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