]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge pull request #5131 from keszybz/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 = NULL; /* just to appease gcc, initialization is not really necessary */
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 assert(unit);
1418
1419 return manager_add_job(m, type, unit, mode, e, ret);
1420 }
1421
1422 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1423 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1424 int r;
1425
1426 assert(m);
1427 assert(type < _JOB_TYPE_MAX);
1428 assert(name);
1429 assert(mode < _JOB_MODE_MAX);
1430
1431 r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1432 if (r < 0)
1433 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1434
1435 return r;
1436 }
1437
1438 Job *manager_get_job(Manager *m, uint32_t id) {
1439 assert(m);
1440
1441 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1442 }
1443
1444 Unit *manager_get_unit(Manager *m, const char *name) {
1445 assert(m);
1446 assert(name);
1447
1448 return hashmap_get(m->units, name);
1449 }
1450
1451 unsigned manager_dispatch_load_queue(Manager *m) {
1452 Unit *u;
1453 unsigned n = 0;
1454
1455 assert(m);
1456
1457 /* Make sure we are not run recursively */
1458 if (m->dispatching_load_queue)
1459 return 0;
1460
1461 m->dispatching_load_queue = true;
1462
1463 /* Dispatches the load queue. Takes a unit from the queue and
1464 * tries to load its data until the queue is empty */
1465
1466 while ((u = m->load_queue)) {
1467 assert(u->in_load_queue);
1468
1469 unit_load(u);
1470 n++;
1471 }
1472
1473 m->dispatching_load_queue = false;
1474 return n;
1475 }
1476
1477 int manager_load_unit_prepare(
1478 Manager *m,
1479 const char *name,
1480 const char *path,
1481 sd_bus_error *e,
1482 Unit **_ret) {
1483
1484 Unit *ret;
1485 UnitType t;
1486 int r;
1487
1488 assert(m);
1489 assert(name || path);
1490 assert(_ret);
1491
1492 /* This will prepare the unit for loading, but not actually
1493 * load anything from disk. */
1494
1495 if (path && !is_path(path))
1496 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1497
1498 if (!name)
1499 name = basename(path);
1500
1501 t = unit_name_to_type(name);
1502
1503 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1504 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1505 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1506
1507 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1508 }
1509
1510 ret = manager_get_unit(m, name);
1511 if (ret) {
1512 *_ret = ret;
1513 return 1;
1514 }
1515
1516 ret = unit_new(m, unit_vtable[t]->object_size);
1517 if (!ret)
1518 return -ENOMEM;
1519
1520 if (path) {
1521 ret->fragment_path = strdup(path);
1522 if (!ret->fragment_path) {
1523 unit_free(ret);
1524 return -ENOMEM;
1525 }
1526 }
1527
1528 r = unit_add_name(ret, name);
1529 if (r < 0) {
1530 unit_free(ret);
1531 return r;
1532 }
1533
1534 unit_add_to_load_queue(ret);
1535 unit_add_to_dbus_queue(ret);
1536 unit_add_to_gc_queue(ret);
1537
1538 *_ret = ret;
1539
1540 return 0;
1541 }
1542
1543 int manager_load_unit(
1544 Manager *m,
1545 const char *name,
1546 const char *path,
1547 sd_bus_error *e,
1548 Unit **_ret) {
1549
1550 int r;
1551
1552 assert(m);
1553 assert(_ret);
1554
1555 /* This will load the service information files, but not actually
1556 * start any services or anything. */
1557
1558 r = manager_load_unit_prepare(m, name, path, e, _ret);
1559 if (r != 0)
1560 return r;
1561
1562 manager_dispatch_load_queue(m);
1563
1564 *_ret = unit_follow_merge(*_ret);
1565
1566 return 0;
1567 }
1568
1569 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1570 Iterator i;
1571 Job *j;
1572
1573 assert(s);
1574 assert(f);
1575
1576 HASHMAP_FOREACH(j, s->jobs, i)
1577 job_dump(j, f, prefix);
1578 }
1579
1580 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1581 Iterator i;
1582 Unit *u;
1583 const char *t;
1584
1585 assert(s);
1586 assert(f);
1587
1588 HASHMAP_FOREACH_KEY(u, t, s->units, i)
1589 if (u->id == t)
1590 unit_dump(u, f, prefix);
1591 }
1592
1593 void manager_clear_jobs(Manager *m) {
1594 Job *j;
1595
1596 assert(m);
1597
1598 while ((j = hashmap_first(m->jobs)))
1599 /* No need to recurse. We're cancelling all jobs. */
1600 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1601 }
1602
1603 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1604 Manager *m = userdata;
1605 Job *j;
1606
1607 assert(source);
1608 assert(m);
1609
1610 while ((j = m->run_queue)) {
1611 assert(j->installed);
1612 assert(j->in_run_queue);
1613
1614 job_run_and_invalidate(j);
1615 }
1616
1617 if (m->n_running_jobs > 0)
1618 manager_watch_jobs_in_progress(m);
1619
1620 if (m->n_on_console > 0)
1621 manager_watch_idle_pipe(m);
1622
1623 return 1;
1624 }
1625
1626 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1627 Job *j;
1628 Unit *u;
1629 unsigned n = 0;
1630
1631 assert(m);
1632
1633 if (m->dispatching_dbus_queue)
1634 return 0;
1635
1636 m->dispatching_dbus_queue = true;
1637
1638 while ((u = m->dbus_unit_queue)) {
1639 assert(u->in_dbus_queue);
1640
1641 bus_unit_send_change_signal(u);
1642 n++;
1643 }
1644
1645 while ((j = m->dbus_job_queue)) {
1646 assert(j->in_dbus_queue);
1647
1648 bus_job_send_change_signal(j);
1649 n++;
1650 }
1651
1652 m->dispatching_dbus_queue = false;
1653
1654 if (m->send_reloading_done) {
1655 m->send_reloading_done = false;
1656
1657 bus_manager_send_reloading(m, false);
1658 }
1659
1660 if (m->queued_message)
1661 bus_send_queued_message(m);
1662
1663 return n;
1664 }
1665
1666 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1667 Manager *m = userdata;
1668 char buf[PATH_MAX+1];
1669 ssize_t n;
1670
1671 n = recv(fd, buf, sizeof(buf), 0);
1672 if (n < 0)
1673 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1674 if (n == 0) {
1675 log_error("Got zero-length cgroups agent message, ignoring.");
1676 return 0;
1677 }
1678 if ((size_t) n >= sizeof(buf)) {
1679 log_error("Got overly long cgroups agent message, ignoring.");
1680 return 0;
1681 }
1682
1683 if (memchr(buf, 0, n)) {
1684 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1685 return 0;
1686 }
1687 buf[n] = 0;
1688
1689 manager_notify_cgroup_empty(m, buf);
1690 bus_forward_agent_released(m, buf);
1691
1692 return 0;
1693 }
1694
1695 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1696 _cleanup_strv_free_ char **tags = NULL;
1697
1698 assert(m);
1699 assert(u);
1700 assert(buf);
1701
1702 tags = strv_split(buf, "\n\r");
1703 if (!tags) {
1704 log_oom();
1705 return;
1706 }
1707
1708 if (UNIT_VTABLE(u)->notify_message)
1709 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1710 else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1711 _cleanup_free_ char *x = NULL, *y = NULL;
1712
1713 x = cescape(buf);
1714 if (x)
1715 y = ellipsize(x, 20, 90);
1716 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1717 }
1718 }
1719
1720 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1721
1722 _cleanup_fdset_free_ FDSet *fds = NULL;
1723 Manager *m = userdata;
1724 char buf[NOTIFY_BUFFER_MAX+1];
1725 struct iovec iovec = {
1726 .iov_base = buf,
1727 .iov_len = sizeof(buf)-1,
1728 };
1729 union {
1730 struct cmsghdr cmsghdr;
1731 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1732 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1733 } control = {};
1734 struct msghdr msghdr = {
1735 .msg_iov = &iovec,
1736 .msg_iovlen = 1,
1737 .msg_control = &control,
1738 .msg_controllen = sizeof(control),
1739 };
1740
1741 struct cmsghdr *cmsg;
1742 struct ucred *ucred = NULL;
1743 Unit *u1, *u2, *u3;
1744 int r, *fd_array = NULL;
1745 unsigned n_fds = 0;
1746 ssize_t n;
1747
1748 assert(m);
1749 assert(m->notify_fd == fd);
1750
1751 if (revents != EPOLLIN) {
1752 log_warning("Got unexpected poll event for notify fd.");
1753 return 0;
1754 }
1755
1756 n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1757 if (n < 0) {
1758 if (IN_SET(errno, EAGAIN, EINTR))
1759 return 0; /* Spurious wakeup, try again */
1760
1761 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1762 * won't take notification messages anymore, but that's still better than busy looping around this:
1763 * being woken up over and over again but being unable to actually read the message off the socket. */
1764 return log_error_errno(errno, "Failed to receive notification message: %m");
1765 }
1766
1767 CMSG_FOREACH(cmsg, &msghdr) {
1768 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1769
1770 fd_array = (int*) CMSG_DATA(cmsg);
1771 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1772
1773 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1774 cmsg->cmsg_type == SCM_CREDENTIALS &&
1775 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1776
1777 ucred = (struct ucred*) CMSG_DATA(cmsg);
1778 }
1779 }
1780
1781 if (n_fds > 0) {
1782 assert(fd_array);
1783
1784 r = fdset_new_array(&fds, fd_array, n_fds);
1785 if (r < 0) {
1786 close_many(fd_array, n_fds);
1787 log_oom();
1788 return 0;
1789 }
1790 }
1791
1792 if (!ucred || ucred->pid <= 0) {
1793 log_warning("Received notify message without valid credentials. Ignoring.");
1794 return 0;
1795 }
1796
1797 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1798 log_warning("Received notify message exceeded maximum size. Ignoring.");
1799 return 0;
1800 }
1801
1802 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1803 * trailing NUL byte in the message, but don't expect it. */
1804 if (n > 1 && memchr(buf, 0, n-1)) {
1805 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1806 return 0;
1807 }
1808
1809 /* Make sure it's NUL-terminated. */
1810 buf[n] = 0;
1811
1812 /* Notify every unit that might be interested, but try
1813 * to avoid notifying the same one multiple times. */
1814 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1815 if (u1)
1816 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1817
1818 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1819 if (u2 && u2 != u1)
1820 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1821
1822 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1823 if (u3 && u3 != u2 && u3 != u1)
1824 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1825
1826 if (!u1 && !u2 && !u3)
1827 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1828
1829 if (fdset_size(fds) > 0)
1830 log_warning("Got extra auxiliary fds with notification message, closing them.");
1831
1832 return 0;
1833 }
1834
1835 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1836 uint64_t iteration;
1837
1838 assert(m);
1839 assert(u);
1840 assert(si);
1841
1842 sd_event_get_iteration(m->event, &iteration);
1843
1844 log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1845
1846 unit_unwatch_pid(u, si->si_pid);
1847
1848 if (UNIT_VTABLE(u)->sigchld_event) {
1849 if (set_size(u->pids) <= 1 ||
1850 iteration != u->sigchldgen ||
1851 unit_main_pid(u) == si->si_pid ||
1852 unit_control_pid(u) == si->si_pid) {
1853 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1854 u->sigchldgen = iteration;
1855 } else
1856 log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1857 }
1858 }
1859
1860 static int manager_dispatch_sigchld(Manager *m) {
1861 assert(m);
1862
1863 for (;;) {
1864 siginfo_t si = {};
1865
1866 /* First we call waitd() for a PID and do not reap the
1867 * zombie. That way we can still access /proc/$PID for
1868 * it while it is a zombie. */
1869 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1870
1871 if (errno == ECHILD)
1872 break;
1873
1874 if (errno == EINTR)
1875 continue;
1876
1877 return -errno;
1878 }
1879
1880 if (si.si_pid <= 0)
1881 break;
1882
1883 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1884 _cleanup_free_ char *name = NULL;
1885 Unit *u1, *u2, *u3;
1886
1887 get_process_comm(si.si_pid, &name);
1888
1889 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1890 si.si_pid, strna(name),
1891 sigchld_code_to_string(si.si_code),
1892 si.si_status,
1893 strna(si.si_code == CLD_EXITED
1894 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1895 : signal_to_string(si.si_status)));
1896
1897 /* And now figure out the unit this belongs
1898 * to, it might be multiple... */
1899 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1900 if (u1)
1901 invoke_sigchld_event(m, u1, &si);
1902 u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1903 if (u2 && u2 != u1)
1904 invoke_sigchld_event(m, u2, &si);
1905 u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1906 if (u3 && u3 != u2 && u3 != u1)
1907 invoke_sigchld_event(m, u3, &si);
1908 }
1909
1910 /* And now, we actually reap the zombie. */
1911 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1912 if (errno == EINTR)
1913 continue;
1914
1915 return -errno;
1916 }
1917 }
1918
1919 return 0;
1920 }
1921
1922 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1923 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1924 int r;
1925
1926 log_debug("Activating special unit %s", name);
1927
1928 r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1929 if (r < 0)
1930 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1931
1932 return r;
1933 }
1934
1935 static void manager_handle_ctrl_alt_del(Manager *m) {
1936 /* If the user presses C-A-D more than
1937 * 7 times within 2s, we reboot/shutdown immediately,
1938 * unless it was disabled in system.conf */
1939
1940 if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
1941 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1942 else
1943 emergency_action(m, m->cad_burst_action, NULL,
1944 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1945 }
1946
1947 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1948 Manager *m = userdata;
1949 ssize_t n;
1950 struct signalfd_siginfo sfsi;
1951 bool sigchld = false;
1952 int r;
1953
1954 assert(m);
1955 assert(m->signal_fd == fd);
1956
1957 if (revents != EPOLLIN) {
1958 log_warning("Got unexpected events from signal file descriptor.");
1959 return 0;
1960 }
1961
1962 for (;;) {
1963 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1964 if (n != sizeof(sfsi)) {
1965 if (n >= 0) {
1966 log_warning("Truncated read from signal fd (%zu bytes)!", n);
1967 return 0;
1968 }
1969
1970 if (IN_SET(errno, EINTR, EAGAIN))
1971 break;
1972
1973 /* We return an error here, which will kill this handler,
1974 * to avoid a busy loop on read error. */
1975 return log_error_errno(errno, "Reading from signal fd failed: %m");
1976 }
1977
1978 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1979 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1980 ? LOG_DEBUG : LOG_INFO,
1981 &sfsi);
1982
1983 switch (sfsi.ssi_signo) {
1984
1985 case SIGCHLD:
1986 sigchld = true;
1987 break;
1988
1989 case SIGTERM:
1990 if (MANAGER_IS_SYSTEM(m)) {
1991 /* This is for compatibility with the
1992 * original sysvinit */
1993 r = verify_run_space_and_log("Refusing to reexecute");
1994 if (r >= 0)
1995 m->exit_code = MANAGER_REEXECUTE;
1996 break;
1997 }
1998
1999 /* Fall through */
2000
2001 case SIGINT:
2002 if (MANAGER_IS_SYSTEM(m)) {
2003 manager_handle_ctrl_alt_del(m);
2004 break;
2005 }
2006
2007 /* Run the exit target if there is one, if not, just exit. */
2008 if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
2009 m->exit_code = MANAGER_EXIT;
2010 return 0;
2011 }
2012
2013 break;
2014
2015 case SIGWINCH:
2016 if (MANAGER_IS_SYSTEM(m))
2017 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2018
2019 /* This is a nop on non-init */
2020 break;
2021
2022 case SIGPWR:
2023 if (MANAGER_IS_SYSTEM(m))
2024 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2025
2026 /* This is a nop on non-init */
2027 break;
2028
2029 case SIGUSR1: {
2030 Unit *u;
2031
2032 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2033
2034 if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2035 log_info("Trying to reconnect to bus...");
2036 bus_init(m, true);
2037 }
2038
2039 if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2040 log_info("Loading D-Bus service...");
2041 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2042 }
2043
2044 break;
2045 }
2046
2047 case SIGUSR2: {
2048 _cleanup_free_ char *dump = NULL;
2049 _cleanup_fclose_ FILE *f = NULL;
2050 size_t size;
2051
2052 f = open_memstream(&dump, &size);
2053 if (!f) {
2054 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2055 break;
2056 }
2057
2058 manager_dump_units(m, f, "\t");
2059 manager_dump_jobs(m, f, "\t");
2060
2061 r = fflush_and_check(f);
2062 if (r < 0) {
2063 log_warning_errno(r, "Failed to write status stream: %m");
2064 break;
2065 }
2066
2067 log_dump(LOG_INFO, dump);
2068 break;
2069 }
2070
2071 case SIGHUP:
2072 r = verify_run_space_and_log("Refusing to reload");
2073 if (r >= 0)
2074 m->exit_code = MANAGER_RELOAD;
2075 break;
2076
2077 default: {
2078
2079 /* Starting SIGRTMIN+0 */
2080 static const char * const target_table[] = {
2081 [0] = SPECIAL_DEFAULT_TARGET,
2082 [1] = SPECIAL_RESCUE_TARGET,
2083 [2] = SPECIAL_EMERGENCY_TARGET,
2084 [3] = SPECIAL_HALT_TARGET,
2085 [4] = SPECIAL_POWEROFF_TARGET,
2086 [5] = SPECIAL_REBOOT_TARGET,
2087 [6] = SPECIAL_KEXEC_TARGET
2088 };
2089
2090 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2091 static const ManagerExitCode code_table[] = {
2092 [0] = MANAGER_HALT,
2093 [1] = MANAGER_POWEROFF,
2094 [2] = MANAGER_REBOOT,
2095 [3] = MANAGER_KEXEC
2096 };
2097
2098 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2099 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2100 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2101 manager_start_target(m, target_table[idx],
2102 (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2103 break;
2104 }
2105
2106 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2107 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2108 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2109 break;
2110 }
2111
2112 switch (sfsi.ssi_signo - SIGRTMIN) {
2113
2114 case 20:
2115 manager_set_show_status(m, SHOW_STATUS_YES);
2116 break;
2117
2118 case 21:
2119 manager_set_show_status(m, SHOW_STATUS_NO);
2120 break;
2121
2122 case 22:
2123 log_set_max_level(LOG_DEBUG);
2124 log_info("Setting log level to debug.");
2125 break;
2126
2127 case 23:
2128 log_set_max_level(LOG_INFO);
2129 log_info("Setting log level to info.");
2130 break;
2131
2132 case 24:
2133 if (MANAGER_IS_USER(m)) {
2134 m->exit_code = MANAGER_EXIT;
2135 return 0;
2136 }
2137
2138 /* This is a nop on init */
2139 break;
2140
2141 case 26:
2142 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2143 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2144 log_notice("Setting log target to journal-or-kmsg.");
2145 break;
2146
2147 case 27:
2148 log_set_target(LOG_TARGET_CONSOLE);
2149 log_notice("Setting log target to console.");
2150 break;
2151
2152 case 28:
2153 log_set_target(LOG_TARGET_KMSG);
2154 log_notice("Setting log target to kmsg.");
2155 break;
2156
2157 default:
2158 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2159 }
2160 }
2161 }
2162 }
2163
2164 if (sigchld)
2165 manager_dispatch_sigchld(m);
2166
2167 return 0;
2168 }
2169
2170 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2171 Manager *m = userdata;
2172 Iterator i;
2173 Unit *u;
2174
2175 assert(m);
2176 assert(m->time_change_fd == fd);
2177
2178 log_struct(LOG_DEBUG,
2179 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2180 LOG_MESSAGE("Time has been changed"),
2181 NULL);
2182
2183 /* Restart the watch */
2184 m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2185 m->time_change_fd = safe_close(m->time_change_fd);
2186
2187 manager_setup_time_change(m);
2188
2189 HASHMAP_FOREACH(u, m->units, i)
2190 if (UNIT_VTABLE(u)->time_change)
2191 UNIT_VTABLE(u)->time_change(u);
2192
2193 return 0;
2194 }
2195
2196 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2197 Manager *m = userdata;
2198
2199 assert(m);
2200 assert(m->idle_pipe[2] == fd);
2201
2202 m->no_console_output = m->n_on_console > 0;
2203
2204 manager_close_idle_pipe(m);
2205
2206 return 0;
2207 }
2208
2209 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2210 Manager *m = userdata;
2211 int r;
2212 uint64_t next;
2213
2214 assert(m);
2215 assert(source);
2216
2217 manager_print_jobs_in_progress(m);
2218
2219 next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2220 r = sd_event_source_set_time(source, next);
2221 if (r < 0)
2222 return r;
2223
2224 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2225 }
2226
2227 int manager_loop(Manager *m) {
2228 int r;
2229
2230 RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2231
2232 assert(m);
2233 m->exit_code = MANAGER_OK;
2234
2235 /* Release the path cache */
2236 m->unit_path_cache = set_free_free(m->unit_path_cache);
2237
2238 manager_check_finished(m);
2239
2240 /* There might still be some zombies hanging around from
2241 * before we were exec()'ed. Let's reap them. */
2242 r = manager_dispatch_sigchld(m);
2243 if (r < 0)
2244 return r;
2245
2246 while (m->exit_code == MANAGER_OK) {
2247 usec_t wait_usec;
2248
2249 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2250 watchdog_ping();
2251
2252 if (!ratelimit_test(&rl)) {
2253 /* Yay, something is going seriously wrong, pause a little */
2254 log_warning("Looping too fast. Throttling execution a little.");
2255 sleep(1);
2256 }
2257
2258 if (manager_dispatch_load_queue(m) > 0)
2259 continue;
2260
2261 if (manager_dispatch_gc_job_queue(m) > 0)
2262 continue;
2263
2264 if (manager_dispatch_gc_unit_queue(m) > 0)
2265 continue;
2266
2267 if (manager_dispatch_cleanup_queue(m) > 0)
2268 continue;
2269
2270 if (manager_dispatch_cgroup_queue(m) > 0)
2271 continue;
2272
2273 if (manager_dispatch_dbus_queue(m) > 0)
2274 continue;
2275
2276 /* Sleep for half the watchdog time */
2277 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2278 wait_usec = m->runtime_watchdog / 2;
2279 if (wait_usec <= 0)
2280 wait_usec = 1;
2281 } else
2282 wait_usec = USEC_INFINITY;
2283
2284 r = sd_event_run(m->event, wait_usec);
2285 if (r < 0)
2286 return log_error_errno(r, "Failed to run event loop: %m");
2287 }
2288
2289 return m->exit_code;
2290 }
2291
2292 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2293 _cleanup_free_ char *n = NULL;
2294 sd_id128_t invocation_id;
2295 Unit *u;
2296 int r;
2297
2298 assert(m);
2299 assert(s);
2300 assert(_u);
2301
2302 r = unit_name_from_dbus_path(s, &n);
2303 if (r < 0)
2304 return r;
2305
2306 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2307 * as invocation ID. */
2308 r = sd_id128_from_string(n, &invocation_id);
2309 if (r >= 0) {
2310 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2311 if (u) {
2312 *_u = u;
2313 return 0;
2314 }
2315
2316 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));
2317 }
2318
2319 /* If this didn't work, we use the suffix as unit name. */
2320 r = manager_load_unit(m, n, NULL, e, &u);
2321 if (r < 0)
2322 return r;
2323
2324 *_u = u;
2325 return 0;
2326 }
2327
2328 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2329 const char *p;
2330 unsigned id;
2331 Job *j;
2332 int r;
2333
2334 assert(m);
2335 assert(s);
2336 assert(_j);
2337
2338 p = startswith(s, "/org/freedesktop/systemd1/job/");
2339 if (!p)
2340 return -EINVAL;
2341
2342 r = safe_atou(p, &id);
2343 if (r < 0)
2344 return r;
2345
2346 j = manager_get_job(m, id);
2347 if (!j)
2348 return -ENOENT;
2349
2350 *_j = j;
2351
2352 return 0;
2353 }
2354
2355 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2356
2357 #ifdef HAVE_AUDIT
2358 _cleanup_free_ char *p = NULL;
2359 const char *msg;
2360 int audit_fd, r;
2361
2362 if (!MANAGER_IS_SYSTEM(m))
2363 return;
2364
2365 audit_fd = get_audit_fd();
2366 if (audit_fd < 0)
2367 return;
2368
2369 /* Don't generate audit events if the service was already
2370 * started and we're just deserializing */
2371 if (MANAGER_IS_RELOADING(m))
2372 return;
2373
2374 if (u->type != UNIT_SERVICE)
2375 return;
2376
2377 r = unit_name_to_prefix_and_instance(u->id, &p);
2378 if (r < 0) {
2379 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2380 return;
2381 }
2382
2383 msg = strjoina("unit=", p);
2384 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2385 if (errno == EPERM)
2386 /* We aren't allowed to send audit messages?
2387 * Then let's not retry again. */
2388 close_audit_fd();
2389 else
2390 log_warning_errno(errno, "Failed to send audit message: %m");
2391 }
2392 #endif
2393
2394 }
2395
2396 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2397 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2398 _cleanup_free_ char *message = NULL;
2399 _cleanup_close_ int fd = -1;
2400 int n = 0;
2401
2402 /* Don't generate plymouth events if the service was already
2403 * started and we're just deserializing */
2404 if (MANAGER_IS_RELOADING(m))
2405 return;
2406
2407 if (!MANAGER_IS_SYSTEM(m))
2408 return;
2409
2410 if (detect_container() > 0)
2411 return;
2412
2413 if (u->type != UNIT_SERVICE &&
2414 u->type != UNIT_MOUNT &&
2415 u->type != UNIT_SWAP)
2416 return;
2417
2418 /* We set SOCK_NONBLOCK here so that we rather drop the
2419 * message then wait for plymouth */
2420 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2421 if (fd < 0) {
2422 log_error_errno(errno, "socket() failed: %m");
2423 return;
2424 }
2425
2426 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2427
2428 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2429 log_error_errno(errno, "connect() failed: %m");
2430 return;
2431 }
2432
2433 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2434 log_oom();
2435 return;
2436 }
2437
2438 errno = 0;
2439 if (write(fd, message, n + 1) != n + 1)
2440 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2441 log_error_errno(errno, "Failed to write Plymouth message: %m");
2442 }
2443
2444 int manager_open_serialization(Manager *m, FILE **_f) {
2445 int fd;
2446 FILE *f;
2447
2448 assert(_f);
2449
2450 fd = open_serialization_fd("systemd-state");
2451 if (fd < 0)
2452 return fd;
2453
2454 f = fdopen(fd, "w+");
2455 if (!f) {
2456 safe_close(fd);
2457 return -errno;
2458 }
2459
2460 *_f = f;
2461 return 0;
2462 }
2463
2464 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2465 Iterator i;
2466 Unit *u;
2467 const char *t;
2468 int r;
2469
2470 assert(m);
2471 assert(f);
2472 assert(fds);
2473
2474 m->n_reloading++;
2475
2476 fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2477 fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2478 fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2479 fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2480
2481 dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2482 dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2483 dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2484 dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2485
2486 if (!in_initrd()) {
2487 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2488 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2489 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2490 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2491 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2492 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2493 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2494 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2495 }
2496
2497 if (!switching_root)
2498 (void) serialize_environment(f, m->environment);
2499
2500 if (m->notify_fd >= 0) {
2501 int copy;
2502
2503 copy = fdset_put_dup(fds, m->notify_fd);
2504 if (copy < 0)
2505 return copy;
2506
2507 fprintf(f, "notify-fd=%i\n", copy);
2508 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2509 }
2510
2511 if (m->cgroups_agent_fd >= 0) {
2512 int copy;
2513
2514 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2515 if (copy < 0)
2516 return copy;
2517
2518 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2519 }
2520
2521 if (m->user_lookup_fds[0] >= 0) {
2522 int copy0, copy1;
2523
2524 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2525 if (copy0 < 0)
2526 return copy0;
2527
2528 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2529 if (copy1 < 0)
2530 return copy1;
2531
2532 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2533 }
2534
2535 bus_track_serialize(m->subscribed, f, "subscribed");
2536
2537 r = dynamic_user_serialize(m, f, fds);
2538 if (r < 0)
2539 return r;
2540
2541 manager_serialize_uid_refs(m, f);
2542 manager_serialize_gid_refs(m, f);
2543
2544 fputc('\n', f);
2545
2546 HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2547 if (u->id != t)
2548 continue;
2549
2550 /* Start marker */
2551 fputs(u->id, f);
2552 fputc('\n', f);
2553
2554 r = unit_serialize(u, f, fds, !switching_root);
2555 if (r < 0) {
2556 m->n_reloading--;
2557 return r;
2558 }
2559 }
2560
2561 assert(m->n_reloading > 0);
2562 m->n_reloading--;
2563
2564 if (ferror(f))
2565 return -EIO;
2566
2567 r = bus_fdset_add_all(m, fds);
2568 if (r < 0)
2569 return r;
2570
2571 return 0;
2572 }
2573
2574 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2575 int r = 0;
2576
2577 assert(m);
2578 assert(f);
2579
2580 log_debug("Deserializing state...");
2581
2582 m->n_reloading++;
2583
2584 for (;;) {
2585 char line[LINE_MAX], *l;
2586 const char *val;
2587
2588 if (!fgets(line, sizeof(line), f)) {
2589 if (feof(f))
2590 r = 0;
2591 else
2592 r = -errno;
2593
2594 goto finish;
2595 }
2596
2597 char_array_0(line);
2598 l = strstrip(line);
2599
2600 if (l[0] == 0)
2601 break;
2602
2603 if ((val = startswith(l, "current-job-id="))) {
2604 uint32_t id;
2605
2606 if (safe_atou32(val, &id) < 0)
2607 log_debug("Failed to parse current job id value %s", val);
2608 else
2609 m->current_job_id = MAX(m->current_job_id, id);
2610
2611 } else if ((val = startswith(l, "n-installed-jobs="))) {
2612 uint32_t n;
2613
2614 if (safe_atou32(val, &n) < 0)
2615 log_debug("Failed to parse installed jobs counter %s", val);
2616 else
2617 m->n_installed_jobs += n;
2618
2619 } else if ((val = startswith(l, "n-failed-jobs="))) {
2620 uint32_t n;
2621
2622 if (safe_atou32(val, &n) < 0)
2623 log_debug("Failed to parse failed jobs counter %s", val);
2624 else
2625 m->n_failed_jobs += n;
2626
2627 } else if ((val = startswith(l, "taint-usr="))) {
2628 int b;
2629
2630 b = parse_boolean(val);
2631 if (b < 0)
2632 log_debug("Failed to parse taint /usr flag %s", val);
2633 else
2634 m->taint_usr = m->taint_usr || b;
2635
2636 } else if ((val = startswith(l, "firmware-timestamp=")))
2637 dual_timestamp_deserialize(val, &m->firmware_timestamp);
2638 else if ((val = startswith(l, "loader-timestamp=")))
2639 dual_timestamp_deserialize(val, &m->loader_timestamp);
2640 else if ((val = startswith(l, "kernel-timestamp=")))
2641 dual_timestamp_deserialize(val, &m->kernel_timestamp);
2642 else if ((val = startswith(l, "initrd-timestamp=")))
2643 dual_timestamp_deserialize(val, &m->initrd_timestamp);
2644 else if ((val = startswith(l, "userspace-timestamp=")))
2645 dual_timestamp_deserialize(val, &m->userspace_timestamp);
2646 else if ((val = startswith(l, "finish-timestamp=")))
2647 dual_timestamp_deserialize(val, &m->finish_timestamp);
2648 else if ((val = startswith(l, "security-start-timestamp=")))
2649 dual_timestamp_deserialize(val, &m->security_start_timestamp);
2650 else if ((val = startswith(l, "security-finish-timestamp=")))
2651 dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2652 else if ((val = startswith(l, "generators-start-timestamp=")))
2653 dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2654 else if ((val = startswith(l, "generators-finish-timestamp=")))
2655 dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2656 else if ((val = startswith(l, "units-load-start-timestamp=")))
2657 dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2658 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2659 dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2660 else if (startswith(l, "env=")) {
2661 r = deserialize_environment(&m->environment, l);
2662 if (r < 0)
2663 return r;
2664
2665 } else if ((val = startswith(l, "notify-fd="))) {
2666 int fd;
2667
2668 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2669 log_debug("Failed to parse notify fd: %s", val);
2670 else {
2671 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2672 safe_close(m->notify_fd);
2673 m->notify_fd = fdset_remove(fds, fd);
2674 }
2675
2676 } else if ((val = startswith(l, "notify-socket="))) {
2677 char *n;
2678
2679 n = strdup(val);
2680 if (!n) {
2681 r = -ENOMEM;
2682 goto finish;
2683 }
2684
2685 free(m->notify_socket);
2686 m->notify_socket = n;
2687
2688 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2689 int fd;
2690
2691 if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2692 log_debug("Failed to parse cgroups agent fd: %s", val);
2693 else {
2694 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2695 safe_close(m->cgroups_agent_fd);
2696 m->cgroups_agent_fd = fdset_remove(fds, fd);
2697 }
2698
2699 } else if ((val = startswith(l, "user-lookup="))) {
2700 int fd0, fd1;
2701
2702 if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2703 log_debug("Failed to parse user lookup fd: %s", val);
2704 else {
2705 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2706 safe_close_pair(m->user_lookup_fds);
2707 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2708 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2709 }
2710
2711 } else if ((val = startswith(l, "dynamic-user=")))
2712 dynamic_user_deserialize_one(m, val, fds);
2713 else if ((val = startswith(l, "destroy-ipc-uid=")))
2714 manager_deserialize_uid_refs_one(m, val);
2715 else if ((val = startswith(l, "destroy-ipc-gid=")))
2716 manager_deserialize_gid_refs_one(m, val);
2717 else if ((val = startswith(l, "subscribed="))) {
2718
2719 if (strv_extend(&m->deserialized_subscribed, val) < 0)
2720 log_oom();
2721
2722 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2723 log_debug("Unknown serialization item '%s'", l);
2724 }
2725
2726 for (;;) {
2727 Unit *u;
2728 char name[UNIT_NAME_MAX+2];
2729
2730 /* Start marker */
2731 if (!fgets(name, sizeof(name), f)) {
2732 if (feof(f))
2733 r = 0;
2734 else
2735 r = -errno;
2736
2737 goto finish;
2738 }
2739
2740 char_array_0(name);
2741
2742 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2743 if (r < 0)
2744 goto finish;
2745
2746 r = unit_deserialize(u, f, fds);
2747 if (r < 0)
2748 goto finish;
2749 }
2750
2751 finish:
2752 if (ferror(f))
2753 r = -EIO;
2754
2755 assert(m->n_reloading > 0);
2756 m->n_reloading--;
2757
2758 return r;
2759 }
2760
2761 int manager_reload(Manager *m) {
2762 int r, q;
2763 _cleanup_fclose_ FILE *f = NULL;
2764 _cleanup_fdset_free_ FDSet *fds = NULL;
2765
2766 assert(m);
2767
2768 r = manager_open_serialization(m, &f);
2769 if (r < 0)
2770 return r;
2771
2772 m->n_reloading++;
2773 bus_manager_send_reloading(m, true);
2774
2775 fds = fdset_new();
2776 if (!fds) {
2777 m->n_reloading--;
2778 return -ENOMEM;
2779 }
2780
2781 r = manager_serialize(m, f, fds, false);
2782 if (r < 0) {
2783 m->n_reloading--;
2784 return r;
2785 }
2786
2787 if (fseeko(f, 0, SEEK_SET) < 0) {
2788 m->n_reloading--;
2789 return -errno;
2790 }
2791
2792 /* From here on there is no way back. */
2793 manager_clear_jobs_and_units(m);
2794 lookup_paths_flush_generator(&m->lookup_paths);
2795 lookup_paths_free(&m->lookup_paths);
2796 dynamic_user_vacuum(m, false);
2797 m->uid_refs = hashmap_free(m->uid_refs);
2798 m->gid_refs = hashmap_free(m->gid_refs);
2799
2800 q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2801 if (q < 0 && r >= 0)
2802 r = q;
2803
2804 q = manager_run_environment_generators(m);
2805 if (q < 0 && r >= 0)
2806 r = q;
2807
2808 /* Find new unit paths */
2809 q = manager_run_generators(m);
2810 if (q < 0 && r >= 0)
2811 r = q;
2812
2813 lookup_paths_reduce(&m->lookup_paths);
2814 manager_build_unit_path_cache(m);
2815
2816 /* First, enumerate what we can from all config files */
2817 manager_enumerate(m);
2818
2819 /* Second, deserialize our stored data */
2820 q = manager_deserialize(m, f, fds);
2821 if (q < 0 && r >= 0)
2822 r = q;
2823
2824 fclose(f);
2825 f = NULL;
2826
2827 /* Re-register notify_fd as event source */
2828 q = manager_setup_notify(m);
2829 if (q < 0 && r >= 0)
2830 r = q;
2831
2832 q = manager_setup_cgroups_agent(m);
2833 if (q < 0 && r >= 0)
2834 r = q;
2835
2836 q = manager_setup_user_lookup_fd(m);
2837 if (q < 0 && r >= 0)
2838 r = q;
2839
2840 /* Third, fire things up! */
2841 manager_coldplug(m);
2842
2843 /* Release any dynamic users no longer referenced */
2844 dynamic_user_vacuum(m, true);
2845
2846 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2847 manager_vacuum_uid_refs(m);
2848 manager_vacuum_gid_refs(m);
2849
2850 /* Sync current state of bus names with our set of listening units */
2851 if (m->api_bus)
2852 manager_sync_bus_names(m, m->api_bus);
2853
2854 assert(m->n_reloading > 0);
2855 m->n_reloading--;
2856
2857 m->send_reloading_done = true;
2858
2859 return r;
2860 }
2861
2862 void manager_reset_failed(Manager *m) {
2863 Unit *u;
2864 Iterator i;
2865
2866 assert(m);
2867
2868 HASHMAP_FOREACH(u, m->units, i)
2869 unit_reset_failed(u);
2870 }
2871
2872 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2873 Unit *u;
2874
2875 assert(m);
2876 assert(name);
2877
2878 /* Returns true if the unit is inactive or going down */
2879 u = manager_get_unit(m, name);
2880 if (!u)
2881 return true;
2882
2883 return unit_inactive_or_pending(u);
2884 }
2885
2886 static void manager_notify_finished(Manager *m) {
2887 char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2888 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2889
2890 if (m->test_run)
2891 return;
2892
2893 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2894
2895 /* Note that m->kernel_usec.monotonic is always at 0,
2896 * and m->firmware_usec.monotonic and
2897 * m->loader_usec.monotonic should be considered
2898 * negative values. */
2899
2900 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2901 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2902 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2903 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2904
2905 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2906
2907 kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2908 initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2909
2910 log_struct(LOG_INFO,
2911 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2912 "KERNEL_USEC="USEC_FMT, kernel_usec,
2913 "INITRD_USEC="USEC_FMT, initrd_usec,
2914 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2915 LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2916 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2917 format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2918 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2919 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2920 NULL);
2921 } else {
2922 kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2923 initrd_usec = 0;
2924
2925 log_struct(LOG_INFO,
2926 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
2927 "KERNEL_USEC="USEC_FMT, kernel_usec,
2928 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2929 LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2930 format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2931 format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2932 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2933 NULL);
2934 }
2935 } else {
2936 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2937 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2938
2939 log_struct(LOG_INFO,
2940 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
2941 "USERSPACE_USEC="USEC_FMT, userspace_usec,
2942 LOG_MESSAGE("Startup finished in %s.",
2943 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2944 NULL);
2945 }
2946
2947 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2948
2949 sd_notifyf(false,
2950 "READY=1\n"
2951 "STATUS=Startup finished in %s.",
2952 format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2953 }
2954
2955 void manager_check_finished(Manager *m) {
2956 assert(m);
2957
2958 if (MANAGER_IS_RELOADING(m))
2959 return;
2960
2961 /* Verify that we are actually running currently. Initially
2962 * the exit code is set to invalid, and during operation it is
2963 * then set to MANAGER_OK */
2964 if (m->exit_code != MANAGER_OK)
2965 return;
2966
2967 if (hashmap_size(m->jobs) > 0) {
2968 if (m->jobs_in_progress_event_source)
2969 /* Ignore any failure, this is only for feedback */
2970 (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2971
2972 return;
2973 }
2974
2975 manager_flip_auto_status(m, false);
2976
2977 /* Notify Type=idle units that we are done now */
2978 manager_close_idle_pipe(m);
2979
2980 /* Turn off confirm spawn now */
2981 m->confirm_spawn = NULL;
2982
2983 /* No need to update ask password status when we're going non-interactive */
2984 manager_close_ask_password(m);
2985
2986 /* This is no longer the first boot */
2987 manager_set_first_boot(m, false);
2988
2989 if (dual_timestamp_is_set(&m->finish_timestamp))
2990 return;
2991
2992 dual_timestamp_get(&m->finish_timestamp);
2993
2994 manager_notify_finished(m);
2995
2996 manager_invalidate_startup_units(m);
2997 }
2998
2999 static bool generator_path_any(const char* const* paths) {
3000 char **path;
3001 bool found = false;
3002
3003 /* Optimize by skipping the whole process by not creating output directories
3004 * if no generators are found. */
3005 STRV_FOREACH(path, (char**) paths)
3006 if (access(*path, F_OK) == 0)
3007 found = true;
3008 else if (errno != ENOENT)
3009 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3010
3011 return found;
3012 }
3013
3014 static const char* system_env_generator_binary_paths[] = {
3015 "/run/systemd/system-environment-generators",
3016 "/etc/systemd/system-environment-generators",
3017 "/usr/local/lib/systemd/system-environment-generators",
3018 SYSTEM_ENV_GENERATOR_PATH,
3019 NULL
3020 };
3021
3022 static const char* user_env_generator_binary_paths[] = {
3023 "/run/systemd/user-environment-generators",
3024 "/etc/systemd/user-environment-generators",
3025 "/usr/local/lib/systemd/user-environment-generators",
3026 USER_ENV_GENERATOR_PATH,
3027 NULL
3028 };
3029
3030 static int manager_run_environment_generators(Manager *m) {
3031 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3032 const char **paths;
3033 void* args[] = {&tmp, &tmp, &m->environment};
3034
3035 if (m->test_run)
3036 return 0;
3037
3038 paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths;
3039
3040 if (!generator_path_any(paths))
3041 return 0;
3042
3043 return execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, args, NULL);
3044 }
3045
3046 static int manager_run_generators(Manager *m) {
3047 _cleanup_strv_free_ char **paths = NULL;
3048 const char *argv[5];
3049 int r;
3050
3051 assert(m);
3052
3053 if (m->test_run)
3054 return 0;
3055
3056 paths = generator_binary_paths(m->unit_file_scope);
3057 if (!paths)
3058 return log_oom();
3059
3060 if (!generator_path_any((const char* const*) paths))
3061 return 0;
3062
3063 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3064 if (r < 0)
3065 goto finish;
3066
3067 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3068 argv[1] = m->lookup_paths.generator;
3069 argv[2] = m->lookup_paths.generator_early;
3070 argv[3] = m->lookup_paths.generator_late;
3071 argv[4] = NULL;
3072
3073 RUN_WITH_UMASK(0022)
3074 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC,
3075 NULL, NULL, (char**) argv);
3076
3077 finish:
3078 lookup_paths_trim_generator(&m->lookup_paths);
3079 return r;
3080 }
3081
3082 int manager_environment_add(Manager *m, char **minus, char **plus) {
3083 char **a = NULL, **b = NULL, **l;
3084 assert(m);
3085
3086 l = m->environment;
3087
3088 if (!strv_isempty(minus)) {
3089 a = strv_env_delete(l, 1, minus);
3090 if (!a)
3091 return -ENOMEM;
3092
3093 l = a;
3094 }
3095
3096 if (!strv_isempty(plus)) {
3097 b = strv_env_merge(2, l, plus);
3098 if (!b) {
3099 strv_free(a);
3100 return -ENOMEM;
3101 }
3102
3103 l = b;
3104 }
3105
3106 if (m->environment != l)
3107 strv_free(m->environment);
3108 if (a != l)
3109 strv_free(a);
3110 if (b != l)
3111 strv_free(b);
3112
3113 m->environment = l;
3114 manager_clean_environment(m);
3115 strv_sort(m->environment);
3116
3117 return 0;
3118 }
3119
3120 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3121 int i;
3122
3123 assert(m);
3124
3125 for (i = 0; i < _RLIMIT_MAX; i++) {
3126 m->rlimit[i] = mfree(m->rlimit[i]);
3127
3128 if (!default_rlimit[i])
3129 continue;
3130
3131 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3132 if (!m->rlimit[i])
3133 return log_oom();
3134 }
3135
3136 return 0;
3137 }
3138
3139 void manager_recheck_journal(Manager *m) {
3140 Unit *u;
3141
3142 assert(m);
3143
3144 if (!MANAGER_IS_SYSTEM(m))
3145 return;
3146
3147 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3148 if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3149 log_close_journal();
3150 return;
3151 }
3152
3153 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3154 if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3155 log_close_journal();
3156 return;
3157 }
3158
3159 /* Hmm, OK, so the socket is fully up and the service is up
3160 * too, then let's make use of the thing. */
3161 log_open();
3162 }
3163
3164 void manager_set_show_status(Manager *m, ShowStatus mode) {
3165 assert(m);
3166 assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3167
3168 if (!MANAGER_IS_SYSTEM(m))
3169 return;
3170
3171 if (m->show_status != mode)
3172 log_debug("%s showing of status.",
3173 mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3174 m->show_status = mode;
3175
3176 if (mode > 0)
3177 (void) touch("/run/systemd/show-status");
3178 else
3179 (void) unlink("/run/systemd/show-status");
3180 }
3181
3182 static bool manager_get_show_status(Manager *m, StatusType type) {
3183 assert(m);
3184
3185 if (!MANAGER_IS_SYSTEM(m))
3186 return false;
3187
3188 if (m->no_console_output)
3189 return false;
3190
3191 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3192 return false;
3193
3194 /* If we cannot find out the status properly, just proceed. */
3195 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3196 return false;
3197
3198 if (m->show_status > 0)
3199 return true;
3200
3201 return false;
3202 }
3203
3204 const char *manager_get_confirm_spawn(Manager *m) {
3205 static int last_errno = 0;
3206 const char *vc = m->confirm_spawn;
3207 struct stat st;
3208 int r;
3209
3210 /* Here's the deal: we want to test the validity of the console but don't want
3211 * PID1 to go through the whole console process which might block. But we also
3212 * want to warn the user only once if something is wrong with the console so we
3213 * cannot do the sanity checks after spawning our children. So here we simply do
3214 * really basic tests to hopefully trap common errors.
3215 *
3216 * If the console suddenly disappear at the time our children will really it
3217 * then they will simply fail to acquire it and a positive answer will be
3218 * assumed. New children will fallback to /dev/console though.
3219 *
3220 * Note: TTYs are devices that can come and go any time, and frequently aren't
3221 * available yet during early boot (consider a USB rs232 dongle...). If for any
3222 * reason the configured console is not ready, we fallback to the default
3223 * console. */
3224
3225 if (!vc || path_equal(vc, "/dev/console"))
3226 return vc;
3227
3228 r = stat(vc, &st);
3229 if (r < 0)
3230 goto fail;
3231
3232 if (!S_ISCHR(st.st_mode)) {
3233 errno = ENOTTY;
3234 goto fail;
3235 }
3236
3237 last_errno = 0;
3238 return vc;
3239 fail:
3240 if (last_errno != errno) {
3241 last_errno = errno;
3242 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3243 }
3244 return "/dev/console";
3245 }
3246
3247 void manager_set_first_boot(Manager *m, bool b) {
3248 assert(m);
3249
3250 if (!MANAGER_IS_SYSTEM(m))
3251 return;
3252
3253 if (m->first_boot != (int) b) {
3254 if (b)
3255 (void) touch("/run/systemd/first-boot");
3256 else
3257 (void) unlink("/run/systemd/first-boot");
3258 }
3259
3260 m->first_boot = b;
3261 }
3262
3263 void manager_disable_confirm_spawn(void) {
3264 (void) touch("/run/systemd/confirm_spawn_disabled");
3265 }
3266
3267 bool manager_is_confirm_spawn_disabled(Manager *m) {
3268 if (!m->confirm_spawn)
3269 return true;
3270
3271 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3272 }
3273
3274 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3275 va_list ap;
3276
3277 /* If m is NULL, assume we're after shutdown and let the messages through. */
3278
3279 if (m && !manager_get_show_status(m, type))
3280 return;
3281
3282 /* XXX We should totally drop the check for ephemeral here
3283 * and thus effectively make 'Type=idle' pointless. */
3284 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3285 return;
3286
3287 va_start(ap, format);
3288 status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3289 va_end(ap);
3290 }
3291
3292 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3293 char p[strlen(path)+1];
3294
3295 assert(m);
3296 assert(path);
3297
3298 strcpy(p, path);
3299 path_kill_slashes(p);
3300
3301 return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3302 }
3303
3304 const char *manager_get_runtime_prefix(Manager *m) {
3305 assert(m);
3306
3307 return MANAGER_IS_SYSTEM(m) ?
3308 "/run" :
3309 getenv("XDG_RUNTIME_DIR");
3310 }
3311
3312 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3313 unsigned size;
3314 int r;
3315
3316 assert(m);
3317 assert(u->manager == m);
3318
3319 size = set_size(m->failed_units);
3320
3321 if (failed) {
3322 r = set_ensure_allocated(&m->failed_units, NULL);
3323 if (r < 0)
3324 return log_oom();
3325
3326 if (set_put(m->failed_units, u) < 0)
3327 return log_oom();
3328 } else
3329 (void) set_remove(m->failed_units, u);
3330
3331 if (set_size(m->failed_units) != size)
3332 bus_manager_send_change_signal(m);
3333
3334 return 0;
3335 }
3336
3337 ManagerState manager_state(Manager *m) {
3338 Unit *u;
3339
3340 assert(m);
3341
3342 /* Did we ever finish booting? If not then we are still starting up */
3343 if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3344
3345 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3346 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3347 return MANAGER_INITIALIZING;
3348
3349 return MANAGER_STARTING;
3350 }
3351
3352 /* Is the special shutdown target queued? If so, we are in shutdown state */
3353 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3354 if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3355 return MANAGER_STOPPING;
3356
3357 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3358 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3359 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3360 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3361 return MANAGER_MAINTENANCE;
3362
3363 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3364 if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3365 (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3366 return MANAGER_MAINTENANCE;
3367
3368 /* Are there any failed units? If so, we are in degraded mode */
3369 if (set_size(m->failed_units) > 0)
3370 return MANAGER_DEGRADED;
3371
3372 return MANAGER_RUNNING;
3373 }
3374
3375 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3376
3377 static void manager_unref_uid_internal(
3378 Manager *m,
3379 Hashmap **uid_refs,
3380 uid_t uid,
3381 bool destroy_now,
3382 int (*_clean_ipc)(uid_t uid)) {
3383
3384 uint32_t c, n;
3385
3386 assert(m);
3387 assert(uid_refs);
3388 assert(uid_is_valid(uid));
3389 assert(_clean_ipc);
3390
3391 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3392 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3393 *
3394 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3395 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3396 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3397 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3398
3399 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3400 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3401
3402 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3403 return;
3404
3405 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3406
3407 n = c & ~DESTROY_IPC_FLAG;
3408 assert(n > 0);
3409 n--;
3410
3411 if (destroy_now && n == 0) {
3412 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3413
3414 if (c & DESTROY_IPC_FLAG) {
3415 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3416 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3417 uid);
3418 (void) _clean_ipc(uid);
3419 }
3420 } else {
3421 c = n | (c & DESTROY_IPC_FLAG);
3422 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3423 }
3424 }
3425
3426 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3427 manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3428 }
3429
3430 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3431 manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3432 }
3433
3434 static int manager_ref_uid_internal(
3435 Manager *m,
3436 Hashmap **uid_refs,
3437 uid_t uid,
3438 bool clean_ipc) {
3439
3440 uint32_t c, n;
3441 int r;
3442
3443 assert(m);
3444 assert(uid_refs);
3445 assert(uid_is_valid(uid));
3446
3447 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3448 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3449
3450 assert_cc(sizeof(uid_t) == sizeof(gid_t));
3451 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3452
3453 if (uid == 0) /* We don't keep track of root, and will never destroy it */
3454 return 0;
3455
3456 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3457 if (r < 0)
3458 return r;
3459
3460 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3461
3462 n = c & ~DESTROY_IPC_FLAG;
3463 n++;
3464
3465 if (n & DESTROY_IPC_FLAG) /* check for overflow */
3466 return -EOVERFLOW;
3467
3468 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3469
3470 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3471 }
3472
3473 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3474 return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3475 }
3476
3477 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3478 return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3479 }
3480
3481 static void manager_vacuum_uid_refs_internal(
3482 Manager *m,
3483 Hashmap **uid_refs,
3484 int (*_clean_ipc)(uid_t uid)) {
3485
3486 Iterator i;
3487 void *p, *k;
3488
3489 assert(m);
3490 assert(uid_refs);
3491 assert(_clean_ipc);
3492
3493 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3494 uint32_t c, n;
3495 uid_t uid;
3496
3497 uid = PTR_TO_UID(k);
3498 c = PTR_TO_UINT32(p);
3499
3500 n = c & ~DESTROY_IPC_FLAG;
3501 if (n > 0)
3502 continue;
3503
3504 if (c & DESTROY_IPC_FLAG) {
3505 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3506 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3507 uid);
3508 (void) _clean_ipc(uid);
3509 }
3510
3511 assert_se(hashmap_remove(*uid_refs, k) == p);
3512 }
3513 }
3514
3515 void manager_vacuum_uid_refs(Manager *m) {
3516 manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3517 }
3518
3519 void manager_vacuum_gid_refs(Manager *m) {
3520 manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3521 }
3522
3523 static void manager_serialize_uid_refs_internal(
3524 Manager *m,
3525 FILE *f,
3526 Hashmap **uid_refs,
3527 const char *field_name) {
3528
3529 Iterator i;
3530 void *p, *k;
3531
3532 assert(m);
3533 assert(f);
3534 assert(uid_refs);
3535 assert(field_name);
3536
3537 /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3538 * of it is better rebuild after a reload/reexec. */
3539
3540 HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3541 uint32_t c;
3542 uid_t uid;
3543
3544 uid = PTR_TO_UID(k);
3545 c = PTR_TO_UINT32(p);
3546
3547 if (!(c & DESTROY_IPC_FLAG))
3548 continue;
3549
3550 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3551 }
3552 }
3553
3554 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3555 manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3556 }
3557
3558 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3559 manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3560 }
3561
3562 static void manager_deserialize_uid_refs_one_internal(
3563 Manager *m,
3564 Hashmap** uid_refs,
3565 const char *value) {
3566
3567 uid_t uid;
3568 uint32_t c;
3569 int r;
3570
3571 assert(m);
3572 assert(uid_refs);
3573 assert(value);
3574
3575 r = parse_uid(value, &uid);
3576 if (r < 0 || uid == 0) {
3577 log_debug("Unable to parse UID reference serialization");
3578 return;
3579 }
3580
3581 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3582 if (r < 0) {
3583 log_oom();
3584 return;
3585 }
3586
3587 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3588 if (c & DESTROY_IPC_FLAG)
3589 return;
3590
3591 c |= DESTROY_IPC_FLAG;
3592
3593 r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3594 if (r < 0) {
3595 log_debug("Failed to add UID reference entry");
3596 return;
3597 }
3598 }
3599
3600 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3601 manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3602 }
3603
3604 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3605 manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3606 }
3607
3608 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3609 struct buffer {
3610 uid_t uid;
3611 gid_t gid;
3612 char unit_name[UNIT_NAME_MAX+1];
3613 } _packed_ buffer;
3614
3615 Manager *m = userdata;
3616 ssize_t l;
3617 size_t n;
3618 Unit *u;
3619
3620 assert_se(source);
3621 assert_se(m);
3622
3623 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3624 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3625 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3626
3627 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3628 if (l < 0) {
3629 if (errno == EINTR || errno == EAGAIN)
3630 return 0;
3631
3632 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3633 }
3634
3635 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3636 log_warning("Received too short user lookup message, ignoring.");
3637 return 0;
3638 }
3639
3640 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3641 log_warning("Received too long user lookup message, ignoring.");
3642 return 0;
3643 }
3644
3645 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3646 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3647 return 0;
3648 }
3649
3650 n = (size_t) l - offsetof(struct buffer, unit_name);
3651 if (memchr(buffer.unit_name, 0, n)) {
3652 log_warning("Received lookup message with embedded NUL character, ignoring.");
3653 return 0;
3654 }
3655
3656 buffer.unit_name[n] = 0;
3657 u = manager_get_unit(m, buffer.unit_name);
3658 if (!u) {
3659 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3660 return 0;
3661 }
3662
3663 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3664
3665 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3666 return 0;
3667 }
3668
3669 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3670 [MANAGER_INITIALIZING] = "initializing",
3671 [MANAGER_STARTING] = "starting",
3672 [MANAGER_RUNNING] = "running",
3673 [MANAGER_DEGRADED] = "degraded",
3674 [MANAGER_MAINTENANCE] = "maintenance",
3675 [MANAGER_STOPPING] = "stopping",
3676 };
3677
3678 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
Unnamed repository; edit this file 'description' to name the repository.