_many(fd_array, n_fds);
1762 log_oom();
1763 return 0;
1764 }
1765 }
1766
1767 if (!ucred || ucred->pid <= 0) {
1768 log_warning("Received notify message without valid credentials. Ignoring.");
1769 return 0;
1770 }
1771
1772 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1773 log_warning("Received notify message exceeded maximum size. Ignoring.");
1774 return 0;
1775 }
1776
1777 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1778 * trailing NUL byte in the message, but don't expect it. */
1779 if (n > 1 && memchr(buf, 0, n-1)) {
1780 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1781 return 0;
1782 }
1783
1784 /* Make sure it's NUL-terminated. */
1785 buf[n] = 0;
1786
1787 /* Notify every unit that might be interested, but try
1788 * to avoid notifying the same one multiple times. */
1789 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1790 if (u1)
1791 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1792
1793 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1794 if (u2 && u2 != u1)
1795 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1796
1797 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1798 if (u3 && u3 != u2 && u3 != u1)
1799 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1800
1801 if (!u1 && !u2 && !u3)
1802 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1803
1804 if (fdset_size(fds) > 0)
1805 log_warning("Got extra auxiliary fds with notification message, closing them.");
1806
1807 return 0;
1808 }
1809
1810 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1811 uint64_t iteration;
1812
1813 assert(m);
1814 assert(u);
1815 assert(si);
1816
1817 sd_event_get_iteration(m->event, &iteration);
1818
1819 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1820
1821 unit_unwatch_pid(u, si->si_pid);
1822
1823 if (UNIT_VTABLE(u)->sigchld_event) {
1824 if (set_size(u->pids) <= 1 ||
1825 iteration != u->sigchldgen ||
1826 unit_main_pid(u) == si->si_pid ||
1827 unit_control_pid(u) == si->si_pid) {
1828 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1829 u->sigchldgen = iteration;
1830 } else
1831 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1832 }
1833 }
1834
1835 static int manager_dispatch_sigchld(Manager *m) {
1836 assert(m);
1837
1838 for (;;) {
1839 siginfo_t si = {};
1840
1841 /* First we call waitd() for a PID and do not reap the
1842 * zombie. That way we can still access /proc/$PID for
1843 * it while it is a zombie. */
1844 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1845
1846 if (errno == ECHILD)
1847 break;
1848
1849 if (errno == EINTR)
1850 continue;
1851
1852 return -errno;
1853 }
1854
1855 if (si.si_pid <= 0)
1856 break;
1857
1858 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1859 _cleanup_free_ char *name = NULL;
1860 Unit *u1, *u2, *u3;
1861
1862 get_process_comm(si.si_pid, &name);
1863
1864 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1865 si.si_pid, strna(name),
1866 sigchld_code_to_string(si.si_code),
1867 si.si_status,
1868 strna(si.si_code == CLD_EXITED
1869 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1870 : signal_to_string(si.si_status)));
1871
1872 /* And now figure out the unit this belongs
1873 * to, it might be multiple... */
1874 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1875 if (u1)
1876 invoke_sigchld_event(m, u1, &si);
1877 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1878 if (u2 && u2 != u1)
1879 invoke_sigchld_event(m, u2, &si);
1880 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1881 if (u3 && u3 != u2 && u3 != u1)
1882 invoke_sigchld_event(m, u3, &si);
1883 }
1884
1885 /* And now, we actually reap the zombie. */
1886 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1887 if (errno == EINTR)
1888 continue;
1889
1890 return -errno;
1891 }
1892 }
1893
1894 return 0;
1895 }
1896
1897 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1898 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1899 int r;
1900
1901 log_debug("Activating special unit %s", name);
1902
1903 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1904 if (r < 0)
1905 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1906
1907 return r;
1908 }
1909
1910 static void manager_handle_ctrl_alt_del(Manager *m) {
1911 /* If the user presses C-A-D more than
1912 * 7 times within 2s, we reboot/shutdown immediately,
1913 * unless it was disabled in system.conf */
1914
1915 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == CAD_BURST_ACTION_IGNORE)
1916 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1917 else {
1918 switch (m->cad_burst_action) {
1919
1920 case CAD_BURST_ACTION_REBOOT:
1921 m->exit_code = MANAGER_REBOOT;
1922 break;
1923
1924 case CAD_BURST_ACTION_POWEROFF:
1925 m->exit_code = MANAGER_POWEROFF;
1926 break;
1927
1928 default:
1929 assert_not_reached("Unknown action.");
1930 }
1931
1932 log_notice("Ctrl-Alt-Del was pressed more than 7 times within 2s, performing immediate %s.",
1933 cad_burst_action_to_string(m->cad_burst_action));
1934 status_printf(NULL, true, false, "Ctrl-Alt-Del was pressed more than 7 times within 2s, performing immediate %s.",
1935 cad_burst_action_to_string(m->cad_burst_action));
1936 }
1937 }
1938
1939 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1940 Manager *m = userdata;
1941 ssize_t n;
1942 struct signalfd_siginfo sfsi;
1943 bool sigchld = false;
1944 int r;
1945
1946 assert(m);
1947 assert(m->signal_fd == fd);
1948
1949 if (revents != EPOLLIN) {
1950 log_warning("Got unexpected events from signal file descriptor.");
1951 return 0;
1952 }
1953
1954 for (;;) {
1955 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1956 if (n != sizeof(sfsi)) {
1957 if (n >= 0) {
1958 log_warning("Truncated read from signal fd (%zu bytes)!", n);
1959 return 0;
1960 }
1961
1962 if (IN_SET(errno, EINTR, EAGAIN))
1963 break;
1964
1965 /* We return an error here, which will kill this handler,
1966 * to avoid a busy loop on read error. */
1967 return log_error_errno(errno, "Reading from signal fd failed: %m");
1968 }
1969
1970 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1971 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1972 ? LOG_DEBUG : LOG_INFO,
1973 &sfsi);
1974
1975 switch (sfsi.ssi_signo) {
1976
1977 case SIGCHLD:
1978 sigchld = true;
1979 break;
1980
1981 case SIGTERM:
1982 if (MANAGER_IS_SYSTEM(m)) {
1983 /* This is for compatibility with the
1984 * original sysvinit */
1985 m->exit_code = MANAGER_REEXECUTE;
1986 break;
1987 }
1988
1989 /* Fall through */
1990
1991 case SIGINT:
1992 if (MANAGER_IS_SYSTEM(m)) {
1993 manager_handle_ctrl_alt_del(m);
1994 break;
1995 }
1996
1997 /* Run the exit target if there is one, if not, just exit. */
1998 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
1999 m->exit_code = MANAGER_EXIT;
2000 return 0;
2001 }
2002
2003 break;
2004
2005 case SIGWINCH:
2006 if (MANAGER_IS_SYSTEM(m))
2007 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2008
2009 /* This is a nop on non-init */
2010 break;
2011
2012 case SIGPWR:
2013 if (MANAGER_IS_SYSTEM(m))
2014 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2015
2016 /* This is a nop on non-init */
2017 break;
2018
2019 case SIGUSR1: {
2020 Unit *u;
2021
2022 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2023
2024 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2025 log_info("Trying to reconnect to bus...");
2026 bus_init(m, true);
2027 }
2028
2029 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2030 log_info("Loading D-Bus service...");
2031 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2032 }
2033
2034 break;
2035 }
2036
2037 case SIGUSR2: {
2038 _cleanup_free_ char *dump = NULL;
2039 _cleanup_fclose_ FILE *f = NULL;
2040 size_t size;
2041
2042 f = open_memstream(&dump, &size);
2043 if (!f) {
2044 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2045 break;
2046 }
2047
2048 manager_dump_units(m, f, "\t");
2049 manager_dump_jobs(m, f, "\t");
2050
2051 r = fflush_and_check(f);
2052 if (r < 0) {
2053 log_warning_errno(r, "Failed to write status stream: %m");
2054 break;
2055 }
2056
2057 log_dump(LOG_INFO, dump);
2058 break;
2059 }
2060
2061 case SIGHUP:
2062 m->exit_code = MANAGER_RELOAD;
2063 break;
2064
2065 default: {
2066
2067 /* Starting SIGRTMIN+0 */
2068 static const char * const target_table[] = {
2069 [0] = SPECIAL_DEFAULT_TARGET,
2070 [1] = SPECIAL_RESCUE_TARGET,
2071 [2] = SPECIAL_EMERGENCY_TARGET,
2072 [3] = SPECIAL_HALT_TARGET,
2073 [4] = SPECIAL_POWEROFF_TARGET,
2074 [5] = SPECIAL_REBOOT_TARGET,
2075 [6] = SPECIAL_KEXEC_TARGET
2076 };
2077
2078 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2079 static const ManagerExitCode code_table[] = {
2080 [0] = MANAGER_HALT,
2081 [1] = MANAGER_POWEROFF,
2082 [2] = MANAGER_REBOOT,
2083 [3] = MANAGER_KEXEC
2084 };
2085
2086 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2087 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2088 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2089 manager_start_target(m, target_table[idx],
2090 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2091 break;
2092 }
2093
2094 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2095 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2096 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2097 break;
2098 }
2099
2100 switch (sfsi.ssi_signo - SIGRTMIN) {
2101
2102 case 20:
2103 manager_set_show_status(m, SHOW_STATUS_YES);
2104 break;
2105
2106 case 21:
2107 manager_set_show_status(m, SHOW_STATUS_NO);
2108 break;
2109
2110 case 22:
2111 log_set_max_level(LOG_DEBUG);
2112 log_info("Setting log level to debug.");
2113 break;
2114
2115 case 23:
2116 log_set_max_level(LOG_INFO);
2117 log_info("Setting log level to info.");
2118 break;
2119
2120 case 24:
2121 if (MANAGER_IS_USER(m)) {
2122 m->exit_code = MANAGER_EXIT;
2123 return 0;
2124 }
2125
2126 /* This is a nop on init */
2127 break;
2128
2129 case 26:
2130 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2131 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2132 log_notice("Setting log target to journal-or-kmsg.");
2133 break;
2134
2135 case 27:
2136 log_set_target(LOG_TARGET_CONSOLE);
2137 log_notice("Setting log target to console.");
2138 break;
2139
2140 case 28:
2141 log_set_target(LOG_TARGET_KMSG);
2142 log_notice("Setting log target to kmsg.");
2143 break;
2144
2145 default:
2146 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2147 }
2148 }
2149 }
2150 }
2151
2152 if (sigchld)
2153 manager_dispatch_sigchld(m);
2154
2155 return 0;
2156 }
2157
2158 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2159 Manager *m = userdata;
2160 Iterator i;
2161 Unit *u;
2162
2163 assert(m);
2164 assert(m->time_change_fd == fd);
2165
2166 log_struct(LOG_INFO,
2167 LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
2168 LOG_MESSAGE("Time has been changed"),
2169 NULL);
2170
2171 /* Restart the watch */
2172 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2173 m->time_change_fd = safe_close(m->time_change_fd);
2174
2175 manager_setup_time_change(m);
2176
2177 HASHMAP_FOREACH(u, m->units, i)
2178 if (UNIT_VTABLE(u)->time_change)
2179 UNIT_VTABLE(u)->time_change(u);
2180
2181 return 0;
2182 }
2183
2184 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2185 Manager *m = userdata;
2186
2187 assert(m);
2188 assert(m->idle_pipe[2] == fd);
2189
2190 m->no_console_output = m->n_on_console > 0;
2191
2192 manager_close_idle_pipe(m);
2193
2194 return 0;
2195 }
2196
2197 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2198 Manager *m = userdata;
2199 int r;
2200 uint64_t next;
2201
2202 assert(m);
2203 assert(source);
2204
2205 manager_print_jobs_in_progress(m);
2206
2207 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2208 r = sd_event_source_set_time(source, next);
2209 if (r < 0)
2210 return r;
2211
2212 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2213 }
2214
2215 int manager_loop(Manager *m) {
2216 int r;
2217
2218 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2219
2220 assert(m);
2221 m->exit_code = MANAGER_OK;
2222
2223 /* Release the path cache */
2224 m->unit_path_cache = set_free_free(m->unit_path_cache);
2225
2226 manager_check_finished(m);
2227
2228 /* There might still be some zombies hanging around from
2229 * before we were exec()'ed. Let's reap them. */
2230 r = manager_dispatch_sigchld(m);
2231 if (r < 0)
2232 return r;
2233
2234 while (m->exit_code == MANAGER_OK) {
2235 usec_t wait_usec;
2236
2237 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2238 watchdog_ping();
2239
2240 if (!ratelimit_test(&rl)) {
2241 /* Yay, something is going seriously wrong, pause a little */
2242 log_warning("Looping too fast. Throttling execution a little.");
2243 sleep(1);
2244 }
2245
2246 if (manager_dispatch_load_queue(m) > 0)
2247 continue;
2248
2249 if (manager_dispatch_gc_queue(m) > 0)
2250 continue;
2251
2252 if (manager_dispatch_cleanup_queue(m) > 0)
2253 continue;
2254
2255 if (manager_dispatch_cgroup_queue(m) > 0)
2256 continue;
2257
2258 if (manager_dispatch_dbus_queue(m) > 0)
2259 continue;
2260
2261 /* Sleep for half the watchdog time */
2262 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2263 wait_usec = m->runtime_watchdog / 2;
2264 if (wait_usec <= 0)
2265 wait_usec = 1;
2266 } else
2267 wait_usec = USEC_INFINITY;
2268
2269 r = sd_event_run(m->event, wait_usec);
2270 if (r < 0)
2271 return log_error_errno(r, "Failed to run event loop: %m");
2272 }
2273
2274 return m->exit_code;
2275 }
2276
2277 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2278 _cleanup_free_ char *n = NULL;
2279 sd_id128_t invocation_id;
2280 Unit *u;
2281 int r;
2282
2283 assert(m);
2284 assert(s);
2285 assert(_u);
2286
2287 r = unit_name_from_dbus_path(s, &n);
2288 if (r < 0)
2289 return r;
2290
2291 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2292 * as invocation ID. */
2293 r = sd_id128_from_string(n, &invocation_id);
2294 if (r >= 0) {
2295 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2296 if (u) {
2297 *_u = u;
2298 return 0;
2299 }
2300
2301 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));
2302 }
2303
2304 /* If this didn't work, we use the suffix as unit name. */
2305 r = manager_load_unit(m, n, NULL, e, &u);
2306 if (r < 0)
2307 return r;
2308
2309 *_u = u;
2310 return 0;
2311 }
2312
2313 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2314 const char *p;
2315 unsigned id;
2316 Job *j;
2317 int r;
2318
2319 assert(m);
2320 assert(s);
2321 assert(_j);
2322
2323 p = startswith(s, "/org/freedesktop/systemd1/job/");
2324 if (!p)
2325 return -EINVAL;
2326
2327 r = safe_atou(p, &id);
2328 if (r < 0)
2329 return r;
2330
2331 j = manager_get_job(m, id);
2332 if (!j)
2333 return -ENOENT;
2334
2335 *_j = j;
2336
2337 return 0;
2338 }
2339
2340 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2341
2342 #ifdef HAVE_AUDIT
2343 _cleanup_free_ char *p = NULL;
2344 const char *msg;
2345 int audit_fd, r;
2346
2347 if (!MANAGER_IS_SYSTEM(m))
2348 return;
2349
2350 audit_fd = get_audit_fd();
2351 if (audit_fd < 0)
2352 return;
2353
2354 /* Don't generate audit events if the service was already
2355 * started and we're just deserializing */
2356 if (MANAGER_IS_RELOADING(m))
2357 return;
2358
2359 if (u->type != UNIT_SERVICE)
2360 return;
2361
2362 r = unit_name_to_prefix_and_instance(u->id, &p);
2363 if (r < 0) {
2364 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2365 return;
2366 }
2367
2368 msg = strjoina("unit=", p);
2369 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2370 if (errno == EPERM)
2371 /* We aren't allowed to send audit messages?
2372 * Then let's not retry again. */
2373 close_audit_fd();
2374 else
2375 log_warning_errno(errno, "Failed to send audit message: %m");
2376 }
2377 #endif
2378
2379 }
2380
2381 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2382 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2383 _cleanup_free_ char *message = NULL;
2384 _cleanup_close_ int fd = -1;
2385 int n = 0;
2386
2387 /* Don't generate plymouth events if the service was already
2388 * started and we're just deserializing */
2389 if (MANAGER_IS_RELOADING(m))
2390 return;
2391
2392 if (!MANAGER_IS_SYSTEM(m))
2393 return;
2394
2395 if (detect_container() > 0)
2396 return;
2397
2398 if (u->type != UNIT_SERVICE &&
2399 u->type != UNIT_MOUNT &&
2400 u->type != UNIT_SWAP)
2401 return;
2402
2403 /* We set SOCK_NONBLOCK here so that we rather drop the
2404 * message then wait for plymouth */
2405 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2406 if (fd < 0) {
2407 log_error_errno(errno, "socket() failed: %m");
2408 return;
2409 }
2410
2411 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2412
2413 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2414 log_error_errno(errno, "connect() failed: %m");
2415 return;
2416 }
2417
2418 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2419 log_oom();
2420 return;
2421 }
2422
2423 errno = 0;
2424 if (write(fd, message, n + 1) != n + 1)
2425 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2426 log_error_errno(errno, "Failed to write Plymouth message: %m");
2427 }
2428
2429 int manager_open_serialization(Manager *m, FILE **_f) {
2430 const char *path;
2431 int fd = -1;
2432 FILE *f;
2433
2434 assert(_f);
2435
2436 path = MANAGER_IS_SYSTEM(m) ? "/run/systemd" : "/tmp";
2437 fd = open_tmpfile_unlinkable(path, O_RDWR|O_CLOEXEC);
2438 if (fd < 0)
2439 return -errno;
2440
2441 log_debug("Serializing state to %s", path);
2442
2443 f = fdopen(fd, "w+");
2444 if (!f) {
2445 safe_close(fd);
2446 return -errno;
2447 }
2448
2449 *_f = f;
2450
2451 return 0;
2452 }
2453
2454 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2455 Iterator i;
2456 Unit *u;
2457 const char *t;
2458 char **e;
2459 int r;
2460
2461 assert(m);
2462 assert(f);
2463 assert(fds);
2464
2465 m->n_reloading++;
2466
2467 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2468 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2469 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2470 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2471
2472 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2473 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2474 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2475 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2476
2477 if (!in_initrd()) {
2478 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2479 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2480 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2481 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2482 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2483 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2484 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2485 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2486 }
2487
2488 if (!switching_root) {
2489 STRV_FOREACH(e, m->environment) {
2490 _cleanup_free_ char *ce;
2491
2492 ce = cescape(*e);
2493 if (!ce)
2494 return -ENOMEM;
2495
2496 fprintf(f, "env=%s\n", *e);
2497 }
2498 }
2499
2500 if (m->notify_fd >= 0) {
2501 int copy;
2502
2503 copy = fdset_put_dup(fds, m->notify_fd);
2504 if (copy < 0)
2505 return copy;
2506
2507 fprintf(f, "notify-fd=%i\n", copy);
2508 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2509 }
2510
2511 if (m->cgroups_agent_fd >= 0) {
2512 int copy;
2513
2514 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2515 if (copy < 0)
2516 return copy;
2517
2518 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2519 }
2520
2521 if (m->user_lookup_fds[0] >= 0) {
2522 int copy0, copy1;
2523
2524 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2525 if (copy0 < 0)
2526 return copy0;
2527
2528 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2529 if (copy1 < 0)
2530 return copy1;
2531
2532 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2533 }
2534
2535 bus_track_serialize(m->subscribed, f, "subscribed");
2536
2537 r = dynamic_user_serialize(m, f, fds);
2538 if (r < 0)
2539 return r;
2540
2541 manager_serialize_uid_refs(m, f);
2542 manager_serialize_gid_refs(m, f);
2543
2544 fputc('\n', f);
2545
2546 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2547 if (u->id != t)
2548 continue;
2549
2550 /* Start marker */
2551 fputs(u->id, f);
2552 fputc('\n', f);
2553
2554 r = unit_serialize(u, f, fds, !switching_root);
2555 if (r < 0) {
2556 m->n_reloading--;
2557 return r;
2558 }
2559 }
2560
2561 assert(m->n_reloading > 0);
2562 m->n_reloading--;
2563
2564 if (ferror(f))
2565 return -EIO;
2566
2567 r = bus_fdset_add_all(m, fds);
2568 if (r < 0)
2569 return r;
2570
2571 return 0;
2572 }
2573
2574 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2575 int r = 0;
2576
2577 assert(m);
2578 assert(f);
2579
2580 log_debug("Deserializing state...");
2581
2582 m->n_reloading++;
2583
2584 for (;;) {
2585 char line[LINE_MAX], *l;
2586
2587 if (!fgets(line, sizeof(line), f)) {
2588 if (feof(f))
2589 r = 0;
2590 else
2591 r = -errno;
2592
2593 goto finish;
2594 }
2595
2596 char_array_0(line);
2597 l = strstrip(line);
2598
2599 if (l[0] == 0)
2600 break;
2601
2602 if (startswith(l, "current-job-id=")) {
2603 uint32_t id;
2604
2605 if (safe_atou32(l+15, &id) < 0)
2606 log_debug("Failed to parse current job id value %s", l+15);
2607 else
2608 m->current_job_id = MAX(m->current_job_id, id);
2609
2610 } else if (startswith(l, "n-installed-jobs=")) {
2611 uint32_t n;
2612
2613 if (safe_atou32(l+17, &n) < 0)
2614 log_debug("Failed to parse installed jobs counter %s", l+17);
2615 else
2616 m->n_installed_jobs += n;
2617
2618 } else if (startswith(l, "n-failed-jobs=")) {
2619 uint32_t n;
2620
2621 if (safe_atou32(l+14, &n) < 0)
2622 log_debug("Failed to parse failed jobs counter %s", l+14);
2623 else
2624 m->n_failed_jobs += n;
2625
2626 } else if (startswith(l, "taint-usr=")) {
2627 int b;
2628
2629 b = parse_boolean(l+10);
2630 if (b < 0)
2631 log_debug("Failed to parse taint /usr flag %s", l+10);
2632 else
2633 m->taint_usr = m->taint_usr || b;
2634
2635 } else if (startswith(l, "firmware-timestamp="))
2636 dual_timestamp_deserialize(l+19, &m->firmware_timestamp);
2637 else if (startswith(l, "loader-timestamp="))
2638 dual_timestamp_deserialize(l+17, &m->loader_timestamp);
2639 else if (startswith(l, "kernel-timestamp="))
2640 dual_timestamp_deserialize(l+17, &m->kernel_timestamp);
2641 else if (startswith(l, "initrd-timestamp="))
2642 dual_timestamp_deserialize(l+17, &m->initrd_timestamp);
2643 else if (startswith(l, "userspace-timestamp="))
2644 dual_timestamp_deserialize(l+20, &m->userspace_timestamp);
2645 else if (startswith(l, "finish-timestamp="))
2646 dual_timestamp_deserialize(l+17, &m->finish_timestamp);
2647 else if (startswith(l, "security-start-timestamp="))
2648 dual_timestamp_deserialize(l+25, &m->security_start_timestamp);
2649 else if (startswith(l, "security-finish-timestamp="))
2650 dual_timestamp_deserialize(l+26, &m->security_finish_timestamp);
2651 else if (startswith(l, "generators-start-timestamp="))
2652 dual_timestamp_deserialize(l+27, &m->generators_start_timestamp);
2653 else if (startswith(l, "generators-finish-timestamp="))
2654 dual_timestamp_deserialize(l+28, &m->generators_finish_timestamp);
2655 else if (startswith(l, "units-load-start-timestamp="))
2656 dual_timestamp_deserialize(l+27, &m->units_load_start_timestamp);
2657 else if (startswith(l, "units-load-finish-timestamp="))
2658 dual_timestamp_deserialize(l+28, &m->units_load_finish_timestamp);
2659 else if (startswith(l, "env=")) {
2660 _cleanup_free_ char *uce = NULL;
2661 char **e;
2662
2663 r = cunescape(l + 4, UNESCAPE_RELAX, &uce);
2664 if (r < 0)
2665 goto finish;
2666
2667 e = strv_env_set(m->environment, uce);
2668 if (!e) {
2669 r = -ENOMEM;
2670 goto finish;
2671 }
2672
2673 strv_free(m->environment);
2674 m->environment = e;
2675
2676 } else if (startswith(l, "notify-fd=")) {
2677 int fd;
2678
2679 if (safe_atoi(l + 10, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2680 log_debug("Failed to parse notify fd: %s", l + 10);
2681 else {
2682 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2683 safe_close(m->notify_fd);
2684 m->notify_fd = fdset_remove(fds, fd);
2685 }
2686
2687 } else if (startswith(l, "notify-socket=")) {
2688 char *n;
2689
2690 n = strdup(l+14);
2691 if (!n) {
2692 r = -ENOMEM;
2693 goto finish;
2694 }
2695
2696 free(m->notify_socket);
2697 m->notify_socket = n;
2698
2699 } else if (startswith(l, "cgroups-agent-fd=")) {
2700 int fd;
2701
2702 if (safe_atoi(l + 17, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2703 log_debug("Failed to parse cgroups agent fd: %s", l + 10);
2704 else {
2705 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2706 safe_close(m->cgroups_agent_fd);
2707 m->cgroups_agent_fd = fdset_remove(fds, fd);
2708 }
2709
2710 } else if (startswith(l, "user-lookup=")) {
2711 int fd0, fd1;
2712
2713 if (sscanf(l + 12, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2714 log_debug("Failed to parse user lookup fd: %s", l + 12);
2715 else {
2716 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2717 safe_close_pair(m->user_lookup_fds);
2718 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2719 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2720 }
2721
2722 } else if (startswith(l, "dynamic-user="))
2723 dynamic_user_deserialize_one(m, l + 13, fds);
2724 else if (startswith(l, "destroy-ipc-uid="))
2725 manager_deserialize_uid_refs_one(m, l + 16);
2726 else if (startswith(l, "destroy-ipc-gid="))
2727 manager_deserialize_gid_refs_one(m, l + 16);
2728 else if (startswith(l, "subscribed=")) {
2729
2730 if (strv_extend(&m->deserialized_subscribed, l+11) < 0)
2731 log_oom();
2732
2733 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2734 log_debug("Unknown serialization item '%s'", l);
2735 }
2736
2737 for (;;) {
2738 Unit *u;
2739 char name[UNIT_NAME_MAX+2];
2740
2741 /* Start marker */
2742 if (!fgets(name, sizeof(name), f)) {
2743 if (feof(f))
2744 r = 0;
2745 else
2746 r = -errno;
2747
2748 goto finish;
2749 }
2750
2751 char_array_0(name);
2752
2753 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2754 if (r < 0)
2755 goto finish;
2756
2757 r = unit_deserialize(u, f, fds);
2758 if (r < 0)
2759 goto finish;
2760 }
2761
2762 finish:
2763 if (ferror(f))
2764 r = -EIO;
2765
2766 assert(m->n_reloading > 0);
2767 m->n_reloading--;
2768
2769 return r;
2770 }
2771
2772 int manager_reload(Manager *m) {
2773 int r, q;
2774 _cleanup_fclose_ FILE *f = NULL;
2775 _cleanup_fdset_free_ FDSet *fds = NULL;
2776
2777 assert(m);
2778
2779 r = manager_open_serialization(m, &f);
2780 if (r < 0)
2781 return r;
2782
2783 m->n_reloading++;
2784 bus_manager_send_reloading(m, true);
2785
2786 fds = fdset_new();
2787 if (!fds) {
2788 m->n_reloading--;
2789 return -ENOMEM;
2790 }
2791
2792 r = manager_serialize(m, f, fds, false);
2793 if (r < 0) {
2794 m->n_reloading--;
2795 return r;
2796 }
2797
2798 if (fseeko(f, 0, SEEK_SET) < 0) {
2799 m->n_reloading--;
2800 return -errno;
2801 }
2802
2803 /* From here on there is no way back. */
2804 manager_clear_jobs_and_units(m);
2805 lookup_paths_flush_generator(&m->lookup_paths);
2806 lookup_paths_free(&m->lookup_paths);
2807 dynamic_user_vacuum(m, false);
2808 m->uid_refs = hashmap_free(m->uid_refs);
2809 m->gid_refs = hashmap_free(m->gid_refs);
2810
2811 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2812 if (q < 0 && r >= 0)
2813 r = q;
2814
2815 /* Find new unit paths */
2816 q = manager_run_generators(m);
2817 if (q < 0 && r >= 0)
2818 r = q;
2819
2820 lookup_paths_reduce(&m->lookup_paths);
2821 manager_build_unit_path_cache(m);
2822
2823 /* First, enumerate what we can from all config files */
2824 manager_enumerate(m);
2825
2826 /* Second, deserialize our stored data */
2827 q = manager_deserialize(m, f, fds);
2828 if (q < 0 && r >= 0)
2829 r = q;
2830
2831 fclose(f);
2832 f = NULL;
2833
2834 /* Re-register notify_fd as event source */
2835 q = manager_setup_notify(m);
2836 if (q < 0 && r >= 0)
2837 r = q;
2838
2839 q = manager_setup_cgroups_agent(m);
2840 if (q < 0 && r >= 0)
2841 r = q;
2842
2843 q = manager_setup_user_lookup_fd(m);
2844 if (q < 0 && r >= 0)
2845 r = q;
2846
2847 /* Third, fire things up! */
2848 manager_coldplug(m);
2849
2850 /* Release any dynamic users no longer referenced */
2851 dynamic_user_vacuum(m, true);
2852
2853 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2854 manager_vacuum_uid_refs(m);
2855 manager_vacuum_gid_refs(m);
2856
2857 /* Sync current state of bus names with our set of listening units */
2858 if (m->api_bus)
2859 manager_sync_bus_names(m, m->api_bus);
2860
2861 assert(m->n_reloading > 0);
2862 m->n_reloading--;
2863
2864 m->send_reloading_done = true;
2865
2866 return r;
2867 }
2868
2869 void manager_reset_failed(Manager *m) {
2870 Unit *u;
2871 Iterator i;
2872
2873 assert(m);
2874
2875 HASHMAP_FOREACH(u, m->units, i)
2876 unit_reset_failed(u);
2877 }
2878
2879 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2880 Unit *u;
2881
2882 assert(m);
2883 assert(name);
2884
2885 /* Returns true if the unit is inactive or going down */
2886 u = manager_get_unit(m, name);
2887 if (!u)
2888 return true;
2889
2890 return unit_inactive_or_pending(u);
2891 }
2892
2893 static void manager_notify_finished(Manager *m) {
2894 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2895 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2896
2897 if (m->test_run)
2898 return;
2899
2900 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2901
2902 /* Note that m->kernel_usec.monotonic is always at 0,
2903 * and m->firmware_usec.monotonic and
2904 * m->loader_usec.monotonic should be considered
2905 * negative values. */
2906
2907 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2908 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2909 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2910 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2911
2912 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2913
2914 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2915 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2916
2917 log_struct(LOG_INFO,
2918 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2919 "KERNEL_USEC="USEC_FMT, kernel_usec,
2920 "INITRD_USEC="USEC_FMT, initrd_usec,
2921 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2922 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2923 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2924 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2925 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2926 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2927 NULL);
2928 } else {
2929 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2930 initrd_usec = 0;
2931
2932 log_struct(LOG_INFO,
2933 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2934 "KERNEL_USEC="USEC_FMT, kernel_usec,
2935 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2936 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2937 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2938 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2939 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2940 NULL);
2941 }
2942 } else {
2943 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2944 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2945
2946 log_struct(LOG_INFO,
2947 LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2948 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2949 LOG_MESSAGE("Startup finished in %s.",
2950 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2951 NULL);
2952 }
2953
2954 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2955
2956 sd_notifyf(false,
2957 "READY=1\n"
2958 "STATUS=Startup finished in %s.",
2959 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2960 }
2961
2962 void manager_check_finished(Manager *m) {
2963 assert(m);
2964
2965 if (MANAGER_IS_RELOADING(m))
2966 return;
2967
2968 /* Verify that we are actually running currently. Initially
2969 * the exit code is set to invalid, and during operation it is
2970 * then set to MANAGER_OK */
2971 if (m->exit_code != MANAGER_OK)
2972 return;
2973
2974 if (hashmap_size(m->jobs) > 0) {
2975 if (m->jobs_in_progress_event_source)
2976 /* Ignore any failure, this is only for feedback */
2977 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2978
2979 return;
2980 }
2981
2982 manager_flip_auto_status(m, false);
2983
2984 /* Notify Type=idle units that we are done now */
2985 manager_close_idle_pipe(m);
2986
2987 /* Turn off confirm spawn now */
2988 m->confirm_spawn = false;
2989
2990 /* No need to update ask password status when we're going non-interactive */
2991 manager_close_ask_password(m);
2992
2993 /* This is no longer the first boot */
2994 manager_set_first_boot(m, false);
2995
2996 if (dual_timestamp_is_set(&m->finish_timestamp))
2997 return;
2998
2999 dual_timestamp_get(&m->finish_timestamp);
3000
3001 manager_notify_finished(m);
3002
3003 manager_invalidate_startup_units(m);
3004 }
3005
3006 static int manager_run_generators(Manager *m) {
3007 _cleanup_strv_free_ char **paths = NULL;
3008 const char *argv[5];
3009 char **path;
3010 int r;
3011
3012 assert(m);
3013
3014 if (m->test_run)
3015 return 0;
3016
3017 paths = generator_binary_paths(m->unit_file_scope);
3018 if (!paths)
3019 return log_oom();
3020
3021 /* Optimize by skipping the whole process by not creating output directories
3022 * if no generators are found. */
3023 STRV_FOREACH(path, paths) {
3024 if (access(*path, F_OK) >= 0)
3025 goto found;
3026 if (errno != ENOENT)
3027 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3028 }
3029
3030 return 0;
3031
3032 found:
3033 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3034 if (r < 0)
3035 goto finish;
3036
3037 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3038 argv[1] = m->lookup_paths.generator;
3039 argv[2] = m->lookup_paths.generator_early;
3040 argv[3] = m->lookup_paths.generator_late;
3041 argv[4] = NULL;
3042
3043 RUN_WITH_UMASK(0022)
3044 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, (char**) argv);
3045
3046 finish:
3047 lookup_paths_trim_generator(&m->lookup_paths);
3048 return r;
3049 }
3050
3051 int manager_environment_add(Manager *m, char **minus, char **plus) {
3052 char **a = NULL, **b = NULL, **l;
3053 assert(m);
3054
3055 l = m->environment;
3056
3057 if (!strv_isempty(minus)) {
3058 a = strv_env_delete(l, 1, minus);
3059 if (!a)
3060 return -ENOMEM;
3061
3062 l = a;
3063 }
3064
3065 if (!strv_isempty(plus)) {
3066 b = strv_env_merge(2, l, plus);
3067 if (!b) {
3068 strv_free(a);
3069 return -ENOMEM;
3070 }
3071
3072 l = b;
3073 }
3074
3075 if (m->environment != l)
3076 strv_free(m->environment);
3077 if (a != l)
3078 strv_free(a);
3079 if (b != l)
3080 strv_free(b);
3081
3082 m->environment = l;
3083 manager_clean_environment(m);
3084 strv_sort(m->environment);
3085
3086 return 0;
3087 }
3088
3089 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3090 int i;
3091
3092 assert(m);
3093
3094 for (i = 0; i < _RLIMIT_MAX; i++) {
3095 m->rlimit[i] = mfree(m->rlimit[i]);
3096
3097 if (!default_rlimit[i])
3098 continue;
3099
3100 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3101 if (!m->rlimit[i])
3102 return -ENOMEM;
3103 }
3104
3105 return 0;
3106 }
3107
3108 void manager_recheck_journal(Manager *m) {
3109 Unit *u;
3110
3111 assert(m);
3112
3113 if (!MANAGER_IS_SYSTEM(m))
3114 return;
3115
3116 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3117 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3118 log_close_journal();
3119 return;
3120 }
3121
3122 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3123 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3124 log_close_journal();
3125 return;
3126 }
3127
3128 /* Hmm, OK, so the socket is fully up and the service is up
3129 * too, then let's make use of the thing. */
3130 log_open();
3131 }
3132
3133 void manager_set_show_status(Manager *m, ShowStatus mode) {
3134 assert(m);
3135 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3136
3137 if (!MANAGER_IS_SYSTEM(m))
3138 return;
3139
3140 if (m->show_status != mode)
3141 log_debug("%s showing of status.",
3142 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3143 m->show_status = mode;
3144
3145 if (mode > 0)
3146 (void) touch("/run/systemd/show-status");
3147 else
3148 (void) unlink("/run/systemd/show-status");
3149 }
3150
3151 static bool manager_get_show_status(Manager *m, StatusType type) {
3152 assert(m);
3153
3154 if (!MANAGER_IS_SYSTEM(m))
3155 return false;
3156
3157 if (m->no_console_output)
3158 return false;
3159
3160 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3161 return false;
3162
3163 /* If we cannot find out the status properly, just proceed. */
3164 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3165 return false;
3166
3167 if (m->show_status > 0)
3168 return true;
3169
3170 return false;
3171 }
3172
3173 void manager_set_first_boot(Manager *m, bool b) {
3174 assert(m);
3175
3176 if (!MANAGER_IS_SYSTEM(m))
3177 return;
3178
3179 if (m->first_boot != (int) b) {
3180 if (b)
3181 (void) touch("/run/systemd/first-boot");
3182 else
3183 (void) unlink("/run/systemd/first-boot");
3184 }
3185
3186 m->first_boot = b;
3187 }
3188
3189 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3190 va_list ap;
3191
3192 /* If m is NULL, assume we're after shutdown and let the messages through. */
3193
3194 if (m && !manager_get_show_status(m, type))
3195 return;
3196
3197 /* XXX We should totally drop the check for ephemeral here
3198 * and thus effectively make 'Type=idle' pointless. */
3199 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3200 return;
3201
3202 va_start(ap, format);
3203 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3204 va_end(ap);
3205 }
3206
3207 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3208 char p[strlen(path)+1];
3209
3210 assert(m);
3211 assert(path);
3212
3213 strcpy(p, path);
3214 path_kill_slashes(p);
3215
3216 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3217 }
3218
3219 const char *manager_get_runtime_prefix(Manager *m) {
3220 assert(m);
3221
3222 return MANAGER_IS_SYSTEM(m) ?
3223 "/run" :
3224 getenv("XDG_RUNTIME_DIR");
3225 }
3226
3227 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3228 unsigned size;
3229 int r;
3230
3231 assert(m);
3232 assert(u->manager == m);
3233
3234 size = set_size(m->failed_units);
3235
3236 if (failed) {
3237 r = set_ensure_allocated(&m->failed_units, NULL);
3238 if (r < 0)
3239 return log_oom();
3240
3241 if (set_put(m->failed_units, u) < 0)
3242 return log_oom();
3243 } else
3244 (void) set_remove(m->failed_units, u);
3245
3246 if (set_size(m->failed_units) != size)
3247 bus_manager_send_change_signal(m);
3248
3249 return 0;
3250 }
3251
3252 ManagerState manager_state(Manager *m) {
3253 Unit *u;
3254
3255 assert(m);
3256
3257 /* Did we ever finish booting? If not then we are still starting up */
3258 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3259
3260 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3261 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3262 return MANAGER_INITIALIZING;
3263
3264 return MANAGER_STARTING;
3265 }
3266
3267 /* Is the special shutdown target queued? If so, we are in shutdown state */
3268 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3269 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3270 return MANAGER_STOPPING;
3271
3272 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3273 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3274 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3275 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3276 return MANAGER_MAINTENANCE;
3277
3278 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3279 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3280 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3281 return MANAGER_MAINTENANCE;
3282
3283 /* Are there any failed units? If so, we are in degraded mode */
3284 if (set_size(m->failed_units) > 0)
3285 return MANAGER_DEGRADED;
3286
3287 return MANAGER_RUNNING;
3288 }
3289
3290 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3291
3292 static void manager_unref_uid_internal(
3293 Manager *m,
3294 Hashmap **uid_refs,
3295 uid_t uid,
3296 bool destroy_now,
3297 int (*_clean_ipc)(uid_t uid)) {
3298
3299 uint32_t c, n;
3300
3301 assert(m);
3302 assert(uid_refs);
3303 assert(uid_is_valid(uid));
3304 assert(_clean_ipc);
3305
3306 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3307 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3308 *
3309 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3310 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3311 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3312 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3313
3314 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3315 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3316
3317 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3318 return;
3319
3320 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3321
3322 n = c & ~DESTROY_IPC_FLAG;
3323 assert(n > 0);
3324 n--;
3325
3326 if (destroy_now && n == 0) {
3327 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3328
3329 if (c & DESTROY_IPC_FLAG) {
3330 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3331 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3332 uid);
3333 (void) _clean_ipc(uid);
3334 }
3335 } else {
3336 c = n | (c & DESTROY_IPC_FLAG);
3337 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3338 }
3339 }
3340
3341 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3342 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3343 }
3344
3345 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3346 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3347 }
3348
3349 static int manager_ref_uid_internal(
3350 Manager *m,
3351 Hashmap **uid_refs,
3352 uid_t uid,
3353 bool clean_ipc) {
3354
3355 uint32_t c, n;
3356 int r;
3357
3358 assert(m);
3359 assert(uid_refs);
3360 assert(uid_is_valid(uid));
3361
3362 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3363 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3364
3365 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3366 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3367
3368 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3369 return 0;
3370
3371 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3372 if (r < 0)
3373 return r;
3374
3375 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3376
3377 n = c & ~DESTROY_IPC_FLAG;
3378 n++;
3379
3380 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3381 return -EOVERFLOW;
3382
3383 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3384
3385 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3386 }
3387
3388 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3389 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3390 }
3391
3392 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3393 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3394 }
3395
3396 static void manager_vacuum_uid_refs_internal(
3397 Manager *m,
3398 Hashmap **uid_refs,
3399 int (*_clean_ipc)(uid_t uid)) {
3400
3401 Iterator i;
3402 void *p, *k;
3403
3404 assert(m);
3405 assert(uid_refs);
3406 assert(_clean_ipc);
3407
3408 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3409 uint32_t c, n;
3410 uid_t uid;
3411
3412 uid = PTR_TO_UID(k);
3413 c = PTR_TO_UINT32(p);
3414
3415 n = c & ~DESTROY_IPC_FLAG;
3416 if (n > 0)
3417 continue;
3418
3419 if (c & DESTROY_IPC_FLAG) {
3420 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3421 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3422 uid);
3423 (void) _clean_ipc(uid);
3424 }
3425
3426 assert_se(hashmap_remove(*uid_refs, k) == p);
3427 }
3428 }
3429
3430 void manager_vacuum_uid_refs(Manager *m) {
3431 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3432 }
3433
3434 void manager_vacuum_gid_refs(Manager *m) {
3435 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3436 }
3437
3438 static void manager_serialize_uid_refs_internal(
3439 Manager *m,
3440 FILE *f,
3441 Hashmap **uid_refs,
3442 const char *field_name) {
3443
3444 Iterator i;
3445 void *p, *k;
3446
3447 assert(m);
3448 assert(f);
3449 assert(uid_refs);
3450 assert(field_name);
3451
3452 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3453 * of it is better rebuild after a reload/reexec. */
3454
3455 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3456 uint32_t c;
3457 uid_t uid;
3458
3459 uid = PTR_TO_UID(k);
3460 c = PTR_TO_UINT32(p);
3461
3462 if (!(c & DESTROY_IPC_FLAG))
3463 continue;
3464
3465 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3466 }
3467 }
3468
3469 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3470 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3471 }
3472
3473 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3474 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3475 }
3476
3477 static void manager_deserialize_uid_refs_one_internal(
3478 Manager *m,
3479 Hashmap** uid_refs,
3480 const char *value) {
3481
3482 uid_t uid;
3483 uint32_t c;
3484 int r;
3485
3486 assert(m);
3487 assert(uid_refs);
3488 assert(value);
3489
3490 r = parse_uid(value, &uid);
3491 if (r < 0 || uid == 0) {
3492 log_debug("Unable to parse UID reference serialization");
3493 return;
3494 }
3495
3496 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3497 if (r < 0) {
3498 log_oom();
3499 return;
3500 }
3501
3502 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3503 if (c & DESTROY_IPC_FLAG)
3504 return;
3505
3506 c |= DESTROY_IPC_FLAG;
3507
3508 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3509 if (r < 0) {
3510 log_debug("Failed to add UID reference entry");
3511 return;
3512 }
3513 }
3514
3515 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3516 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3517 }
3518
3519 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3520 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3521 }
3522
3523 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3524 struct buffer {
3525 uid_t uid;
3526 gid_t gid;
3527 char unit_name[UNIT_NAME_MAX+1];
3528 } _packed_ buffer;
3529
3530 Manager *m = userdata;
3531 ssize_t l;
3532 size_t n;
3533 Unit *u;
3534
3535 assert_se(source);
3536 assert_se(m);
3537
3538 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3539 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3540 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3541
3542 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3543 if (l < 0) {
3544 if (errno == EINTR || errno == EAGAIN)
3545 return 0;
3546
3547 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3548 }
3549
3550 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3551 log_warning("Received too short user lookup message, ignoring.");
3552 return 0;
3553 }
3554
3555 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3556 log_warning("Received too long user lookup message, ignoring.");
3557 return 0;
3558 }
3559
3560 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3561 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3562 return 0;
3563 }
3564
3565 n = (size_t) l - offsetof(struct buffer, unit_name);
3566 if (memchr(buffer.unit_name, 0, n)) {
3567 log_warning("Received lookup message with embedded NUL character, ignoring.");
3568 return 0;
3569 }
3570
3571 buffer.unit_name[n] = 0;
3572 u = manager_get_unit(m, buffer.unit_name);
3573 if (!u) {
3574 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3575 return 0;
3576 }
3577
3578 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3579
3580 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3581 return 0;
3582 }
3583
3584 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3585 [MANAGER_INITIALIZING] = "initializing",
3586 [MANAGER_STARTING] = "starting",
3587 [MANAGER_RUNNING] = "running",
3588 [MANAGER_DEGRADED] = "degraded",
3589 [MANAGER_MAINTENANCE] = "maintenance",
3590 [MANAGER_STOPPING] = "stopping",
3591 };
3592
3593 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
3594
3595 static const char *const cad_burst_action_table[_CAD_BURST_ACTION_MAX] = {
3596 [CAD_BURST_ACTION_IGNORE] = "ignore",
3597 [CAD_BURST_ACTION_REBOOT] = "reboot-force",
3598 [CAD_BURST_ACTION_POWEROFF] = "poweroff-force",
3599 };
3600
3601 DEFINE_STRING_TABLE_LOOKUP(cad_burst_action, CADBurstAction);
Unnamed repository; edit this file 'description' to name the repository.