]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #5449 from keszybz/blkd-error-handling
[thirdparty/systemd.git] / src / core / manager.c
1 /***
2 This file is part of systemd.
3
4 Copyright 2010 Lennart Poettering
5
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
10
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
18 ***/
19
20 #include <errno.h>
21 #include <fcntl.h>
22 #include <linux/kd.h>
23 #include <signal.h>
24 #include <string.h>
25 #include <sys/epoll.h>
26 #include <sys/inotify.h>
27 #include <sys/ioctl.h>
28 #include <sys/reboot.h>
29 #include <sys/timerfd.h>
30 #include <sys/wait.h>
31 #include <unistd.h>
32
33 #ifdef HAVE_AUDIT
34 #include <libaudit.h>
35 #endif
36
37 #include "sd-daemon.h"
38 #include "sd-messages.h"
39
40 #include "alloc-util.h"
41 #include "audit-fd.h"
42 #include "boot-timestamps.h"
43 #include "bus-common-errors.h"
44 #include "bus-error.h"
45 #include "bus-kernel.h"
46 #include "bus-util.h"
47 #include "clean-ipc.h"
48 #include "dbus-job.h"
49 #include "dbus-manager.h"
50 #include "dbus-unit.h"
51 #include "dbus.h"
52 #include "dirent-util.h"
53 #include "env-util.h"
54 #include "escape.h"
55 #include "exec-util.h"
56 #include "exit-status.h"
57 #include "fd-util.h"
58 #include "fileio.h"
59 #include "fs-util.h"
60 #include "hashmap.h"
61 #include "io-util.h"
62 #include "locale-setup.h"
63 #include "log.h"
64 #include "macro.h"
65 #include "manager.h"
66 #include "missing.h"
67 #include "mkdir.h"
68 #include "parse-util.h"
69 #include "path-lookup.h"
70 #include "path-util.h"
71 #include "process-util.h"
72 #include "ratelimit.h"
73 #include "rm-rf.h"
74 #include "signal-util.h"
75 #include "special.h"
76 #include "stat-util.h"
77 #include "string-table.h"
78 #include "string-util.h"
79 #include "strv.h"
80 #include "terminal-util.h"
81 #include "time-util.h"
82 #include "transaction.h"
83 #include "umask-util.h"
84 #include "unit-name.h"
85 #include "user-util.h"
86 #include "util.h"
87 #include "virt.h"
88 #include "watchdog.h"
89
90 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
91 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
92
93 /* Initial delay and the interval for printing status messages about running jobs */
94 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
97
98 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
99 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
100 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
101 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
102 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
103 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
104 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
105 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
106 static int manager_run_environment_generators(Manager *m);
107 static int manager_run_generators(Manager *m);
108
109 static void manager_watch_jobs_in_progress(Manager *m) {
110 usec_t next;
111 int r;
112
113 assert(m);
114
115 /* We do not want to show the cylon animation if the user
116 * needs to confirm service executions otherwise confirmation
117 * messages will be screwed by the cylon animation. */
118 if (!manager_is_confirm_spawn_disabled(m))
119 return;
120
121 if (m->jobs_in_progress_event_source)
122 return;
123
124 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
125 r = sd_event_add_time(
126 m->event,
127 &m->jobs_in_progress_event_source,
128 CLOCK_MONOTONIC,
129 next, 0,
130 manager_dispatch_jobs_in_progress, m);
131 if (r < 0)
132 return;
133
134 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
135 }
136
137 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
138
139 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
140 char *p = buffer;
141
142 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
143 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
144
145 if (pos > 1) {
146 if (pos > 2)
147 p = mempset(p, ' ', pos-2);
148 if (log_get_show_color())
149 p = stpcpy(p, ANSI_RED);
150 *p++ = '*';
151 }
152
153 if (pos > 0 && pos <= width) {
154 if (log_get_show_color())
155 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
156 *p++ = '*';
157 }
158
159 if (log_get_show_color())
160 p = stpcpy(p, ANSI_NORMAL);
161
162 if (pos < width) {
163 if (log_get_show_color())
164 p = stpcpy(p, ANSI_RED);
165 *p++ = '*';
166 if (pos < width-1)
167 p = mempset(p, ' ', width-1-pos);
168 if (log_get_show_color())
169 strcpy(p, ANSI_NORMAL);
170 }
171 }
172
173 void manager_flip_auto_status(Manager *m, bool enable) {
174 assert(m);
175
176 if (enable) {
177 if (m->show_status == SHOW_STATUS_AUTO)
178 manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
179 } else {
180 if (m->show_status == SHOW_STATUS_TEMPORARY)
181 manager_set_show_status(m, SHOW_STATUS_AUTO);
182 }
183 }
184
185 static void manager_print_jobs_in_progress(Manager *m) {
186 _cleanup_free_ char *job_of_n = NULL;
187 Iterator i;
188 Job *j;
189 unsigned counter = 0, print_nr;
190 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
191 unsigned cylon_pos;
192 char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
193 uint64_t x;
194
195 assert(m);
196 assert(m->n_running_jobs > 0);
197
198 manager_flip_auto_status(m, true);
199
200 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
201
202 HASHMAP_FOREACH(j, m->jobs, i)
203 if (j->state == JOB_RUNNING && counter++ == print_nr)
204 break;
205
206 /* m->n_running_jobs must be consistent with the contents of m->jobs,
207 * so the above loop must have succeeded in finding j. */
208 assert(counter == print_nr + 1);
209 assert(j);
210
211 cylon_pos = m->jobs_in_progress_iteration % 14;
212 if (cylon_pos >= 8)
213 cylon_pos = 14 - cylon_pos;
214 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
215
216 m->jobs_in_progress_iteration++;
217
218 if (m->n_running_jobs > 1) {
219 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
220 job_of_n = NULL;
221 }
222
223 format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
224 if (job_get_timeout(j, &x) > 0)
225 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
226
227 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
228 "%sA %s job is running for %s (%s / %s)",
229 strempty(job_of_n),
230 job_type_to_string(j->type),
231 unit_description(j->unit),
232 time, limit);
233 }
234
235 static int have_ask_password(void) {
236 _cleanup_closedir_ DIR *dir;
237 struct dirent *de;
238
239 dir = opendir("/run/systemd/ask-password");
240 if (!dir) {
241 if (errno == ENOENT)
242 return false;
243 else
244 return -errno;
245 }
246
247 FOREACH_DIRENT_ALL(de, dir, return -errno) {
248 if (startswith(de->d_name, "ask."))
249 return true;
250 }
251 return false;
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's or the parents' exported variables.
536 *
537 * The initial passed environment 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 if (!m->environment)
551 return -ENOMEM;
552
553 manager_clean_environment(m);
554 strv_sort(m->environment);
555
556 return 0;
557 }
558
559 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
560 Manager *m;
561 int r;
562
563 assert(_m);
564 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
565
566 m = new0(Manager, 1);
567 if (!m)
568 return -ENOMEM;
569
570 m->unit_file_scope = scope;
571 m->exit_code = _MANAGER_EXIT_CODE_INVALID;
572 m->default_timer_accuracy_usec = USEC_PER_MINUTE;
573 m->default_tasks_accounting = true;
574 m->default_tasks_max = UINT64_MAX;
575
576 #ifdef ENABLE_EFI
577 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
578 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
579 #endif
580
581 /* Prepare log fields we can use for structured logging */
582 if (MANAGER_IS_SYSTEM(m)) {
583 m->unit_log_field = "UNIT=";
584 m->unit_log_format_string = "UNIT=%s";
585
586 m->invocation_log_field = "INVOCATION_ID=";
587 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
588 } else {
589 m->unit_log_field = "USER_UNIT=";
590 m->unit_log_format_string = "USER_UNIT=%s";
591
592 m->invocation_log_field = "USER_INVOCATION_ID=";
593 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
594 }
595
596 m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
597
598 m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
599 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
600 m->ask_password_inotify_fd = -1;
601
602 m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
603
604 m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
605
606 m->have_ask_password = -EINVAL; /* we don't know */
607 m->first_boot = -1;
608
609 m->test_run = test_run;
610
611 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
612 RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
613
614 r = manager_default_environment(m);
615 if (r < 0)
616 goto fail;
617
618 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
619 if (r < 0)
620 goto fail;
621
622 r = hashmap_ensure_allocated(&m->jobs, NULL);
623 if (r < 0)
624 goto fail;
625
626 r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
627 if (r < 0)
628 goto fail;
629
630 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
631 if (r < 0)
632 goto fail;
633
634 r = sd_event_default(&m->event);
635 if (r < 0)
636 goto fail;
637
638 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
639 if (r < 0)
640 goto fail;
641
642 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
643 if (r < 0)
644 goto fail;
645
646 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
647 if (r < 0)
648 goto fail;
649
650 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
651
652 r = manager_setup_signals(m);
653 if (r < 0)
654 goto fail;
655
656 r = manager_setup_cgroup(m);
657 if (r < 0)
658 goto fail;
659
660 r = manager_setup_time_change(m);
661 if (r < 0)
662 goto fail;
663
664 m->udev = udev_new();
665 if (!m->udev) {
666 r = -ENOMEM;
667 goto fail;
668 }
669
670 /* Note that we do not set up the notify fd here. We do that after deserialization,
671 * since they might have gotten serialized across the reexec. */
672
673 m->taint_usr = dir_is_empty("/usr") > 0;
674
675 *_m = m;
676 return 0;
677
678 fail:
679 manager_free(m);
680 return r;
681 }
682
683 static int manager_setup_notify(Manager *m) {
684 int r;
685
686 if (m->test_run)
687 return 0;
688
689 if (m->notify_fd < 0) {
690 _cleanup_close_ int fd = -1;
691 union sockaddr_union sa = {
692 .sa.sa_family = AF_UNIX,
693 };
694 static const int one = 1;
695 const char *e;
696
697 /* First free all secondary fields */
698 m->notify_socket = mfree(m->notify_socket);
699 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
700
701 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
702 if (fd < 0)
703 return log_error_errno(errno, "Failed to allocate notification socket: %m");
704
705 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
706
707 e = manager_get_runtime_prefix(m);
708 if (!e) {
709 log_error("Failed to determine runtime prefix.");
710 return -EINVAL;
711 }
712
713 m->notify_socket = strappend(e, "/systemd/notify");
714 if (!m->notify_socket)
715 return log_oom();
716
717 (void) mkdir_parents_label(m->notify_socket, 0755);
718 (void) unlink(m->notify_socket);
719
720 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
721 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
722 if (r < 0)
723 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
724
725 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
726 if (r < 0)
727 return log_error_errno(errno, "SO_PASSCRED failed: %m");
728
729 m->notify_fd = fd;
730 fd = -1;
731
732 log_debug("Using notification socket %s", m->notify_socket);
733 }
734
735 if (!m->notify_event_source) {
736 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
737 if (r < 0)
738 return log_error_errno(r, "Failed to allocate notify event source: %m");
739
740 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
741 * service an exit message belongs. */
742 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
743 if (r < 0)
744 return log_error_errno(r, "Failed to set priority of notify event source: %m");
745
746 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
747 }
748
749 return 0;
750 }
751
752 static int manager_setup_cgroups_agent(Manager *m) {
753
754 static const union sockaddr_union sa = {
755 .un.sun_family = AF_UNIX,
756 .un.sun_path = "/run/systemd/cgroups-agent",
757 };
758 int r;
759
760 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
761 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
762 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
763 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
764 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
765 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
766 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
767 * we thus won't lose messages.
768 *
769 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
770 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
771 * bus for these messages. */
772
773 if (m->test_run)
774 return 0;
775
776 if (!MANAGER_IS_SYSTEM(m))
777 return 0;
778
779 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
780 if (r < 0)
781 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
782 if (r > 0) /* We don't need this anymore on the unified hierarchy */
783 return 0;
784
785 if (m->cgroups_agent_fd < 0) {
786 _cleanup_close_ int fd = -1;
787
788 /* First free all secondary fields */
789 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
790
791 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
792 if (fd < 0)
793 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
794
795 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
796
797 (void) unlink(sa.un.sun_path);
798
799 /* Only allow root to connect to this socket */
800 RUN_WITH_UMASK(0077)
801 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
802 if (r < 0)
803 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
804
805 m->cgroups_agent_fd = fd;
806 fd = -1;
807 }
808
809 if (!m->cgroups_agent_event_source) {
810 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
811 if (r < 0)
812 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
813
814 /* Process cgroups notifications early, but after having processed service notification messages or
815 * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
816 * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
817 * cgroup inotify for the unified cgroup stuff. */
818 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
819 if (r < 0)
820 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
821
822 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
823 }
824
825 return 0;
826 }
827
828 static int manager_setup_user_lookup_fd(Manager *m) {
829 int r;
830
831 assert(m);
832
833 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
834 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
835 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
836 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
837 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
838 * hence we establish this communication channel so that forked off processes can pass their UID/GID
839 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
840 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
841 * this purpose.
842 *
843 * You might wonder why we need a communication channel for this that is independent of the usual notification
844 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
845 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
846 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
847 *
848 * Note that this function is called under two circumstances: when we first initialize (in which case we
849 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
850 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
851
852 if (m->user_lookup_fds[0] < 0) {
853
854 /* Free all secondary fields */
855 safe_close_pair(m->user_lookup_fds);
856 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
857
858 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
859 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
860
861 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
862 }
863
864 if (!m->user_lookup_event_source) {
865 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
866 if (r < 0)
867 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
868
869 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
870 * resolutions */
871 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
872 if (r < 0)
873 return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
874
875 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
876 }
877
878 return 0;
879 }
880
881 static int manager_connect_bus(Manager *m, bool reexecuting) {
882 bool try_bus_connect;
883
884 assert(m);
885
886 if (m->test_run)
887 return 0;
888
889 try_bus_connect =
890 reexecuting ||
891 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
892
893 /* Try to connect to the buses, if possible. */
894 return bus_init(m, try_bus_connect);
895 }
896
897 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
898 Unit *u;
899 unsigned n = 0;
900
901 assert(m);
902
903 while ((u = m->cleanup_queue)) {
904 assert(u->in_cleanup_queue);
905
906 unit_free(u);
907 n++;
908 }
909
910 return n;
911 }
912
913 enum {
914 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
915 GC_OFFSET_UNSURE, /* No clue */
916 GC_OFFSET_GOOD, /* We still need this unit */
917 GC_OFFSET_BAD, /* We don't need this unit anymore */
918 _GC_OFFSET_MAX
919 };
920
921 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
922 Iterator i;
923 Unit *other;
924
925 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
926
927 /* Recursively mark referenced units as GOOD as well */
928 SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
929 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
930 unit_gc_mark_good(other, gc_marker);
931 }
932
933 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
934 Iterator i;
935 Unit *other;
936 bool is_bad;
937
938 assert(u);
939
940 if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
941 u->gc_marker == gc_marker + GC_OFFSET_BAD ||
942 u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
943 u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
944 return;
945
946 if (u->in_cleanup_queue)
947 goto bad;
948
949 if (unit_check_gc(u))
950 goto good;
951
952 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
953
954 is_bad = true;
955
956 SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
957 unit_gc_sweep(other, gc_marker);
958
959 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
960 goto good;
961
962 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
963 is_bad = false;
964 }
965
966 if (is_bad)
967 goto bad;
968
969 /* We were unable to find anything out about this entry, so
970 * let's investigate it later */
971 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
972 unit_add_to_gc_queue(u);
973 return;
974
975 bad:
976 /* We definitely know that this one is not useful anymore, so
977 * let's mark it for deletion */
978 u->gc_marker = gc_marker + GC_OFFSET_BAD;
979 unit_add_to_cleanup_queue(u);
980 return;
981
982 good:
983 unit_gc_mark_good(u, gc_marker);
984 }
985
986 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
987 unsigned n = 0, gc_marker;
988 Unit *u;
989
990 assert(m);
991
992 /* log_debug("Running GC..."); */
993
994 m->gc_marker += _GC_OFFSET_MAX;
995 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
996 m->gc_marker = 1;
997
998 gc_marker = m->gc_marker;
999
1000 while ((u = m->gc_unit_queue)) {
1001 assert(u->in_gc_queue);
1002
1003 unit_gc_sweep(u, gc_marker);
1004
1005 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1006 u->in_gc_queue = false;
1007
1008 n++;
1009
1010 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1011 u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1012 if (u->id)
1013 log_unit_debug(u, "Collecting.");
1014 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1015 unit_add_to_cleanup_queue(u);
1016 }
1017 }
1018
1019 return n;
1020 }
1021
1022 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1023 unsigned n = 0;
1024 Job *j;
1025
1026 assert(m);
1027
1028 while ((j = m->gc_job_queue)) {
1029 assert(j->in_gc_queue);
1030
1031 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1032 j->in_gc_queue = false;
1033
1034 n++;
1035
1036 if (job_check_gc(j))
1037 continue;
1038
1039 log_unit_debug(j->unit, "Collecting job.");
1040 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1041 }
1042
1043 return n;
1044 }
1045
1046 static void manager_clear_jobs_and_units(Manager *m) {
1047 Unit *u;
1048
1049 assert(m);
1050
1051 while ((u = hashmap_first(m->units)))
1052 unit_free(u);
1053
1054 manager_dispatch_cleanup_queue(m);
1055
1056 assert(!m->load_queue);
1057 assert(!m->run_queue);
1058 assert(!m->dbus_unit_queue);
1059 assert(!m->dbus_job_queue);
1060 assert(!m->cleanup_queue);
1061 assert(!m->gc_unit_queue);
1062 assert(!m->gc_job_queue);
1063
1064 assert(hashmap_isempty(m->jobs));
1065 assert(hashmap_isempty(m->units));
1066
1067 m->n_on_console = 0;
1068 m->n_running_jobs = 0;
1069 }
1070
1071 Manager* manager_free(Manager *m) {
1072 UnitType c;
1073 int i;
1074
1075 if (!m)
1076 return NULL;
1077
1078 manager_clear_jobs_and_units(m);
1079
1080 for (c = 0; c < _UNIT_TYPE_MAX; c++)
1081 if (unit_vtable[c]->shutdown)
1082 unit_vtable[c]->shutdown(m);
1083
1084 /* If we reexecute ourselves, we keep the root cgroup
1085 * around */
1086 manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1087
1088 lookup_paths_flush_generator(&m->lookup_paths);
1089
1090 bus_done(m);
1091
1092 dynamic_user_vacuum(m, false);
1093 hashmap_free(m->dynamic_users);
1094
1095 hashmap_free(m->units);
1096 hashmap_free(m->units_by_invocation_id);
1097 hashmap_free(m->jobs);
1098 hashmap_free(m->watch_pids1);
1099 hashmap_free(m->watch_pids2);
1100 hashmap_free(m->watch_bus);
1101
1102 set_free(m->startup_units);
1103 set_free(m->failed_units);
1104
1105 sd_event_source_unref(m->signal_event_source);
1106 sd_event_source_unref(m->notify_event_source);
1107 sd_event_source_unref(m->cgroups_agent_event_source);
1108 sd_event_source_unref(m->time_change_event_source);
1109 sd_event_source_unref(m->jobs_in_progress_event_source);
1110 sd_event_source_unref(m->run_queue_event_source);
1111 sd_event_source_unref(m->user_lookup_event_source);
1112
1113 safe_close(m->signal_fd);
1114 safe_close(m->notify_fd);
1115 safe_close(m->cgroups_agent_fd);
1116 safe_close(m->time_change_fd);
1117 safe_close_pair(m->user_lookup_fds);
1118
1119 manager_close_ask_password(m);
1120
1121 manager_close_idle_pipe(m);
1122
1123 udev_unref(m->udev);
1124 sd_event_unref(m->event);
1125
1126 free(m->notify_socket);
1127
1128 lookup_paths_free(&m->lookup_paths);
1129 strv_free(m->environment);
1130
1131 hashmap_free(m->cgroup_unit);
1132 set_free_free(m->unit_path_cache);
1133
1134 free(m->switch_root);
1135 free(m->switch_root_init);
1136
1137 for (i = 0; i < _RLIMIT_MAX; i++)
1138 m->rlimit[i] = mfree(m->rlimit[i]);
1139
1140 assert(hashmap_isempty(m->units_requiring_mounts_for));
1141 hashmap_free(m->units_requiring_mounts_for);
1142
1143 hashmap_free(m->uid_refs);
1144 hashmap_free(m->gid_refs);
1145
1146 return mfree(m);
1147 }
1148
1149 void manager_enumerate(Manager *m) {
1150 UnitType c;
1151
1152 assert(m);
1153
1154 /* Let's ask every type to load all units from disk/kernel
1155 * that it might know */
1156 for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1157 if (!unit_type_supported(c)) {
1158 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1159 continue;
1160 }
1161
1162 if (!unit_vtable[c]->enumerate)
1163 continue;
1164
1165 unit_vtable[c]->enumerate(m);
1166 }
1167
1168 manager_dispatch_load_queue(m);
1169 }
1170
1171 static void manager_coldplug(Manager *m) {
1172 Iterator i;
1173 Unit *u;
1174 char *k;
1175 int r;
1176
1177 assert(m);
1178
1179 /* Then, let's set up their initial state. */
1180 HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1181
1182 /* ignore aliases */
1183 if (u->id != k)
1184 continue;
1185
1186 r = unit_coldplug(u);
1187 if (r < 0)
1188 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1189 }
1190 }
1191
1192 static void manager_build_unit_path_cache(Manager *m) {
1193 char **i;
1194 int r;
1195
1196 assert(m);
1197
1198 set_free_free(m->unit_path_cache);
1199
1200 m->unit_path_cache = set_new(&string_hash_ops);
1201 if (!m->unit_path_cache) {
1202 r = -ENOMEM;
1203 goto fail;
1204 }
1205
1206 /* This simply builds a list of files we know exist, so that
1207 * we don't always have to go to disk */
1208
1209 STRV_FOREACH(i, m->lookup_paths.search_path) {
1210 _cleanup_closedir_ DIR *d = NULL;
1211 struct dirent *de;
1212
1213 d = opendir(*i);
1214 if (!d) {
1215 if (errno != ENOENT)
1216 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1217 continue;
1218 }
1219
1220 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1221 char *p;
1222
1223 p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1224 if (!p) {
1225 r = -ENOMEM;
1226 goto fail;
1227 }
1228
1229 r = set_consume(m->unit_path_cache, p);
1230 if (r < 0)
1231 goto fail;
1232 }
1233 }
1234
1235 return;
1236
1237 fail:
1238 log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1239 m->unit_path_cache = set_free_free(m->unit_path_cache);
1240 }
1241
1242 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1243 Iterator i;
1244 Unit *u;
1245
1246 assert(m);
1247
1248 HASHMAP_FOREACH(u, m->units, i) {
1249
1250 if (fdset_size(fds) <= 0)
1251 break;
1252
1253 if (!UNIT_VTABLE(u)->distribute_fds)
1254 continue;
1255
1256 UNIT_VTABLE(u)->distribute_fds(u, fds);
1257 }
1258 }
1259
1260 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1261 int r, q;
1262
1263 assert(m);
1264
1265 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1266 if (r < 0)
1267 return r;
1268
1269 r = manager_run_environment_generators(m);
1270 if (r < 0)
1271 return r;
1272
1273 /* Make sure the transient directory always exists, so that it remains in the search path */
1274 if (!m->test_run) {
1275 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1276 if (r < 0)
1277 return r;
1278 }
1279
1280 dual_timestamp_get(&m->generators_start_timestamp);
1281 r = manager_run_generators(m);
1282 dual_timestamp_get(&m->generators_finish_timestamp);
1283 if (r < 0)
1284 return r;
1285
1286 lookup_paths_reduce(&m->lookup_paths);
1287 manager_build_unit_path_cache(m);
1288
1289 /* If we will deserialize make sure that during enumeration
1290 * this is already known, so we increase the counter here
1291 * already */
1292 if (serialization)
1293 m->n_reloading++;
1294
1295 /* First, enumerate what we can from all config files */
1296 dual_timestamp_get(&m->units_load_start_timestamp);
1297 manager_enumerate(m);
1298 dual_timestamp_get(&m->units_load_finish_timestamp);
1299
1300 /* Second, deserialize if there is something to deserialize */
1301 if (serialization)
1302 r = manager_deserialize(m, serialization, fds);
1303
1304 /* Any fds left? Find some unit which wants them. This is
1305 * useful to allow container managers to pass some file
1306 * descriptors to us pre-initialized. This enables
1307 * socket-based activation of entire containers. */
1308 manager_distribute_fds(m, fds);
1309
1310 /* We might have deserialized the notify fd, but if we didn't
1311 * then let's create the bus now */
1312 q = manager_setup_notify(m);
1313 if (q < 0 && r == 0)
1314 r = q;
1315
1316 q = manager_setup_cgroups_agent(m);
1317 if (q < 0 && r == 0)
1318 r = q;
1319
1320 q = manager_setup_user_lookup_fd(m);
1321 if (q < 0 && r == 0)
1322 r = q;
1323
1324 /* Let's connect to the bus now. */
1325 (void) manager_connect_bus(m, !!serialization);
1326
1327 (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1328 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1329
1330 /* Third, fire things up! */
1331 manager_coldplug(m);
1332
1333 /* Release any dynamic users no longer referenced */
1334 dynamic_user_vacuum(m, true);
1335
1336 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1337 manager_vacuum_uid_refs(m);
1338 manager_vacuum_gid_refs(m);
1339
1340 if (serialization) {
1341 assert(m->n_reloading > 0);
1342 m->n_reloading--;
1343
1344 /* Let's wait for the UnitNew/JobNew messages being
1345 * sent, before we notify that the reload is
1346 * finished */
1347 m->send_reloading_done = true;
1348 }
1349
1350 return r;
1351 }
1352
1353 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1354 int r;
1355 Transaction *tr;
1356
1357 assert(m);
1358 assert(type < _JOB_TYPE_MAX);
1359 assert(unit);
1360 assert(mode < _JOB_MODE_MAX);
1361
1362 if (mode == JOB_ISOLATE && type != JOB_START)
1363 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1364
1365 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1366 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1367
1368 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1369
1370 type = job_type_collapse(type, unit);
1371
1372 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1373 if (!tr)
1374 return -ENOMEM;
1375
1376 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1377 mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1378 mode == JOB_IGNORE_DEPENDENCIES, e);
1379 if (r < 0)
1380 goto tr_abort;
1381
1382 if (mode == JOB_ISOLATE) {
1383 r = transaction_add_isolate_jobs(tr, m);
1384 if (r < 0)
1385 goto tr_abort;
1386 }
1387
1388 r = transaction_activate(tr, m, mode, e);
1389 if (r < 0)
1390 goto tr_abort;
1391
1392 log_unit_debug(unit,
1393 "Enqueued job %s/%s as %u", unit->id,
1394 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1395
1396 if (_ret)
1397 *_ret = tr->anchor_job;
1398
1399 transaction_free(tr);
1400 return 0;
1401
1402 tr_abort:
1403 transaction_abort(tr);
1404 transaction_free(tr);
1405 return r;
1406 }
1407
1408 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1409 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1410 int r;
1411
1412 assert(m);
1413 assert(type < _JOB_TYPE_MAX);
1414 assert(name);
1415 assert(mode < _JOB_MODE_MAX);
1416
1417 r = manager_load_unit(m, name, NULL, NULL, &unit);
1418 if (r < 0)
1419 return r;
1420 assert(unit);
1421
1422 return manager_add_job(m, type, unit, mode, e, ret);
1423 }
1424
1425 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1426 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1427 int r;
1428
1429 assert(m);
1430 assert(type < _JOB_TYPE_MAX);
1431 assert(name);
1432 assert(mode < _JOB_MODE_MAX);
1433
1434 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1435 if (r < 0)
1436 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1437
1438 return r;
1439 }
1440
1441 Job *manager_get_job(Manager *m, uint32_t id) {
1442 assert(m);
1443
1444 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1445 }
1446
1447 Unit *manager_get_unit(Manager *m, const char *name) {
1448 assert(m);
1449 assert(name);
1450
1451 return hashmap_get(m->units, name);
1452 }
1453
1454 unsigned manager_dispatch_load_queue(Manager *m) {
1455 Unit *u;
1456 unsigned n = 0;
1457
1458 assert(m);
1459
1460 /* Make sure we are not run recursively */
1461 if (m->dispatching_load_queue)
1462 return 0;
1463
1464 m->dispatching_load_queue = true;
1465
1466 /* Dispatches the load queue. Takes a unit from the queue and
1467 * tries to load its data until the queue is empty */
1468
1469 while ((u = m->load_queue)) {
1470 assert(u->in_load_queue);
1471
1472 unit_load(u);
1473 n++;
1474 }
1475
1476 m->dispatching_load_queue = false;
1477 return n;
1478 }
1479
1480 int manager_load_unit_prepare(
1481 Manager *m,
1482 const char *name,
1483 const char *path,
1484 sd_bus_error *e,
1485 Unit **_ret) {
1486
1487 Unit *ret;
1488 UnitType t;
1489 int r;
1490
1491 assert(m);
1492 assert(name || path);
1493 assert(_ret);
1494
1495 /* This will prepare the unit for loading, but not actually
1496 * load anything from disk. */
1497
1498 if (path && !is_path(path))
1499 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1500
1501 if (!name)
1502 name = basename(path);
1503
1504 t = unit_name_to_type(name);
1505
1506 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1507 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1508 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1509
1510 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1511 }
1512
1513 ret = manager_get_unit(m, name);
1514 if (ret) {
1515 *_ret = ret;
1516 return 1;
1517 }
1518
1519 ret = unit_new(m, unit_vtable[t]->object_size);
1520 if (!ret)
1521 return -ENOMEM;
1522
1523 if (path) {
1524 ret->fragment_path = strdup(path);
1525 if (!ret->fragment_path) {
1526 unit_free(ret);
1527 return -ENOMEM;
1528 }
1529 }
1530
1531 r = unit_add_name(ret, name);
1532 if (r < 0) {
1533 unit_free(ret);
1534 return r;
1535 }
1536
1537 unit_add_to_load_queue(ret);
1538 unit_add_to_dbus_queue(ret);
1539 unit_add_to_gc_queue(ret);
1540
1541 *_ret = ret;
1542
1543 return 0;
1544 }
1545
1546 int manager_load_unit(
1547 Manager *m,
1548 const char *name,
1549 const char *path,
1550 sd_bus_error *e,
1551 Unit **_ret) {
1552
1553 int r;
1554
1555 assert(m);
1556 assert(_ret);
1557
1558 /* This will load the service information files, but not actually
1559 * start any services or anything. */
1560
1561 r = manager_load_unit_prepare(m, name, path, e, _ret);
1562 if (r != 0)
1563 return r;
1564
1565 manager_dispatch_load_queue(m);
1566
1567 *_ret = unit_follow_merge(*_ret);
1568
1569 return 0;
1570 }
1571
1572 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1573 Iterator i;
1574 Job *j;
1575
1576 assert(s);
1577 assert(f);
1578
1579 HASHMAP_FOREACH(j, s->jobs, i)
1580 job_dump(j, f, prefix);
1581 }
1582
1583 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1584 Iterator i;
1585 Unit *u;
1586 const char *t;
1587
1588 assert(s);
1589 assert(f);
1590
1591 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1592 if (u->id == t)
1593 unit_dump(u, f, prefix);
1594 }
1595
1596 void manager_clear_jobs(Manager *m) {
1597 Job *j;
1598
1599 assert(m);
1600
1601 while ((j = hashmap_first(m->jobs)))
1602 /* No need to recurse. We're cancelling all jobs. */
1603 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1604 }
1605
1606 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1607 Manager *m = userdata;
1608 Job *j;
1609
1610 assert(source);
1611 assert(m);
1612
1613 while ((j = m->run_queue)) {
1614 assert(j->installed);
1615 assert(j->in_run_queue);
1616
1617 job_run_and_invalidate(j);
1618 }
1619
1620 if (m->n_running_jobs > 0)
1621 manager_watch_jobs_in_progress(m);
1622
1623 if (m->n_on_console > 0)
1624 manager_watch_idle_pipe(m);
1625
1626 return 1;
1627 }
1628
1629 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1630 Job *j;
1631 Unit *u;
1632 unsigned n = 0;
1633
1634 assert(m);
1635
1636 if (m->dispatching_dbus_queue)
1637 return 0;
1638
1639 m->dispatching_dbus_queue = true;
1640
1641 while ((u = m->dbus_unit_queue)) {
1642 assert(u->in_dbus_queue);
1643
1644 bus_unit_send_change_signal(u);
1645 n++;
1646 }
1647
1648 while ((j = m->dbus_job_queue)) {
1649 assert(j->in_dbus_queue);
1650
1651 bus_job_send_change_signal(j);
1652 n++;
1653 }
1654
1655 m->dispatching_dbus_queue = false;
1656
1657 if (m->send_reloading_done) {
1658 m->send_reloading_done = false;
1659
1660 bus_manager_send_reloading(m, false);
1661 }
1662
1663 if (m->queued_message)
1664 bus_send_queued_message(m);
1665
1666 return n;
1667 }
1668
1669 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1670 Manager *m = userdata;
1671 char buf[PATH_MAX+1];
1672 ssize_t n;
1673
1674 n = recv(fd, buf, sizeof(buf), 0);
1675 if (n < 0)
1676 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1677 if (n == 0) {
1678 log_error("Got zero-length cgroups agent message, ignoring.");
1679 return 0;
1680 }
1681 if ((size_t) n >= sizeof(buf)) {
1682 log_error("Got overly long cgroups agent message, ignoring.");
1683 return 0;
1684 }
1685
1686 if (memchr(buf, 0, n)) {
1687 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1688 return 0;
1689 }
1690 buf[n] = 0;
1691
1692 manager_notify_cgroup_empty(m, buf);
1693 bus_forward_agent_released(m, buf);
1694
1695 return 0;
1696 }
1697
1698 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1699 _cleanup_strv_free_ char **tags = NULL;
1700
1701 assert(m);
1702 assert(u);
1703 assert(buf);
1704
1705 tags = strv_split(buf, "\n\r");
1706 if (!tags) {
1707 log_oom();
1708 return;
1709 }
1710
1711 if (UNIT_VTABLE(u)->notify_message)
1712 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1713 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1714 _cleanup_free_ char *x = NULL, *y = NULL;
1715
1716 x = cescape(buf);
1717 if (x)
1718 y = ellipsize(x, 20, 90);
1719 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1720 }
1721 }
1722
1723 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1724
1725 _cleanup_fdset_free_ FDSet *fds = NULL;
1726 Manager *m = userdata;
1727 char buf[NOTIFY_BUFFER_MAX+1];
1728 struct iovec iovec = {
1729 .iov_base = buf,
1730 .iov_len = sizeof(buf)-1,
1731 };
1732 union {
1733 struct cmsghdr cmsghdr;
1734 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1735 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1736 } control = {};
1737 struct msghdr msghdr = {
1738 .msg_iov = &iovec,
1739 .msg_iovlen = 1,
1740 .msg_control = &control,
1741 .msg_controllen = sizeof(control),
1742 };
1743
1744 struct cmsghdr *cmsg;
1745 struct ucred *ucred = NULL;
1746 Unit *u1, *u2, *u3;
1747 int r, *fd_array = NULL;
1748 unsigned n_fds = 0;
1749 ssize_t n;
1750
1751 assert(m);
1752 assert(m->notify_fd == fd);
1753
1754 if (revents != EPOLLIN) {
1755 log_warning("Got unexpected poll event for notify fd.");
1756 return 0;
1757 }
1758
1759 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1760 if (n < 0) {
1761 if (IN_SET(errno, EAGAIN, EINTR))
1762 return 0; /* Spurious wakeup, try again */
1763
1764 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1765 * won't take notification messages anymore, but that's still better than busy looping around this:
1766 * being woken up over and over again but being unable to actually read the message off the socket. */
1767 return log_error_errno(errno, "Failed to receive notification message: %m");
1768 }
1769
1770 CMSG_FOREACH(cmsg, &msghdr) {
1771 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1772
1773 fd_array = (int*) CMSG_DATA(cmsg);
1774 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1775
1776 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1777 cmsg->cmsg_type == SCM_CREDENTIALS &&
1778 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1779
1780 ucred = (struct ucred*) CMSG_DATA(cmsg);
1781 }
1782 }
1783
1784 if (n_fds > 0) {
1785 assert(fd_array);
1786
1787 r = fdset_new_array(&fds, fd_array, n_fds);
1788 if (r < 0) {
1789 close_many(fd_array, n_fds);
1790 log_oom();
1791 return 0;
1792 }
1793 }
1794
1795 if (!ucred || ucred->pid <= 0) {
1796 log_warning("Received notify message without valid credentials. Ignoring.");
1797 return 0;
1798 }
1799
1800 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1801 log_warning("Received notify message exceeded maximum size. Ignoring.");
1802 return 0;
1803 }
1804
1805 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1806 * trailing NUL byte in the message, but don't expect it. */
1807 if (n > 1 && memchr(buf, 0, n-1)) {
1808 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1809 return 0;
1810 }
1811
1812 /* Make sure it's NUL-terminated. */
1813 buf[n] = 0;
1814
1815 /* Notify every unit that might be interested, but try
1816 * to avoid notifying the same one multiple times. */
1817 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1818 if (u1)
1819 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1820
1821 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1822 if (u2 && u2 != u1)
1823 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1824
1825 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1826 if (u3 && u3 != u2 && u3 != u1)
1827 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1828
1829 if (!u1 && !u2 && !u3)
1830 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1831
1832 if (fdset_size(fds) > 0)
1833 log_warning("Got extra auxiliary fds with notification message, closing them.");
1834
1835 return 0;
1836 }
1837
1838 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1839 uint64_t iteration;
1840
1841 assert(m);
1842 assert(u);
1843 assert(si);
1844
1845 sd_event_get_iteration(m->event, &iteration);
1846
1847 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1848
1849 unit_unwatch_pid(u, si->si_pid);
1850
1851 if (UNIT_VTABLE(u)->sigchld_event) {
1852 if (set_size(u->pids) <= 1 ||
1853 iteration != u->sigchldgen ||
1854 unit_main_pid(u) == si->si_pid ||
1855 unit_control_pid(u) == si->si_pid) {
1856 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1857 u->sigchldgen = iteration;
1858 } else
1859 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1860 }
1861 }
1862
1863 static int manager_dispatch_sigchld(Manager *m) {
1864 assert(m);
1865
1866 for (;;) {
1867 siginfo_t si = {};
1868
1869 /* First we call waitd() for a PID and do not reap the
1870 * zombie. That way we can still access /proc/$PID for
1871 * it while it is a zombie. */
1872 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1873
1874 if (errno == ECHILD)
1875 break;
1876
1877 if (errno == EINTR)
1878 continue;
1879
1880 return -errno;
1881 }
1882
1883 if (si.si_pid <= 0)
1884 break;
1885
1886 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1887 _cleanup_free_ char *name = NULL;
1888 Unit *u1, *u2, *u3;
1889
1890 get_process_comm(si.si_pid, &name);
1891
1892 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1893 si.si_pid, strna(name),
1894 sigchld_code_to_string(si.si_code),
1895 si.si_status,
1896 strna(si.si_code == CLD_EXITED
1897 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1898 : signal_to_string(si.si_status)));
1899
1900 /* And now figure out the unit this belongs
1901 * to, it might be multiple... */
1902 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1903 if (u1)
1904 invoke_sigchld_event(m, u1, &si);
1905 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1906 if (u2 && u2 != u1)
1907 invoke_sigchld_event(m, u2, &si);
1908 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1909 if (u3 && u3 != u2 && u3 != u1)
1910 invoke_sigchld_event(m, u3, &si);
1911 }
1912
1913 /* And now, we actually reap the zombie. */
1914 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1915 if (errno == EINTR)
1916 continue;
1917
1918 return -errno;
1919 }
1920 }
1921
1922 return 0;
1923 }
1924
1925 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1926 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1927 int r;
1928
1929 log_debug("Activating special unit %s", name);
1930
1931 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1932 if (r < 0)
1933 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1934
1935 return r;
1936 }
1937
1938 static void manager_handle_ctrl_alt_del(Manager *m) {
1939 /* If the user presses C-A-D more than
1940 * 7 times within 2s, we reboot/shutdown immediately,
1941 * unless it was disabled in system.conf */
1942
1943 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
1944 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1945 else
1946 emergency_action(m, m->cad_burst_action, NULL,
1947 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1948 }
1949
1950 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1951 Manager *m = userdata;
1952 ssize_t n;
1953 struct signalfd_siginfo sfsi;
1954 bool sigchld = false;
1955 int r;
1956
1957 assert(m);
1958 assert(m->signal_fd == fd);
1959
1960 if (revents != EPOLLIN) {
1961 log_warning("Got unexpected events from signal file descriptor.");
1962 return 0;
1963 }
1964
1965 for (;;) {
1966 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1967 if (n != sizeof(sfsi)) {
1968 if (n >= 0) {
1969 log_warning("Truncated read from signal fd (%zu bytes)!", n);
1970 return 0;
1971 }
1972
1973 if (IN_SET(errno, EINTR, EAGAIN))
1974 break;
1975
1976 /* We return an error here, which will kill this handler,
1977 * to avoid a busy loop on read error. */
1978 return log_error_errno(errno, "Reading from signal fd failed: %m");
1979 }
1980
1981 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1982 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1983 ? LOG_DEBUG : LOG_INFO,
1984 &sfsi);
1985
1986 switch (sfsi.ssi_signo) {
1987
1988 case SIGCHLD:
1989 sigchld = true;
1990 break;
1991
1992 case SIGTERM:
1993 if (MANAGER_IS_SYSTEM(m)) {
1994 /* This is for compatibility with the
1995 * original sysvinit */
1996 r = verify_run_space_and_log("Refusing to reexecute");
1997 if (r >= 0)
1998 m->exit_code = MANAGER_REEXECUTE;
1999 break;
2000 }
2001
2002 /* Fall through */
2003
2004 case SIGINT:
2005 if (MANAGER_IS_SYSTEM(m)) {
2006 manager_handle_ctrl_alt_del(m);
2007 break;
2008 }
2009
2010 /* Run the exit target if there is one, if not, just exit. */
2011 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
2012 m->exit_code = MANAGER_EXIT;
2013 return 0;
2014 }
2015
2016 break;
2017
2018 case SIGWINCH:
2019 if (MANAGER_IS_SYSTEM(m))
2020 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2021
2022 /* This is a nop on non-init */
2023 break;
2024
2025 case SIGPWR:
2026 if (MANAGER_IS_SYSTEM(m))
2027 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2028
2029 /* This is a nop on non-init */
2030 break;
2031
2032 case SIGUSR1: {
2033 Unit *u;
2034
2035 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2036
2037 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2038 log_info("Trying to reconnect to bus...");
2039 bus_init(m, true);
2040 }
2041
2042 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2043 log_info("Loading D-Bus service...");
2044 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2045 }
2046
2047 break;
2048 }
2049
2050 case SIGUSR2: {
2051 _cleanup_free_ char *dump = NULL;
2052 _cleanup_fclose_ FILE *f = NULL;
2053 size_t size;
2054
2055 f = open_memstream(&dump, &size);
2056 if (!f) {
2057 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2058 break;
2059 }
2060
2061 manager_dump_units(m, f, "\t");
2062 manager_dump_jobs(m, f, "\t");
2063
2064 r = fflush_and_check(f);
2065 if (r < 0) {
2066 log_warning_errno(r, "Failed to write status stream: %m");
2067 break;
2068 }
2069
2070 log_dump(LOG_INFO, dump);
2071 break;
2072 }
2073
2074 case SIGHUP:
2075 r = verify_run_space_and_log("Refusing to reload");
2076 if (r >= 0)
2077 m->exit_code = MANAGER_RELOAD;
2078 break;
2079
2080 default: {
2081
2082 /* Starting SIGRTMIN+0 */
2083 static const char * const target_table[] = {
2084 [0] = SPECIAL_DEFAULT_TARGET,
2085 [1] = SPECIAL_RESCUE_TARGET,
2086 [2] = SPECIAL_EMERGENCY_TARGET,
2087 [3] = SPECIAL_HALT_TARGET,
2088 [4] = SPECIAL_POWEROFF_TARGET,
2089 [5] = SPECIAL_REBOOT_TARGET,
2090 [6] = SPECIAL_KEXEC_TARGET
2091 };
2092
2093 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2094 static const ManagerExitCode code_table[] = {
2095 [0] = MANAGER_HALT,
2096 [1] = MANAGER_POWEROFF,
2097 [2] = MANAGER_REBOOT,
2098 [3] = MANAGER_KEXEC
2099 };
2100
2101 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2102 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2103 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2104 manager_start_target(m, target_table[idx],
2105 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2106 break;
2107 }
2108
2109 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2110 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2111 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2112 break;
2113 }
2114
2115 switch (sfsi.ssi_signo - SIGRTMIN) {
2116
2117 case 20:
2118 manager_set_show_status(m, SHOW_STATUS_YES);
2119 break;
2120
2121 case 21:
2122 manager_set_show_status(m, SHOW_STATUS_NO);
2123 break;
2124
2125 case 22:
2126 log_set_max_level(LOG_DEBUG);
2127 log_info("Setting log level to debug.");
2128 break;
2129
2130 case 23:
2131 log_set_max_level(LOG_INFO);
2132 log_info("Setting log level to info.");
2133 break;
2134
2135 case 24:
2136 if (MANAGER_IS_USER(m)) {
2137 m->exit_code = MANAGER_EXIT;
2138 return 0;
2139 }
2140
2141 /* This is a nop on init */
2142 break;
2143
2144 case 26:
2145 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2146 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2147 log_notice("Setting log target to journal-or-kmsg.");
2148 break;
2149
2150 case 27:
2151 log_set_target(LOG_TARGET_CONSOLE);
2152 log_notice("Setting log target to console.");
2153 break;
2154
2155 case 28:
2156 log_set_target(LOG_TARGET_KMSG);
2157 log_notice("Setting log target to kmsg.");
2158 break;
2159
2160 default:
2161 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2162 }
2163 }
2164 }
2165 }
2166
2167 if (sigchld)
2168 manager_dispatch_sigchld(m);
2169
2170 return 0;
2171 }
2172
2173 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2174 Manager *m = userdata;
2175 Iterator i;
2176 Unit *u;
2177
2178 assert(m);
2179 assert(m->time_change_fd == fd);
2180
2181 log_struct(LOG_DEBUG,
2182 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2183 LOG_MESSAGE("Time has been changed"),
2184 NULL);
2185
2186 /* Restart the watch */
2187 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2188 m->time_change_fd = safe_close(m->time_change_fd);
2189
2190 manager_setup_time_change(m);
2191
2192 HASHMAP_FOREACH(u, m->units, i)
2193 if (UNIT_VTABLE(u)->time_change)
2194 UNIT_VTABLE(u)->time_change(u);
2195
2196 return 0;
2197 }
2198
2199 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2200 Manager *m = userdata;
2201
2202 assert(m);
2203 assert(m->idle_pipe[2] == fd);
2204
2205 m->no_console_output = m->n_on_console > 0;
2206
2207 manager_close_idle_pipe(m);
2208
2209 return 0;
2210 }
2211
2212 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2213 Manager *m = userdata;
2214 int r;
2215 uint64_t next;
2216
2217 assert(m);
2218 assert(source);
2219
2220 manager_print_jobs_in_progress(m);
2221
2222 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2223 r = sd_event_source_set_time(source, next);
2224 if (r < 0)
2225 return r;
2226
2227 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2228 }
2229
2230 int manager_loop(Manager *m) {
2231 int r;
2232
2233 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2234
2235 assert(m);
2236 m->exit_code = MANAGER_OK;
2237
2238 /* Release the path cache */
2239 m->unit_path_cache = set_free_free(m->unit_path_cache);
2240
2241 manager_check_finished(m);
2242
2243 /* There might still be some zombies hanging around from
2244 * before we were exec()'ed. Let's reap them. */
2245 r = manager_dispatch_sigchld(m);
2246 if (r < 0)
2247 return r;
2248
2249 while (m->exit_code == MANAGER_OK) {
2250 usec_t wait_usec;
2251
2252 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2253 watchdog_ping();
2254
2255 if (!ratelimit_test(&rl)) {
2256 /* Yay, something is going seriously wrong, pause a little */
2257 log_warning("Looping too fast. Throttling execution a little.");
2258 sleep(1);
2259 }
2260
2261 if (manager_dispatch_load_queue(m) > 0)
2262 continue;
2263
2264 if (manager_dispatch_gc_job_queue(m) > 0)
2265 continue;
2266
2267 if (manager_dispatch_gc_unit_queue(m) > 0)
2268 continue;
2269
2270 if (manager_dispatch_cleanup_queue(m) > 0)
2271 continue;
2272
2273 if (manager_dispatch_cgroup_queue(m) > 0)
2274 continue;
2275
2276 if (manager_dispatch_dbus_queue(m) > 0)
2277 continue;
2278
2279 /* Sleep for half the watchdog time */
2280 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2281 wait_usec = m->runtime_watchdog / 2;
2282 if (wait_usec <= 0)
2283 wait_usec = 1;
2284 } else
2285 wait_usec = USEC_INFINITY;
2286
2287 r = sd_event_run(m->event, wait_usec);
2288 if (r < 0)
2289 return log_error_errno(r, "Failed to run event loop: %m");
2290 }
2291
2292 return m->exit_code;
2293 }
2294
2295 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2296 _cleanup_free_ char *n = NULL;
2297 sd_id128_t invocation_id;
2298 Unit *u;
2299 int r;
2300
2301 assert(m);
2302 assert(s);
2303 assert(_u);
2304
2305 r = unit_name_from_dbus_path(s, &n);
2306 if (r < 0)
2307 return r;
2308
2309 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2310 * as invocation ID. */
2311 r = sd_id128_from_string(n, &invocation_id);
2312 if (r >= 0) {
2313 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2314 if (u) {
2315 *_u = u;
2316 return 0;
2317 }
2318
2319 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));
2320 }
2321
2322 /* If this didn't work, we use the suffix as unit name. */
2323 r = manager_load_unit(m, n, NULL, e, &u);
2324 if (r < 0)
2325 return r;
2326
2327 *_u = u;
2328 return 0;
2329 }
2330
2331 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2332 const char *p;
2333 unsigned id;
2334 Job *j;
2335 int r;
2336
2337 assert(m);
2338 assert(s);
2339 assert(_j);
2340
2341 p = startswith(s, "/org/freedesktop/systemd1/job/");
2342 if (!p)
2343 return -EINVAL;
2344
2345 r = safe_atou(p, &id);
2346 if (r < 0)
2347 return r;
2348
2349 j = manager_get_job(m, id);
2350 if (!j)
2351 return -ENOENT;
2352
2353 *_j = j;
2354
2355 return 0;
2356 }
2357
2358 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2359
2360 #ifdef HAVE_AUDIT
2361 _cleanup_free_ char *p = NULL;
2362 const char *msg;
2363 int audit_fd, r;
2364
2365 if (!MANAGER_IS_SYSTEM(m))
2366 return;
2367
2368 audit_fd = get_audit_fd();
2369 if (audit_fd < 0)
2370 return;
2371
2372 /* Don't generate audit events if the service was already
2373 * started and we're just deserializing */
2374 if (MANAGER_IS_RELOADING(m))
2375 return;
2376
2377 if (u->type != UNIT_SERVICE)
2378 return;
2379
2380 r = unit_name_to_prefix_and_instance(u->id, &p);
2381 if (r < 0) {
2382 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2383 return;
2384 }
2385
2386 msg = strjoina("unit=", p);
2387 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2388 if (errno == EPERM)
2389 /* We aren't allowed to send audit messages?
2390 * Then let's not retry again. */
2391 close_audit_fd();
2392 else
2393 log_warning_errno(errno, "Failed to send audit message: %m");
2394 }
2395 #endif
2396
2397 }
2398
2399 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2400 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2401 _cleanup_free_ char *message = NULL;
2402 _cleanup_close_ int fd = -1;
2403 int n = 0;
2404
2405 /* Don't generate plymouth events if the service was already
2406 * started and we're just deserializing */
2407 if (MANAGER_IS_RELOADING(m))
2408 return;
2409
2410 if (!MANAGER_IS_SYSTEM(m))
2411 return;
2412
2413 if (detect_container() > 0)
2414 return;
2415
2416 if (u->type != UNIT_SERVICE &&
2417 u->type != UNIT_MOUNT &&
2418 u->type != UNIT_SWAP)
2419 return;
2420
2421 /* We set SOCK_NONBLOCK here so that we rather drop the
2422 * message then wait for plymouth */
2423 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2424 if (fd < 0) {
2425 log_error_errno(errno, "socket() failed: %m");
2426 return;
2427 }
2428
2429 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2430
2431 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2432 log_error_errno(errno, "connect() failed: %m");
2433 return;
2434 }
2435
2436 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2437 log_oom();
2438 return;
2439 }
2440
2441 errno = 0;
2442 if (write(fd, message, n + 1) != n + 1)
2443 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2444 log_error_errno(errno, "Failed to write Plymouth message: %m");
2445 }
2446
2447 int manager_open_serialization(Manager *m, FILE **_f) {
2448 int fd;
2449 FILE *f;
2450
2451 assert(_f);
2452
2453 fd = open_serialization_fd("systemd-state");
2454 if (fd < 0)
2455 return fd;
2456
2457 f = fdopen(fd, "w+");
2458 if (!f) {
2459 safe_close(fd);
2460 return -errno;
2461 }
2462
2463 *_f = f;
2464 return 0;
2465 }
2466
2467 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2468 Iterator i;
2469 Unit *u;
2470 const char *t;
2471 int r;
2472
2473 assert(m);
2474 assert(f);
2475 assert(fds);
2476
2477 m->n_reloading++;
2478
2479 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2480 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2481 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2482 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2483
2484 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2485 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2486 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2487 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2488
2489 if (!in_initrd()) {
2490 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2491 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2492 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2493 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2494 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2495 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2496 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2497 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2498 }
2499
2500 if (!switching_root)
2501 (void) serialize_environment(f, m->environment);
2502
2503 if (m->notify_fd >= 0) {
2504 int copy;
2505
2506 copy = fdset_put_dup(fds, m->notify_fd);
2507 if (copy < 0)
2508 return copy;
2509
2510 fprintf(f, "notify-fd=%i\n", copy);
2511 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2512 }
2513
2514 if (m->cgroups_agent_fd >= 0) {
2515 int copy;
2516
2517 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2518 if (copy < 0)
2519 return copy;
2520
2521 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2522 }
2523
2524 if (m->user_lookup_fds[0] >= 0) {
2525 int copy0, copy1;
2526
2527 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2528 if (copy0 < 0)
2529 return copy0;
2530
2531 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2532 if (copy1 < 0)
2533 return copy1;
2534
2535 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2536 }
2537
2538 bus_track_serialize(m->subscribed, f, "subscribed");
2539
2540 r = dynamic_user_serialize(m, f, fds);
2541 if (r < 0)
2542 return r;
2543
2544 manager_serialize_uid_refs(m, f);
2545 manager_serialize_gid_refs(m, f);
2546
2547 fputc('\n', f);
2548
2549 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2550 if (u->id != t)
2551 continue;
2552
2553 /* Start marker */
2554 fputs(u->id, f);
2555 fputc('\n', f);
2556
2557 r = unit_serialize(u, f, fds, !switching_root);
2558 if (r < 0) {
2559 m->n_reloading--;
2560 return r;
2561 }
2562 }
2563
2564 assert(m->n_reloading > 0);
2565 m->n_reloading--;
2566
2567 if (ferror(f))
2568 return -EIO;
2569
2570 r = bus_fdset_add_all(m, fds);
2571 if (r < 0)
2572 return r;
2573
2574 return 0;
2575 }
2576
2577 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2578 int r = 0;
2579
2580 assert(m);
2581 assert(f);
2582
2583 log_debug("Deserializing state...");
2584
2585 m->n_reloading++;
2586
2587 for (;;) {
2588 char line[LINE_MAX], *l;
2589 const char *val;
2590
2591 if (!fgets(line, sizeof(line), f)) {
2592 if (feof(f))
2593 r = 0;
2594 else
2595 r = -errno;
2596
2597 goto finish;
2598 }
2599
2600 char_array_0(line);
2601 l = strstrip(line);
2602
2603 if (l[0] == 0)
2604 break;
2605
2606 if ((val = startswith(l, "current-job-id="))) {
2607 uint32_t id;
2608
2609 if (safe_atou32(val, &id) < 0)
2610 log_debug("Failed to parse current job id value %s", val);
2611 else
2612 m->current_job_id = MAX(m->current_job_id, id);
2613
2614 } else if ((val = startswith(l, "n-installed-jobs="))) {
2615 uint32_t n;
2616
2617 if (safe_atou32(val, &n) < 0)
2618 log_debug("Failed to parse installed jobs counter %s", val);
2619 else
2620 m->n_installed_jobs += n;
2621
2622 } else if ((val = startswith(l, "n-failed-jobs="))) {
2623 uint32_t n;
2624
2625 if (safe_atou32(val, &n) < 0)
2626 log_debug("Failed to parse failed jobs counter %s", val);
2627 else
2628 m->n_failed_jobs += n;
2629
2630 } else if ((val = startswith(l, "taint-usr="))) {
2631 int b;
2632
2633 b = parse_boolean(val);
2634 if (b < 0)
2635 log_debug("Failed to parse taint /usr flag %s", val);
2636 else
2637 m->taint_usr = m->taint_usr || b;
2638
2639 } else if ((val = startswith(l, "firmware-timestamp=")))
2640 dual_timestamp_deserialize(val, &m->firmware_timestamp);
2641 else if ((val = startswith(l, "loader-timestamp=")))
2642 dual_timestamp_deserialize(val, &m->loader_timestamp);
2643 else if ((val = startswith(l, "kernel-timestamp=")))
2644 dual_timestamp_deserialize(val, &m->kernel_timestamp);
2645 else if ((val = startswith(l, "initrd-timestamp=")))
2646 dual_timestamp_deserialize(val, &m->initrd_timestamp);
2647 else if ((val = startswith(l, "userspace-timestamp=")))
2648 dual_timestamp_deserialize(val, &m->userspace_timestamp);
2649 else if ((val = startswith(l, "finish-timestamp=")))
2650 dual_timestamp_deserialize(val, &m->finish_timestamp);
2651 else if ((val = startswith(l, "security-start-timestamp=")))
2652 dual_timestamp_deserialize(val, &m->security_start_timestamp);
2653 else if ((val = startswith(l, "security-finish-timestamp=")))
2654 dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2655 else if ((val = startswith(l, "generators-start-timestamp=")))
2656 dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2657 else if ((val = startswith(l, "generators-finish-timestamp=")))
2658 dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2659 else if ((val = startswith(l, "units-load-start-timestamp=")))
2660 dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2661 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2662 dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2663 else if (startswith(l, "env=")) {
2664 r = deserialize_environment(&m->environment, l);
2665 if (r < 0)
2666 return r;
2667
2668 } else if ((val = startswith(l, "notify-fd="))) {
2669 int fd;
2670
2671 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2672 log_debug("Failed to parse notify fd: %s", val);
2673 else {
2674 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2675 safe_close(m->notify_fd);
2676 m->notify_fd = fdset_remove(fds, fd);
2677 }
2678
2679 } else if ((val = startswith(l, "notify-socket="))) {
2680 char *n;
2681
2682 n = strdup(val);
2683 if (!n) {
2684 r = -ENOMEM;
2685 goto finish;
2686 }
2687
2688 free(m->notify_socket);
2689 m->notify_socket = n;
2690
2691 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2692 int fd;
2693
2694 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2695 log_debug("Failed to parse cgroups agent fd: %s", val);
2696 else {
2697 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2698 safe_close(m->cgroups_agent_fd);
2699 m->cgroups_agent_fd = fdset_remove(fds, fd);
2700 }
2701
2702 } else if ((val = startswith(l, "user-lookup="))) {
2703 int fd0, fd1;
2704
2705 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2706 log_debug("Failed to parse user lookup fd: %s", val);
2707 else {
2708 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2709 safe_close_pair(m->user_lookup_fds);
2710 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2711 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2712 }
2713
2714 } else if ((val = startswith(l, "dynamic-user=")))
2715 dynamic_user_deserialize_one(m, val, fds);
2716 else if ((val = startswith(l, "destroy-ipc-uid=")))
2717 manager_deserialize_uid_refs_one(m, val);
2718 else if ((val = startswith(l, "destroy-ipc-gid=")))
2719 manager_deserialize_gid_refs_one(m, val);
2720 else if ((val = startswith(l, "subscribed="))) {
2721
2722 if (strv_extend(&m->deserialized_subscribed, val) < 0)
2723 log_oom();
2724
2725 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2726 log_debug("Unknown serialization item '%s'", l);
2727 }
2728
2729 for (;;) {
2730 Unit *u;
2731 char name[UNIT_NAME_MAX+2];
2732
2733 /* Start marker */
2734 if (!fgets(name, sizeof(name), f)) {
2735 if (feof(f))
2736 r = 0;
2737 else
2738 r = -errno;
2739
2740 goto finish;
2741 }
2742
2743 char_array_0(name);
2744
2745 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2746 if (r < 0)
2747 goto finish;
2748
2749 r = unit_deserialize(u, f, fds);
2750 if (r < 0)
2751 goto finish;
2752 }
2753
2754 finish:
2755 if (ferror(f))
2756 r = -EIO;
2757
2758 assert(m->n_reloading > 0);
2759 m->n_reloading--;
2760
2761 return r;
2762 }
2763
2764 int manager_reload(Manager *m) {
2765 int r, q;
2766 _cleanup_fclose_ FILE *f = NULL;
2767 _cleanup_fdset_free_ FDSet *fds = NULL;
2768
2769 assert(m);
2770
2771 r = manager_open_serialization(m, &f);
2772 if (r < 0)
2773 return r;
2774
2775 m->n_reloading++;
2776 bus_manager_send_reloading(m, true);
2777
2778 fds = fdset_new();
2779 if (!fds) {
2780 m->n_reloading--;
2781 return -ENOMEM;
2782 }
2783
2784 r = manager_serialize(m, f, fds, false);
2785 if (r < 0) {
2786 m->n_reloading--;
2787 return r;
2788 }
2789
2790 if (fseeko(f, 0, SEEK_SET) < 0) {
2791 m->n_reloading--;
2792 return -errno;
2793 }
2794
2795 /* From here on there is no way back. */
2796 manager_clear_jobs_and_units(m);
2797 lookup_paths_flush_generator(&m->lookup_paths);
2798 lookup_paths_free(&m->lookup_paths);
2799 dynamic_user_vacuum(m, false);
2800 m->uid_refs = hashmap_free(m->uid_refs);
2801 m->gid_refs = hashmap_free(m->gid_refs);
2802
2803 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2804 if (q < 0 && r >= 0)
2805 r = q;
2806
2807 q = manager_run_environment_generators(m);
2808 if (q < 0 && r >= 0)
2809 r = q;
2810
2811 /* Find new unit paths */
2812 q = manager_run_generators(m);
2813 if (q < 0 && r >= 0)
2814 r = q;
2815
2816 lookup_paths_reduce(&m->lookup_paths);
2817 manager_build_unit_path_cache(m);
2818
2819 /* First, enumerate what we can from all config files */
2820 manager_enumerate(m);
2821
2822 /* Second, deserialize our stored data */
2823 q = manager_deserialize(m, f, fds);
2824 if (q < 0 && r >= 0)
2825 r = q;
2826
2827 fclose(f);
2828 f = NULL;
2829
2830 /* Re-register notify_fd as event source */
2831 q = manager_setup_notify(m);
2832 if (q < 0 && r >= 0)
2833 r = q;
2834
2835 q = manager_setup_cgroups_agent(m);
2836 if (q < 0 && r >= 0)
2837 r = q;
2838
2839 q = manager_setup_user_lookup_fd(m);
2840 if (q < 0 && r >= 0)
2841 r = q;
2842
2843 /* Third, fire things up! */
2844 manager_coldplug(m);
2845
2846 /* Release any dynamic users no longer referenced */
2847 dynamic_user_vacuum(m, true);
2848
2849 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2850 manager_vacuum_uid_refs(m);
2851 manager_vacuum_gid_refs(m);
2852
2853 /* Sync current state of bus names with our set of listening units */
2854 if (m->api_bus)
2855 manager_sync_bus_names(m, m->api_bus);
2856
2857 assert(m->n_reloading > 0);
2858 m->n_reloading--;
2859
2860 m->send_reloading_done = true;
2861
2862 return r;
2863 }
2864
2865 void manager_reset_failed(Manager *m) {
2866 Unit *u;
2867 Iterator i;
2868
2869 assert(m);
2870
2871 HASHMAP_FOREACH(u, m->units, i)
2872 unit_reset_failed(u);
2873 }
2874
2875 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2876 Unit *u;
2877
2878 assert(m);
2879 assert(name);
2880
2881 /* Returns true if the unit is inactive or going down */
2882 u = manager_get_unit(m, name);
2883 if (!u)
2884 return true;
2885
2886 return unit_inactive_or_pending(u);
2887 }
2888
2889 static void manager_notify_finished(Manager *m) {
2890 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2891 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2892
2893 if (m->test_run)
2894 return;
2895
2896 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2897
2898 /* Note that m->kernel_usec.monotonic is always at 0,
2899 * and m->firmware_usec.monotonic and
2900 * m->loader_usec.monotonic should be considered
2901 * negative values. */
2902
2903 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2904 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2905 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2906 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2907
2908 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2909
2910 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2911 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2912
2913 log_struct(LOG_INFO,
2914 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2915 "KERNEL_USEC="USEC_FMT, kernel_usec,
2916 "INITRD_USEC="USEC_FMT, initrd_usec,
2917 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2918 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2919 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2920 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2921 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2922 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2923 NULL);
2924 } else {
2925 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2926 initrd_usec = 0;
2927
2928 log_struct(LOG_INFO,
2929 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2930 "KERNEL_USEC="USEC_FMT, kernel_usec,
2931 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2932 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2933 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2934 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2935 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2936 NULL);
2937 }
2938 } else {
2939 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2940 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2941
2942 log_struct(LOG_INFO,
2943 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
2944 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2945 LOG_MESSAGE("Startup finished in %s.",
2946 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2947 NULL);
2948 }
2949
2950 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2951
2952 sd_notifyf(false,
2953 "READY=1\n"
2954 "STATUS=Startup finished in %s.",
2955 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2956 }
2957
2958 void manager_check_finished(Manager *m) {
2959 assert(m);
2960
2961 if (MANAGER_IS_RELOADING(m))
2962 return;
2963
2964 /* Verify that we are actually running currently. Initially
2965 * the exit code is set to invalid, and during operation it is
2966 * then set to MANAGER_OK */
2967 if (m->exit_code != MANAGER_OK)
2968 return;
2969
2970 if (hashmap_size(m->jobs) > 0) {
2971 if (m->jobs_in_progress_event_source)
2972 /* Ignore any failure, this is only for feedback */
2973 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2974
2975 return;
2976 }
2977
2978 manager_flip_auto_status(m, false);
2979
2980 /* Notify Type=idle units that we are done now */
2981 manager_close_idle_pipe(m);
2982
2983 /* Turn off confirm spawn now */
2984 m->confirm_spawn = NULL;
2985
2986 /* No need to update ask password status when we're going non-interactive */
2987 manager_close_ask_password(m);
2988
2989 /* This is no longer the first boot */
2990 manager_set_first_boot(m, false);
2991
2992 if (dual_timestamp_is_set(&m->finish_timestamp))
2993 return;
2994
2995 dual_timestamp_get(&m->finish_timestamp);
2996
2997 manager_notify_finished(m);
2998
2999 manager_invalidate_startup_units(m);
3000 }
3001
3002 static bool generator_path_any(const char* const* paths) {
3003 char **path;
3004 bool found = false;
3005
3006 /* Optimize by skipping the whole process by not creating output directories
3007 * if no generators are found. */
3008 STRV_FOREACH(path, (char**) paths)
3009 if (access(*path, F_OK) == 0)
3010 found = true;
3011 else if (errno != ENOENT)
3012 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3013
3014 return found;
3015 }
3016
3017 static const char* system_env_generator_binary_paths[] = {
3018 "/run/systemd/system-environment-generators",
3019 "/etc/systemd/system-environment-generators",
3020 "/usr/local/lib/systemd/system-environment-generators",
3021 SYSTEM_ENV_GENERATOR_PATH,
3022 NULL
3023 };
3024
3025 static const char* user_env_generator_binary_paths[] = {
3026 "/run/systemd/user-environment-generators",
3027 "/etc/systemd/user-environment-generators",
3028 "/usr/local/lib/systemd/user-environment-generators",
3029 USER_ENV_GENERATOR_PATH,
3030 NULL
3031 };
3032
3033 static int manager_run_environment_generators(Manager *m) {
3034 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3035 const char **paths;
3036 void* args[] = {&tmp, &tmp, &m->environment};
3037
3038 if (m->test_run)
3039 return 0;
3040
3041 paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths;
3042
3043 if (!generator_path_any(paths))
3044 return 0;
3045
3046 return execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, args, NULL);
3047 }
3048
3049 static int manager_run_generators(Manager *m) {
3050 _cleanup_strv_free_ char **paths = NULL;
3051 const char *argv[5];
3052 int r;
3053
3054 assert(m);
3055
3056 if (m->test_run)
3057 return 0;
3058
3059 paths = generator_binary_paths(m->unit_file_scope);
3060 if (!paths)
3061 return log_oom();
3062
3063 if (!generator_path_any((const char* const*) paths))
3064 return 0;
3065
3066 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3067 if (r < 0)
3068 goto finish;
3069
3070 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3071 argv[1] = m->lookup_paths.generator;
3072 argv[2] = m->lookup_paths.generator_early;
3073 argv[3] = m->lookup_paths.generator_late;
3074 argv[4] = NULL;
3075
3076 RUN_WITH_UMASK(0022)
3077 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC,
3078 NULL, NULL, (char**) argv);
3079
3080 finish:
3081 lookup_paths_trim_generator(&m->lookup_paths);
3082 return r;
3083 }
3084
3085 int manager_environment_add(Manager *m, char **minus, char **plus) {
3086 char **a = NULL, **b = NULL, **l;
3087 assert(m);
3088
3089 l = m->environment;
3090
3091 if (!strv_isempty(minus)) {
3092 a = strv_env_delete(l, 1, minus);
3093 if (!a)
3094 return -ENOMEM;
3095
3096 l = a;
3097 }
3098
3099 if (!strv_isempty(plus)) {
3100 b = strv_env_merge(2, l, plus);
3101 if (!b) {
3102 strv_free(a);
3103 return -ENOMEM;
3104 }
3105
3106 l = b;
3107 }
3108
3109 if (m->environment != l)
3110 strv_free(m->environment);
3111 if (a != l)
3112 strv_free(a);
3113 if (b != l)
3114 strv_free(b);
3115
3116 m->environment = l;
3117 manager_clean_environment(m);
3118 strv_sort(m->environment);
3119
3120 return 0;
3121 }
3122
3123 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3124 int i;
3125
3126 assert(m);
3127
3128 for (i = 0; i < _RLIMIT_MAX; i++) {
3129 m->rlimit[i] = mfree(m->rlimit[i]);
3130
3131 if (!default_rlimit[i])
3132 continue;
3133
3134 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3135 if (!m->rlimit[i])
3136 return log_oom();
3137 }
3138
3139 return 0;
3140 }
3141
3142 void manager_recheck_journal(Manager *m) {
3143 Unit *u;
3144
3145 assert(m);
3146
3147 if (!MANAGER_IS_SYSTEM(m))
3148 return;
3149
3150 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3151 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3152 log_close_journal();
3153 return;
3154 }
3155
3156 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3157 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3158 log_close_journal();
3159 return;
3160 }
3161
3162 /* Hmm, OK, so the socket is fully up and the service is up
3163 * too, then let's make use of the thing. */
3164 log_open();
3165 }
3166
3167 void manager_set_show_status(Manager *m, ShowStatus mode) {
3168 assert(m);
3169 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3170
3171 if (!MANAGER_IS_SYSTEM(m))
3172 return;
3173
3174 if (m->show_status != mode)
3175 log_debug("%s showing of status.",
3176 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3177 m->show_status = mode;
3178
3179 if (mode > 0)
3180 (void) touch("/run/systemd/show-status");
3181 else
3182 (void) unlink("/run/systemd/show-status");
3183 }
3184
3185 static bool manager_get_show_status(Manager *m, StatusType type) {
3186 assert(m);
3187
3188 if (!MANAGER_IS_SYSTEM(m))
3189 return false;
3190
3191 if (m->no_console_output)
3192 return false;
3193
3194 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3195 return false;
3196
3197 /* If we cannot find out the status properly, just proceed. */
3198 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3199 return false;
3200
3201 if (m->show_status > 0)
3202 return true;
3203
3204 return false;
3205 }
3206
3207 const char *manager_get_confirm_spawn(Manager *m) {
3208 static int last_errno = 0;
3209 const char *vc = m->confirm_spawn;
3210 struct stat st;
3211 int r;
3212
3213 /* Here's the deal: we want to test the validity of the console but don't want
3214 * PID1 to go through the whole console process which might block. But we also
3215 * want to warn the user only once if something is wrong with the console so we
3216 * cannot do the sanity checks after spawning our children. So here we simply do
3217 * really basic tests to hopefully trap common errors.
3218 *
3219 * If the console suddenly disappear at the time our children will really it
3220 * then they will simply fail to acquire it and a positive answer will be
3221 * assumed. New children will fallback to /dev/console though.
3222 *
3223 * Note: TTYs are devices that can come and go any time, and frequently aren't
3224 * available yet during early boot (consider a USB rs232 dongle...). If for any
3225 * reason the configured console is not ready, we fallback to the default
3226 * console. */
3227
3228 if (!vc || path_equal(vc, "/dev/console"))
3229 return vc;
3230
3231 r = stat(vc, &st);
3232 if (r < 0)
3233 goto fail;
3234
3235 if (!S_ISCHR(st.st_mode)) {
3236 errno = ENOTTY;
3237 goto fail;
3238 }
3239
3240 last_errno = 0;
3241 return vc;
3242 fail:
3243 if (last_errno != errno) {
3244 last_errno = errno;
3245 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3246 }
3247 return "/dev/console";
3248 }
3249
3250 void manager_set_first_boot(Manager *m, bool b) {
3251 assert(m);
3252
3253 if (!MANAGER_IS_SYSTEM(m))
3254 return;
3255
3256 if (m->first_boot != (int) b) {
3257 if (b)
3258 (void) touch("/run/systemd/first-boot");
3259 else
3260 (void) unlink("/run/systemd/first-boot");
3261 }
3262
3263 m->first_boot = b;
3264 }
3265
3266 void manager_disable_confirm_spawn(void) {
3267 (void) touch("/run/systemd/confirm_spawn_disabled");
3268 }
3269
3270 bool manager_is_confirm_spawn_disabled(Manager *m) {
3271 if (!m->confirm_spawn)
3272 return true;
3273
3274 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3275 }
3276
3277 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3278 va_list ap;
3279
3280 /* If m is NULL, assume we're after shutdown and let the messages through. */
3281
3282 if (m && !manager_get_show_status(m, type))
3283 return;
3284
3285 /* XXX We should totally drop the check for ephemeral here
3286 * and thus effectively make 'Type=idle' pointless. */
3287 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3288 return;
3289
3290 va_start(ap, format);
3291 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3292 va_end(ap);
3293 }
3294
3295 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3296 char p[strlen(path)+1];
3297
3298 assert(m);
3299 assert(path);
3300
3301 strcpy(p, path);
3302 path_kill_slashes(p);
3303
3304 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3305 }
3306
3307 const char *manager_get_runtime_prefix(Manager *m) {
3308 assert(m);
3309
3310 return MANAGER_IS_SYSTEM(m) ?
3311 "/run" :
3312 getenv("XDG_RUNTIME_DIR");
3313 }
3314
3315 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3316 unsigned size;
3317 int r;
3318
3319 assert(m);
3320 assert(u->manager == m);
3321
3322 size = set_size(m->failed_units);
3323
3324 if (failed) {
3325 r = set_ensure_allocated(&m->failed_units, NULL);
3326 if (r < 0)
3327 return log_oom();
3328
3329 if (set_put(m->failed_units, u) < 0)
3330 return log_oom();
3331 } else
3332 (void) set_remove(m->failed_units, u);
3333
3334 if (set_size(m->failed_units) != size)
3335 bus_manager_send_change_signal(m);
3336
3337 return 0;
3338 }
3339
3340 ManagerState manager_state(Manager *m) {
3341 Unit *u;
3342
3343 assert(m);
3344
3345 /* Did we ever finish booting? If not then we are still starting up */
3346 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3347
3348 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3349 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3350 return MANAGER_INITIALIZING;
3351
3352 return MANAGER_STARTING;
3353 }
3354
3355 /* Is the special shutdown target queued? If so, we are in shutdown state */
3356 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3357 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3358 return MANAGER_STOPPING;
3359
3360 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3361 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3362 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3363 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3364 return MANAGER_MAINTENANCE;
3365
3366 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3367 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3368 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3369 return MANAGER_MAINTENANCE;
3370
3371 /* Are there any failed units? If so, we are in degraded mode */
3372 if (set_size(m->failed_units) > 0)
3373 return MANAGER_DEGRADED;
3374
3375 return MANAGER_RUNNING;
3376 }
3377
3378 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3379
3380 static void manager_unref_uid_internal(
3381 Manager *m,
3382 Hashmap **uid_refs,
3383 uid_t uid,
3384 bool destroy_now,
3385 int (*_clean_ipc)(uid_t uid)) {
3386
3387 uint32_t c, n;
3388
3389 assert(m);
3390 assert(uid_refs);
3391 assert(uid_is_valid(uid));
3392 assert(_clean_ipc);
3393
3394 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3395 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3396 *
3397 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3398 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3399 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3400 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3401
3402 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3403 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3404
3405 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3406 return;
3407
3408 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3409
3410 n = c & ~DESTROY_IPC_FLAG;
3411 assert(n > 0);
3412 n--;
3413
3414 if (destroy_now && n == 0) {
3415 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3416
3417 if (c & DESTROY_IPC_FLAG) {
3418 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3419 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3420 uid);
3421 (void) _clean_ipc(uid);
3422 }
3423 } else {
3424 c = n | (c & DESTROY_IPC_FLAG);
3425 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3426 }
3427 }
3428
3429 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3430 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3431 }
3432
3433 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3434 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3435 }
3436
3437 static int manager_ref_uid_internal(
3438 Manager *m,
3439 Hashmap **uid_refs,
3440 uid_t uid,
3441 bool clean_ipc) {
3442
3443 uint32_t c, n;
3444 int r;
3445
3446 assert(m);
3447 assert(uid_refs);
3448 assert(uid_is_valid(uid));
3449
3450 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3451 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3452
3453 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3454 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3455
3456 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3457 return 0;
3458
3459 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3460 if (r < 0)
3461 return r;
3462
3463 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3464
3465 n = c & ~DESTROY_IPC_FLAG;
3466 n++;
3467
3468 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3469 return -EOVERFLOW;
3470
3471 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3472
3473 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3474 }
3475
3476 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3477 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3478 }
3479
3480 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3481 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3482 }
3483
3484 static void manager_vacuum_uid_refs_internal(
3485 Manager *m,
3486 Hashmap **uid_refs,
3487 int (*_clean_ipc)(uid_t uid)) {
3488
3489 Iterator i;
3490 void *p, *k;
3491
3492 assert(m);
3493 assert(uid_refs);
3494 assert(_clean_ipc);
3495
3496 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3497 uint32_t c, n;
3498 uid_t uid;
3499
3500 uid = PTR_TO_UID(k);
3501 c = PTR_TO_UINT32(p);
3502
3503 n = c & ~DESTROY_IPC_FLAG;
3504 if (n > 0)
3505 continue;
3506
3507 if (c & DESTROY_IPC_FLAG) {
3508 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3509 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3510 uid);
3511 (void) _clean_ipc(uid);
3512 }
3513
3514 assert_se(hashmap_remove(*uid_refs, k) == p);
3515 }
3516 }
3517
3518 void manager_vacuum_uid_refs(Manager *m) {
3519 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3520 }
3521
3522 void manager_vacuum_gid_refs(Manager *m) {
3523 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3524 }
3525
3526 static void manager_serialize_uid_refs_internal(
3527 Manager *m,
3528 FILE *f,
3529 Hashmap **uid_refs,
3530 const char *field_name) {
3531
3532 Iterator i;
3533 void *p, *k;
3534
3535 assert(m);
3536 assert(f);
3537 assert(uid_refs);
3538 assert(field_name);
3539
3540 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3541 * of it is better rebuild after a reload/reexec. */
3542
3543 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3544 uint32_t c;
3545 uid_t uid;
3546
3547 uid = PTR_TO_UID(k);
3548 c = PTR_TO_UINT32(p);
3549
3550 if (!(c & DESTROY_IPC_FLAG))
3551 continue;
3552
3553 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3554 }
3555 }
3556
3557 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3558 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3559 }
3560
3561 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3562 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3563 }
3564
3565 static void manager_deserialize_uid_refs_one_internal(
3566 Manager *m,
3567 Hashmap** uid_refs,
3568 const char *value) {
3569
3570 uid_t uid;
3571 uint32_t c;
3572 int r;
3573
3574 assert(m);
3575 assert(uid_refs);
3576 assert(value);
3577
3578 r = parse_uid(value, &uid);
3579 if (r < 0 || uid == 0) {
3580 log_debug("Unable to parse UID reference serialization");
3581 return;
3582 }
3583
3584 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3585 if (r < 0) {
3586 log_oom();
3587 return;
3588 }
3589
3590 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3591 if (c & DESTROY_IPC_FLAG)
3592 return;
3593
3594 c |= DESTROY_IPC_FLAG;
3595
3596 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3597 if (r < 0) {
3598 log_debug("Failed to add UID reference entry");
3599 return;
3600 }
3601 }
3602
3603 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3604 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3605 }
3606
3607 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3608 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3609 }
3610
3611 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3612 struct buffer {
3613 uid_t uid;
3614 gid_t gid;
3615 char unit_name[UNIT_NAME_MAX+1];
3616 } _packed_ buffer;
3617
3618 Manager *m = userdata;
3619 ssize_t l;
3620 size_t n;
3621 Unit *u;
3622
3623 assert_se(source);
3624 assert_se(m);
3625
3626 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3627 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3628 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3629
3630 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3631 if (l < 0) {
3632 if (errno == EINTR || errno == EAGAIN)
3633 return 0;
3634
3635 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3636 }
3637
3638 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3639 log_warning("Received too short user lookup message, ignoring.");
3640 return 0;
3641 }
3642
3643 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3644 log_warning("Received too long user lookup message, ignoring.");
3645 return 0;
3646 }
3647
3648 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3649 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3650 return 0;
3651 }
3652
3653 n = (size_t) l - offsetof(struct buffer, unit_name);
3654 if (memchr(buffer.unit_name, 0, n)) {
3655 log_warning("Received lookup message with embedded NUL character, ignoring.");
3656 return 0;
3657 }
3658
3659 buffer.unit_name[n] = 0;
3660 u = manager_get_unit(m, buffer.unit_name);
3661 if (!u) {
3662 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3663 return 0;
3664 }
3665
3666 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3667
3668 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3669 return 0;
3670 }
3671
3672 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3673 [MANAGER_INITIALIZING] = "initializing",
3674 [MANAGER_STARTING] = "starting",
3675 [MANAGER_RUNNING] = "running",
3676 [MANAGER_DEGRADED] = "degraded",
3677 [MANAGER_MAINTENANCE] = "maintenance",
3678 [MANAGER_STOPPING] = "stopping",
3679 };
3680
3681 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
Unnamed repository; edit this file 'description' to name the repository.