]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/manager.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
core: Make sure cgroup_oom_queue is flushed on manager exit
[thirdparty/systemd.git] / src / core / manager.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <linux/kd.h>
6 #include <sys/epoll.h>
7 #include <sys/inotify.h>
8 #include <sys/ioctl.h>
9 #include <sys/reboot.h>
10 #include <sys/timerfd.h>
11 #include <sys/wait.h>
12 #include <unistd.h>
13
14 #if HAVE_AUDIT
15 #include <libaudit.h>
16 #endif
17
18 #include "sd-daemon.h"
19 #include "sd-messages.h"
20 #include "sd-path.h"
21
22 #include "all-units.h"
23 #include "alloc-util.h"
24 #include "audit-fd.h"
25 #include "boot-timestamps.h"
26 #include "bus-common-errors.h"
27 #include "bus-error.h"
28 #include "bus-kernel.h"
29 #include "bus-util.h"
30 #include "clean-ipc.h"
31 #include "clock-util.h"
32 #include "core-varlink.h"
33 #include "creds-util.h"
34 #include "dbus-job.h"
35 #include "dbus-manager.h"
36 #include "dbus-unit.h"
37 #include "dbus.h"
38 #include "def.h"
39 #include "dirent-util.h"
40 #include "env-util.h"
41 #include "escape.h"
42 #include "exec-util.h"
43 #include "execute.h"
44 #include "exit-status.h"
45 #include "fd-util.h"
46 #include "fileio.h"
47 #include "fs-util.h"
48 #include "generator-setup.h"
49 #include "hashmap.h"
50 #include "install.h"
51 #include "io-util.h"
52 #include "label.h"
53 #include "load-fragment.h"
54 #include "locale-setup.h"
55 #include "log.h"
56 #include "macro.h"
57 #include "manager.h"
58 #include "manager-dump.h"
59 #include "manager-serialize.h"
60 #include "memory-util.h"
61 #include "mkdir.h"
62 #include "parse-util.h"
63 #include "path-lookup.h"
64 #include "path-util.h"
65 #include "process-util.h"
66 #include "ratelimit.h"
67 #include "rlimit-util.h"
68 #include "rm-rf.h"
69 #include "selinux-util.h"
70 #include "signal-util.h"
71 #include "socket-util.h"
72 #include "special.h"
73 #include "stat-util.h"
74 #include "string-table.h"
75 #include "string-util.h"
76 #include "strv.h"
77 #include "strxcpyx.h"
78 #include "sysctl-util.h"
79 #include "syslog-util.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 "virt.h"
87 #include "watchdog.h"
88
89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
91
92 /* Initial delay and the interval for printing status messages about running jobs */
93 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
94 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
95 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
96 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
97
98 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
99 * the queue gets more empty. */
100 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
101
102 /* How many units and jobs to process of the bus queue before returning to the event loop. */
103 #define MANAGER_BUS_MESSAGE_BUDGET 100U
104
105 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
106 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
107 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
108 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
109 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
110 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
111 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
112 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
113 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata);
114 static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata);
115 static int manager_run_environment_generators(Manager *m);
116 static int manager_run_generators(Manager *m);
117 static void manager_vacuum(Manager *m);
118
119 static usec_t manager_watch_jobs_next_time(Manager *m) {
120 usec_t timeout;
121
122 if (MANAGER_IS_USER(m))
123 /* Let the user manager without a timeout show status quickly, so the system manager can make
124 * use of it, if it wants to. */
125 timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3;
126 else if (show_status_on(m->show_status))
127 /* When status is on, just use the usual timeout. */
128 timeout = JOBS_IN_PROGRESS_WAIT_USEC;
129 else
130 timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC;
131
132 return usec_add(now(CLOCK_MONOTONIC), timeout);
133 }
134
135 static void manager_watch_jobs_in_progress(Manager *m) {
136 usec_t next;
137 int r;
138
139 assert(m);
140
141 /* We do not want to show the cylon animation if the user
142 * needs to confirm service executions otherwise confirmation
143 * messages will be screwed by the cylon animation. */
144 if (!manager_is_confirm_spawn_disabled(m))
145 return;
146
147 if (m->jobs_in_progress_event_source)
148 return;
149
150 next = manager_watch_jobs_next_time(m);
151 r = sd_event_add_time(
152 m->event,
153 &m->jobs_in_progress_event_source,
154 CLOCK_MONOTONIC,
155 next, 0,
156 manager_dispatch_jobs_in_progress, m);
157 if (r < 0)
158 return;
159
160 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
161 }
162
163 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
164
165 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
166 char *p = buffer;
167
168 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
169 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
170
171 if (pos > 1) {
172 if (pos > 2)
173 p = mempset(p, ' ', pos-2);
174 if (log_get_show_color())
175 p = stpcpy(p, ANSI_RED);
176 *p++ = '*';
177 }
178
179 if (pos > 0 && pos <= width) {
180 if (log_get_show_color())
181 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
182 *p++ = '*';
183 }
184
185 if (log_get_show_color())
186 p = stpcpy(p, ANSI_NORMAL);
187
188 if (pos < width) {
189 if (log_get_show_color())
190 p = stpcpy(p, ANSI_RED);
191 *p++ = '*';
192 if (pos < width-1)
193 p = mempset(p, ' ', width-1-pos);
194 if (log_get_show_color())
195 strcpy(p, ANSI_NORMAL);
196 }
197 }
198
199 static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) {
200 assert(m);
201
202 if (enable) {
203 if (m->show_status == SHOW_STATUS_AUTO)
204 manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason);
205 } else {
206 if (m->show_status == SHOW_STATUS_TEMPORARY)
207 manager_set_show_status(m, SHOW_STATUS_AUTO, reason);
208 }
209 }
210
211 static void manager_print_jobs_in_progress(Manager *m) {
212 Job *j;
213 unsigned counter = 0, print_nr;
214 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
215 unsigned cylon_pos;
216 uint64_t x;
217
218 assert(m);
219 assert(m->n_running_jobs > 0);
220
221 manager_flip_auto_status(m, true, "delay");
222
223 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
224
225 HASHMAP_FOREACH(j, m->jobs)
226 if (j->state == JOB_RUNNING && counter++ == print_nr)
227 break;
228
229 /* m->n_running_jobs must be consistent with the contents of m->jobs,
230 * so the above loop must have succeeded in finding j. */
231 assert(counter == print_nr + 1);
232 assert(j);
233
234 cylon_pos = m->jobs_in_progress_iteration % 14;
235 if (cylon_pos >= 8)
236 cylon_pos = 14 - cylon_pos;
237 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
238
239 m->jobs_in_progress_iteration++;
240
241 char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
242 if (m->n_running_jobs > 1)
243 xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs);
244
245 bool have_timeout = job_get_timeout(j, &x) > 0;
246
247 /* We want to use enough information for the user to identify previous lines talking about the same
248 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
249 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
250 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
251 * second argument to unit_status_string(). */
252 const char *ident = unit_status_string(j->unit, NULL);
253
254 const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
255 const char *limit = have_timeout ? FORMAT_TIMESPAN(x - j->begin_usec, 1*USEC_PER_SEC) : "no limit";
256
257 if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION)
258 /* When using 'Description', we effectively don't have enough space to show the nested status
259 * without ellipsization, so let's not even try. */
260 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
261 "%sA %s job is running for %s (%s / %s)",
262 job_of_n,
263 job_type_to_string(j->type),
264 ident,
265 time, limit);
266 else {
267 const char *status_text = unit_status_text(j->unit);
268
269 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
270 "%sJob %s/%s running (%s / %s)%s%s",
271 job_of_n,
272 ident,
273 job_type_to_string(j->type),
274 time, limit,
275 status_text ? ": " : "",
276 strempty(status_text));
277 }
278
279 sd_notifyf(false,
280 "STATUS=%sUser job %s/%s running (%s / %s)...",
281 job_of_n,
282 ident,
283 job_type_to_string(j->type),
284 time, limit);
285 m->status_ready = false;
286 }
287
288 static int have_ask_password(void) {
289 _cleanup_closedir_ DIR *dir = NULL;
290 struct dirent *de;
291
292 dir = opendir("/run/systemd/ask-password");
293 if (!dir) {
294 if (errno == ENOENT)
295 return false;
296 else
297 return -errno;
298 }
299
300 FOREACH_DIRENT_ALL(de, dir, return -errno) {
301 if (startswith(de->d_name, "ask."))
302 return true;
303 }
304 return false;
305 }
306
307 static int manager_dispatch_ask_password_fd(sd_event_source *source,
308 int fd, uint32_t revents, void *userdata) {
309 Manager *m = userdata;
310
311 assert(m);
312
313 (void) flush_fd(fd);
314
315 m->have_ask_password = have_ask_password();
316 if (m->have_ask_password < 0)
317 /* Log error but continue. Negative have_ask_password
318 * is treated as unknown status. */
319 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
320
321 return 0;
322 }
323
324 static void manager_close_ask_password(Manager *m) {
325 assert(m);
326
327 m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source);
328 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
329 m->have_ask_password = -EINVAL;
330 }
331
332 static int manager_check_ask_password(Manager *m) {
333 int r;
334
335 assert(m);
336
337 if (!m->ask_password_event_source) {
338 assert(m->ask_password_inotify_fd < 0);
339
340 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
341
342 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
343 if (m->ask_password_inotify_fd < 0)
344 return log_error_errno(errno, "Failed to create inotify object: %m");
345
346 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
347 "/run/systemd/ask-password",
348 IN_CREATE|IN_DELETE|IN_MOVE);
349 if (r < 0) {
350 manager_close_ask_password(m);
351 return r;
352 }
353
354 r = sd_event_add_io(m->event, &m->ask_password_event_source,
355 m->ask_password_inotify_fd, EPOLLIN,
356 manager_dispatch_ask_password_fd, m);
357 if (r < 0) {
358 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
359 manager_close_ask_password(m);
360 return r;
361 }
362
363 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
364
365 /* Queries might have been added meanwhile... */
366 manager_dispatch_ask_password_fd(m->ask_password_event_source,
367 m->ask_password_inotify_fd, EPOLLIN, m);
368 }
369
370 return m->have_ask_password;
371 }
372
373 static int manager_watch_idle_pipe(Manager *m) {
374 int r;
375
376 assert(m);
377
378 if (m->idle_pipe_event_source)
379 return 0;
380
381 if (m->idle_pipe[2] < 0)
382 return 0;
383
384 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
385 if (r < 0)
386 return log_error_errno(r, "Failed to watch idle pipe: %m");
387
388 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
389
390 return 0;
391 }
392
393 static void manager_close_idle_pipe(Manager *m) {
394 assert(m);
395
396 m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source);
397
398 safe_close_pair(m->idle_pipe);
399 safe_close_pair(m->idle_pipe + 2);
400 }
401
402 static int manager_setup_time_change(Manager *m) {
403 int r;
404
405 assert(m);
406
407 if (MANAGER_IS_TEST_RUN(m))
408 return 0;
409
410 m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source);
411 m->time_change_fd = safe_close(m->time_change_fd);
412
413 m->time_change_fd = time_change_fd();
414 if (m->time_change_fd < 0)
415 return log_error_errno(m->time_change_fd, "Failed to create timer change timer fd: %m");
416
417 r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
418 if (r < 0)
419 return log_error_errno(r, "Failed to create time change event source: %m");
420
421 /* Schedule this slightly earlier than the .timer event sources */
422 r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1);
423 if (r < 0)
424 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
425
426 (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
427
428 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
429
430 return 0;
431 }
432
433 static int manager_read_timezone_stat(Manager *m) {
434 struct stat st;
435 bool changed;
436
437 assert(m);
438
439 /* Read the current stat() data of /etc/localtime so that we detect changes */
440 if (lstat("/etc/localtime", &st) < 0) {
441 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
442 changed = m->etc_localtime_accessible;
443 m->etc_localtime_accessible = false;
444 } else {
445 usec_t k;
446
447 k = timespec_load(&st.st_mtim);
448 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
449
450 m->etc_localtime_mtime = k;
451 m->etc_localtime_accessible = true;
452 }
453
454 return changed;
455 }
456
457 static int manager_setup_timezone_change(Manager *m) {
458 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
459 int r;
460
461 assert(m);
462
463 if (MANAGER_IS_TEST_RUN(m))
464 return 0;
465
466 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
467 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
468 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
469 * went to zero and all fds to it are closed.
470 *
471 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
472 * correctly.
473 *
474 * Note that we create the new event source first here, before releasing the old one. This should optimize
475 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
476
477 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
478 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
479 if (r == -ENOENT) {
480 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
481 * O_CREATE or by rename() */
482
483 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
484 r = sd_event_add_inotify(m->event, &new_event, "/etc",
485 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
486 }
487 if (r < 0)
488 return log_error_errno(r, "Failed to create timezone change event source: %m");
489
490 /* Schedule this slightly earlier than the .timer event sources */
491 r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1);
492 if (r < 0)
493 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
494
495 sd_event_source_unref(m->timezone_change_event_source);
496 m->timezone_change_event_source = TAKE_PTR(new_event);
497
498 return 0;
499 }
500
501 static int enable_special_signals(Manager *m) {
502 _cleanup_close_ int fd = -1;
503
504 assert(m);
505
506 if (MANAGER_IS_TEST_RUN(m))
507 return 0;
508
509 /* Enable that we get SIGINT on control-alt-del. In containers
510 * this will fail with EPERM (older) or EINVAL (newer), so
511 * ignore that. */
512 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
513 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
514
515 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
516 if (fd < 0) {
517 /* Support systems without virtual console */
518 if (fd != -ENOENT)
519 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
520 } else {
521 /* Enable that we get SIGWINCH on kbrequest */
522 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
523 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
524 }
525
526 return 0;
527 }
528
529 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
530
531 static int manager_setup_signals(Manager *m) {
532 struct sigaction sa = {
533 .sa_handler = SIG_DFL,
534 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
535 };
536 sigset_t mask;
537 int r;
538
539 assert(m);
540
541 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
542
543 /* We make liberal use of realtime signals here. On
544 * Linux/glibc we have 30 of them (with the exception of Linux
545 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
546 * (aka SIGRTMAX). */
547
548 assert_se(sigemptyset(&mask) == 0);
549 sigset_add_many(&mask,
550 SIGCHLD, /* Child died */
551 SIGTERM, /* Reexecute daemon */
552 SIGHUP, /* Reload configuration */
553 SIGUSR1, /* systemd: reconnect to D-Bus */
554 SIGUSR2, /* systemd: dump status */
555 SIGINT, /* Kernel sends us this on control-alt-del */
556 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
557 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
558
559 SIGRTMIN+0, /* systemd: start default.target */
560 SIGRTMIN+1, /* systemd: isolate rescue.target */
561 SIGRTMIN+2, /* systemd: isolate emergency.target */
562 SIGRTMIN+3, /* systemd: start halt.target */
563 SIGRTMIN+4, /* systemd: start poweroff.target */
564 SIGRTMIN+5, /* systemd: start reboot.target */
565 SIGRTMIN+6, /* systemd: start kexec.target */
566
567 /* ... space for more special targets ... */
568
569 SIGRTMIN+13, /* systemd: Immediate halt */
570 SIGRTMIN+14, /* systemd: Immediate poweroff */
571 SIGRTMIN+15, /* systemd: Immediate reboot */
572 SIGRTMIN+16, /* systemd: Immediate kexec */
573
574 /* ... space for more immediate system state changes ... */
575
576 SIGRTMIN+20, /* systemd: enable status messages */
577 SIGRTMIN+21, /* systemd: disable status messages */
578 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
579 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
580 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
581 SIGRTMIN+25, /* systemd: reexecute manager */
582
583 /* Apparently Linux on hppa had fewer RT signals until v3.18,
584 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
585 * see commit v3.17-7614-g1f25df2eff.
586 *
587 * We cannot unconditionally make use of those signals here,
588 * so let's use a runtime check. Since these commands are
589 * accessible by different means and only really a safety
590 * net, the missing functionality on hppa shouldn't matter.
591 */
592
593 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
594 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
595 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
596 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
597
598 /* ... one free signal here SIGRTMIN+30 ... */
599 -1);
600 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
601
602 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
603 if (m->signal_fd < 0)
604 return -errno;
605
606 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
607 if (r < 0)
608 return r;
609
610 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
611
612 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
613 * notify processing can still figure out to which process/service a message belongs, before we reap the
614 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
615 * status information before detecting that there's no process in a cgroup. */
616 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
617 if (r < 0)
618 return r;
619
620 if (MANAGER_IS_SYSTEM(m))
621 return enable_special_signals(m);
622
623 return 0;
624 }
625
626 static char** sanitize_environment(char **l) {
627
628 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
629 strv_env_unset_many(
630 l,
631 "CACHE_DIRECTORY",
632 "CONFIGURATION_DIRECTORY",
633 "CREDENTIALS_DIRECTORY",
634 "EXIT_CODE",
635 "EXIT_STATUS",
636 "INVOCATION_ID",
637 "JOURNAL_STREAM",
638 "LISTEN_FDNAMES",
639 "LISTEN_FDS",
640 "LISTEN_PID",
641 "LOGS_DIRECTORY",
642 "MAINPID",
643 "MANAGERPID",
644 "NOTIFY_SOCKET",
645 "PIDFILE",
646 "REMOTE_ADDR",
647 "REMOTE_PORT",
648 "RUNTIME_DIRECTORY",
649 "SERVICE_RESULT",
650 "STATE_DIRECTORY",
651 "WATCHDOG_PID",
652 "WATCHDOG_USEC",
653 NULL);
654
655 /* Let's order the environment alphabetically, just to make it pretty */
656 return strv_sort(l);
657 }
658
659 int manager_default_environment(Manager *m) {
660 int r;
661
662 assert(m);
663
664 m->transient_environment = strv_free(m->transient_environment);
665
666 if (MANAGER_IS_SYSTEM(m)) {
667 /* The system manager always starts with a clean
668 * environment for its children. It does not import
669 * the kernel's or the parents' exported variables.
670 *
671 * The initial passed environment is untouched to keep
672 * /proc/self/environ valid; it is used for tagging
673 * the init process inside containers. */
674 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
675 if (!m->transient_environment)
676 return log_oom();
677
678 /* Import locale variables LC_*= from configuration */
679 (void) locale_setup(&m->transient_environment);
680 } else {
681 /* The user manager passes its own environment along to its children, except for $PATH. */
682 m->transient_environment = strv_copy(environ);
683 if (!m->transient_environment)
684 return log_oom();
685
686 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
687 if (r < 0)
688 return log_oom();
689 }
690
691 sanitize_environment(m->transient_environment);
692
693 return 0;
694 }
695
696 static int manager_setup_prefix(Manager *m) {
697 struct table_entry {
698 uint64_t type;
699 const char *suffix;
700 };
701
702 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
703 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
704 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
705 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
706 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
707 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
708 };
709
710 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
711 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
712 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
713 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
714 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
715 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
716 };
717
718 assert(m);
719
720 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
721 int r;
722
723 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
724 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
725 if (r < 0)
726 return log_warning_errno(r, "Failed to lookup %s path: %m",
727 exec_directory_type_to_string(i));
728 }
729
730 return 0;
731 }
732
733 static void manager_free_unit_name_maps(Manager *m) {
734 m->unit_id_map = hashmap_free(m->unit_id_map);
735 m->unit_name_map = hashmap_free(m->unit_name_map);
736 m->unit_path_cache = set_free(m->unit_path_cache);
737 m->unit_cache_timestamp_hash = 0;
738 }
739
740 static int manager_setup_run_queue(Manager *m) {
741 int r;
742
743 assert(m);
744 assert(!m->run_queue_event_source);
745
746 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
747 if (r < 0)
748 return r;
749
750 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
751 if (r < 0)
752 return r;
753
754 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
755 if (r < 0)
756 return r;
757
758 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
759
760 return 0;
761 }
762
763 static int manager_setup_sigchld_event_source(Manager *m) {
764 int r;
765
766 assert(m);
767 assert(!m->sigchld_event_source);
768
769 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
770 if (r < 0)
771 return r;
772
773 r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7);
774 if (r < 0)
775 return r;
776
777 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
778 if (r < 0)
779 return r;
780
781 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
782
783 return 0;
784 }
785
786 int manager_new(UnitFileScope scope, ManagerTestRunFlags test_run_flags, Manager **_m) {
787 _cleanup_(manager_freep) Manager *m = NULL;
788 const char *e;
789 int r;
790
791 assert(_m);
792 assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
793
794 m = new(Manager, 1);
795 if (!m)
796 return -ENOMEM;
797
798 *m = (Manager) {
799 .unit_file_scope = scope,
800 .objective = _MANAGER_OBJECTIVE_INVALID,
801
802 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
803
804 .default_timer_accuracy_usec = USEC_PER_MINUTE,
805 .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
806 .default_tasks_accounting = true,
807 .default_tasks_max = TASKS_MAX_UNSET,
808 .default_timeout_start_usec = DEFAULT_TIMEOUT_USEC,
809 .default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC,
810 .default_restart_usec = DEFAULT_RESTART_USEC,
811
812 .original_log_level = -1,
813 .original_log_target = _LOG_TARGET_INVALID,
814
815 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
816 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
817 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
818
819 .show_status_overridden = _SHOW_STATUS_INVALID,
820
821 .notify_fd = -1,
822 .cgroups_agent_fd = -1,
823 .signal_fd = -1,
824 .time_change_fd = -1,
825 .user_lookup_fds = { -1, -1 },
826 .private_listen_fd = -1,
827 .dev_autofs_fd = -1,
828 .cgroup_inotify_fd = -1,
829 .pin_cgroupfs_fd = -1,
830 .ask_password_inotify_fd = -1,
831 .idle_pipe = { -1, -1, -1, -1},
832
833 /* start as id #1, so that we can leave #0 around as "null-like" value */
834 .current_job_id = 1,
835
836 .have_ask_password = -EINVAL, /* we don't know */
837 .first_boot = -1,
838 .test_run_flags = test_run_flags,
839
840 .default_oom_policy = OOM_STOP,
841 };
842
843 #if ENABLE_EFI
844 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
845 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
846 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
847 m->timestamps + MANAGER_TIMESTAMP_LOADER);
848 #endif
849
850 /* Prepare log fields we can use for structured logging */
851 if (MANAGER_IS_SYSTEM(m)) {
852 m->unit_log_field = "UNIT=";
853 m->unit_log_format_string = "UNIT=%s";
854
855 m->invocation_log_field = "INVOCATION_ID=";
856 m->invocation_log_format_string = "INVOCATION_ID=%s";
857 } else {
858 m->unit_log_field = "USER_UNIT=";
859 m->unit_log_format_string = "USER_UNIT=%s";
860
861 m->invocation_log_field = "USER_INVOCATION_ID=";
862 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
863 }
864
865 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
866 m->ctrl_alt_del_ratelimit = (RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
867
868 r = manager_default_environment(m);
869 if (r < 0)
870 return r;
871
872 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
873 if (r < 0)
874 return r;
875
876 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
877 if (r < 0)
878 return r;
879
880 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
881 if (r < 0)
882 return r;
883
884 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
885 if (r < 0)
886 return r;
887
888 r = manager_setup_prefix(m);
889 if (r < 0)
890 return r;
891
892 r = get_credentials_dir(&e);
893 if (r >= 0) {
894 m->received_credentials = strdup(e);
895 if (!m->received_credentials)
896 return -ENOMEM;
897 }
898
899 r = sd_event_default(&m->event);
900 if (r < 0)
901 return r;
902
903 r = manager_setup_run_queue(m);
904 if (r < 0)
905 return r;
906
907 if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) {
908 m->cgroup_root = strdup("");
909 if (!m->cgroup_root)
910 return -ENOMEM;
911 } else {
912 r = manager_setup_signals(m);
913 if (r < 0)
914 return r;
915
916 r = manager_setup_cgroup(m);
917 if (r < 0)
918 return r;
919
920 r = manager_setup_time_change(m);
921 if (r < 0)
922 return r;
923
924 r = manager_read_timezone_stat(m);
925 if (r < 0)
926 return r;
927
928 (void) manager_setup_timezone_change(m);
929
930 r = manager_setup_sigchld_event_source(m);
931 if (r < 0)
932 return r;
933 }
934
935 if (test_run_flags == 0) {
936 if (MANAGER_IS_SYSTEM(m))
937 r = mkdir_label("/run/systemd/units", 0755);
938 else {
939 _cleanup_free_ char *units_path = NULL;
940 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
941 if (r < 0)
942 return r;
943 r = mkdir_p_label(units_path, 0755);
944 }
945
946 if (r < 0 && r != -EEXIST)
947 return r;
948 }
949
950 m->taint_usr =
951 !in_initrd() &&
952 dir_is_empty("/usr") > 0;
953
954 /* Note that we do not set up the notify fd here. We do that after deserialization,
955 * since they might have gotten serialized across the reexec. */
956
957 *_m = TAKE_PTR(m);
958
959 return 0;
960 }
961
962 static int manager_setup_notify(Manager *m) {
963 int r;
964
965 if (MANAGER_IS_TEST_RUN(m))
966 return 0;
967
968 if (m->notify_fd < 0) {
969 _cleanup_close_ int fd = -1;
970 union sockaddr_union sa;
971 socklen_t sa_len;
972
973 /* First free all secondary fields */
974 m->notify_socket = mfree(m->notify_socket);
975 m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source);
976
977 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
978 if (fd < 0)
979 return log_error_errno(errno, "Failed to allocate notification socket: %m");
980
981 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
982
983 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
984 if (!m->notify_socket)
985 return log_oom();
986
987 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
988 if (r < 0)
989 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
990 m->notify_socket);
991 sa_len = r;
992
993 (void) mkdir_parents_label(m->notify_socket, 0755);
994 (void) sockaddr_un_unlink(&sa.un);
995
996 r = mac_selinux_bind(fd, &sa.sa, sa_len);
997 if (r < 0)
998 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
999
1000 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1001 if (r < 0)
1002 return log_error_errno(r, "SO_PASSCRED failed: %m");
1003
1004 m->notify_fd = TAKE_FD(fd);
1005
1006 log_debug("Using notification socket %s", m->notify_socket);
1007 }
1008
1009 if (!m->notify_event_source) {
1010 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
1011 if (r < 0)
1012 return log_error_errno(r, "Failed to allocate notify event source: %m");
1013
1014 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1015 * service an exit message belongs. */
1016 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8);
1017 if (r < 0)
1018 return log_error_errno(r, "Failed to set priority of notify event source: %m");
1019
1020 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
1021 }
1022
1023 return 0;
1024 }
1025
1026 static int manager_setup_cgroups_agent(Manager *m) {
1027
1028 static const union sockaddr_union sa = {
1029 .un.sun_family = AF_UNIX,
1030 .un.sun_path = "/run/systemd/cgroups-agent",
1031 };
1032 int r;
1033
1034 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1035 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1036 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1037 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1038 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1039 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1040 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1041 * we thus won't lose messages.
1042 *
1043 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1044 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1045 * bus for these messages. */
1046
1047 if (MANAGER_IS_TEST_RUN(m))
1048 return 0;
1049
1050 if (!MANAGER_IS_SYSTEM(m))
1051 return 0;
1052
1053 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1054 if (r < 0)
1055 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1056 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1057 return 0;
1058
1059 if (m->cgroups_agent_fd < 0) {
1060 _cleanup_close_ int fd = -1;
1061
1062 /* First free all secondary fields */
1063 m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source);
1064
1065 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1066 if (fd < 0)
1067 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1068
1069 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1070
1071 (void) sockaddr_un_unlink(&sa.un);
1072
1073 /* Only allow root to connect to this socket */
1074 RUN_WITH_UMASK(0077)
1075 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1076 if (r < 0)
1077 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1078
1079 m->cgroups_agent_fd = TAKE_FD(fd);
1080 }
1081
1082 if (!m->cgroups_agent_event_source) {
1083 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1084 if (r < 0)
1085 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1086
1087 /* Process cgroups notifications early. Note that when the agent notification is received
1088 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1089 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1090 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9);
1091 if (r < 0)
1092 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1093
1094 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1095 }
1096
1097 return 0;
1098 }
1099
1100 static int manager_setup_user_lookup_fd(Manager *m) {
1101 int r;
1102
1103 assert(m);
1104
1105 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1106 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1107 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1108 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1109 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1110 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1111 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1112 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1113 * this purpose.
1114 *
1115 * You might wonder why we need a communication channel for this that is independent of the usual notification
1116 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1117 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1118 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1119 *
1120 * Note that this function is called under two circumstances: when we first initialize (in which case we
1121 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1122 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1123
1124 if (m->user_lookup_fds[0] < 0) {
1125
1126 /* Free all secondary fields */
1127 safe_close_pair(m->user_lookup_fds);
1128 m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source);
1129
1130 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1131 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1132
1133 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1134 }
1135
1136 if (!m->user_lookup_event_source) {
1137 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1138 if (r < 0)
1139 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1140
1141 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1142 * resolutions */
1143 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11);
1144 if (r < 0)
1145 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1146
1147 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1148 }
1149
1150 return 0;
1151 }
1152
1153 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1154 Unit *u;
1155 unsigned n = 0;
1156
1157 assert(m);
1158
1159 while ((u = m->cleanup_queue)) {
1160 assert(u->in_cleanup_queue);
1161
1162 unit_free(u);
1163 n++;
1164 }
1165
1166 return n;
1167 }
1168
1169 enum {
1170 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1171 GC_OFFSET_UNSURE, /* No clue */
1172 GC_OFFSET_GOOD, /* We still need this unit */
1173 GC_OFFSET_BAD, /* We don't need this unit anymore */
1174 _GC_OFFSET_MAX
1175 };
1176
1177 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1178 Unit *other;
1179
1180 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1181
1182 /* Recursively mark referenced units as GOOD as well */
1183 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES)
1184 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1185 unit_gc_mark_good(other, gc_marker);
1186 }
1187
1188 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1189 Unit *other;
1190 bool is_bad;
1191
1192 assert(u);
1193
1194 if (IN_SET(u->gc_marker - gc_marker,
1195 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1196 return;
1197
1198 if (u->in_cleanup_queue)
1199 goto bad;
1200
1201 if (!unit_may_gc(u))
1202 goto good;
1203
1204 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1205
1206 is_bad = true;
1207
1208 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) {
1209 unit_gc_sweep(other, gc_marker);
1210
1211 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1212 goto good;
1213
1214 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1215 is_bad = false;
1216 }
1217
1218 const UnitRef *ref;
1219 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1220 unit_gc_sweep(ref->source, gc_marker);
1221
1222 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1223 goto good;
1224
1225 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1226 is_bad = false;
1227 }
1228
1229 if (is_bad)
1230 goto bad;
1231
1232 /* We were unable to find anything out about this entry, so
1233 * let's investigate it later */
1234 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1235 unit_add_to_gc_queue(u);
1236 return;
1237
1238 bad:
1239 /* We definitely know that this one is not useful anymore, so
1240 * let's mark it for deletion */
1241 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1242 unit_add_to_cleanup_queue(u);
1243 return;
1244
1245 good:
1246 unit_gc_mark_good(u, gc_marker);
1247 }
1248
1249 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1250 unsigned n = 0, gc_marker;
1251 Unit *u;
1252
1253 assert(m);
1254
1255 /* log_debug("Running GC..."); */
1256
1257 m->gc_marker += _GC_OFFSET_MAX;
1258 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1259 m->gc_marker = 1;
1260
1261 gc_marker = m->gc_marker;
1262
1263 while ((u = m->gc_unit_queue)) {
1264 assert(u->in_gc_queue);
1265
1266 unit_gc_sweep(u, gc_marker);
1267
1268 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1269 u->in_gc_queue = false;
1270
1271 n++;
1272
1273 if (IN_SET(u->gc_marker - gc_marker,
1274 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1275 if (u->id)
1276 log_unit_debug(u, "Collecting.");
1277 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1278 unit_add_to_cleanup_queue(u);
1279 }
1280 }
1281
1282 return n;
1283 }
1284
1285 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1286 unsigned n = 0;
1287 Job *j;
1288
1289 assert(m);
1290
1291 while ((j = m->gc_job_queue)) {
1292 assert(j->in_gc_queue);
1293
1294 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1295 j->in_gc_queue = false;
1296
1297 n++;
1298
1299 if (!job_may_gc(j))
1300 continue;
1301
1302 log_unit_debug(j->unit, "Collecting job.");
1303 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1304 }
1305
1306 return n;
1307 }
1308
1309 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1310 unsigned n = 0;
1311 Unit *u;
1312 int r;
1313
1314 assert(m);
1315
1316 while ((u = m->stop_when_unneeded_queue)) {
1317 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1318
1319 assert(u->in_stop_when_unneeded_queue);
1320 LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
1321 u->in_stop_when_unneeded_queue = false;
1322
1323 n++;
1324
1325 if (!unit_is_unneeded(u))
1326 continue;
1327
1328 log_unit_debug(u, "Unit is not needed anymore.");
1329
1330 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1331 * service being unnecessary after a while. */
1332
1333 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1334 log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1335 continue;
1336 }
1337
1338 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1339 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1340 if (r < 0)
1341 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1342 }
1343
1344 return n;
1345 }
1346
1347 static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) {
1348 unsigned n = 0;
1349 Unit *u;
1350 int r;
1351
1352 assert(m);
1353
1354 while ((u = m->start_when_upheld_queue)) {
1355 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1356 Unit *culprit = NULL;
1357
1358 assert(u->in_start_when_upheld_queue);
1359 LIST_REMOVE(start_when_upheld_queue, m->start_when_upheld_queue, u);
1360 u->in_start_when_upheld_queue = false;
1361
1362 n++;
1363
1364 if (!unit_is_upheld_by_active(u, &culprit))
1365 continue;
1366
1367 log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id);
1368
1369 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1370 * service being unnecessary after a while. */
1371
1372 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1373 log_unit_warning(u, "Unit needs to be started because active unit %s upholds it, but not starting since we tried this too often recently.", culprit->id);
1374 continue;
1375 }
1376
1377 r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL);
1378 if (r < 0)
1379 log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r));
1380 }
1381
1382 return n;
1383 }
1384
1385 static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) {
1386 unsigned n = 0;
1387 Unit *u;
1388 int r;
1389
1390 assert(m);
1391
1392 while ((u = m->stop_when_bound_queue)) {
1393 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1394 Unit *culprit = NULL;
1395
1396 assert(u->in_stop_when_bound_queue);
1397 LIST_REMOVE(stop_when_bound_queue, m->stop_when_bound_queue, u);
1398 u->in_stop_when_bound_queue = false;
1399
1400 n++;
1401
1402 if (!unit_is_bound_by_inactive(u, &culprit))
1403 continue;
1404
1405 log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id);
1406
1407 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1408 * service being unnecessary after a while. */
1409
1410 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1411 log_unit_warning(u, "Unit needs to be stopped because it is bound to inactive unit %s it, but not stopping since we tried this too often recently.", culprit->id);
1412 continue;
1413 }
1414
1415 r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL);
1416 if (r < 0)
1417 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1418 }
1419
1420 return n;
1421 }
1422
1423 static void manager_clear_jobs_and_units(Manager *m) {
1424 Unit *u;
1425
1426 assert(m);
1427
1428 while ((u = hashmap_first(m->units)))
1429 unit_free(u);
1430
1431 manager_dispatch_cleanup_queue(m);
1432
1433 assert(!m->load_queue);
1434 assert(prioq_isempty(m->run_queue));
1435 assert(!m->dbus_unit_queue);
1436 assert(!m->dbus_job_queue);
1437 assert(!m->cleanup_queue);
1438 assert(!m->gc_unit_queue);
1439 assert(!m->gc_job_queue);
1440 assert(!m->cgroup_realize_queue);
1441 assert(!m->cgroup_empty_queue);
1442 assert(!m->cgroup_oom_queue);
1443 assert(!m->target_deps_queue);
1444 assert(!m->stop_when_unneeded_queue);
1445 assert(!m->start_when_upheld_queue);
1446 assert(!m->stop_when_bound_queue);
1447
1448 assert(hashmap_isempty(m->jobs));
1449 assert(hashmap_isempty(m->units));
1450
1451 m->n_on_console = 0;
1452 m->n_running_jobs = 0;
1453 m->n_installed_jobs = 0;
1454 m->n_failed_jobs = 0;
1455 }
1456
1457 Manager* manager_free(Manager *m) {
1458 if (!m)
1459 return NULL;
1460
1461 manager_clear_jobs_and_units(m);
1462
1463 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1464 if (unit_vtable[c]->shutdown)
1465 unit_vtable[c]->shutdown(m);
1466
1467 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1468 manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1469
1470 lookup_paths_flush_generator(&m->lookup_paths);
1471
1472 bus_done(m);
1473 manager_varlink_done(m);
1474
1475 exec_runtime_vacuum(m);
1476 hashmap_free(m->exec_runtime_by_id);
1477
1478 dynamic_user_vacuum(m, false);
1479 hashmap_free(m->dynamic_users);
1480
1481 hashmap_free(m->units);
1482 hashmap_free(m->units_by_invocation_id);
1483 hashmap_free(m->jobs);
1484 hashmap_free(m->watch_pids);
1485 hashmap_free(m->watch_bus);
1486
1487 prioq_free(m->run_queue);
1488
1489 set_free(m->startup_units);
1490 set_free(m->failed_units);
1491
1492 sd_event_source_unref(m->signal_event_source);
1493 sd_event_source_unref(m->sigchld_event_source);
1494 sd_event_source_unref(m->notify_event_source);
1495 sd_event_source_unref(m->cgroups_agent_event_source);
1496 sd_event_source_unref(m->time_change_event_source);
1497 sd_event_source_unref(m->timezone_change_event_source);
1498 sd_event_source_unref(m->jobs_in_progress_event_source);
1499 sd_event_source_unref(m->run_queue_event_source);
1500 sd_event_source_unref(m->user_lookup_event_source);
1501
1502 safe_close(m->signal_fd);
1503 safe_close(m->notify_fd);
1504 safe_close(m->cgroups_agent_fd);
1505 safe_close(m->time_change_fd);
1506 safe_close_pair(m->user_lookup_fds);
1507
1508 manager_close_ask_password(m);
1509
1510 manager_close_idle_pipe(m);
1511
1512 sd_event_unref(m->event);
1513
1514 free(m->notify_socket);
1515
1516 lookup_paths_free(&m->lookup_paths);
1517 strv_free(m->transient_environment);
1518 strv_free(m->client_environment);
1519
1520 hashmap_free(m->cgroup_unit);
1521 manager_free_unit_name_maps(m);
1522
1523 free(m->switch_root);
1524 free(m->switch_root_init);
1525
1526 rlimit_free_all(m->rlimit);
1527
1528 assert(hashmap_isempty(m->units_requiring_mounts_for));
1529 hashmap_free(m->units_requiring_mounts_for);
1530
1531 hashmap_free(m->uid_refs);
1532 hashmap_free(m->gid_refs);
1533
1534 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1535 m->prefix[dt] = mfree(m->prefix[dt]);
1536 free(m->received_credentials);
1537
1538 return mfree(m);
1539 }
1540
1541 static void manager_enumerate_perpetual(Manager *m) {
1542 assert(m);
1543
1544 if (m->test_run_flags == MANAGER_TEST_RUN_MINIMAL)
1545 return;
1546
1547 /* Let's ask every type to load all units from disk/kernel that it might know */
1548 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1549 if (!unit_type_supported(c)) {
1550 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1551 continue;
1552 }
1553
1554 if (unit_vtable[c]->enumerate_perpetual)
1555 unit_vtable[c]->enumerate_perpetual(m);
1556 }
1557 }
1558
1559 static void manager_enumerate(Manager *m) {
1560 assert(m);
1561
1562 if (m->test_run_flags == MANAGER_TEST_RUN_MINIMAL)
1563 return;
1564
1565 /* Let's ask every type to load all units from disk/kernel that it might know */
1566 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1567 if (!unit_type_supported(c)) {
1568 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1569 continue;
1570 }
1571
1572 if (unit_vtable[c]->enumerate)
1573 unit_vtable[c]->enumerate(m);
1574 }
1575
1576 manager_dispatch_load_queue(m);
1577 }
1578
1579 static void manager_coldplug(Manager *m) {
1580 Unit *u;
1581 char *k;
1582 int r;
1583
1584 assert(m);
1585
1586 log_debug("Invoking unit coldplug() handlers…");
1587
1588 /* Let's place the units back into their deserialized state */
1589 HASHMAP_FOREACH_KEY(u, k, m->units) {
1590
1591 /* ignore aliases */
1592 if (u->id != k)
1593 continue;
1594
1595 r = unit_coldplug(u);
1596 if (r < 0)
1597 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1598 }
1599 }
1600
1601 static void manager_catchup(Manager *m) {
1602 Unit *u;
1603 char *k;
1604
1605 assert(m);
1606
1607 log_debug("Invoking unit catchup() handlers…");
1608
1609 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1610 HASHMAP_FOREACH_KEY(u, k, m->units) {
1611
1612 /* ignore aliases */
1613 if (u->id != k)
1614 continue;
1615
1616 unit_catchup(u);
1617 }
1618 }
1619
1620 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1621 Unit *u;
1622
1623 assert(m);
1624
1625 HASHMAP_FOREACH(u, m->units) {
1626
1627 if (fdset_size(fds) <= 0)
1628 break;
1629
1630 if (!UNIT_VTABLE(u)->distribute_fds)
1631 continue;
1632
1633 UNIT_VTABLE(u)->distribute_fds(u, fds);
1634 }
1635 }
1636
1637 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1638 Unit *u;
1639
1640 assert(m);
1641
1642 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1643 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1644 * rather than the current one. */
1645
1646 if (MANAGER_IS_TEST_RUN(m))
1647 return false;
1648
1649 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1650 if (!u)
1651 return false;
1652 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1653 return false;
1654
1655 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1656 if (!u)
1657 return false;
1658 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), SERVICE_RUNNING, SERVICE_RELOAD))
1659 return false;
1660
1661 return true;
1662 }
1663
1664 static void manager_setup_bus(Manager *m) {
1665 assert(m);
1666
1667 /* Let's set up our private bus connection now, unconditionally */
1668 (void) bus_init_private(m);
1669
1670 /* If we are in --user mode also connect to the system bus now */
1671 if (MANAGER_IS_USER(m))
1672 (void) bus_init_system(m);
1673
1674 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1675 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1676 (void) bus_init_api(m);
1677
1678 if (MANAGER_IS_SYSTEM(m))
1679 (void) bus_init_system(m);
1680 }
1681 }
1682
1683 static void manager_preset_all(Manager *m) {
1684 int r;
1685
1686 assert(m);
1687
1688 if (m->first_boot <= 0)
1689 return;
1690
1691 if (!MANAGER_IS_SYSTEM(m))
1692 return;
1693
1694 if (MANAGER_IS_TEST_RUN(m))
1695 return;
1696
1697 /* If this is the first boot, and we are in the host system, then preset everything */
1698 r = unit_file_preset_all(UNIT_FILE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0);
1699 if (r < 0)
1700 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1701 "Failed to populate /etc with preset unit settings, ignoring: %m");
1702 else
1703 log_info("Populated /etc with preset unit settings.");
1704 }
1705
1706 static void manager_ready(Manager *m) {
1707 assert(m);
1708
1709 /* After having loaded everything, do the final round of catching up with what might have changed */
1710
1711 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1712
1713 /* It might be safe to log to the journal now and connect to dbus */
1714 manager_recheck_journal(m);
1715 manager_recheck_dbus(m);
1716
1717 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1718 manager_catchup(m);
1719
1720 m->honor_device_enumeration = true;
1721 }
1722
1723 Manager* manager_reloading_start(Manager *m) {
1724 m->n_reloading++;
1725 return m;
1726 }
1727 void manager_reloading_stopp(Manager **m) {
1728 if (*m) {
1729 assert((*m)->n_reloading > 0);
1730 (*m)->n_reloading--;
1731 }
1732 }
1733
1734 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1735 int r;
1736
1737 assert(m);
1738
1739 /* If we are running in test mode, we still want to run the generators,
1740 * but we should not touch the real generator directories. */
1741 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope,
1742 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1743 NULL);
1744 if (r < 0)
1745 return log_error_errno(r, "Failed to initialize path lookup table: %m");
1746
1747 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1748 r = manager_run_environment_generators(m);
1749 if (r >= 0)
1750 r = manager_run_generators(m);
1751 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1752 if (r < 0)
1753 return r;
1754
1755 manager_preset_all(m);
1756
1757 lookup_paths_log(&m->lookup_paths);
1758
1759 {
1760 /* This block is (optionally) done with the reloading counter bumped */
1761 _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1762
1763 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1764 * counter here already */
1765 if (serialization)
1766 reloading = manager_reloading_start(m);
1767
1768 /* First, enumerate what we can from all config files */
1769 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1770 manager_enumerate_perpetual(m);
1771 manager_enumerate(m);
1772 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1773
1774 /* Second, deserialize if there is something to deserialize */
1775 if (serialization) {
1776 r = manager_deserialize(m, serialization, fds);
1777 if (r < 0)
1778 return log_error_errno(r, "Deserialization failed: %m");
1779 }
1780
1781 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1782 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1783 * containers. */
1784 manager_distribute_fds(m, fds);
1785
1786 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1787 r = manager_setup_notify(m);
1788 if (r < 0)
1789 /* No sense to continue without notifications, our children would fail anyway. */
1790 return r;
1791
1792 r = manager_setup_cgroups_agent(m);
1793 if (r < 0)
1794 /* Likewise, no sense to continue without empty cgroup notifications. */
1795 return r;
1796
1797 r = manager_setup_user_lookup_fd(m);
1798 if (r < 0)
1799 /* This shouldn't fail, except if things are really broken. */
1800 return r;
1801
1802 /* Connect to the bus if we are good for it */
1803 manager_setup_bus(m);
1804
1805 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1806 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1807 if (r < 0)
1808 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
1809 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1810
1811 r = manager_varlink_init(m);
1812 if (r < 0)
1813 log_warning_errno(r, "Failed to set up Varlink server, ignoring: %m");
1814
1815 /* Third, fire things up! */
1816 manager_coldplug(m);
1817
1818 /* Clean up runtime objects */
1819 manager_vacuum(m);
1820
1821 if (serialization)
1822 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1823 * reload is finished */
1824 m->send_reloading_done = true;
1825 }
1826
1827 manager_ready(m);
1828
1829 return 0;
1830 }
1831
1832 int manager_add_job(
1833 Manager *m,
1834 JobType type,
1835 Unit *unit,
1836 JobMode mode,
1837 Set *affected_jobs,
1838 sd_bus_error *error,
1839 Job **ret) {
1840
1841 Transaction *tr;
1842 int r;
1843
1844 assert(m);
1845 assert(type < _JOB_TYPE_MAX);
1846 assert(unit);
1847 assert(mode < _JOB_MODE_MAX);
1848
1849 if (mode == JOB_ISOLATE && type != JOB_START)
1850 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1851
1852 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1853 return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1854
1855 if (mode == JOB_TRIGGERING && type != JOB_STOP)
1856 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
1857
1858 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1859
1860 type = job_type_collapse(type, unit);
1861
1862 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1863 if (!tr)
1864 return -ENOMEM;
1865
1866 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1867 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1868 mode == JOB_IGNORE_DEPENDENCIES, error);
1869 if (r < 0)
1870 goto tr_abort;
1871
1872 if (mode == JOB_ISOLATE) {
1873 r = transaction_add_isolate_jobs(tr, m);
1874 if (r < 0)
1875 goto tr_abort;
1876 }
1877
1878 if (mode == JOB_TRIGGERING) {
1879 r = transaction_add_triggering_jobs(tr, unit);
1880 if (r < 0)
1881 goto tr_abort;
1882 }
1883
1884 r = transaction_activate(tr, m, mode, affected_jobs, error);
1885 if (r < 0)
1886 goto tr_abort;
1887
1888 log_unit_debug(unit,
1889 "Enqueued job %s/%s as %u", unit->id,
1890 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1891
1892 if (ret)
1893 *ret = tr->anchor_job;
1894
1895 transaction_free(tr);
1896 return 0;
1897
1898 tr_abort:
1899 transaction_abort(tr);
1900 transaction_free(tr);
1901 return r;
1902 }
1903
1904 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
1905 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1906 int r;
1907
1908 assert(m);
1909 assert(type < _JOB_TYPE_MAX);
1910 assert(name);
1911 assert(mode < _JOB_MODE_MAX);
1912
1913 r = manager_load_unit(m, name, NULL, NULL, &unit);
1914 if (r < 0)
1915 return r;
1916 assert(unit);
1917
1918 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
1919 }
1920
1921 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
1922 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1923 int r;
1924
1925 assert(m);
1926 assert(type < _JOB_TYPE_MAX);
1927 assert(name);
1928 assert(mode < _JOB_MODE_MAX);
1929
1930 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
1931 if (r < 0)
1932 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1933
1934 return r;
1935 }
1936
1937 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1938 int r;
1939 Transaction *tr;
1940
1941 assert(m);
1942 assert(unit);
1943 assert(mode < _JOB_MODE_MAX);
1944 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1945
1946 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1947 if (!tr)
1948 return -ENOMEM;
1949
1950 /* We need an anchor job */
1951 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
1952 if (r < 0)
1953 goto tr_abort;
1954
1955 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1956 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
1957
1958 r = transaction_activate(tr, m, mode, NULL, e);
1959 if (r < 0)
1960 goto tr_abort;
1961
1962 transaction_free(tr);
1963 return 0;
1964
1965 tr_abort:
1966 transaction_abort(tr);
1967 transaction_free(tr);
1968 return r;
1969 }
1970
1971 Job *manager_get_job(Manager *m, uint32_t id) {
1972 assert(m);
1973
1974 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1975 }
1976
1977 Unit *manager_get_unit(Manager *m, const char *name) {
1978 assert(m);
1979 assert(name);
1980
1981 return hashmap_get(m->units, name);
1982 }
1983
1984 static int manager_dispatch_target_deps_queue(Manager *m) {
1985 Unit *u;
1986 int r = 0;
1987
1988 assert(m);
1989
1990 while ((u = m->target_deps_queue)) {
1991 _cleanup_free_ Unit **targets = NULL;
1992 int n_targets;
1993
1994 assert(u->in_target_deps_queue);
1995
1996 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
1997 u->in_target_deps_queue = false;
1998
1999 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2000 * dependencies, and we can't have it that hash tables we iterate through are modified while
2001 * we are iterating through them. */
2002 n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets);
2003 if (n_targets < 0)
2004 return n_targets;
2005
2006 for (int i = 0; i < n_targets; i++) {
2007 r = unit_add_default_target_dependency(u, targets[i]);
2008 if (r < 0)
2009 return r;
2010 }
2011 }
2012
2013 return r;
2014 }
2015
2016 unsigned manager_dispatch_load_queue(Manager *m) {
2017 Unit *u;
2018 unsigned n = 0;
2019
2020 assert(m);
2021
2022 /* Make sure we are not run recursively */
2023 if (m->dispatching_load_queue)
2024 return 0;
2025
2026 m->dispatching_load_queue = true;
2027
2028 /* Dispatches the load queue. Takes a unit from the queue and
2029 * tries to load its data until the queue is empty */
2030
2031 while ((u = m->load_queue)) {
2032 assert(u->in_load_queue);
2033
2034 unit_load(u);
2035 n++;
2036 }
2037
2038 m->dispatching_load_queue = false;
2039
2040 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2041 * should be loaded and have aliases resolved */
2042 (void) manager_dispatch_target_deps_queue(m);
2043
2044 return n;
2045 }
2046
2047 bool manager_unit_cache_should_retry_load(Unit *u) {
2048 assert(u);
2049
2050 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2051 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2052 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2053 if (u->load_state != UNIT_NOT_FOUND)
2054 return false;
2055
2056 /* The cache has been updated since the last time we tried to load the unit. There might be new
2057 * fragment paths to read. */
2058 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
2059 return true;
2060
2061 /* The cache needs to be updated because there are modifications on disk. */
2062 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
2063 }
2064
2065 int manager_load_unit_prepare(
2066 Manager *m,
2067 const char *name,
2068 const char *path,
2069 sd_bus_error *e,
2070 Unit **_ret) {
2071
2072 _cleanup_(unit_freep) Unit *cleanup_ret = NULL;
2073 Unit *ret;
2074 UnitType t;
2075 int r;
2076
2077 assert(m);
2078 assert(_ret);
2079
2080 /* This will prepare the unit for loading, but not actually load anything from disk. */
2081
2082 if (path && !path_is_absolute(path))
2083 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
2084
2085 if (!name) {
2086 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2087 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2088 * but this cannot be possible in any code path (See #6119). */
2089 assert_se(path);
2090 name = basename(path);
2091 }
2092
2093 t = unit_name_to_type(name);
2094
2095 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
2096 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
2097 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
2098
2099 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
2100 }
2101
2102 ret = manager_get_unit(m, name);
2103 if (ret) {
2104 /* The time-based cache allows to start new units without daemon-reload,
2105 * but if they are already referenced (because of dependencies or ordering)
2106 * then we have to force a load of the fragment. As an optimization, check
2107 * first if anything in the usual paths was modified since the last time
2108 * the cache was loaded. Also check if the last time an attempt to load the
2109 * unit was made was before the most recent cache refresh, so that we know
2110 * we need to try again — even if the cache is current, it might have been
2111 * updated in a different context before we had a chance to retry loading
2112 * this particular unit. */
2113 if (manager_unit_cache_should_retry_load(ret))
2114 ret->load_state = UNIT_STUB;
2115 else {
2116 *_ret = ret;
2117 return 1;
2118 }
2119 } else {
2120 ret = cleanup_ret = unit_new(m, unit_vtable[t]->object_size);
2121 if (!ret)
2122 return -ENOMEM;
2123 }
2124
2125 if (path) {
2126 r = free_and_strdup(&ret->fragment_path, path);
2127 if (r < 0)
2128 return r;
2129 }
2130
2131 r = unit_add_name(ret, name);
2132 if (r < 0)
2133 return r;
2134
2135 unit_add_to_load_queue(ret);
2136 unit_add_to_dbus_queue(ret);
2137 unit_add_to_gc_queue(ret);
2138
2139 *_ret = ret;
2140 cleanup_ret = NULL;
2141
2142 return 0;
2143 }
2144
2145 int manager_load_unit(
2146 Manager *m,
2147 const char *name,
2148 const char *path,
2149 sd_bus_error *e,
2150 Unit **_ret) {
2151
2152 int r;
2153
2154 assert(m);
2155 assert(_ret);
2156
2157 /* This will load the service information files, but not actually
2158 * start any services or anything. */
2159
2160 r = manager_load_unit_prepare(m, name, path, e, _ret);
2161 if (r != 0)
2162 return r;
2163
2164 manager_dispatch_load_queue(m);
2165
2166 *_ret = unit_follow_merge(*_ret);
2167 return 0;
2168 }
2169
2170 int manager_load_startable_unit_or_warn(
2171 Manager *m,
2172 const char *name,
2173 const char *path,
2174 Unit **ret) {
2175
2176 /* Load a unit, make sure it loaded fully and is not masked. */
2177
2178 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2179 Unit *unit;
2180 int r;
2181
2182 r = manager_load_unit(m, name, path, &error, &unit);
2183 if (r < 0)
2184 return log_error_errno(r, "Failed to load %s %s: %s",
2185 name ? "unit" : "unit file", name ?: path,
2186 bus_error_message(&error, r));
2187
2188 r = bus_unit_validate_load_state(unit, &error);
2189 if (r < 0)
2190 return log_error_errno(r, "%s", bus_error_message(&error, r));
2191
2192 *ret = unit;
2193 return 0;
2194 }
2195
2196 void manager_clear_jobs(Manager *m) {
2197 Job *j;
2198
2199 assert(m);
2200
2201 while ((j = hashmap_first(m->jobs)))
2202 /* No need to recurse. We're cancelling all jobs. */
2203 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2204 }
2205
2206 void manager_unwatch_pid(Manager *m, pid_t pid) {
2207 assert(m);
2208
2209 /* First let's drop the unit keyed as "pid". */
2210 (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid));
2211
2212 /* Then, let's also drop the array keyed by -pid. */
2213 free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid)));
2214 }
2215
2216 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2217 Manager *m = userdata;
2218 Job *j;
2219
2220 assert(source);
2221 assert(m);
2222
2223 while ((j = prioq_peek(m->run_queue))) {
2224 assert(j->installed);
2225 assert(j->in_run_queue);
2226
2227 (void) job_run_and_invalidate(j);
2228 }
2229
2230 if (m->n_running_jobs > 0)
2231 manager_watch_jobs_in_progress(m);
2232
2233 if (m->n_on_console > 0)
2234 manager_watch_idle_pipe(m);
2235
2236 return 1;
2237 }
2238
2239 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2240 unsigned n = 0, budget;
2241 Unit *u;
2242 Job *j;
2243
2244 assert(m);
2245
2246 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2247 * as we can. There's no point in throttling generation of signals in that case. */
2248 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2249 budget = UINT_MAX; /* infinite budget in this case */
2250 else {
2251 /* Anything to do at all? */
2252 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2253 return 0;
2254
2255 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2256 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2257 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2258 return 0;
2259
2260 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2261 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2262 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2263 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2264 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2265 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2266 * connections it will be counted five times. This difference in counting ("references"
2267 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2268 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2269 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2270 * currently chosen much higher than the "budget". */
2271 budget = MANAGER_BUS_MESSAGE_BUDGET;
2272 }
2273
2274 while (budget != 0 && (u = m->dbus_unit_queue)) {
2275
2276 assert(u->in_dbus_queue);
2277
2278 bus_unit_send_change_signal(u);
2279 n++;
2280
2281 if (budget != UINT_MAX)
2282 budget--;
2283 }
2284
2285 while (budget != 0 && (j = m->dbus_job_queue)) {
2286 assert(j->in_dbus_queue);
2287
2288 bus_job_send_change_signal(j);
2289 n++;
2290
2291 if (budget != UINT_MAX)
2292 budget--;
2293 }
2294
2295 if (m->send_reloading_done) {
2296 m->send_reloading_done = false;
2297 bus_manager_send_reloading(m, false);
2298 n++;
2299 }
2300
2301 if (m->pending_reload_message) {
2302 bus_send_pending_reload_message(m);
2303 n++;
2304 }
2305
2306 return n;
2307 }
2308
2309 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2310 Manager *m = userdata;
2311 char buf[PATH_MAX];
2312 ssize_t n;
2313
2314 n = recv(fd, buf, sizeof(buf), 0);
2315 if (n < 0)
2316 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2317 if (n == 0) {
2318 log_error("Got zero-length cgroups agent message, ignoring.");
2319 return 0;
2320 }
2321 if ((size_t) n >= sizeof(buf)) {
2322 log_error("Got overly long cgroups agent message, ignoring.");
2323 return 0;
2324 }
2325
2326 if (memchr(buf, 0, n)) {
2327 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2328 return 0;
2329 }
2330 buf[n] = 0;
2331
2332 manager_notify_cgroup_empty(m, buf);
2333 (void) bus_forward_agent_released(m, buf);
2334
2335 return 0;
2336 }
2337
2338 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2339
2340 /* nothing else must be sent when using BARRIER=1 */
2341 if (strv_contains(tags, "BARRIER=1")) {
2342 if (strv_length(tags) == 1) {
2343 if (fdset_size(fds) != 1)
2344 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2345 } else
2346 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2347
2348 /* Drop the message if BARRIER=1 was found */
2349 return true;
2350 }
2351
2352 return false;
2353 }
2354
2355 static void manager_invoke_notify_message(
2356 Manager *m,
2357 Unit *u,
2358 const struct ucred *ucred,
2359 char * const *tags,
2360 FDSet *fds) {
2361
2362 assert(m);
2363 assert(u);
2364 assert(ucred);
2365 assert(tags);
2366
2367 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2368 return;
2369 u->notifygen = m->notifygen;
2370
2371 if (UNIT_VTABLE(u)->notify_message)
2372 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2373
2374 else if (DEBUG_LOGGING) {
2375 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2376
2377 buf = strv_join(tags, ", ");
2378 if (buf)
2379 x = ellipsize(buf, 20, 90);
2380 if (x)
2381 y = cescape(x);
2382
2383 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2384 }
2385 }
2386
2387 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2388
2389 _cleanup_fdset_free_ FDSet *fds = NULL;
2390 Manager *m = userdata;
2391 char buf[NOTIFY_BUFFER_MAX+1];
2392 struct iovec iovec = {
2393 .iov_base = buf,
2394 .iov_len = sizeof(buf)-1,
2395 };
2396 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2397 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2398 struct msghdr msghdr = {
2399 .msg_iov = &iovec,
2400 .msg_iovlen = 1,
2401 .msg_control = &control,
2402 .msg_controllen = sizeof(control),
2403 };
2404
2405 struct cmsghdr *cmsg;
2406 struct ucred *ucred = NULL;
2407 _cleanup_free_ Unit **array_copy = NULL;
2408 _cleanup_strv_free_ char **tags = NULL;
2409 Unit *u1, *u2, **array;
2410 int r, *fd_array = NULL;
2411 size_t n_fds = 0;
2412 bool found = false;
2413 ssize_t n;
2414
2415 assert(m);
2416 assert(m->notify_fd == fd);
2417
2418 if (revents != EPOLLIN) {
2419 log_warning("Got unexpected poll event for notify fd.");
2420 return 0;
2421 }
2422
2423 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2424 if (IN_SET(n, -EAGAIN, -EINTR))
2425 return 0; /* Spurious wakeup, try again */
2426 if (n == -EXFULL) {
2427 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2428 return 0;
2429 }
2430 if (n < 0)
2431 /* If this is any other, real error, then let's stop processing this socket. This of course
2432 * means we won't take notification messages anymore, but that's still better than busy
2433 * looping around this: being woken up over and over again but being unable to actually read
2434 * the message off the socket. */
2435 return log_error_errno(n, "Failed to receive notification message: %m");
2436
2437 CMSG_FOREACH(cmsg, &msghdr) {
2438 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2439
2440 assert(!fd_array);
2441 fd_array = (int*) CMSG_DATA(cmsg);
2442 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2443
2444 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2445 cmsg->cmsg_type == SCM_CREDENTIALS &&
2446 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2447
2448 assert(!ucred);
2449 ucred = (struct ucred*) CMSG_DATA(cmsg);
2450 }
2451 }
2452
2453 if (n_fds > 0) {
2454 assert(fd_array);
2455
2456 r = fdset_new_array(&fds, fd_array, n_fds);
2457 if (r < 0) {
2458 close_many(fd_array, n_fds);
2459 log_oom();
2460 return 0;
2461 }
2462 }
2463
2464 if (!ucred || !pid_is_valid(ucred->pid)) {
2465 log_warning("Received notify message without valid credentials. Ignoring.");
2466 return 0;
2467 }
2468
2469 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2470 log_warning("Received notify message exceeded maximum size. Ignoring.");
2471 return 0;
2472 }
2473
2474 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2475 * trailing NUL byte in the message, but don't expect it. */
2476 if (n > 1 && memchr(buf, 0, n-1)) {
2477 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2478 return 0;
2479 }
2480
2481 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2482 buf[n] = 0;
2483 tags = strv_split_newlines(buf);
2484 if (!tags) {
2485 log_oom();
2486 return 0;
2487 }
2488
2489 /* possibly a barrier fd, let's see */
2490 if (manager_process_barrier_fd(tags, fds))
2491 return 0;
2492
2493 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2494 m->notifygen++;
2495
2496 /* Notify every unit that might be interested, which might be multiple. */
2497 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
2498 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid));
2499 array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid));
2500 if (array) {
2501 size_t k = 0;
2502
2503 while (array[k])
2504 k++;
2505
2506 array_copy = newdup(Unit*, array, k+1);
2507 if (!array_copy)
2508 log_oom();
2509 }
2510 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2511 * make sure we only invoke each unit's handler once. */
2512 if (u1) {
2513 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2514 found = true;
2515 }
2516 if (u2) {
2517 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2518 found = true;
2519 }
2520 if (array_copy)
2521 for (size_t i = 0; array_copy[i]; i++) {
2522 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2523 found = true;
2524 }
2525
2526 if (!found)
2527 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2528
2529 if (fdset_size(fds) > 0)
2530 log_warning("Got extra auxiliary fds with notification message, closing them.");
2531
2532 return 0;
2533 }
2534
2535 static void manager_invoke_sigchld_event(
2536 Manager *m,
2537 Unit *u,
2538 const siginfo_t *si) {
2539
2540 assert(m);
2541 assert(u);
2542 assert(si);
2543
2544 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2545 if (u->sigchldgen == m->sigchldgen)
2546 return;
2547 u->sigchldgen = m->sigchldgen;
2548
2549 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2550 unit_unwatch_pid(u, si->si_pid);
2551
2552 if (UNIT_VTABLE(u)->sigchld_event)
2553 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2554 }
2555
2556 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2557 Manager *m = userdata;
2558 siginfo_t si = {};
2559 int r;
2560
2561 assert(source);
2562 assert(m);
2563
2564 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2565 * while it is a zombie. */
2566
2567 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2568
2569 if (errno != ECHILD)
2570 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2571
2572 goto turn_off;
2573 }
2574
2575 if (si.si_pid <= 0)
2576 goto turn_off;
2577
2578 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2579 _cleanup_free_ Unit **array_copy = NULL;
2580 _cleanup_free_ char *name = NULL;
2581 Unit *u1, *u2, **array;
2582
2583 (void) get_process_comm(si.si_pid, &name);
2584
2585 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2586 si.si_pid, strna(name),
2587 sigchld_code_to_string(si.si_code),
2588 si.si_status,
2589 strna(si.si_code == CLD_EXITED
2590 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2591 : signal_to_string(si.si_status)));
2592
2593 /* Increase the generation counter used for filtering out duplicate unit invocations */
2594 m->sigchldgen++;
2595
2596 /* And now figure out the unit this belongs to, it might be multiple... */
2597 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2598 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid));
2599 array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid));
2600 if (array) {
2601 size_t n = 0;
2602
2603 /* Count how many entries the array has */
2604 while (array[n])
2605 n++;
2606
2607 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2608 array_copy = newdup(Unit*, array, n+1);
2609 if (!array_copy)
2610 log_oom();
2611 }
2612
2613 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2614 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2615 * each iteration. */
2616 if (u1) {
2617 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2618 * We only do this for the cgroup the PID belonged to. */
2619 (void) unit_check_oom(u1);
2620
2621 /* This only logs for now. In the future when the interface for kills/notifications
2622 * is more stable we can extend service results table similar to how kernel oom kills
2623 * are managed. */
2624 (void) unit_check_oomd_kill(u1);
2625
2626 manager_invoke_sigchld_event(m, u1, &si);
2627 }
2628 if (u2)
2629 manager_invoke_sigchld_event(m, u2, &si);
2630 if (array_copy)
2631 for (size_t i = 0; array_copy[i]; i++)
2632 manager_invoke_sigchld_event(m, array_copy[i], &si);
2633 }
2634
2635 /* And now, we actually reap the zombie. */
2636 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2637 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2638 return 0;
2639 }
2640
2641 return 0;
2642
2643 turn_off:
2644 /* All children processed for now, turn off event source */
2645
2646 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2647 if (r < 0)
2648 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2649
2650 return 0;
2651 }
2652
2653 static void manager_start_special(Manager *m, const char *name, JobMode mode) {
2654 Job *job;
2655
2656 if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0)
2657 return;
2658
2659 const char *s = unit_status_string(job->unit, NULL);
2660
2661 log_info("Activating special unit %s...", s);
2662
2663 sd_notifyf(false,
2664 "STATUS=Activating special unit %s...", s);
2665 m->status_ready = false;
2666 }
2667
2668 static void manager_handle_ctrl_alt_del(Manager *m) {
2669 /* If the user presses C-A-D more than
2670 * 7 times within 2s, we reboot/shutdown immediately,
2671 * unless it was disabled in system.conf */
2672
2673 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2674 manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2675 else
2676 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2677 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2678 }
2679
2680 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2681 Manager *m = userdata;
2682 ssize_t n;
2683 struct signalfd_siginfo sfsi;
2684 int r;
2685
2686 assert(m);
2687 assert(m->signal_fd == fd);
2688
2689 if (revents != EPOLLIN) {
2690 log_warning("Got unexpected events from signal file descriptor.");
2691 return 0;
2692 }
2693
2694 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2695 if (n != sizeof(sfsi)) {
2696 if (n >= 0) {
2697 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n);
2698 return 0;
2699 }
2700
2701 if (IN_SET(errno, EINTR, EAGAIN))
2702 return 0;
2703
2704 /* We return an error here, which will kill this handler,
2705 * to avoid a busy loop on read error. */
2706 return log_error_errno(errno, "Reading from signal fd failed: %m");
2707 }
2708
2709 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2710 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2711 ? LOG_DEBUG : LOG_INFO,
2712 &sfsi);
2713
2714 switch (sfsi.ssi_signo) {
2715
2716 case SIGCHLD:
2717 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2718 if (r < 0)
2719 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2720
2721 break;
2722
2723 case SIGTERM:
2724 if (MANAGER_IS_SYSTEM(m)) {
2725 /* This is for compatibility with the original sysvinit */
2726 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2727 break;
2728
2729 m->objective = MANAGER_REEXECUTE;
2730 break;
2731 }
2732
2733 _fallthrough_;
2734 case SIGINT:
2735 if (MANAGER_IS_SYSTEM(m))
2736 manager_handle_ctrl_alt_del(m);
2737 else
2738 manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY);
2739 break;
2740
2741 case SIGWINCH:
2742 /* This is a nop on non-init */
2743 if (MANAGER_IS_SYSTEM(m))
2744 manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2745
2746 break;
2747
2748 case SIGPWR:
2749 /* This is a nop on non-init */
2750 if (MANAGER_IS_SYSTEM(m))
2751 manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2752
2753 break;
2754
2755 case SIGUSR1:
2756 if (manager_dbus_is_running(m, false)) {
2757 log_info("Trying to reconnect to bus...");
2758
2759 (void) bus_init_api(m);
2760
2761 if (MANAGER_IS_SYSTEM(m))
2762 (void) bus_init_system(m);
2763 } else
2764 manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2765
2766 break;
2767
2768 case SIGUSR2: {
2769 _cleanup_free_ char *dump = NULL;
2770
2771 r = manager_get_dump_string(m, &dump);
2772 if (r < 0) {
2773 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2774 break;
2775 }
2776
2777 log_dump(LOG_INFO, dump);
2778 break;
2779 }
2780
2781 case SIGHUP:
2782 if (verify_run_space_and_log("Refusing to reload") < 0)
2783 break;
2784
2785 m->objective = MANAGER_RELOAD;
2786 break;
2787
2788 default: {
2789
2790 /* Starting SIGRTMIN+0 */
2791 static const struct {
2792 const char *target;
2793 JobMode mode;
2794 } target_table[] = {
2795 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2796 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2797 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2798 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2799 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2800 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2801 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
2802 };
2803
2804 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2805 static const ManagerObjective objective_table[] = {
2806 [0] = MANAGER_HALT,
2807 [1] = MANAGER_POWEROFF,
2808 [2] = MANAGER_REBOOT,
2809 [3] = MANAGER_KEXEC,
2810 };
2811
2812 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2813 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2814 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2815 manager_start_special(m, target_table[idx].target, target_table[idx].mode);
2816 break;
2817 }
2818
2819 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2820 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
2821 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
2822 break;
2823 }
2824
2825 switch (sfsi.ssi_signo - SIGRTMIN) {
2826
2827 case 20:
2828 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
2829 break;
2830
2831 case 21:
2832 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
2833 break;
2834
2835 case 22:
2836 manager_override_log_level(m, LOG_DEBUG);
2837 break;
2838
2839 case 23:
2840 manager_restore_original_log_level(m);
2841 break;
2842
2843 case 24:
2844 if (MANAGER_IS_USER(m)) {
2845 m->objective = MANAGER_EXIT;
2846 return 0;
2847 }
2848
2849 /* This is a nop on init */
2850 break;
2851
2852 case 25:
2853 m->objective = MANAGER_REEXECUTE;
2854 break;
2855
2856 case 26:
2857 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2858 manager_restore_original_log_target(m);
2859 break;
2860
2861 case 27:
2862 manager_override_log_target(m, LOG_TARGET_CONSOLE);
2863 break;
2864
2865 case 28:
2866 manager_override_log_target(m, LOG_TARGET_KMSG);
2867 break;
2868
2869 default:
2870 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2871 }
2872 }}
2873
2874 return 0;
2875 }
2876
2877 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2878 Manager *m = userdata;
2879 Unit *u;
2880
2881 assert(m);
2882 assert(m->time_change_fd == fd);
2883
2884 log_struct(LOG_DEBUG,
2885 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2886 LOG_MESSAGE("Time has been changed"));
2887
2888 /* Restart the watch */
2889 (void) manager_setup_time_change(m);
2890
2891 HASHMAP_FOREACH(u, m->units)
2892 if (UNIT_VTABLE(u)->time_change)
2893 UNIT_VTABLE(u)->time_change(u);
2894
2895 return 0;
2896 }
2897
2898 static int manager_dispatch_timezone_change(
2899 sd_event_source *source,
2900 const struct inotify_event *e,
2901 void *userdata) {
2902
2903 Manager *m = userdata;
2904 int changed;
2905 Unit *u;
2906
2907 assert(m);
2908
2909 log_debug("inotify event for /etc/localtime");
2910
2911 changed = manager_read_timezone_stat(m);
2912 if (changed <= 0)
2913 return changed;
2914
2915 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2916 (void) manager_setup_timezone_change(m);
2917
2918 /* Read the new timezone */
2919 tzset();
2920
2921 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
2922
2923 HASHMAP_FOREACH(u, m->units)
2924 if (UNIT_VTABLE(u)->timezone_change)
2925 UNIT_VTABLE(u)->timezone_change(u);
2926
2927 return 0;
2928 }
2929
2930 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2931 Manager *m = userdata;
2932
2933 assert(m);
2934 assert(m->idle_pipe[2] == fd);
2935
2936 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2937 * now turn off any further console output if there's at least one service that needs console access, so that
2938 * from now on our own output should not spill into that service's output anymore. After all, we support
2939 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2940 * exclusively without our interference. */
2941 m->no_console_output = m->n_on_console > 0;
2942
2943 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2944 * by closing the pipes towards them, which is what they are waiting for. */
2945 manager_close_idle_pipe(m);
2946
2947 return 0;
2948 }
2949
2950 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2951 Manager *m = userdata;
2952 int r;
2953
2954 assert(m);
2955 assert(source);
2956
2957 manager_print_jobs_in_progress(m);
2958
2959 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
2960 if (r < 0)
2961 return r;
2962
2963 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2964 }
2965
2966 int manager_loop(Manager *m) {
2967 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
2968 int r;
2969
2970 assert(m);
2971 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
2972
2973 manager_check_finished(m);
2974
2975 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
2976 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2977 if (r < 0)
2978 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
2979
2980 while (m->objective == MANAGER_OK) {
2981 usec_t wait_usec, watchdog_usec;
2982
2983 watchdog_usec = manager_get_watchdog(m, WATCHDOG_RUNTIME);
2984 if (m->runtime_watchdog_running)
2985 (void) watchdog_ping();
2986 else if (timestamp_is_set(watchdog_usec))
2987 manager_retry_runtime_watchdog(m);
2988
2989 if (!ratelimit_below(&rl)) {
2990 /* Yay, something is going seriously wrong, pause a little */
2991 log_warning("Looping too fast. Throttling execution a little.");
2992 sleep(1);
2993 }
2994
2995 if (manager_dispatch_load_queue(m) > 0)
2996 continue;
2997
2998 if (manager_dispatch_gc_job_queue(m) > 0)
2999 continue;
3000
3001 if (manager_dispatch_gc_unit_queue(m) > 0)
3002 continue;
3003
3004 if (manager_dispatch_cleanup_queue(m) > 0)
3005 continue;
3006
3007 if (manager_dispatch_cgroup_realize_queue(m) > 0)
3008 continue;
3009
3010 if (manager_dispatch_start_when_upheld_queue(m) > 0)
3011 continue;
3012
3013 if (manager_dispatch_stop_when_bound_queue(m) > 0)
3014 continue;
3015
3016 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
3017 continue;
3018
3019 if (manager_dispatch_dbus_queue(m) > 0)
3020 continue;
3021
3022 /* Sleep for watchdog runtime wait time */
3023 if (timestamp_is_set(watchdog_usec))
3024 wait_usec = watchdog_runtime_wait();
3025 else
3026 wait_usec = USEC_INFINITY;
3027
3028 r = sd_event_run(m->event, wait_usec);
3029 if (r < 0)
3030 return log_error_errno(r, "Failed to run event loop: %m");
3031 }
3032
3033 return m->objective;
3034 }
3035
3036 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
3037 _cleanup_free_ char *n = NULL;
3038 sd_id128_t invocation_id;
3039 Unit *u;
3040 int r;
3041
3042 assert(m);
3043 assert(s);
3044 assert(_u);
3045
3046 r = unit_name_from_dbus_path(s, &n);
3047 if (r < 0)
3048 return r;
3049
3050 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
3051 * as invocation ID. */
3052 r = sd_id128_from_string(n, &invocation_id);
3053 if (r >= 0) {
3054 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3055 if (u) {
3056 *_u = u;
3057 return 0;
3058 }
3059
3060 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3061 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3062 SD_ID128_FORMAT_VAL(invocation_id));
3063 }
3064
3065 /* If this didn't work, we check if this is a unit name */
3066 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3067 _cleanup_free_ char *nn = NULL;
3068
3069 nn = cescape(n);
3070 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3071 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3072 }
3073
3074 r = manager_load_unit(m, n, NULL, e, &u);
3075 if (r < 0)
3076 return r;
3077
3078 *_u = u;
3079 return 0;
3080 }
3081
3082 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3083 const char *p;
3084 unsigned id;
3085 Job *j;
3086 int r;
3087
3088 assert(m);
3089 assert(s);
3090 assert(_j);
3091
3092 p = startswith(s, "/org/freedesktop/systemd1/job/");
3093 if (!p)
3094 return -EINVAL;
3095
3096 r = safe_atou(p, &id);
3097 if (r < 0)
3098 return r;
3099
3100 j = manager_get_job(m, id);
3101 if (!j)
3102 return -ENOENT;
3103
3104 *_j = j;
3105
3106 return 0;
3107 }
3108
3109 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3110
3111 #if HAVE_AUDIT
3112 _cleanup_free_ char *p = NULL;
3113 const char *msg;
3114 int audit_fd, r;
3115
3116 if (!MANAGER_IS_SYSTEM(m))
3117 return;
3118
3119 audit_fd = get_audit_fd();
3120 if (audit_fd < 0)
3121 return;
3122
3123 /* Don't generate audit events if the service was already
3124 * started and we're just deserializing */
3125 if (MANAGER_IS_RELOADING(m))
3126 return;
3127
3128 if (u->type != UNIT_SERVICE)
3129 return;
3130
3131 r = unit_name_to_prefix_and_instance(u->id, &p);
3132 if (r < 0) {
3133 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
3134 return;
3135 }
3136
3137 msg = strjoina("unit=", p);
3138 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3139 if (errno == EPERM)
3140 /* We aren't allowed to send audit messages?
3141 * Then let's not retry again. */
3142 close_audit_fd();
3143 else
3144 log_warning_errno(errno, "Failed to send audit message: %m");
3145 }
3146 #endif
3147
3148 }
3149
3150 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3151 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
3152 _cleanup_free_ char *message = NULL;
3153 _cleanup_close_ int fd = -1;
3154 int n = 0;
3155
3156 /* Don't generate plymouth events if the service was already
3157 * started and we're just deserializing */
3158 if (MANAGER_IS_RELOADING(m))
3159 return;
3160
3161 if (!MANAGER_IS_SYSTEM(m))
3162 return;
3163
3164 if (detect_container() > 0)
3165 return;
3166
3167 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
3168 return;
3169
3170 /* We set SOCK_NONBLOCK here so that we rather drop the
3171 * message then wait for plymouth */
3172 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3173 if (fd < 0) {
3174 log_error_errno(errno, "socket() failed: %m");
3175 return;
3176 }
3177
3178 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
3179 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3180 log_error_errno(errno, "connect() failed: %m");
3181 return;
3182 }
3183
3184 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0)
3185 return (void) log_oom();
3186
3187 errno = 0;
3188 if (write(fd, message, n + 1) != n + 1)
3189 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3190 log_error_errno(errno, "Failed to write Plymouth message: %m");
3191 }
3192
3193 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3194 assert(m);
3195
3196 if (MANAGER_IS_USER(m))
3197 return USEC_INFINITY;
3198
3199 if (timestamp_is_set(m->watchdog_overridden[t]))
3200 return m->watchdog_overridden[t];
3201
3202 return m->watchdog[t];
3203 }
3204
3205 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3206 int r = 0;
3207
3208 assert(m);
3209
3210 if (MANAGER_IS_USER(m))
3211 return;
3212
3213 if (m->watchdog[t] == timeout)
3214 return;
3215
3216 if (t == WATCHDOG_RUNTIME)
3217 if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME])) {
3218 if (timestamp_is_set(timeout)) {
3219 r = watchdog_set_timeout(&timeout);
3220
3221 if (r >= 0)
3222 m->runtime_watchdog_running = true;
3223 } else {
3224 watchdog_close(true);
3225 m->runtime_watchdog_running = false;
3226 }
3227 }
3228
3229 m->watchdog[t] = timeout;
3230 }
3231
3232 int manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3233 int r = 0;
3234
3235 assert(m);
3236
3237 if (MANAGER_IS_USER(m))
3238 return 0;
3239
3240 if (m->watchdog_overridden[t] == timeout)
3241 return 0;
3242
3243 if (t == WATCHDOG_RUNTIME) {
3244 usec_t *p;
3245
3246 p = timestamp_is_set(timeout) ? &timeout : &m->watchdog[t];
3247 if (timestamp_is_set(*p)) {
3248 r = watchdog_set_timeout(p);
3249
3250 if (r >= 0)
3251 m->runtime_watchdog_running = true;
3252 } else {
3253 watchdog_close(true);
3254 m->runtime_watchdog_running = false;
3255 }
3256 }
3257
3258 m->watchdog_overridden[t] = timeout;
3259
3260 return 0;
3261 }
3262
3263 void manager_retry_runtime_watchdog(Manager *m) {
3264 int r = 0;
3265
3266 assert(m);
3267
3268 if (timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME]))
3269 r = watchdog_set_timeout(&m->watchdog_overridden[WATCHDOG_RUNTIME]);
3270 else
3271 r = watchdog_set_timeout(&m->watchdog[WATCHDOG_RUNTIME]);
3272
3273 if (r >= 0)
3274 m->runtime_watchdog_running = true;
3275 }
3276
3277 int manager_reload(Manager *m) {
3278 _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3279 _cleanup_fdset_free_ FDSet *fds = NULL;
3280 _cleanup_fclose_ FILE *f = NULL;
3281 int r;
3282
3283 assert(m);
3284
3285 r = manager_open_serialization(m, &f);
3286 if (r < 0)
3287 return log_error_errno(r, "Failed to create serialization file: %m");
3288
3289 fds = fdset_new();
3290 if (!fds)
3291 return log_oom();
3292
3293 /* We are officially in reload mode from here on. */
3294 reloading = manager_reloading_start(m);
3295
3296 r = manager_serialize(m, f, fds, false);
3297 if (r < 0)
3298 return r;
3299
3300 if (fseeko(f, 0, SEEK_SET) < 0)
3301 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3302
3303 /* 💀 This is the point of no return, from here on there is no way back. 💀 */
3304 reloading = NULL;
3305
3306 bus_manager_send_reloading(m, true);
3307
3308 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3309 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3310 * it. */
3311
3312 manager_clear_jobs_and_units(m);
3313 lookup_paths_flush_generator(&m->lookup_paths);
3314 lookup_paths_free(&m->lookup_paths);
3315 exec_runtime_vacuum(m);
3316 dynamic_user_vacuum(m, false);
3317 m->uid_refs = hashmap_free(m->uid_refs);
3318 m->gid_refs = hashmap_free(m->gid_refs);
3319
3320 r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
3321 if (r < 0)
3322 log_warning_errno(r, "Failed to initialize path lookup table, ignoring: %m");
3323
3324 (void) manager_run_environment_generators(m);
3325 (void) manager_run_generators(m);
3326
3327 lookup_paths_log(&m->lookup_paths);
3328
3329 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3330 manager_free_unit_name_maps(m);
3331
3332 /* First, enumerate what we can from kernel and suchlike */
3333 manager_enumerate_perpetual(m);
3334 manager_enumerate(m);
3335
3336 /* Second, deserialize our stored data */
3337 r = manager_deserialize(m, f, fds);
3338 if (r < 0)
3339 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3340
3341 /* We don't need the serialization anymore */
3342 f = safe_fclose(f);
3343
3344 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3345 (void) manager_setup_notify(m);
3346 (void) manager_setup_cgroups_agent(m);
3347 (void) manager_setup_user_lookup_fd(m);
3348
3349 /* Third, fire things up! */
3350 manager_coldplug(m);
3351
3352 /* Clean up runtime objects no longer referenced */
3353 manager_vacuum(m);
3354
3355 /* Clean up deserialized tracked clients */
3356 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3357
3358 /* Consider the reload process complete now. */
3359 assert(m->n_reloading > 0);
3360 m->n_reloading--;
3361
3362 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3363 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3364 * let's always set the flag here for safety. */
3365 m->honor_device_enumeration = true;
3366
3367 manager_ready(m);
3368
3369 m->send_reloading_done = true;
3370 return 0;
3371 }
3372
3373 void manager_reset_failed(Manager *m) {
3374 Unit *u;
3375
3376 assert(m);
3377
3378 HASHMAP_FOREACH(u, m->units)
3379 unit_reset_failed(u);
3380 }
3381
3382 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3383 Unit *u;
3384
3385 assert(m);
3386 assert(name);
3387
3388 /* Returns true if the unit is inactive or going down */
3389 u = manager_get_unit(m, name);
3390 if (!u)
3391 return true;
3392
3393 return unit_inactive_or_pending(u);
3394 }
3395
3396 static void log_taint_string(Manager *m) {
3397 _cleanup_free_ char *taint = NULL;
3398
3399 assert(m);
3400
3401 if (MANAGER_IS_USER(m) || m->taint_logged)
3402 return;
3403
3404 m->taint_logged = true; /* only check for taint once */
3405
3406 taint = manager_taint_string(m);
3407 if (isempty(taint))
3408 return;
3409
3410 log_struct(LOG_NOTICE,
3411 LOG_MESSAGE("System is tainted: %s", taint),
3412 "TAINT=%s", taint,
3413 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3414 }
3415
3416 static void manager_notify_finished(Manager *m) {
3417 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3418
3419 if (MANAGER_IS_TEST_RUN(m))
3420 return;
3421
3422 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3423 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3424 = {};
3425 char *p = buf;
3426 size_t size = sizeof buf;
3427
3428 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3429 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3430 * negative values. */
3431
3432 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3433 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3434 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3435 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3436
3437 if (firmware_usec > 0)
3438 size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC));
3439 if (loader_usec > 0)
3440 size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC));
3441
3442 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3443
3444 /* The initrd case on bare-metal */
3445 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3446 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3447
3448 log_struct(LOG_INFO,
3449 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3450 "KERNEL_USEC="USEC_FMT, kernel_usec,
3451 "INITRD_USEC="USEC_FMT, initrd_usec,
3452 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3453 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3454 buf,
3455 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3456 FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC),
3457 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3458 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3459 } else {
3460 /* The initrd-less case on bare-metal */
3461
3462 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3463 initrd_usec = 0;
3464
3465 log_struct(LOG_INFO,
3466 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3467 "KERNEL_USEC="USEC_FMT, kernel_usec,
3468 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3469 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3470 buf,
3471 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3472 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3473 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3474 }
3475 } else {
3476 /* The container and --user case */
3477 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3478 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3479
3480 log_struct(LOG_INFO,
3481 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3482 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3483 LOG_MESSAGE("Startup finished in %s.",
3484 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3485 }
3486
3487 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3488
3489 log_taint_string(m);
3490 }
3491
3492 static void user_manager_send_ready(Manager *m) {
3493 assert(m);
3494
3495 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3496 if (!MANAGER_IS_USER(m) || m->ready_sent)
3497 return;
3498
3499 sd_notifyf(false,
3500 "READY=1\n"
3501 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3502 m->ready_sent = true;
3503 m->status_ready = false;
3504 }
3505
3506 static void manager_send_ready(Manager *m) {
3507 if (m->ready_sent && m->status_ready)
3508 /* Skip the notification if nothing changed. */
3509 return;
3510
3511 sd_notifyf(false,
3512 "%sSTATUS=Ready.",
3513 m->ready_sent ? "READY=1\n" : "");
3514 m->ready_sent = m->status_ready = true;
3515 }
3516
3517 static void manager_check_basic_target(Manager *m) {
3518 Unit *u;
3519
3520 assert(m);
3521
3522 /* Small shortcut */
3523 if (m->ready_sent && m->taint_logged)
3524 return;
3525
3526 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3527 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3528 return;
3529
3530 /* For user managers, send out READY=1 as soon as we reach basic.target */
3531 user_manager_send_ready(m);
3532
3533 /* Log the taint string as soon as we reach basic.target */
3534 log_taint_string(m);
3535 }
3536
3537 void manager_check_finished(Manager *m) {
3538 assert(m);
3539
3540 if (MANAGER_IS_RELOADING(m))
3541 return;
3542
3543 /* Verify that we have entered the event loop already, and not left it again. */
3544 if (!MANAGER_IS_RUNNING(m))
3545 return;
3546
3547 manager_check_basic_target(m);
3548
3549 if (hashmap_size(m->jobs) > 0) {
3550 if (m->jobs_in_progress_event_source)
3551 /* Ignore any failure, this is only for feedback */
3552 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
3553 manager_watch_jobs_next_time(m));
3554 return;
3555 }
3556
3557 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3558 kill the hashmap if it is relatively large. */
3559 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
3560 m->jobs = hashmap_free(m->jobs);
3561
3562 manager_send_ready(m);
3563
3564 if (MANAGER_IS_FINISHED(m))
3565 return;
3566
3567 manager_flip_auto_status(m, false, "boot finished");
3568
3569 /* Notify Type=idle units that we are done now */
3570 manager_close_idle_pipe(m);
3571
3572 /* Turn off confirm spawn now */
3573 m->confirm_spawn = NULL;
3574
3575 /* No need to update ask password status when we're going non-interactive */
3576 manager_close_ask_password(m);
3577
3578 /* This is no longer the first boot */
3579 manager_set_first_boot(m, false);
3580
3581 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH);
3582
3583 manager_notify_finished(m);
3584
3585 manager_invalidate_startup_units(m);
3586 }
3587
3588 static bool generator_path_any(const char* const* paths) {
3589 char **path;
3590 bool found = false;
3591
3592 /* Optimize by skipping the whole process by not creating output directories
3593 * if no generators are found. */
3594 STRV_FOREACH(path, (char**) paths)
3595 if (access(*path, F_OK) == 0)
3596 found = true;
3597 else if (errno != ENOENT)
3598 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3599
3600 return found;
3601 }
3602
3603 static int manager_run_environment_generators(Manager *m) {
3604 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3605 _cleanup_strv_free_ char **paths = NULL;
3606 void* args[] = {
3607 [STDOUT_GENERATE] = &tmp,
3608 [STDOUT_COLLECT] = &tmp,
3609 [STDOUT_CONSUME] = &m->transient_environment,
3610 };
3611 int r;
3612
3613 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3614 return 0;
3615
3616 paths = env_generator_binary_paths(MANAGER_IS_SYSTEM(m));
3617 if (!paths)
3618 return log_oom();
3619
3620 if (!generator_path_any((const char* const*) paths))
3621 return 0;
3622
3623 RUN_WITH_UMASK(0022)
3624 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
3625 args, NULL, m->transient_environment,
3626 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3627 return r;
3628 }
3629
3630 static int manager_run_generators(Manager *m) {
3631 _cleanup_strv_free_ char **paths = NULL;
3632 const char *argv[5];
3633 int r;
3634
3635 assert(m);
3636
3637 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3638 return 0;
3639
3640 paths = generator_binary_paths(m->unit_file_scope);
3641 if (!paths)
3642 return log_oom();
3643
3644 if (!generator_path_any((const char* const*) paths))
3645 return 0;
3646
3647 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3648 if (r < 0) {
3649 log_error_errno(r, "Failed to create generator directories: %m");
3650 goto finish;
3651 }
3652
3653 argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3654 argv[1] = m->lookup_paths.generator;
3655 argv[2] = m->lookup_paths.generator_early;
3656 argv[3] = m->lookup_paths.generator_late;
3657 argv[4] = NULL;
3658
3659 RUN_WITH_UMASK(0022)
3660 (void) execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, NULL, NULL,
3661 (char**) argv, m->transient_environment,
3662 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3663
3664 r = 0;
3665
3666 finish:
3667 lookup_paths_trim_generator(&m->lookup_paths);
3668 return r;
3669 }
3670
3671 int manager_transient_environment_add(Manager *m, char **plus) {
3672 char **a;
3673
3674 assert(m);
3675
3676 if (strv_isempty(plus))
3677 return 0;
3678
3679 a = strv_env_merge(2, m->transient_environment, plus);
3680 if (!a)
3681 return log_oom();
3682
3683 sanitize_environment(a);
3684
3685 return strv_free_and_replace(m->transient_environment, a);
3686 }
3687
3688 int manager_client_environment_modify(
3689 Manager *m,
3690 char **minus,
3691 char **plus) {
3692
3693 char **a = NULL, **b = NULL, **l;
3694
3695 assert(m);
3696
3697 if (strv_isempty(minus) && strv_isempty(plus))
3698 return 0;
3699
3700 l = m->client_environment;
3701
3702 if (!strv_isempty(minus)) {
3703 a = strv_env_delete(l, 1, minus);
3704 if (!a)
3705 return -ENOMEM;
3706
3707 l = a;
3708 }
3709
3710 if (!strv_isempty(plus)) {
3711 b = strv_env_merge(2, l, plus);
3712 if (!b) {
3713 strv_free(a);
3714 return -ENOMEM;
3715 }
3716
3717 l = b;
3718 }
3719
3720 if (m->client_environment != l)
3721 strv_free(m->client_environment);
3722
3723 if (a != l)
3724 strv_free(a);
3725 if (b != l)
3726 strv_free(b);
3727
3728 m->client_environment = sanitize_environment(l);
3729 return 0;
3730 }
3731
3732 int manager_get_effective_environment(Manager *m, char ***ret) {
3733 char **l;
3734
3735 assert(m);
3736 assert(ret);
3737
3738 l = strv_env_merge(2, m->transient_environment, m->client_environment);
3739 if (!l)
3740 return -ENOMEM;
3741
3742 *ret = l;
3743 return 0;
3744 }
3745
3746 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3747 assert(m);
3748
3749 for (unsigned i = 0; i < _RLIMIT_MAX; i++) {
3750 m->rlimit[i] = mfree(m->rlimit[i]);
3751
3752 if (!default_rlimit[i])
3753 continue;
3754
3755 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3756 if (!m->rlimit[i])
3757 return log_oom();
3758 }
3759
3760 return 0;
3761 }
3762
3763 void manager_recheck_dbus(Manager *m) {
3764 assert(m);
3765
3766 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3767 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3768 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3769 * while in the user instance we can assume it's already there. */
3770
3771 if (MANAGER_IS_RELOADING(m))
3772 return; /* don't check while we are reloading… */
3773
3774 if (manager_dbus_is_running(m, false)) {
3775 (void) bus_init_api(m);
3776
3777 if (MANAGER_IS_SYSTEM(m))
3778 (void) bus_init_system(m);
3779 } else {
3780 (void) bus_done_api(m);
3781
3782 if (MANAGER_IS_SYSTEM(m))
3783 (void) bus_done_system(m);
3784 }
3785 }
3786
3787 static bool manager_journal_is_running(Manager *m) {
3788 Unit *u;
3789
3790 assert(m);
3791
3792 if (MANAGER_IS_TEST_RUN(m))
3793 return false;
3794
3795 /* If we are the user manager we can safely assume that the journal is up */
3796 if (!MANAGER_IS_SYSTEM(m))
3797 return true;
3798
3799 /* Check that the socket is not only up, but in RUNNING state */
3800 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3801 if (!u)
3802 return false;
3803 if (SOCKET(u)->state != SOCKET_RUNNING)
3804 return false;
3805
3806 /* Similar, check if the daemon itself is fully up, too */
3807 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3808 if (!u)
3809 return false;
3810 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
3811 return false;
3812
3813 return true;
3814 }
3815
3816 void disable_printk_ratelimit(void) {
3817 /* Disable kernel's printk ratelimit.
3818 *
3819 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
3820 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
3821 * setting takes precedence. */
3822 int r;
3823
3824 r = sysctl_write("kernel/printk_devkmsg", "on");
3825 if (r < 0)
3826 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
3827 }
3828
3829 void manager_recheck_journal(Manager *m) {
3830
3831 assert(m);
3832
3833 /* Don't bother with this unless we are in the special situation of being PID 1 */
3834 if (getpid_cached() != 1)
3835 return;
3836
3837 /* Don't check this while we are reloading, things might still change */
3838 if (MANAGER_IS_RELOADING(m))
3839 return;
3840
3841 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3842 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3843 * an activation ourselves we can't fulfill. */
3844 log_set_prohibit_ipc(!manager_journal_is_running(m));
3845 log_open();
3846 }
3847
3848 static ShowStatus manager_get_show_status(Manager *m) {
3849 assert(m);
3850
3851 if (MANAGER_IS_USER(m))
3852 return _SHOW_STATUS_INVALID;
3853
3854 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
3855 return m->show_status_overridden;
3856
3857 return m->show_status;
3858 }
3859
3860 bool manager_get_show_status_on(Manager *m) {
3861 assert(m);
3862
3863 return show_status_on(manager_get_show_status(m));
3864 }
3865
3866 static void set_show_status_marker(bool b) {
3867 if (b)
3868 (void) touch("/run/systemd/show-status");
3869 else
3870 (void) unlink("/run/systemd/show-status");
3871 }
3872
3873 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
3874 assert(m);
3875 assert(reason);
3876 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
3877
3878 if (MANAGER_IS_USER(m))
3879 return;
3880
3881 if (mode == m->show_status)
3882 return;
3883
3884 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
3885 bool enabled;
3886
3887 enabled = show_status_on(mode);
3888 log_debug("%s (%s) showing of status (%s).",
3889 enabled ? "Enabling" : "Disabling",
3890 strna(show_status_to_string(mode)),
3891 reason);
3892
3893 set_show_status_marker(enabled);
3894 }
3895
3896 m->show_status = mode;
3897 }
3898
3899 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
3900 assert(m);
3901 assert(mode < _SHOW_STATUS_MAX);
3902
3903 if (MANAGER_IS_USER(m))
3904 return;
3905
3906 if (mode == m->show_status_overridden)
3907 return;
3908
3909 m->show_status_overridden = mode;
3910
3911 if (mode == _SHOW_STATUS_INVALID)
3912 mode = m->show_status;
3913
3914 log_debug("%s (%s) showing of status (%s).",
3915 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
3916 strna(show_status_to_string(mode)),
3917 reason);
3918
3919 set_show_status_marker(show_status_on(mode));
3920 }
3921
3922 const char *manager_get_confirm_spawn(Manager *m) {
3923 static int last_errno = 0;
3924 struct stat st;
3925 int r;
3926
3927 assert(m);
3928
3929 /* Here's the deal: we want to test the validity of the console but don't want
3930 * PID1 to go through the whole console process which might block. But we also
3931 * want to warn the user only once if something is wrong with the console so we
3932 * cannot do the sanity checks after spawning our children. So here we simply do
3933 * really basic tests to hopefully trap common errors.
3934 *
3935 * If the console suddenly disappear at the time our children will really it
3936 * then they will simply fail to acquire it and a positive answer will be
3937 * assumed. New children will fall back to /dev/console though.
3938 *
3939 * Note: TTYs are devices that can come and go any time, and frequently aren't
3940 * available yet during early boot (consider a USB rs232 dongle...). If for any
3941 * reason the configured console is not ready, we fall back to the default
3942 * console. */
3943
3944 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
3945 return m->confirm_spawn;
3946
3947 if (stat(m->confirm_spawn, &st) < 0) {
3948 r = -errno;
3949 goto fail;
3950 }
3951
3952 if (!S_ISCHR(st.st_mode)) {
3953 r = -ENOTTY;
3954 goto fail;
3955 }
3956
3957 last_errno = 0;
3958 return m->confirm_spawn;
3959
3960 fail:
3961 if (last_errno != r)
3962 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
3963
3964 return "/dev/console";
3965 }
3966
3967 void manager_set_first_boot(Manager *m, bool b) {
3968 assert(m);
3969
3970 if (!MANAGER_IS_SYSTEM(m))
3971 return;
3972
3973 if (m->first_boot != (int) b) {
3974 if (b)
3975 (void) touch("/run/systemd/first-boot");
3976 else
3977 (void) unlink("/run/systemd/first-boot");
3978 }
3979
3980 m->first_boot = b;
3981 }
3982
3983 void manager_disable_confirm_spawn(void) {
3984 (void) touch("/run/systemd/confirm_spawn_disabled");
3985 }
3986
3987 bool manager_is_confirm_spawn_disabled(Manager *m) {
3988 if (!m->confirm_spawn)
3989 return true;
3990
3991 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3992 }
3993
3994 static bool manager_should_show_status(Manager *m, StatusType type) {
3995 assert(m);
3996
3997 if (!MANAGER_IS_SYSTEM(m))
3998 return false;
3999
4000 if (m->no_console_output)
4001 return false;
4002
4003 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4004 return false;
4005
4006 /* If we cannot find out the status properly, just proceed. */
4007 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4008 return false;
4009
4010 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4011 return true;
4012
4013 return manager_get_show_status_on(m);
4014 }
4015
4016 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4017 va_list ap;
4018
4019 /* If m is NULL, assume we're after shutdown and let the messages through. */
4020
4021 if (m && !manager_should_show_status(m, type))
4022 return;
4023
4024 /* XXX We should totally drop the check for ephemeral here
4025 * and thus effectively make 'Type=idle' pointless. */
4026 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4027 return;
4028
4029 va_start(ap, format);
4030 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4031 va_end(ap);
4032 }
4033
4034 Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
4035 assert(m);
4036 assert(path);
4037
4038 if (path_equal(path, "/"))
4039 path = "";
4040
4041 return hashmap_get(m->units_requiring_mounts_for, path);
4042 }
4043
4044 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4045 unsigned size;
4046 int r;
4047
4048 assert(m);
4049 assert(u->manager == m);
4050
4051 size = set_size(m->failed_units);
4052
4053 if (failed) {
4054 r = set_ensure_put(&m->failed_units, NULL, u);
4055 if (r < 0)
4056 return log_oom();
4057 } else
4058 (void) set_remove(m->failed_units, u);
4059
4060 if (set_size(m->failed_units) != size)
4061 bus_manager_send_change_signal(m);
4062
4063 return 0;
4064 }
4065
4066 ManagerState manager_state(Manager *m) {
4067 Unit *u;
4068
4069 assert(m);
4070
4071 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4072 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4073 if (u && unit_active_or_pending(u))
4074 return MANAGER_STOPPING;
4075
4076 /* Did we ever finish booting? If not then we are still starting up */
4077 if (!MANAGER_IS_FINISHED(m)) {
4078
4079 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4080 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4081 return MANAGER_INITIALIZING;
4082
4083 return MANAGER_STARTING;
4084 }
4085
4086 if (MANAGER_IS_SYSTEM(m)) {
4087 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4088 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4089 if (u && unit_active_or_pending(u))
4090 return MANAGER_MAINTENANCE;
4091
4092 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4093 if (u && unit_active_or_pending(u))
4094 return MANAGER_MAINTENANCE;
4095 }
4096
4097 /* Are there any failed units? If so, we are in degraded mode */
4098 if (set_size(m->failed_units) > 0)
4099 return MANAGER_DEGRADED;
4100
4101 return MANAGER_RUNNING;
4102 }
4103
4104 static void manager_unref_uid_internal(
4105 Hashmap *uid_refs,
4106 uid_t uid,
4107 bool destroy_now,
4108 int (*_clean_ipc)(uid_t uid)) {
4109
4110 uint32_t c, n;
4111
4112 assert(uid_is_valid(uid));
4113 assert(_clean_ipc);
4114
4115 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4116 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4117 *
4118 * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
4119 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4120 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4121 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4122
4123 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4124 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4125
4126 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4127 return;
4128
4129 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4130
4131 n = c & ~DESTROY_IPC_FLAG;
4132 assert(n > 0);
4133 n--;
4134
4135 if (destroy_now && n == 0) {
4136 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4137
4138 if (c & DESTROY_IPC_FLAG) {
4139 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4140 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4141 uid);
4142 (void) _clean_ipc(uid);
4143 }
4144 } else {
4145 c = n | (c & DESTROY_IPC_FLAG);
4146 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4147 }
4148 }
4149
4150 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4151 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4152 }
4153
4154 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4155 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4156 }
4157
4158 static int manager_ref_uid_internal(
4159 Hashmap **uid_refs,
4160 uid_t uid,
4161 bool clean_ipc) {
4162
4163 uint32_t c, n;
4164 int r;
4165
4166 assert(uid_refs);
4167 assert(uid_is_valid(uid));
4168
4169 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4170 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4171
4172 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4173 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4174
4175 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4176 return 0;
4177
4178 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4179 if (r < 0)
4180 return r;
4181
4182 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4183
4184 n = c & ~DESTROY_IPC_FLAG;
4185 n++;
4186
4187 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4188 return -EOVERFLOW;
4189
4190 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4191
4192 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4193 }
4194
4195 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4196 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4197 }
4198
4199 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4200 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4201 }
4202
4203 static void manager_vacuum_uid_refs_internal(
4204 Hashmap *uid_refs,
4205 int (*_clean_ipc)(uid_t uid)) {
4206
4207 void *p, *k;
4208
4209 assert(_clean_ipc);
4210
4211 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4212 uint32_t c, n;
4213 uid_t uid;
4214
4215 uid = PTR_TO_UID(k);
4216 c = PTR_TO_UINT32(p);
4217
4218 n = c & ~DESTROY_IPC_FLAG;
4219 if (n > 0)
4220 continue;
4221
4222 if (c & DESTROY_IPC_FLAG) {
4223 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4224 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4225 uid);
4226 (void) _clean_ipc(uid);
4227 }
4228
4229 assert_se(hashmap_remove(uid_refs, k) == p);
4230 }
4231 }
4232
4233 static void manager_vacuum_uid_refs(Manager *m) {
4234 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4235 }
4236
4237 static void manager_vacuum_gid_refs(Manager *m) {
4238 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4239 }
4240
4241 static void manager_vacuum(Manager *m) {
4242 assert(m);
4243
4244 /* Release any dynamic users no longer referenced */
4245 dynamic_user_vacuum(m, true);
4246
4247 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4248 manager_vacuum_uid_refs(m);
4249 manager_vacuum_gid_refs(m);
4250
4251 /* Release any runtimes no longer referenced */
4252 exec_runtime_vacuum(m);
4253 }
4254
4255 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4256 struct buffer {
4257 uid_t uid;
4258 gid_t gid;
4259 char unit_name[UNIT_NAME_MAX+1];
4260 } _packed_ buffer;
4261
4262 Manager *m = userdata;
4263 ssize_t l;
4264 size_t n;
4265 Unit *u;
4266
4267 assert_se(source);
4268 assert_se(m);
4269
4270 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4271 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4272 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4273
4274 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4275 if (l < 0) {
4276 if (IN_SET(errno, EINTR, EAGAIN))
4277 return 0;
4278
4279 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4280 }
4281
4282 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4283 log_warning("Received too short user lookup message, ignoring.");
4284 return 0;
4285 }
4286
4287 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4288 log_warning("Received too long user lookup message, ignoring.");
4289 return 0;
4290 }
4291
4292 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4293 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4294 return 0;
4295 }
4296
4297 n = (size_t) l - offsetof(struct buffer, unit_name);
4298 if (memchr(buffer.unit_name, 0, n)) {
4299 log_warning("Received lookup message with embedded NUL character, ignoring.");
4300 return 0;
4301 }
4302
4303 buffer.unit_name[n] = 0;
4304 u = manager_get_unit(m, buffer.unit_name);
4305 if (!u) {
4306 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4307 return 0;
4308 }
4309
4310 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4311
4312 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4313 return 0;
4314 }
4315
4316 char *manager_taint_string(Manager *m) {
4317 _cleanup_free_ char *destination = NULL, *overflowuid = NULL, *overflowgid = NULL;
4318 char *buf, *e;
4319 int r;
4320
4321 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad".
4322 * Only things that are detected at runtime should be tagged
4323 * here. For stuff that is set during compilation, emit a warning
4324 * in the configuration phase. */
4325
4326 assert(m);
4327
4328 buf = new(char, sizeof("split-usr:"
4329 "cgroups-missing:"
4330 "cgrousv1:"
4331 "local-hwclock:"
4332 "var-run-bad:"
4333 "overflowuid-not-65534:"
4334 "overflowgid-not-65534:"));
4335 if (!buf)
4336 return NULL;
4337
4338 e = buf;
4339 buf[0] = 0;
4340
4341 if (m->taint_usr)
4342 e = stpcpy(e, "split-usr:");
4343
4344 if (access("/proc/cgroups", F_OK) < 0)
4345 e = stpcpy(e, "cgroups-missing:");
4346
4347 if (cg_all_unified() == 0)
4348 e = stpcpy(e, "cgroupsv1:");
4349
4350 if (clock_is_localtime(NULL) > 0)
4351 e = stpcpy(e, "local-hwclock:");
4352
4353 r = readlink_malloc("/var/run", &destination);
4354 if (r < 0 || !PATH_IN_SET(destination, "../run", "/run"))
4355 e = stpcpy(e, "var-run-bad:");
4356
4357 r = read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid);
4358 if (r >= 0 && !streq(overflowuid, "65534"))
4359 e = stpcpy(e, "overflowuid-not-65534:");
4360
4361 r = read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid);
4362 if (r >= 0 && !streq(overflowgid, "65534"))
4363 e = stpcpy(e, "overflowgid-not-65534:");
4364
4365 /* remove the last ':' */
4366 if (e != buf)
4367 e[-1] = 0;
4368
4369 return buf;
4370 }
4371
4372 void manager_ref_console(Manager *m) {
4373 assert(m);
4374
4375 m->n_on_console++;
4376 }
4377
4378 void manager_unref_console(Manager *m) {
4379
4380 assert(m->n_on_console > 0);
4381 m->n_on_console--;
4382
4383 if (m->n_on_console == 0)
4384 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4385 }
4386
4387 void manager_override_log_level(Manager *m, int level) {
4388 _cleanup_free_ char *s = NULL;
4389 assert(m);
4390
4391 if (!m->log_level_overridden) {
4392 m->original_log_level = log_get_max_level();
4393 m->log_level_overridden = true;
4394 }
4395
4396 (void) log_level_to_string_alloc(level, &s);
4397 log_info("Setting log level to %s.", strna(s));
4398
4399 log_set_max_level(level);
4400 }
4401
4402 void manager_restore_original_log_level(Manager *m) {
4403 _cleanup_free_ char *s = NULL;
4404 assert(m);
4405
4406 if (!m->log_level_overridden)
4407 return;
4408
4409 (void) log_level_to_string_alloc(m->original_log_level, &s);
4410 log_info("Restoring log level to original (%s).", strna(s));
4411
4412 log_set_max_level(m->original_log_level);
4413 m->log_level_overridden = false;
4414 }
4415
4416 void manager_override_log_target(Manager *m, LogTarget target) {
4417 assert(m);
4418
4419 if (!m->log_target_overridden) {
4420 m->original_log_target = log_get_target();
4421 m->log_target_overridden = true;
4422 }
4423
4424 log_info("Setting log target to %s.", log_target_to_string(target));
4425 log_set_target(target);
4426 }
4427
4428 void manager_restore_original_log_target(Manager *m) {
4429 assert(m);
4430
4431 if (!m->log_target_overridden)
4432 return;
4433
4434 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4435
4436 log_set_target(m->original_log_target);
4437 m->log_target_overridden = false;
4438 }
4439
4440 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4441 if (in_initrd() &&
4442 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4443 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4444 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4445 return s;
4446 }
4447
4448 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
4449 [MANAGER_INITIALIZING] = "initializing",
4450 [MANAGER_STARTING] = "starting",
4451 [MANAGER_RUNNING] = "running",
4452 [MANAGER_DEGRADED] = "degraded",
4453 [MANAGER_MAINTENANCE] = "maintenance",
4454 [MANAGER_STOPPING] = "stopping",
4455 };
4456
4457 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
4458
4459 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
4460 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
4461 [MANAGER_TIMESTAMP_LOADER] = "loader",
4462 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
4463 [MANAGER_TIMESTAMP_INITRD] = "initrd",
4464 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
4465 [MANAGER_TIMESTAMP_FINISH] = "finish",
4466 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
4467 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
4468 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
4469 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
4470 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
4471 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
4472 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
4473 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
4474 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
4475 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
4476 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
4477 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
4478 };
4479
4480 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
4481
4482 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
4483 [OOM_CONTINUE] = "continue",
4484 [OOM_STOP] = "stop",
4485 [OOM_KILL] = "kill",
4486 };
4487
4488 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
Unnamed repository; edit this file 'description' to name the repository.