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/utsname.h>
12 #include <sys/wait.h>
13 #include <unistd.h>
14
15 #if HAVE_AUDIT
16 #include <libaudit.h>
17 #endif
18
19 #include "sd-daemon.h"
20 #include "sd-messages.h"
21 #include "sd-path.h"
22
23 #include "all-units.h"
24 #include "alloc-util.h"
25 #include "audit-fd.h"
26 #include "boot-timestamps.h"
27 #include "bus-common-errors.h"
28 #include "bus-error.h"
29 #include "bus-kernel.h"
30 #include "bus-util.h"
31 #include "clean-ipc.h"
32 #include "clock-util.h"
33 #include "core-varlink.h"
34 #include "creds-util.h"
35 #include "dbus-job.h"
36 #include "dbus-manager.h"
37 #include "dbus-unit.h"
38 #include "dbus.h"
39 #include "def.h"
40 #include "dirent-util.h"
41 #include "env-util.h"
42 #include "escape.h"
43 #include "event-util.h"
44 #include "exec-util.h"
45 #include "execute.h"
46 #include "exit-status.h"
47 #include "fd-util.h"
48 #include "fileio.h"
49 #include "generator-setup.h"
50 #include "hashmap.h"
51 #include "inotify-util.h"
52 #include "install.h"
53 #include "io-util.h"
54 #include "label.h"
55 #include "load-fragment.h"
56 #include "locale-setup.h"
57 #include "log.h"
58 #include "macro.h"
59 #include "manager.h"
60 #include "manager-dump.h"
61 #include "manager-serialize.h"
62 #include "memory-util.h"
63 #include "mkdir-label.h"
64 #include "parse-util.h"
65 #include "path-lookup.h"
66 #include "path-util.h"
67 #include "process-util.h"
68 #include "ratelimit.h"
69 #include "rlimit-util.h"
70 #include "rm-rf.h"
71 #include "selinux-util.h"
72 #include "signal-util.h"
73 #include "socket-util.h"
74 #include "special.h"
75 #include "stat-util.h"
76 #include "string-table.h"
77 #include "string-util.h"
78 #include "strv.h"
79 #include "strxcpyx.h"
80 #include "sysctl-util.h"
81 #include "syslog-util.h"
82 #include "terminal-util.h"
83 #include "time-util.h"
84 #include "transaction.h"
85 #include "uid-range.h"
86 #include "umask-util.h"
87 #include "unit-name.h"
88 #include "user-util.h"
89 #include "virt.h"
90 #include "watchdog.h"
91
92 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
93 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
94
95 /* Initial delay and the interval for printing status messages about running jobs */
96 #define JOBS_IN_PROGRESS_WAIT_USEC (2*USEC_PER_SEC)
97 #define JOBS_IN_PROGRESS_QUIET_WAIT_USEC (25*USEC_PER_SEC)
98 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
99 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
100
101 /* If there are more than 1K bus messages queue across our API and direct buses, then let's not add more on top until
102 * the queue gets more empty. */
103 #define MANAGER_BUS_BUSY_THRESHOLD 1024LU
104
105 /* How many units and jobs to process of the bus queue before returning to the event loop. */
106 #define MANAGER_BUS_MESSAGE_BUDGET 100U
107
108 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
109 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
110 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
111 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
112 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
113 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
114 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
115 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
116 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata);
117 static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata);
118 static int manager_run_environment_generators(Manager *m);
119 static int manager_run_generators(Manager *m);
120 static void manager_vacuum(Manager *m);
121
manager_watch_jobs_next_time(Manager * m)122 static usec_t manager_watch_jobs_next_time(Manager *m) {
123 usec_t timeout;
124
125 if (MANAGER_IS_USER(m))
126 /* Let the user manager without a timeout show status quickly, so the system manager can make
127 * use of it, if it wants to. */
128 timeout = JOBS_IN_PROGRESS_WAIT_USEC * 2 / 3;
129 else if (show_status_on(m->show_status))
130 /* When status is on, just use the usual timeout. */
131 timeout = JOBS_IN_PROGRESS_WAIT_USEC;
132 else
133 timeout = JOBS_IN_PROGRESS_QUIET_WAIT_USEC;
134
135 return usec_add(now(CLOCK_MONOTONIC), timeout);
136 }
137
manager_watch_jobs_in_progress(Manager * m)138 static void manager_watch_jobs_in_progress(Manager *m) {
139 usec_t next;
140 int r;
141
142 assert(m);
143
144 /* We do not want to show the cylon animation if the user
145 * needs to confirm service executions otherwise confirmation
146 * messages will be screwed by the cylon animation. */
147 if (!manager_is_confirm_spawn_disabled(m))
148 return;
149
150 if (m->jobs_in_progress_event_source)
151 return;
152
153 next = manager_watch_jobs_next_time(m);
154 r = sd_event_add_time(
155 m->event,
156 &m->jobs_in_progress_event_source,
157 CLOCK_MONOTONIC,
158 next, 0,
159 manager_dispatch_jobs_in_progress, m);
160 if (r < 0)
161 return;
162
163 (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
164 }
165
166 #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL))
167
draw_cylon(char buffer[],size_t buflen,unsigned width,unsigned pos)168 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
169 char *p = buffer;
170
171 assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
172 assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
173
174 if (pos > 1) {
175 if (pos > 2)
176 p = mempset(p, ' ', pos-2);
177 if (log_get_show_color())
178 p = stpcpy(p, ANSI_RED);
179 *p++ = '*';
180 }
181
182 if (pos > 0 && pos <= width) {
183 if (log_get_show_color())
184 p = stpcpy(p, ANSI_HIGHLIGHT_RED);
185 *p++ = '*';
186 }
187
188 if (log_get_show_color())
189 p = stpcpy(p, ANSI_NORMAL);
190
191 if (pos < width) {
192 if (log_get_show_color())
193 p = stpcpy(p, ANSI_RED);
194 *p++ = '*';
195 if (pos < width-1)
196 p = mempset(p, ' ', width-1-pos);
197 if (log_get_show_color())
198 strcpy(p, ANSI_NORMAL);
199 }
200 }
201
manager_flip_auto_status(Manager * m,bool enable,const char * reason)202 static void manager_flip_auto_status(Manager *m, bool enable, const char *reason) {
203 assert(m);
204
205 if (enable) {
206 if (m->show_status == SHOW_STATUS_AUTO)
207 manager_set_show_status(m, SHOW_STATUS_TEMPORARY, reason);
208 } else {
209 if (m->show_status == SHOW_STATUS_TEMPORARY)
210 manager_set_show_status(m, SHOW_STATUS_AUTO, reason);
211 }
212 }
213
manager_print_jobs_in_progress(Manager * m)214 static void manager_print_jobs_in_progress(Manager *m) {
215 Job *j;
216 unsigned counter = 0, print_nr;
217 char cylon[6 + CYLON_BUFFER_EXTRA + 1];
218 unsigned cylon_pos;
219 uint64_t x;
220
221 assert(m);
222 assert(m->n_running_jobs > 0);
223
224 manager_flip_auto_status(m, true, "delay");
225
226 print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
227
228 HASHMAP_FOREACH(j, m->jobs)
229 if (j->state == JOB_RUNNING && counter++ == print_nr)
230 break;
231
232 /* m->n_running_jobs must be consistent with the contents of m->jobs,
233 * so the above loop must have succeeded in finding j. */
234 assert(counter == print_nr + 1);
235 assert(j);
236
237 cylon_pos = m->jobs_in_progress_iteration % 14;
238 if (cylon_pos >= 8)
239 cylon_pos = 14 - cylon_pos;
240 draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
241
242 m->jobs_in_progress_iteration++;
243
244 char job_of_n[STRLEN("( of ) ") + DECIMAL_STR_MAX(unsigned)*2] = "";
245 if (m->n_running_jobs > 1)
246 xsprintf(job_of_n, "(%u of %u) ", counter, m->n_running_jobs);
247
248 bool have_timeout = job_get_timeout(j, &x) > 0;
249
250 /* We want to use enough information for the user to identify previous lines talking about the same
251 * unit, but keep the message as short as possible. So if 'Starting foo.service' or 'Starting
252 * foo.service - Description' were used, 'foo.service' is enough here. On the other hand, if we used
253 * 'Starting Description' before, then we shall also use 'Description' here. So we pass NULL as the
254 * second argument to unit_status_string(). */
255 const char *ident = unit_status_string(j->unit, NULL);
256
257 const char *time = FORMAT_TIMESPAN(now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
258 const char *limit = have_timeout ? FORMAT_TIMESPAN(x - j->begin_usec, 1*USEC_PER_SEC) : "no limit";
259
260 if (m->status_unit_format == STATUS_UNIT_FORMAT_DESCRIPTION)
261 /* When using 'Description', we effectively don't have enough space to show the nested status
262 * without ellipsization, so let's not even try. */
263 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
264 "%sA %s job is running for %s (%s / %s)",
265 job_of_n,
266 job_type_to_string(j->type),
267 ident,
268 time, limit);
269 else {
270 const char *status_text = unit_status_text(j->unit);
271
272 manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
273 "%sJob %s/%s running (%s / %s)%s%s",
274 job_of_n,
275 ident,
276 job_type_to_string(j->type),
277 time, limit,
278 status_text ? ": " : "",
279 strempty(status_text));
280 }
281
282 sd_notifyf(false,
283 "STATUS=%sUser job %s/%s running (%s / %s)...",
284 job_of_n,
285 ident,
286 job_type_to_string(j->type),
287 time, limit);
288 m->status_ready = false;
289 }
290
have_ask_password(void)291 static int have_ask_password(void) {
292 _cleanup_closedir_ DIR *dir = NULL;
293
294 dir = opendir("/run/systemd/ask-password");
295 if (!dir) {
296 if (errno == ENOENT)
297 return false;
298 else
299 return -errno;
300 }
301
302 FOREACH_DIRENT_ALL(de, dir, return -errno)
303 if (startswith(de->d_name, "ask."))
304 return true;
305 return false;
306 }
307
manager_dispatch_ask_password_fd(sd_event_source * source,int fd,uint32_t revents,void * userdata)308 static int manager_dispatch_ask_password_fd(sd_event_source *source,
309 int fd, uint32_t revents, void *userdata) {
310 Manager *m = userdata;
311
312 assert(m);
313
314 (void) flush_fd(fd);
315
316 m->have_ask_password = have_ask_password();
317 if (m->have_ask_password < 0)
318 /* Log error but continue. Negative have_ask_password
319 * is treated as unknown status. */
320 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
321
322 return 0;
323 }
324
manager_close_ask_password(Manager * m)325 static void manager_close_ask_password(Manager *m) {
326 assert(m);
327
328 m->ask_password_event_source = sd_event_source_disable_unref(m->ask_password_event_source);
329 m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
330 m->have_ask_password = -EINVAL;
331 }
332
manager_check_ask_password(Manager * m)333 static int manager_check_ask_password(Manager *m) {
334 int r;
335
336 assert(m);
337
338 if (!m->ask_password_event_source) {
339 assert(m->ask_password_inotify_fd < 0);
340
341 (void) mkdir_p_label("/run/systemd/ask-password", 0755);
342
343 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
344 if (m->ask_password_inotify_fd < 0)
345 return log_error_errno(errno, "Failed to create inotify object: %m");
346
347 r = inotify_add_watch_and_warn(m->ask_password_inotify_fd,
348 "/run/systemd/ask-password",
349 IN_CREATE|IN_DELETE|IN_MOVE);
350 if (r < 0) {
351 manager_close_ask_password(m);
352 return r;
353 }
354
355 r = sd_event_add_io(m->event, &m->ask_password_event_source,
356 m->ask_password_inotify_fd, EPOLLIN,
357 manager_dispatch_ask_password_fd, m);
358 if (r < 0) {
359 log_error_errno(r, "Failed to add event source for /run/systemd/ask-password: %m");
360 manager_close_ask_password(m);
361 return r;
362 }
363
364 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
365
366 /* Queries might have been added meanwhile... */
367 manager_dispatch_ask_password_fd(m->ask_password_event_source,
368 m->ask_password_inotify_fd, EPOLLIN, m);
369 }
370
371 return m->have_ask_password;
372 }
373
manager_watch_idle_pipe(Manager * m)374 static int manager_watch_idle_pipe(Manager *m) {
375 int r;
376
377 assert(m);
378
379 if (m->idle_pipe_event_source)
380 return 0;
381
382 if (m->idle_pipe[2] < 0)
383 return 0;
384
385 r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
386 if (r < 0)
387 return log_error_errno(r, "Failed to watch idle pipe: %m");
388
389 (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
390
391 return 0;
392 }
393
manager_close_idle_pipe(Manager * m)394 static void manager_close_idle_pipe(Manager *m) {
395 assert(m);
396
397 m->idle_pipe_event_source = sd_event_source_disable_unref(m->idle_pipe_event_source);
398
399 safe_close_pair(m->idle_pipe);
400 safe_close_pair(m->idle_pipe + 2);
401 }
402
manager_setup_time_change(Manager * m)403 static int manager_setup_time_change(Manager *m) {
404 int r;
405
406 assert(m);
407
408 if (MANAGER_IS_TEST_RUN(m))
409 return 0;
410
411 m->time_change_event_source = sd_event_source_disable_unref(m->time_change_event_source);
412
413 r = event_add_time_change(m->event, &m->time_change_event_source, manager_dispatch_time_change_fd, m);
414 if (r < 0)
415 return log_error_errno(r, "Failed to create time change event source: %m");
416
417 /* Schedule this slightly earlier than the .timer event sources */
418 r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1);
419 if (r < 0)
420 return log_error_errno(r, "Failed to set priority of time change event sources: %m");
421
422 log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
423
424 return 0;
425 }
426
manager_read_timezone_stat(Manager * m)427 static int manager_read_timezone_stat(Manager *m) {
428 struct stat st;
429 bool changed;
430
431 assert(m);
432
433 /* Read the current stat() data of /etc/localtime so that we detect changes */
434 if (lstat("/etc/localtime", &st) < 0) {
435 log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m");
436 changed = m->etc_localtime_accessible;
437 m->etc_localtime_accessible = false;
438 } else {
439 usec_t k;
440
441 k = timespec_load(&st.st_mtim);
442 changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime;
443
444 m->etc_localtime_mtime = k;
445 m->etc_localtime_accessible = true;
446 }
447
448 return changed;
449 }
450
manager_setup_timezone_change(Manager * m)451 static int manager_setup_timezone_change(Manager *m) {
452 _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL;
453 int r;
454
455 assert(m);
456
457 if (MANAGER_IS_TEST_RUN(m))
458 return 0;
459
460 /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even
461 * though another link might be kept), renames, and file close operations after writing. Note we don't bother
462 * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count
463 * went to zero and all fds to it are closed.
464 *
465 * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases
466 * correctly.
467 *
468 * Note that we create the new event source first here, before releasing the old one. This should optimize
469 * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */
470
471 r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime",
472 IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m);
473 if (r == -ENOENT) {
474 /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by
475 * O_CREATE or by rename() */
476
477 log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead.");
478 r = sd_event_add_inotify(m->event, &new_event, "/etc",
479 IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m);
480 }
481 if (r < 0)
482 return log_error_errno(r, "Failed to create timezone change event source: %m");
483
484 /* Schedule this slightly earlier than the .timer event sources */
485 r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1);
486 if (r < 0)
487 return log_error_errno(r, "Failed to set priority of timezone change event sources: %m");
488
489 sd_event_source_unref(m->timezone_change_event_source);
490 m->timezone_change_event_source = TAKE_PTR(new_event);
491
492 return 0;
493 }
494
enable_special_signals(Manager * m)495 static int enable_special_signals(Manager *m) {
496 _cleanup_close_ int fd = -1;
497
498 assert(m);
499
500 if (MANAGER_IS_TEST_RUN(m))
501 return 0;
502
503 /* Enable that we get SIGINT on control-alt-del. In containers
504 * this will fail with EPERM (older) or EINVAL (newer), so
505 * ignore that. */
506 if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL))
507 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
508
509 fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
510 if (fd < 0) {
511 /* Support systems without virtual console */
512 if (fd != -ENOENT)
513 log_warning_errno(errno, "Failed to open /dev/tty0: %m");
514 } else {
515 /* Enable that we get SIGWINCH on kbrequest */
516 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
517 log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
518 }
519
520 return 0;
521 }
522
523 #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1)
524
manager_setup_signals(Manager * m)525 static int manager_setup_signals(Manager *m) {
526 struct sigaction sa = {
527 .sa_handler = SIG_DFL,
528 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
529 };
530 sigset_t mask;
531 int r;
532
533 assert(m);
534
535 assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
536
537 /* We make liberal use of realtime signals here. On
538 * Linux/glibc we have 30 of them (with the exception of Linux
539 * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
540 * (aka SIGRTMAX). */
541
542 assert_se(sigemptyset(&mask) == 0);
543 sigset_add_many(&mask,
544 SIGCHLD, /* Child died */
545 SIGTERM, /* Reexecute daemon */
546 SIGHUP, /* Reload configuration */
547 SIGUSR1, /* systemd: reconnect to D-Bus */
548 SIGUSR2, /* systemd: dump status */
549 SIGINT, /* Kernel sends us this on control-alt-del */
550 SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */
551 SIGPWR, /* Some kernel drivers and upsd send us this on power failure */
552
553 SIGRTMIN+0, /* systemd: start default.target */
554 SIGRTMIN+1, /* systemd: isolate rescue.target */
555 SIGRTMIN+2, /* systemd: isolate emergency.target */
556 SIGRTMIN+3, /* systemd: start halt.target */
557 SIGRTMIN+4, /* systemd: start poweroff.target */
558 SIGRTMIN+5, /* systemd: start reboot.target */
559 SIGRTMIN+6, /* systemd: start kexec.target */
560
561 /* ... space for more special targets ... */
562
563 SIGRTMIN+13, /* systemd: Immediate halt */
564 SIGRTMIN+14, /* systemd: Immediate poweroff */
565 SIGRTMIN+15, /* systemd: Immediate reboot */
566 SIGRTMIN+16, /* systemd: Immediate kexec */
567
568 /* ... space for more immediate system state changes ... */
569
570 SIGRTMIN+20, /* systemd: enable status messages */
571 SIGRTMIN+21, /* systemd: disable status messages */
572 SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
573 SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
574 SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
575 SIGRTMIN+25, /* systemd: reexecute manager */
576
577 /* Apparently Linux on hppa had fewer RT signals until v3.18,
578 * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered,
579 * see commit v3.17-7614-g1f25df2eff.
580 *
581 * We cannot unconditionally make use of those signals here,
582 * so let's use a runtime check. Since these commands are
583 * accessible by different means and only really a safety
584 * net, the missing functionality on hppa shouldn't matter.
585 */
586
587 RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */
588 RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */
589 RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */
590 RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */
591
592 /* ... one free signal here SIGRTMIN+30 ... */
593 -1);
594 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
595
596 m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
597 if (m->signal_fd < 0)
598 return -errno;
599
600 r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
601 if (r < 0)
602 return r;
603
604 (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
605
606 /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
607 * notify processing can still figure out to which process/service a message belongs, before we reap the
608 * process. Also, process this before handling cgroup notifications, so that we always collect child exit
609 * status information before detecting that there's no process in a cgroup. */
610 r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
611 if (r < 0)
612 return r;
613
614 if (MANAGER_IS_SYSTEM(m))
615 return enable_special_signals(m);
616
617 return 0;
618 }
619
sanitize_environment(char ** l)620 static char** sanitize_environment(char **l) {
621
622 /* Let's remove some environment variables that we need ourselves to communicate with our clients */
623 strv_env_unset_many(
624 l,
625 "CACHE_DIRECTORY",
626 "CONFIGURATION_DIRECTORY",
627 "CREDENTIALS_DIRECTORY",
628 "EXIT_CODE",
629 "EXIT_STATUS",
630 "INVOCATION_ID",
631 "JOURNAL_STREAM",
632 "LISTEN_FDNAMES",
633 "LISTEN_FDS",
634 "LISTEN_PID",
635 "LOGS_DIRECTORY",
636 "MAINPID",
637 "MANAGERPID",
638 "NOTIFY_SOCKET",
639 "PIDFILE",
640 "REMOTE_ADDR",
641 "REMOTE_PORT",
642 "RUNTIME_DIRECTORY",
643 "SERVICE_RESULT",
644 "STATE_DIRECTORY",
645 "WATCHDOG_PID",
646 "WATCHDOG_USEC",
647 NULL);
648
649 /* Let's order the environment alphabetically, just to make it pretty */
650 return strv_sort(l);
651 }
652
manager_default_environment(Manager * m)653 int manager_default_environment(Manager *m) {
654 int r;
655
656 assert(m);
657
658 m->transient_environment = strv_free(m->transient_environment);
659
660 if (MANAGER_IS_SYSTEM(m)) {
661 /* The system manager always starts with a clean
662 * environment for its children. It does not import
663 * the kernel's or the parents' exported variables.
664 *
665 * The initial passed environment is untouched to keep
666 * /proc/self/environ valid; it is used for tagging
667 * the init process inside containers. */
668 m->transient_environment = strv_new("PATH=" DEFAULT_PATH);
669 if (!m->transient_environment)
670 return log_oom();
671
672 /* Import locale variables LC_*= from configuration */
673 (void) locale_setup(&m->transient_environment);
674 } else {
675 /* The user manager passes its own environment along to its children, except for $PATH. */
676 m->transient_environment = strv_copy(environ);
677 if (!m->transient_environment)
678 return log_oom();
679
680 r = strv_env_replace_strdup(&m->transient_environment, "PATH=" DEFAULT_USER_PATH);
681 if (r < 0)
682 return log_oom();
683 }
684
685 sanitize_environment(m->transient_environment);
686
687 return 0;
688 }
689
manager_setup_prefix(Manager * m)690 static int manager_setup_prefix(Manager *m) {
691 struct table_entry {
692 uint64_t type;
693 const char *suffix;
694 };
695
696 static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = {
697 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL },
698 [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL },
699 [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL },
700 [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL },
701 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL },
702 };
703
704 static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = {
705 [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL },
706 [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL },
707 [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL },
708 [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" },
709 [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL },
710 };
711
712 assert(m);
713
714 const struct table_entry *p = MANAGER_IS_SYSTEM(m) ? paths_system : paths_user;
715 int r;
716
717 for (ExecDirectoryType i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) {
718 r = sd_path_lookup(p[i].type, p[i].suffix, &m->prefix[i]);
719 if (r < 0)
720 return log_warning_errno(r, "Failed to lookup %s path: %m",
721 exec_directory_type_to_string(i));
722 }
723
724 return 0;
725 }
726
manager_free_unit_name_maps(Manager * m)727 static void manager_free_unit_name_maps(Manager *m) {
728 m->unit_id_map = hashmap_free(m->unit_id_map);
729 m->unit_name_map = hashmap_free(m->unit_name_map);
730 m->unit_path_cache = set_free(m->unit_path_cache);
731 m->unit_cache_timestamp_hash = 0;
732 }
733
manager_setup_run_queue(Manager * m)734 static int manager_setup_run_queue(Manager *m) {
735 int r;
736
737 assert(m);
738 assert(!m->run_queue_event_source);
739
740 r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
741 if (r < 0)
742 return r;
743
744 r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
745 if (r < 0)
746 return r;
747
748 r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
749 if (r < 0)
750 return r;
751
752 (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
753
754 return 0;
755 }
756
manager_setup_sigchld_event_source(Manager * m)757 static int manager_setup_sigchld_event_source(Manager *m) {
758 int r;
759
760 assert(m);
761 assert(!m->sigchld_event_source);
762
763 r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m);
764 if (r < 0)
765 return r;
766
767 r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7);
768 if (r < 0)
769 return r;
770
771 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
772 if (r < 0)
773 return r;
774
775 (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld");
776
777 return 0;
778 }
779
manager_find_credentials_dirs(Manager * m)780 static int manager_find_credentials_dirs(Manager *m) {
781 const char *e;
782 int r;
783
784 assert(m);
785
786 r = get_credentials_dir(&e);
787 if (r < 0) {
788 if (r != -ENXIO)
789 log_debug_errno(r, "Failed to determine credentials directory, ignoring: %m");
790 } else {
791 m->received_credentials_directory = strdup(e);
792 if (!m->received_credentials_directory)
793 return -ENOMEM;
794 }
795
796 r = get_encrypted_credentials_dir(&e);
797 if (r < 0) {
798 if (r != -ENXIO)
799 log_debug_errno(r, "Failed to determine encrypted credentials directory, ignoring: %m");
800 } else {
801 m->received_encrypted_credentials_directory = strdup(e);
802 if (!m->received_encrypted_credentials_directory)
803 return -ENOMEM;
804 }
805
806 return 0;
807 }
808
manager_new(LookupScope scope,ManagerTestRunFlags test_run_flags,Manager ** _m)809 int manager_new(LookupScope scope, ManagerTestRunFlags test_run_flags, Manager **_m) {
810 _cleanup_(manager_freep) Manager *m = NULL;
811 int r;
812
813 assert(_m);
814 assert(IN_SET(scope, LOOKUP_SCOPE_SYSTEM, LOOKUP_SCOPE_USER));
815
816 m = new(Manager, 1);
817 if (!m)
818 return -ENOMEM;
819
820 *m = (Manager) {
821 .unit_file_scope = scope,
822 .objective = _MANAGER_OBJECTIVE_INVALID,
823
824 .status_unit_format = STATUS_UNIT_FORMAT_DEFAULT,
825
826 .default_timer_accuracy_usec = USEC_PER_MINUTE,
827 .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT,
828 .default_tasks_accounting = true,
829 .default_tasks_max = TASKS_MAX_UNSET,
830 .default_timeout_start_usec = DEFAULT_TIMEOUT_USEC,
831 .default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC,
832 .default_restart_usec = DEFAULT_RESTART_USEC,
833
834 .original_log_level = -1,
835 .original_log_target = _LOG_TARGET_INVALID,
836
837 .watchdog_overridden[WATCHDOG_RUNTIME] = USEC_INFINITY,
838 .watchdog_overridden[WATCHDOG_REBOOT] = USEC_INFINITY,
839 .watchdog_overridden[WATCHDOG_KEXEC] = USEC_INFINITY,
840 .watchdog_overridden[WATCHDOG_PRETIMEOUT] = USEC_INFINITY,
841
842 .show_status_overridden = _SHOW_STATUS_INVALID,
843
844 .notify_fd = -1,
845 .cgroups_agent_fd = -1,
846 .signal_fd = -1,
847 .user_lookup_fds = { -1, -1 },
848 .private_listen_fd = -1,
849 .dev_autofs_fd = -1,
850 .cgroup_inotify_fd = -1,
851 .pin_cgroupfs_fd = -1,
852 .ask_password_inotify_fd = -1,
853 .idle_pipe = { -1, -1, -1, -1},
854
855 /* start as id #1, so that we can leave #0 around as "null-like" value */
856 .current_job_id = 1,
857
858 .have_ask_password = -EINVAL, /* we don't know */
859 .first_boot = -1,
860 .test_run_flags = test_run_flags,
861
862 .default_oom_policy = OOM_STOP,
863 };
864
865 #if ENABLE_EFI
866 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
867 boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE,
868 m->timestamps + MANAGER_TIMESTAMP_FIRMWARE,
869 m->timestamps + MANAGER_TIMESTAMP_LOADER);
870 #endif
871
872 /* Prepare log fields we can use for structured logging */
873 if (MANAGER_IS_SYSTEM(m)) {
874 m->unit_log_field = "UNIT=";
875 m->unit_log_format_string = "UNIT=%s";
876
877 m->invocation_log_field = "INVOCATION_ID=";
878 m->invocation_log_format_string = "INVOCATION_ID=%s";
879 } else {
880 m->unit_log_field = "USER_UNIT=";
881 m->unit_log_format_string = "USER_UNIT=%s";
882
883 m->invocation_log_field = "USER_INVOCATION_ID=";
884 m->invocation_log_format_string = "USER_INVOCATION_ID=%s";
885 }
886
887 /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
888 m->ctrl_alt_del_ratelimit = (RateLimit) { .interval = 2 * USEC_PER_SEC, .burst = 7 };
889
890 r = manager_default_environment(m);
891 if (r < 0)
892 return r;
893
894 r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
895 if (r < 0)
896 return r;
897
898 r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops);
899 if (r < 0)
900 return r;
901
902 r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
903 if (r < 0)
904 return r;
905
906 r = prioq_ensure_allocated(&m->run_queue, compare_job_priority);
907 if (r < 0)
908 return r;
909
910 r = manager_setup_prefix(m);
911 if (r < 0)
912 return r;
913
914 r = manager_find_credentials_dirs(m);
915 if (r < 0)
916 return r;
917
918 r = sd_event_default(&m->event);
919 if (r < 0)
920 return r;
921
922 r = manager_setup_run_queue(m);
923 if (r < 0)
924 return r;
925
926 if (FLAGS_SET(test_run_flags, MANAGER_TEST_RUN_MINIMAL)) {
927 m->cgroup_root = strdup("");
928 if (!m->cgroup_root)
929 return -ENOMEM;
930 } else {
931 r = manager_setup_signals(m);
932 if (r < 0)
933 return r;
934
935 r = manager_setup_cgroup(m);
936 if (r < 0)
937 return r;
938
939 r = manager_setup_time_change(m);
940 if (r < 0)
941 return r;
942
943 r = manager_read_timezone_stat(m);
944 if (r < 0)
945 return r;
946
947 (void) manager_setup_timezone_change(m);
948
949 r = manager_setup_sigchld_event_source(m);
950 if (r < 0)
951 return r;
952
953 #if HAVE_LIBBPF
954 if (MANAGER_IS_SYSTEM(m) && lsm_bpf_supported(/* initialize = */ true)) {
955 r = lsm_bpf_setup(m);
956 if (r < 0)
957 log_warning_errno(r, "Failed to setup LSM BPF, ignoring: %m");
958 }
959 #endif
960 }
961
962 if (test_run_flags == 0) {
963 if (MANAGER_IS_SYSTEM(m))
964 r = mkdir_label("/run/systemd/units", 0755);
965 else {
966 _cleanup_free_ char *units_path = NULL;
967 r = xdg_user_runtime_dir(&units_path, "/systemd/units");
968 if (r < 0)
969 return r;
970 r = mkdir_p_label(units_path, 0755);
971 }
972
973 if (r < 0 && r != -EEXIST)
974 return r;
975 }
976
977 m->taint_usr =
978 !in_initrd() &&
979 dir_is_empty("/usr", /* ignore_hidden_or_backup= */ false) > 0;
980
981 /* Note that we do not set up the notify fd here. We do that after deserialization,
982 * since they might have gotten serialized across the reexec. */
983
984 *_m = TAKE_PTR(m);
985
986 return 0;
987 }
988
manager_setup_notify(Manager * m)989 static int manager_setup_notify(Manager *m) {
990 int r;
991
992 if (MANAGER_IS_TEST_RUN(m))
993 return 0;
994
995 if (m->notify_fd < 0) {
996 _cleanup_close_ int fd = -1;
997 union sockaddr_union sa;
998 socklen_t sa_len;
999
1000 /* First free all secondary fields */
1001 m->notify_socket = mfree(m->notify_socket);
1002 m->notify_event_source = sd_event_source_disable_unref(m->notify_event_source);
1003
1004 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1005 if (fd < 0)
1006 return log_error_errno(errno, "Failed to allocate notification socket: %m");
1007
1008 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
1009
1010 m->notify_socket = path_join(m->prefix[EXEC_DIRECTORY_RUNTIME], "systemd/notify");
1011 if (!m->notify_socket)
1012 return log_oom();
1013
1014 r = sockaddr_un_set_path(&sa.un, m->notify_socket);
1015 if (r < 0)
1016 return log_error_errno(r, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.",
1017 m->notify_socket);
1018 sa_len = r;
1019
1020 (void) mkdir_parents_label(m->notify_socket, 0755);
1021 (void) sockaddr_un_unlink(&sa.un);
1022
1023 r = mac_selinux_bind(fd, &sa.sa, sa_len);
1024 if (r < 0)
1025 return log_error_errno(r, "bind(%s) failed: %m", m->notify_socket);
1026
1027 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1028 if (r < 0)
1029 return log_error_errno(r, "SO_PASSCRED failed: %m");
1030
1031 m->notify_fd = TAKE_FD(fd);
1032
1033 log_debug("Using notification socket %s", m->notify_socket);
1034 }
1035
1036 if (!m->notify_event_source) {
1037 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
1038 if (r < 0)
1039 return log_error_errno(r, "Failed to allocate notify event source: %m");
1040
1041 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
1042 * service an exit message belongs. */
1043 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8);
1044 if (r < 0)
1045 return log_error_errno(r, "Failed to set priority of notify event source: %m");
1046
1047 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
1048 }
1049
1050 return 0;
1051 }
1052
manager_setup_cgroups_agent(Manager * m)1053 static int manager_setup_cgroups_agent(Manager *m) {
1054
1055 static const union sockaddr_union sa = {
1056 .un.sun_family = AF_UNIX,
1057 .un.sun_path = "/run/systemd/cgroups-agent",
1058 };
1059 int r;
1060
1061 /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
1062 * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
1063 * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
1064 * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
1065 * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
1066 * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
1067 * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
1068 * we thus won't lose messages.
1069 *
1070 * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
1071 * to it. The system instance hence listens on this special socket, but the user instances listen on the system
1072 * bus for these messages. */
1073
1074 if (MANAGER_IS_TEST_RUN(m))
1075 return 0;
1076
1077 if (!MANAGER_IS_SYSTEM(m))
1078 return 0;
1079
1080 r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
1081 if (r < 0)
1082 return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m");
1083 if (r > 0) /* We don't need this anymore on the unified hierarchy */
1084 return 0;
1085
1086 if (m->cgroups_agent_fd < 0) {
1087 _cleanup_close_ int fd = -1;
1088
1089 /* First free all secondary fields */
1090 m->cgroups_agent_event_source = sd_event_source_disable_unref(m->cgroups_agent_event_source);
1091
1092 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1093 if (fd < 0)
1094 return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
1095
1096 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
1097
1098 (void) sockaddr_un_unlink(&sa.un);
1099
1100 /* Only allow root to connect to this socket */
1101 RUN_WITH_UMASK(0077)
1102 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
1103 if (r < 0)
1104 return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
1105
1106 m->cgroups_agent_fd = TAKE_FD(fd);
1107 }
1108
1109 if (!m->cgroups_agent_event_source) {
1110 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
1111 if (r < 0)
1112 return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
1113
1114 /* Process cgroups notifications early. Note that when the agent notification is received
1115 * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than
1116 * that. Also see handling of cgroup inotify for the unified cgroup stuff. */
1117 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9);
1118 if (r < 0)
1119 return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
1120
1121 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
1122 }
1123
1124 return 0;
1125 }
1126
manager_setup_user_lookup_fd(Manager * m)1127 static int manager_setup_user_lookup_fd(Manager *m) {
1128 int r;
1129
1130 assert(m);
1131
1132 /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
1133 * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
1134 * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
1135 * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
1136 * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
1137 * hence we establish this communication channel so that forked off processes can pass their UID/GID
1138 * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
1139 * datagram, along with their unit name, so that we can share one communication socket pair among all units for
1140 * this purpose.
1141 *
1142 * You might wonder why we need a communication channel for this that is independent of the usual notification
1143 * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
1144 * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
1145 * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
1146 *
1147 * Note that this function is called under two circumstances: when we first initialize (in which case we
1148 * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
1149 * (in which case the socket pair already exists but we still need to allocate the event source for it). */
1150
1151 if (m->user_lookup_fds[0] < 0) {
1152
1153 /* Free all secondary fields */
1154 safe_close_pair(m->user_lookup_fds);
1155 m->user_lookup_event_source = sd_event_source_disable_unref(m->user_lookup_event_source);
1156
1157 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
1158 return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
1159
1160 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
1161 }
1162
1163 if (!m->user_lookup_event_source) {
1164 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
1165 if (r < 0)
1166 return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
1167
1168 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
1169 * resolutions */
1170 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11);
1171 if (r < 0)
1172 return log_error_errno(errno, "Failed to set priority of user lookup event source: %m");
1173
1174 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
1175 }
1176
1177 return 0;
1178 }
1179
manager_dispatch_cleanup_queue(Manager * m)1180 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
1181 Unit *u;
1182 unsigned n = 0;
1183
1184 assert(m);
1185
1186 while ((u = m->cleanup_queue)) {
1187 assert(u->in_cleanup_queue);
1188
1189 unit_free(u);
1190 n++;
1191 }
1192
1193 return n;
1194 }
1195
1196 enum {
1197 GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */
1198 GC_OFFSET_UNSURE, /* No clue */
1199 GC_OFFSET_GOOD, /* We still need this unit */
1200 GC_OFFSET_BAD, /* We don't need this unit anymore */
1201 _GC_OFFSET_MAX
1202 };
1203
unit_gc_mark_good(Unit * u,unsigned gc_marker)1204 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
1205 Unit *other;
1206
1207 u->gc_marker = gc_marker + GC_OFFSET_GOOD;
1208
1209 /* Recursively mark referenced units as GOOD as well */
1210 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCES)
1211 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
1212 unit_gc_mark_good(other, gc_marker);
1213 }
1214
unit_gc_sweep(Unit * u,unsigned gc_marker)1215 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
1216 Unit *other;
1217 bool is_bad;
1218
1219 assert(u);
1220
1221 if (IN_SET(u->gc_marker - gc_marker,
1222 GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH))
1223 return;
1224
1225 if (u->in_cleanup_queue)
1226 goto bad;
1227
1228 if (!unit_may_gc(u))
1229 goto good;
1230
1231 u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
1232
1233 is_bad = true;
1234
1235 UNIT_FOREACH_DEPENDENCY(other, u, UNIT_ATOM_REFERENCED_BY) {
1236 unit_gc_sweep(other, gc_marker);
1237
1238 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
1239 goto good;
1240
1241 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
1242 is_bad = false;
1243 }
1244
1245 LIST_FOREACH(refs_by_target, ref, u->refs_by_target) {
1246 unit_gc_sweep(ref->source, gc_marker);
1247
1248 if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD)
1249 goto good;
1250
1251 if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD)
1252 is_bad = false;
1253 }
1254
1255 if (is_bad)
1256 goto bad;
1257
1258 /* We were unable to find anything out about this entry, so
1259 * let's investigate it later */
1260 u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
1261 unit_add_to_gc_queue(u);
1262 return;
1263
1264 bad:
1265 /* We definitely know that this one is not useful anymore, so
1266 * let's mark it for deletion */
1267 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1268 unit_add_to_cleanup_queue(u);
1269 return;
1270
1271 good:
1272 unit_gc_mark_good(u, gc_marker);
1273 }
1274
manager_dispatch_gc_unit_queue(Manager * m)1275 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
1276 unsigned n = 0, gc_marker;
1277 Unit *u;
1278
1279 assert(m);
1280
1281 /* log_debug("Running GC..."); */
1282
1283 m->gc_marker += _GC_OFFSET_MAX;
1284 if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
1285 m->gc_marker = 1;
1286
1287 gc_marker = m->gc_marker;
1288
1289 while ((u = m->gc_unit_queue)) {
1290 assert(u->in_gc_queue);
1291
1292 unit_gc_sweep(u, gc_marker);
1293
1294 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1295 u->in_gc_queue = false;
1296
1297 n++;
1298
1299 if (IN_SET(u->gc_marker - gc_marker,
1300 GC_OFFSET_BAD, GC_OFFSET_UNSURE)) {
1301 if (u->id)
1302 log_unit_debug(u, "Collecting.");
1303 u->gc_marker = gc_marker + GC_OFFSET_BAD;
1304 unit_add_to_cleanup_queue(u);
1305 }
1306 }
1307
1308 return n;
1309 }
1310
manager_dispatch_gc_job_queue(Manager * m)1311 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1312 unsigned n = 0;
1313 Job *j;
1314
1315 assert(m);
1316
1317 while ((j = m->gc_job_queue)) {
1318 assert(j->in_gc_queue);
1319
1320 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1321 j->in_gc_queue = false;
1322
1323 n++;
1324
1325 if (!job_may_gc(j))
1326 continue;
1327
1328 log_unit_debug(j->unit, "Collecting job.");
1329 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1330 }
1331
1332 return n;
1333 }
1334
manager_dispatch_stop_when_unneeded_queue(Manager * m)1335 static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) {
1336 unsigned n = 0;
1337 Unit *u;
1338 int r;
1339
1340 assert(m);
1341
1342 while ((u = m->stop_when_unneeded_queue)) {
1343 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1344
1345 assert(u->in_stop_when_unneeded_queue);
1346 LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u);
1347 u->in_stop_when_unneeded_queue = false;
1348
1349 n++;
1350
1351 if (!unit_is_unneeded(u))
1352 continue;
1353
1354 log_unit_debug(u, "Unit is not needed anymore.");
1355
1356 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1357 * service being unnecessary after a while. */
1358
1359 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1360 log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
1361 continue;
1362 }
1363
1364 /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
1365 r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
1366 if (r < 0)
1367 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1368 }
1369
1370 return n;
1371 }
1372
manager_dispatch_start_when_upheld_queue(Manager * m)1373 static unsigned manager_dispatch_start_when_upheld_queue(Manager *m) {
1374 unsigned n = 0;
1375 Unit *u;
1376 int r;
1377
1378 assert(m);
1379
1380 while ((u = m->start_when_upheld_queue)) {
1381 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1382 Unit *culprit = NULL;
1383
1384 assert(u->in_start_when_upheld_queue);
1385 LIST_REMOVE(start_when_upheld_queue, m->start_when_upheld_queue, u);
1386 u->in_start_when_upheld_queue = false;
1387
1388 n++;
1389
1390 if (!unit_is_upheld_by_active(u, &culprit))
1391 continue;
1392
1393 log_unit_debug(u, "Unit is started because upheld by active unit %s.", culprit->id);
1394
1395 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1396 * service being unnecessary after a while. */
1397
1398 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1399 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);
1400 continue;
1401 }
1402
1403 r = manager_add_job(u->manager, JOB_START, u, JOB_FAIL, NULL, &error, NULL);
1404 if (r < 0)
1405 log_unit_warning_errno(u, r, "Failed to enqueue start job, ignoring: %s", bus_error_message(&error, r));
1406 }
1407
1408 return n;
1409 }
1410
manager_dispatch_stop_when_bound_queue(Manager * m)1411 static unsigned manager_dispatch_stop_when_bound_queue(Manager *m) {
1412 unsigned n = 0;
1413 Unit *u;
1414 int r;
1415
1416 assert(m);
1417
1418 while ((u = m->stop_when_bound_queue)) {
1419 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1420 Unit *culprit = NULL;
1421
1422 assert(u->in_stop_when_bound_queue);
1423 LIST_REMOVE(stop_when_bound_queue, m->stop_when_bound_queue, u);
1424 u->in_stop_when_bound_queue = false;
1425
1426 n++;
1427
1428 if (!unit_is_bound_by_inactive(u, &culprit))
1429 continue;
1430
1431 log_unit_debug(u, "Unit is stopped because bound to inactive unit %s.", culprit->id);
1432
1433 /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the
1434 * service being unnecessary after a while. */
1435
1436 if (!ratelimit_below(&u->auto_start_stop_ratelimit)) {
1437 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);
1438 continue;
1439 }
1440
1441 r = manager_add_job(u->manager, JOB_STOP, u, JOB_REPLACE, NULL, &error, NULL);
1442 if (r < 0)
1443 log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
1444 }
1445
1446 return n;
1447 }
1448
manager_clear_jobs_and_units(Manager * m)1449 static void manager_clear_jobs_and_units(Manager *m) {
1450 Unit *u;
1451
1452 assert(m);
1453
1454 while ((u = hashmap_first(m->units)))
1455 unit_free(u);
1456
1457 manager_dispatch_cleanup_queue(m);
1458
1459 assert(!m->load_queue);
1460 assert(prioq_isempty(m->run_queue));
1461 assert(!m->dbus_unit_queue);
1462 assert(!m->dbus_job_queue);
1463 assert(!m->cleanup_queue);
1464 assert(!m->gc_unit_queue);
1465 assert(!m->gc_job_queue);
1466 assert(!m->cgroup_realize_queue);
1467 assert(!m->cgroup_empty_queue);
1468 assert(!m->cgroup_oom_queue);
1469 assert(!m->target_deps_queue);
1470 assert(!m->stop_when_unneeded_queue);
1471 assert(!m->start_when_upheld_queue);
1472 assert(!m->stop_when_bound_queue);
1473
1474 assert(hashmap_isempty(m->jobs));
1475 assert(hashmap_isempty(m->units));
1476
1477 m->n_on_console = 0;
1478 m->n_running_jobs = 0;
1479 m->n_installed_jobs = 0;
1480 m->n_failed_jobs = 0;
1481 }
1482
manager_free(Manager * m)1483 Manager* manager_free(Manager *m) {
1484 if (!m)
1485 return NULL;
1486
1487 manager_clear_jobs_and_units(m);
1488
1489 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++)
1490 if (unit_vtable[c]->shutdown)
1491 unit_vtable[c]->shutdown(m);
1492
1493 /* Keep the cgroup hierarchy in place except when we know we are going down for good */
1494 manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC));
1495
1496 lookup_paths_flush_generator(&m->lookup_paths);
1497
1498 bus_done(m);
1499 manager_varlink_done(m);
1500
1501 exec_runtime_vacuum(m);
1502 hashmap_free(m->exec_runtime_by_id);
1503
1504 dynamic_user_vacuum(m, false);
1505 hashmap_free(m->dynamic_users);
1506
1507 hashmap_free(m->units);
1508 hashmap_free(m->units_by_invocation_id);
1509 hashmap_free(m->jobs);
1510 hashmap_free(m->watch_pids);
1511 hashmap_free(m->watch_bus);
1512
1513 prioq_free(m->run_queue);
1514
1515 set_free(m->startup_units);
1516 set_free(m->failed_units);
1517
1518 sd_event_source_unref(m->signal_event_source);
1519 sd_event_source_unref(m->sigchld_event_source);
1520 sd_event_source_unref(m->notify_event_source);
1521 sd_event_source_unref(m->cgroups_agent_event_source);
1522 sd_event_source_unref(m->time_change_event_source);
1523 sd_event_source_unref(m->timezone_change_event_source);
1524 sd_event_source_unref(m->jobs_in_progress_event_source);
1525 sd_event_source_unref(m->run_queue_event_source);
1526 sd_event_source_unref(m->user_lookup_event_source);
1527
1528 safe_close(m->signal_fd);
1529 safe_close(m->notify_fd);
1530 safe_close(m->cgroups_agent_fd);
1531 safe_close_pair(m->user_lookup_fds);
1532
1533 manager_close_ask_password(m);
1534
1535 manager_close_idle_pipe(m);
1536
1537 sd_event_unref(m->event);
1538
1539 free(m->notify_socket);
1540
1541 lookup_paths_free(&m->lookup_paths);
1542 strv_free(m->transient_environment);
1543 strv_free(m->client_environment);
1544
1545 hashmap_free(m->cgroup_unit);
1546 manager_free_unit_name_maps(m);
1547
1548 free(m->switch_root);
1549 free(m->switch_root_init);
1550
1551 rlimit_free_all(m->rlimit);
1552
1553 assert(hashmap_isempty(m->units_requiring_mounts_for));
1554 hashmap_free(m->units_requiring_mounts_for);
1555
1556 hashmap_free(m->uid_refs);
1557 hashmap_free(m->gid_refs);
1558
1559 for (ExecDirectoryType dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++)
1560 m->prefix[dt] = mfree(m->prefix[dt]);
1561 free(m->received_credentials_directory);
1562 free(m->received_encrypted_credentials_directory);
1563
1564 free(m->watchdog_pretimeout_governor);
1565 free(m->watchdog_pretimeout_governor_overridden);
1566
1567 #if BPF_FRAMEWORK
1568 lsm_bpf_destroy(m->restrict_fs);
1569 #endif
1570
1571 return mfree(m);
1572 }
1573
manager_enumerate_perpetual(Manager * m)1574 static void manager_enumerate_perpetual(Manager *m) {
1575 assert(m);
1576
1577 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1578 return;
1579
1580 /* Let's ask every type to load all units from disk/kernel that it might know */
1581 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1582 if (!unit_type_supported(c)) {
1583 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1584 continue;
1585 }
1586
1587 if (unit_vtable[c]->enumerate_perpetual)
1588 unit_vtable[c]->enumerate_perpetual(m);
1589 }
1590 }
1591
manager_enumerate(Manager * m)1592 static void manager_enumerate(Manager *m) {
1593 assert(m);
1594
1595 if (FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL))
1596 return;
1597
1598 /* Let's ask every type to load all units from disk/kernel that it might know */
1599 for (UnitType c = 0; c < _UNIT_TYPE_MAX; c++) {
1600 if (!unit_type_supported(c)) {
1601 log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1602 continue;
1603 }
1604
1605 if (unit_vtable[c]->enumerate)
1606 unit_vtable[c]->enumerate(m);
1607 }
1608
1609 manager_dispatch_load_queue(m);
1610 }
1611
manager_coldplug(Manager * m)1612 static void manager_coldplug(Manager *m) {
1613 Unit *u;
1614 char *k;
1615 int r;
1616
1617 assert(m);
1618
1619 log_debug("Invoking unit coldplug() handlers…");
1620
1621 /* Let's place the units back into their deserialized state */
1622 HASHMAP_FOREACH_KEY(u, k, m->units) {
1623
1624 /* ignore aliases */
1625 if (u->id != k)
1626 continue;
1627
1628 r = unit_coldplug(u);
1629 if (r < 0)
1630 log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1631 }
1632 }
1633
manager_catchup(Manager * m)1634 static void manager_catchup(Manager *m) {
1635 Unit *u;
1636 char *k;
1637
1638 assert(m);
1639
1640 log_debug("Invoking unit catchup() handlers…");
1641
1642 /* Let's catch up on any state changes that happened while we were reloading/reexecing */
1643 HASHMAP_FOREACH_KEY(u, k, m->units) {
1644
1645 /* ignore aliases */
1646 if (u->id != k)
1647 continue;
1648
1649 unit_catchup(u);
1650 }
1651 }
1652
manager_distribute_fds(Manager * m,FDSet * fds)1653 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1654 Unit *u;
1655
1656 assert(m);
1657
1658 HASHMAP_FOREACH(u, m->units) {
1659
1660 if (fdset_size(fds) <= 0)
1661 break;
1662
1663 if (!UNIT_VTABLE(u)->distribute_fds)
1664 continue;
1665
1666 UNIT_VTABLE(u)->distribute_fds(u, fds);
1667 }
1668 }
1669
manager_dbus_is_running(Manager * m,bool deserialized)1670 static bool manager_dbus_is_running(Manager *m, bool deserialized) {
1671 Unit *u;
1672
1673 assert(m);
1674
1675 /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket
1676 * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit
1677 * rather than the current one. */
1678
1679 if (MANAGER_IS_TEST_RUN(m))
1680 return false;
1681
1682 u = manager_get_unit(m, SPECIAL_DBUS_SOCKET);
1683 if (!u)
1684 return false;
1685 if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING)
1686 return false;
1687
1688 u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
1689 if (!u)
1690 return false;
1691 if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), SERVICE_RUNNING, SERVICE_RELOAD))
1692 return false;
1693
1694 return true;
1695 }
1696
manager_setup_bus(Manager * m)1697 static void manager_setup_bus(Manager *m) {
1698 assert(m);
1699
1700 /* Let's set up our private bus connection now, unconditionally */
1701 (void) bus_init_private(m);
1702
1703 /* If we are in --user mode also connect to the system bus now */
1704 if (MANAGER_IS_USER(m))
1705 (void) bus_init_system(m);
1706
1707 /* Let's connect to the bus now, but only if the unit is supposed to be up */
1708 if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) {
1709 (void) bus_init_api(m);
1710
1711 if (MANAGER_IS_SYSTEM(m))
1712 (void) bus_init_system(m);
1713 }
1714 }
1715
manager_preset_all(Manager * m)1716 static void manager_preset_all(Manager *m) {
1717 int r;
1718
1719 assert(m);
1720
1721 if (m->first_boot <= 0)
1722 return;
1723
1724 if (!MANAGER_IS_SYSTEM(m))
1725 return;
1726
1727 if (MANAGER_IS_TEST_RUN(m))
1728 return;
1729
1730 /* If this is the first boot, and we are in the host system, then preset everything */
1731 r = unit_file_preset_all(LOOKUP_SCOPE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0);
1732 if (r < 0)
1733 log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r,
1734 "Failed to populate /etc with preset unit settings, ignoring: %m");
1735 else
1736 log_info("Populated /etc with preset unit settings.");
1737 }
1738
manager_ready(Manager * m)1739 static void manager_ready(Manager *m) {
1740 assert(m);
1741
1742 /* After having loaded everything, do the final round of catching up with what might have changed */
1743
1744 m->objective = MANAGER_OK; /* Tell everyone we are up now */
1745
1746 /* It might be safe to log to the journal now and connect to dbus */
1747 manager_recheck_journal(m);
1748 manager_recheck_dbus(m);
1749
1750 /* Let's finally catch up with any changes that took place while we were reloading/reexecing */
1751 manager_catchup(m);
1752
1753 /* Create a file which will indicate when the manager started loading units the last time. */
1754 if (MANAGER_IS_SYSTEM(m))
1755 (void) touch_file("/run/systemd/systemd-units-load", false,
1756 m->timestamps[MANAGER_TIMESTAMP_UNITS_LOAD].realtime ?: now(CLOCK_REALTIME),
1757 UID_INVALID, GID_INVALID, 0444);
1758
1759 m->honor_device_enumeration = true;
1760 }
1761
manager_reloading_start(Manager * m)1762 Manager* manager_reloading_start(Manager *m) {
1763 m->n_reloading++;
1764 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_UNITS_LOAD);
1765 return m;
1766 }
1767
manager_reloading_stopp(Manager ** m)1768 void manager_reloading_stopp(Manager **m) {
1769 if (*m) {
1770 assert((*m)->n_reloading > 0);
1771 (*m)->n_reloading--;
1772 }
1773 }
1774
manager_startup(Manager * m,FILE * serialization,FDSet * fds,const char * root)1775 int manager_startup(Manager *m, FILE *serialization, FDSet *fds, const char *root) {
1776 int r;
1777
1778 assert(m);
1779
1780 /* If we are running in test mode, we still want to run the generators,
1781 * but we should not touch the real generator directories. */
1782 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope,
1783 MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0,
1784 root);
1785 if (r < 0)
1786 return r;
1787
1788 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START));
1789 r = manager_run_environment_generators(m);
1790 if (r >= 0)
1791 r = manager_run_generators(m);
1792 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH));
1793 if (r < 0)
1794 return r;
1795
1796 manager_preset_all(m);
1797
1798 lookup_paths_log(&m->lookup_paths);
1799
1800 {
1801 /* This block is (optionally) done with the reloading counter bumped */
1802 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
1803
1804 /* If we will deserialize make sure that during enumeration this is already known, so we increase the
1805 * counter here already */
1806 if (serialization)
1807 reloading = manager_reloading_start(m);
1808
1809 /* First, enumerate what we can from all config files */
1810 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START));
1811 manager_enumerate_perpetual(m);
1812 manager_enumerate(m);
1813 dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH));
1814
1815 /* Second, deserialize if there is something to deserialize */
1816 if (serialization) {
1817 r = manager_deserialize(m, serialization, fds);
1818 if (r < 0)
1819 return log_error_errno(r, "Deserialization failed: %m");
1820 }
1821
1822 /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass
1823 * some file descriptors to us pre-initialized. This enables socket-based activation of entire
1824 * containers. */
1825 manager_distribute_fds(m, fds);
1826
1827 /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */
1828 r = manager_setup_notify(m);
1829 if (r < 0)
1830 /* No sense to continue without notifications, our children would fail anyway. */
1831 return r;
1832
1833 r = manager_setup_cgroups_agent(m);
1834 if (r < 0)
1835 /* Likewise, no sense to continue without empty cgroup notifications. */
1836 return r;
1837
1838 r = manager_setup_user_lookup_fd(m);
1839 if (r < 0)
1840 /* This shouldn't fail, except if things are really broken. */
1841 return r;
1842
1843 /* Connect to the bus if we are good for it */
1844 manager_setup_bus(m);
1845
1846 /* Now that we are connected to all possible buses, let's deserialize who is tracking us. */
1847 r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1848 if (r < 0)
1849 log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m");
1850 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1851
1852 r = manager_varlink_init(m);
1853 if (r < 0)
1854 log_warning_errno(r, "Failed to set up Varlink, ignoring: %m");
1855
1856 /* Third, fire things up! */
1857 manager_coldplug(m);
1858
1859 /* Clean up runtime objects */
1860 manager_vacuum(m);
1861
1862 if (serialization)
1863 /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the
1864 * reload is finished */
1865 m->send_reloading_done = true;
1866 }
1867
1868 manager_ready(m);
1869
1870 return 0;
1871 }
1872
manager_add_job(Manager * m,JobType type,Unit * unit,JobMode mode,Set * affected_jobs,sd_bus_error * error,Job ** ret)1873 int manager_add_job(
1874 Manager *m,
1875 JobType type,
1876 Unit *unit,
1877 JobMode mode,
1878 Set *affected_jobs,
1879 sd_bus_error *error,
1880 Job **ret) {
1881
1882 Transaction *tr;
1883 int r;
1884
1885 assert(m);
1886 assert(type < _JOB_TYPE_MAX);
1887 assert(unit);
1888 assert(mode < _JOB_MODE_MAX);
1889
1890 if (mode == JOB_ISOLATE && type != JOB_START)
1891 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1892
1893 if (mode == JOB_ISOLATE && !unit->allow_isolate)
1894 return sd_bus_error_set(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1895
1896 if (mode == JOB_TRIGGERING && type != JOB_STOP)
1897 return sd_bus_error_set(error, SD_BUS_ERROR_INVALID_ARGS, "--job-mode=triggering is only valid for stop.");
1898
1899 log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1900
1901 type = job_type_collapse(type, unit);
1902
1903 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1904 if (!tr)
1905 return -ENOMEM;
1906
1907 r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1908 IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS),
1909 mode == JOB_IGNORE_DEPENDENCIES, error);
1910 if (r < 0)
1911 goto tr_abort;
1912
1913 if (mode == JOB_ISOLATE) {
1914 r = transaction_add_isolate_jobs(tr, m);
1915 if (r < 0)
1916 goto tr_abort;
1917 }
1918
1919 if (mode == JOB_TRIGGERING) {
1920 r = transaction_add_triggering_jobs(tr, unit);
1921 if (r < 0)
1922 goto tr_abort;
1923 }
1924
1925 r = transaction_activate(tr, m, mode, affected_jobs, error);
1926 if (r < 0)
1927 goto tr_abort;
1928
1929 log_unit_debug(unit,
1930 "Enqueued job %s/%s as %u", unit->id,
1931 job_type_to_string(type), (unsigned) tr->anchor_job->id);
1932
1933 if (ret)
1934 *ret = tr->anchor_job;
1935
1936 transaction_free(tr);
1937 return 0;
1938
1939 tr_abort:
1940 transaction_abort(tr);
1941 transaction_free(tr);
1942 return r;
1943 }
1944
manager_add_job_by_name(Manager * m,JobType type,const char * name,JobMode mode,Set * affected_jobs,sd_bus_error * e,Job ** ret)1945 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) {
1946 Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */
1947 int r;
1948
1949 assert(m);
1950 assert(type < _JOB_TYPE_MAX);
1951 assert(name);
1952 assert(mode < _JOB_MODE_MAX);
1953
1954 r = manager_load_unit(m, name, NULL, NULL, &unit);
1955 if (r < 0)
1956 return r;
1957 assert(unit);
1958
1959 return manager_add_job(m, type, unit, mode, affected_jobs, e, ret);
1960 }
1961
manager_add_job_by_name_and_warn(Manager * m,JobType type,const char * name,JobMode mode,Set * affected_jobs,Job ** ret)1962 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) {
1963 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1964 int r;
1965
1966 assert(m);
1967 assert(type < _JOB_TYPE_MAX);
1968 assert(name);
1969 assert(mode < _JOB_MODE_MAX);
1970
1971 r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret);
1972 if (r < 0)
1973 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1974
1975 return r;
1976 }
1977
manager_propagate_reload(Manager * m,Unit * unit,JobMode mode,sd_bus_error * e)1978 int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) {
1979 int r;
1980 Transaction *tr;
1981
1982 assert(m);
1983 assert(unit);
1984 assert(mode < _JOB_MODE_MAX);
1985 assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */
1986
1987 tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1988 if (!tr)
1989 return -ENOMEM;
1990
1991 /* We need an anchor job */
1992 r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e);
1993 if (r < 0)
1994 goto tr_abort;
1995
1996 /* Failure in adding individual dependencies is ignored, so this always succeeds. */
1997 transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e);
1998
1999 r = transaction_activate(tr, m, mode, NULL, e);
2000 if (r < 0)
2001 goto tr_abort;
2002
2003 transaction_free(tr);
2004 return 0;
2005
2006 tr_abort:
2007 transaction_abort(tr);
2008 transaction_free(tr);
2009 return r;
2010 }
2011
manager_get_job(Manager * m,uint32_t id)2012 Job *manager_get_job(Manager *m, uint32_t id) {
2013 assert(m);
2014
2015 return hashmap_get(m->jobs, UINT32_TO_PTR(id));
2016 }
2017
manager_get_unit(Manager * m,const char * name)2018 Unit *manager_get_unit(Manager *m, const char *name) {
2019 assert(m);
2020 assert(name);
2021
2022 return hashmap_get(m->units, name);
2023 }
2024
manager_dispatch_target_deps_queue(Manager * m)2025 static int manager_dispatch_target_deps_queue(Manager *m) {
2026 Unit *u;
2027 int r = 0;
2028
2029 assert(m);
2030
2031 while ((u = m->target_deps_queue)) {
2032 _cleanup_free_ Unit **targets = NULL;
2033 int n_targets;
2034
2035 assert(u->in_target_deps_queue);
2036
2037 LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
2038 u->in_target_deps_queue = false;
2039
2040 /* Take an "atomic" snapshot of dependencies here, as the call below will likely modify the
2041 * dependencies, and we can't have it that hash tables we iterate through are modified while
2042 * we are iterating through them. */
2043 n_targets = unit_get_dependency_array(u, UNIT_ATOM_DEFAULT_TARGET_DEPENDENCIES, &targets);
2044 if (n_targets < 0)
2045 return n_targets;
2046
2047 for (int i = 0; i < n_targets; i++) {
2048 r = unit_add_default_target_dependency(u, targets[i]);
2049 if (r < 0)
2050 return r;
2051 }
2052 }
2053
2054 return r;
2055 }
2056
manager_dispatch_load_queue(Manager * m)2057 unsigned manager_dispatch_load_queue(Manager *m) {
2058 Unit *u;
2059 unsigned n = 0;
2060
2061 assert(m);
2062
2063 /* Make sure we are not run recursively */
2064 if (m->dispatching_load_queue)
2065 return 0;
2066
2067 m->dispatching_load_queue = true;
2068
2069 /* Dispatches the load queue. Takes a unit from the queue and
2070 * tries to load its data until the queue is empty */
2071
2072 while ((u = m->load_queue)) {
2073 assert(u->in_load_queue);
2074
2075 unit_load(u);
2076 n++;
2077 }
2078
2079 m->dispatching_load_queue = false;
2080
2081 /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about
2082 * should be loaded and have aliases resolved */
2083 (void) manager_dispatch_target_deps_queue(m);
2084
2085 return n;
2086 }
2087
manager_unit_cache_should_retry_load(Unit * u)2088 bool manager_unit_cache_should_retry_load(Unit *u) {
2089 assert(u);
2090
2091 /* Automatic reloading from disk only applies to units which were not found sometime in the past, and
2092 * the not-found stub is kept pinned in the unit graph by dependencies. For units that were
2093 * previously loaded, we don't do automatic reloading, and daemon-reload is necessary to update. */
2094 if (u->load_state != UNIT_NOT_FOUND)
2095 return false;
2096
2097 /* The cache has been updated since the last time we tried to load the unit. There might be new
2098 * fragment paths to read. */
2099 if (u->manager->unit_cache_timestamp_hash != u->fragment_not_found_timestamp_hash)
2100 return true;
2101
2102 /* The cache needs to be updated because there are modifications on disk. */
2103 return !lookup_paths_timestamp_hash_same(&u->manager->lookup_paths, u->manager->unit_cache_timestamp_hash, NULL);
2104 }
2105
manager_load_unit_prepare(Manager * m,const char * name,const char * path,sd_bus_error * e,Unit ** _ret)2106 int manager_load_unit_prepare(
2107 Manager *m,
2108 const char *name,
2109 const char *path,
2110 sd_bus_error *e,
2111 Unit **_ret) {
2112
2113 _cleanup_(unit_freep) Unit *cleanup_ret = NULL;
2114 Unit *ret;
2115 UnitType t;
2116 int r;
2117
2118 assert(m);
2119 assert(_ret);
2120
2121 /* This will prepare the unit for loading, but not actually load anything from disk. */
2122
2123 if (path && !path_is_absolute(path))
2124 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
2125
2126 if (!name) {
2127 /* 'name' and 'path' must not both be null. Check here 'path' using assert_se() to
2128 * workaround a bug in gcc that generates a -Wnonnull warning when calling basename(),
2129 * but this cannot be possible in any code path (See #6119). */
2130 assert_se(path);
2131 name = basename(path);
2132 }
2133
2134 t = unit_name_to_type(name);
2135
2136 if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
2137 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
2138 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
2139
2140 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
2141 }
2142
2143 ret = manager_get_unit(m, name);
2144 if (ret) {
2145 /* The time-based cache allows to start new units without daemon-reload,
2146 * but if they are already referenced (because of dependencies or ordering)
2147 * then we have to force a load of the fragment. As an optimization, check
2148 * first if anything in the usual paths was modified since the last time
2149 * the cache was loaded. Also check if the last time an attempt to load the
2150 * unit was made was before the most recent cache refresh, so that we know
2151 * we need to try again — even if the cache is current, it might have been
2152 * updated in a different context before we had a chance to retry loading
2153 * this particular unit. */
2154 if (manager_unit_cache_should_retry_load(ret))
2155 ret->load_state = UNIT_STUB;
2156 else {
2157 *_ret = ret;
2158 return 1;
2159 }
2160 } else {
2161 ret = cleanup_ret = unit_new(m, unit_vtable[t]->object_size);
2162 if (!ret)
2163 return -ENOMEM;
2164 }
2165
2166 if (path) {
2167 r = free_and_strdup(&ret->fragment_path, path);
2168 if (r < 0)
2169 return r;
2170 }
2171
2172 r = unit_add_name(ret, name);
2173 if (r < 0)
2174 return r;
2175
2176 unit_add_to_load_queue(ret);
2177 unit_add_to_dbus_queue(ret);
2178 unit_add_to_gc_queue(ret);
2179
2180 *_ret = ret;
2181 cleanup_ret = NULL;
2182
2183 return 0;
2184 }
2185
manager_load_unit(Manager * m,const char * name,const char * path,sd_bus_error * e,Unit ** _ret)2186 int manager_load_unit(
2187 Manager *m,
2188 const char *name,
2189 const char *path,
2190 sd_bus_error *e,
2191 Unit **_ret) {
2192
2193 int r;
2194
2195 assert(m);
2196 assert(_ret);
2197
2198 /* This will load the service information files, but not actually
2199 * start any services or anything. */
2200
2201 r = manager_load_unit_prepare(m, name, path, e, _ret);
2202 if (r != 0)
2203 return r;
2204
2205 manager_dispatch_load_queue(m);
2206
2207 *_ret = unit_follow_merge(*_ret);
2208 return 0;
2209 }
2210
manager_load_startable_unit_or_warn(Manager * m,const char * name,const char * path,Unit ** ret)2211 int manager_load_startable_unit_or_warn(
2212 Manager *m,
2213 const char *name,
2214 const char *path,
2215 Unit **ret) {
2216
2217 /* Load a unit, make sure it loaded fully and is not masked. */
2218
2219 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2220 Unit *unit;
2221 int r;
2222
2223 r = manager_load_unit(m, name, path, &error, &unit);
2224 if (r < 0)
2225 return log_error_errno(r, "Failed to load %s %s: %s",
2226 name ? "unit" : "unit file", name ?: path,
2227 bus_error_message(&error, r));
2228
2229 r = bus_unit_validate_load_state(unit, &error);
2230 if (r < 0)
2231 return log_error_errno(r, "%s", bus_error_message(&error, r));
2232
2233 *ret = unit;
2234 return 0;
2235 }
2236
manager_clear_jobs(Manager * m)2237 void manager_clear_jobs(Manager *m) {
2238 Job *j;
2239
2240 assert(m);
2241
2242 while ((j = hashmap_first(m->jobs)))
2243 /* No need to recurse. We're cancelling all jobs. */
2244 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
2245 }
2246
manager_unwatch_pid(Manager * m,pid_t pid)2247 void manager_unwatch_pid(Manager *m, pid_t pid) {
2248 assert(m);
2249
2250 /* First let's drop the unit keyed as "pid". */
2251 (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid));
2252
2253 /* Then, let's also drop the array keyed by -pid. */
2254 free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid)));
2255 }
2256
manager_dispatch_run_queue(sd_event_source * source,void * userdata)2257 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
2258 Manager *m = userdata;
2259 Job *j;
2260
2261 assert(source);
2262 assert(m);
2263
2264 while ((j = prioq_peek(m->run_queue))) {
2265 assert(j->installed);
2266 assert(j->in_run_queue);
2267
2268 (void) job_run_and_invalidate(j);
2269 }
2270
2271 if (m->n_running_jobs > 0)
2272 manager_watch_jobs_in_progress(m);
2273
2274 if (m->n_on_console > 0)
2275 manager_watch_idle_pipe(m);
2276
2277 return 1;
2278 }
2279
manager_trigger_run_queue(Manager * m)2280 void manager_trigger_run_queue(Manager *m) {
2281 int r;
2282
2283 assert(m);
2284
2285 r = sd_event_source_set_enabled(
2286 m->run_queue_event_source,
2287 prioq_isempty(m->run_queue) ? SD_EVENT_OFF : SD_EVENT_ONESHOT);
2288 if (r < 0)
2289 log_warning_errno(r, "Failed to enable job run queue event source, ignoring: %m");
2290 }
2291
manager_dispatch_dbus_queue(Manager * m)2292 static unsigned manager_dispatch_dbus_queue(Manager *m) {
2293 unsigned n = 0, budget;
2294 Unit *u;
2295 Job *j;
2296
2297 assert(m);
2298
2299 /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly
2300 * as we can. There's no point in throttling generation of signals in that case. */
2301 if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message)
2302 budget = UINT_MAX; /* infinite budget in this case */
2303 else {
2304 /* Anything to do at all? */
2305 if (!m->dbus_unit_queue && !m->dbus_job_queue)
2306 return 0;
2307
2308 /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's
2309 * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */
2310 if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD)
2311 return 0;
2312
2313 /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't
2314 * overly full before this call we shouldn't increase it in size too wildly in one step, and we
2315 * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of
2316 * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message,
2317 * regardless how many buses/direct connections it is enqueued on, while the "threshold" is applied to
2318 * each queued instance of bus message, i.e. if the same message is enqueued to five buses/direct
2319 * connections it will be counted five times. This difference in counting ("references"
2320 * vs. "instances") is primarily a result of the fact that it's easier to implement it this way,
2321 * however it also reflects the thinking that the "threshold" should put a limit on used queue memory,
2322 * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is
2323 * currently chosen much higher than the "budget". */
2324 budget = MANAGER_BUS_MESSAGE_BUDGET;
2325 }
2326
2327 while (budget != 0 && (u = m->dbus_unit_queue)) {
2328
2329 assert(u->in_dbus_queue);
2330
2331 bus_unit_send_change_signal(u);
2332 n++;
2333
2334 if (budget != UINT_MAX)
2335 budget--;
2336 }
2337
2338 while (budget != 0 && (j = m->dbus_job_queue)) {
2339 assert(j->in_dbus_queue);
2340
2341 bus_job_send_change_signal(j);
2342 n++;
2343
2344 if (budget != UINT_MAX)
2345 budget--;
2346 }
2347
2348 if (m->send_reloading_done) {
2349 m->send_reloading_done = false;
2350 bus_manager_send_reloading(m, false);
2351 n++;
2352 }
2353
2354 if (m->pending_reload_message) {
2355 bus_send_pending_reload_message(m);
2356 n++;
2357 }
2358
2359 return n;
2360 }
2361
manager_dispatch_cgroups_agent_fd(sd_event_source * source,int fd,uint32_t revents,void * userdata)2362 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2363 Manager *m = userdata;
2364 char buf[PATH_MAX];
2365 ssize_t n;
2366
2367 n = recv(fd, buf, sizeof(buf), 0);
2368 if (n < 0)
2369 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
2370 if (n == 0) {
2371 log_error("Got zero-length cgroups agent message, ignoring.");
2372 return 0;
2373 }
2374 if ((size_t) n >= sizeof(buf)) {
2375 log_error("Got overly long cgroups agent message, ignoring.");
2376 return 0;
2377 }
2378
2379 if (memchr(buf, 0, n)) {
2380 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
2381 return 0;
2382 }
2383 buf[n] = 0;
2384
2385 manager_notify_cgroup_empty(m, buf);
2386 (void) bus_forward_agent_released(m, buf);
2387
2388 return 0;
2389 }
2390
manager_process_barrier_fd(char * const * tags,FDSet * fds)2391 static bool manager_process_barrier_fd(char * const *tags, FDSet *fds) {
2392
2393 /* nothing else must be sent when using BARRIER=1 */
2394 if (strv_contains(tags, "BARRIER=1")) {
2395 if (strv_length(tags) == 1) {
2396 if (fdset_size(fds) != 1)
2397 log_warning("Got incorrect number of fds with BARRIER=1, closing them.");
2398 } else
2399 log_warning("Extra notification messages sent with BARRIER=1, ignoring everything.");
2400
2401 /* Drop the message if BARRIER=1 was found */
2402 return true;
2403 }
2404
2405 return false;
2406 }
2407
manager_invoke_notify_message(Manager * m,Unit * u,const struct ucred * ucred,char * const * tags,FDSet * fds)2408 static void manager_invoke_notify_message(
2409 Manager *m,
2410 Unit *u,
2411 const struct ucred *ucred,
2412 char * const *tags,
2413 FDSet *fds) {
2414
2415 assert(m);
2416 assert(u);
2417 assert(ucred);
2418 assert(tags);
2419
2420 if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */
2421 return;
2422 u->notifygen = m->notifygen;
2423
2424 if (UNIT_VTABLE(u)->notify_message)
2425 UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds);
2426
2427 else if (DEBUG_LOGGING) {
2428 _cleanup_free_ char *buf = NULL, *x = NULL, *y = NULL;
2429
2430 buf = strv_join(tags, ", ");
2431 if (buf)
2432 x = ellipsize(buf, 20, 90);
2433 if (x)
2434 y = cescape(x);
2435
2436 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
2437 }
2438 }
2439
manager_dispatch_notify_fd(sd_event_source * source,int fd,uint32_t revents,void * userdata)2440 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2441
2442 _cleanup_fdset_free_ FDSet *fds = NULL;
2443 Manager *m = userdata;
2444 char buf[NOTIFY_BUFFER_MAX+1];
2445 struct iovec iovec = {
2446 .iov_base = buf,
2447 .iov_len = sizeof(buf)-1,
2448 };
2449 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
2450 CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
2451 struct msghdr msghdr = {
2452 .msg_iov = &iovec,
2453 .msg_iovlen = 1,
2454 .msg_control = &control,
2455 .msg_controllen = sizeof(control),
2456 };
2457
2458 struct cmsghdr *cmsg;
2459 struct ucred *ucred = NULL;
2460 _cleanup_free_ Unit **array_copy = NULL;
2461 _cleanup_strv_free_ char **tags = NULL;
2462 Unit *u1, *u2, **array;
2463 int r, *fd_array = NULL;
2464 size_t n_fds = 0;
2465 bool found = false;
2466 ssize_t n;
2467
2468 assert(m);
2469 assert(m->notify_fd == fd);
2470
2471 if (revents != EPOLLIN) {
2472 log_warning("Got unexpected poll event for notify fd.");
2473 return 0;
2474 }
2475
2476 n = recvmsg_safe(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
2477 if (n < 0) {
2478 if (ERRNO_IS_TRANSIENT(n))
2479 return 0; /* Spurious wakeup, try again */
2480 if (n == -EXFULL) {
2481 log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
2482 return 0;
2483 }
2484 /* If this is any other, real error, then let's stop processing this socket. This of course
2485 * means we won't take notification messages anymore, but that's still better than busy
2486 * looping around this: being woken up over and over again but being unable to actually read
2487 * the message off the socket. */
2488 return log_error_errno(n, "Failed to receive notification message: %m");
2489 }
2490
2491 CMSG_FOREACH(cmsg, &msghdr) {
2492 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
2493
2494 assert(!fd_array);
2495 fd_array = (int*) CMSG_DATA(cmsg);
2496 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
2497
2498 } else if (cmsg->cmsg_level == SOL_SOCKET &&
2499 cmsg->cmsg_type == SCM_CREDENTIALS &&
2500 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
2501
2502 assert(!ucred);
2503 ucred = (struct ucred*) CMSG_DATA(cmsg);
2504 }
2505 }
2506
2507 if (n_fds > 0) {
2508 assert(fd_array);
2509
2510 r = fdset_new_array(&fds, fd_array, n_fds);
2511 if (r < 0) {
2512 close_many(fd_array, n_fds);
2513 log_oom();
2514 return 0;
2515 }
2516 }
2517
2518 if (!ucred || !pid_is_valid(ucred->pid)) {
2519 log_warning("Received notify message without valid credentials. Ignoring.");
2520 return 0;
2521 }
2522
2523 if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
2524 log_warning("Received notify message exceeded maximum size. Ignoring.");
2525 return 0;
2526 }
2527
2528 /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
2529 * trailing NUL byte in the message, but don't expect it. */
2530 if (n > 1 && memchr(buf, 0, n-1)) {
2531 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
2532 return 0;
2533 }
2534
2535 /* Make sure it's NUL-terminated, then parse it to obtain the tags list */
2536 buf[n] = 0;
2537 tags = strv_split_newlines(buf);
2538 if (!tags) {
2539 log_oom();
2540 return 0;
2541 }
2542
2543 /* possibly a barrier fd, let's see */
2544 if (manager_process_barrier_fd(tags, fds))
2545 return 0;
2546
2547 /* Increase the generation counter used for filtering out duplicate unit invocations. */
2548 m->notifygen++;
2549
2550 /* Notify every unit that might be interested, which might be multiple. */
2551 u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
2552 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid));
2553 array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid));
2554 if (array) {
2555 size_t k = 0;
2556
2557 while (array[k])
2558 k++;
2559
2560 array_copy = newdup(Unit*, array, k+1);
2561 if (!array_copy)
2562 log_oom();
2563 }
2564 /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units
2565 * make sure we only invoke each unit's handler once. */
2566 if (u1) {
2567 manager_invoke_notify_message(m, u1, ucred, tags, fds);
2568 found = true;
2569 }
2570 if (u2) {
2571 manager_invoke_notify_message(m, u2, ucred, tags, fds);
2572 found = true;
2573 }
2574 if (array_copy)
2575 for (size_t i = 0; array_copy[i]; i++) {
2576 manager_invoke_notify_message(m, array_copy[i], ucred, tags, fds);
2577 found = true;
2578 }
2579
2580 if (!found)
2581 log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid);
2582
2583 if (fdset_size(fds) > 0)
2584 log_warning("Got extra auxiliary fds with notification message, closing them.");
2585
2586 return 0;
2587 }
2588
manager_invoke_sigchld_event(Manager * m,Unit * u,const siginfo_t * si)2589 static void manager_invoke_sigchld_event(
2590 Manager *m,
2591 Unit *u,
2592 const siginfo_t *si) {
2593
2594 assert(m);
2595 assert(u);
2596 assert(si);
2597
2598 /* Already invoked the handler of this unit in this iteration? Then don't process this again */
2599 if (u->sigchldgen == m->sigchldgen)
2600 return;
2601 u->sigchldgen = m->sigchldgen;
2602
2603 log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id);
2604 unit_unwatch_pid(u, si->si_pid);
2605
2606 if (UNIT_VTABLE(u)->sigchld_event)
2607 UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
2608 }
2609
manager_dispatch_sigchld(sd_event_source * source,void * userdata)2610 static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) {
2611 Manager *m = userdata;
2612 siginfo_t si = {};
2613 int r;
2614
2615 assert(source);
2616 assert(m);
2617
2618 /* First we call waitid() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it
2619 * while it is a zombie. */
2620
2621 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
2622
2623 if (errno != ECHILD)
2624 log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m");
2625
2626 goto turn_off;
2627 }
2628
2629 if (si.si_pid <= 0)
2630 goto turn_off;
2631
2632 if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) {
2633 _cleanup_free_ Unit **array_copy = NULL;
2634 _cleanup_free_ char *name = NULL;
2635 Unit *u1, *u2, **array;
2636
2637 (void) get_process_comm(si.si_pid, &name);
2638
2639 log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
2640 si.si_pid, strna(name),
2641 sigchld_code_to_string(si.si_code),
2642 si.si_status,
2643 strna(si.si_code == CLD_EXITED
2644 ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
2645 : signal_to_string(si.si_status)));
2646
2647 /* Increase the generation counter used for filtering out duplicate unit invocations */
2648 m->sigchldgen++;
2649
2650 /* And now figure out the unit this belongs to, it might be multiple... */
2651 u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
2652 u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid));
2653 array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid));
2654 if (array) {
2655 size_t n = 0;
2656
2657 /* Count how many entries the array has */
2658 while (array[n])
2659 n++;
2660
2661 /* Make a copy of the array so that we don't trip up on the array changing beneath us */
2662 array_copy = newdup(Unit*, array, n+1);
2663 if (!array_copy)
2664 log_oom();
2665 }
2666
2667 /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but
2668 * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for
2669 * each iteration. */
2670 if (u1) {
2671 /* We check for oom condition, in case we got SIGCHLD before the oom notification.
2672 * We only do this for the cgroup the PID belonged to. */
2673 (void) unit_check_oom(u1);
2674
2675 /* We check if systemd-oomd performed a kill so that we log and notify appropriately */
2676 (void) unit_check_oomd_kill(u1);
2677
2678 manager_invoke_sigchld_event(m, u1, &si);
2679 }
2680 if (u2)
2681 manager_invoke_sigchld_event(m, u2, &si);
2682 if (array_copy)
2683 for (size_t i = 0; array_copy[i]; i++)
2684 manager_invoke_sigchld_event(m, array_copy[i], &si);
2685 }
2686
2687 /* And now, we actually reap the zombie. */
2688 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
2689 log_error_errno(errno, "Failed to dequeue child, ignoring: %m");
2690 return 0;
2691 }
2692
2693 return 0;
2694
2695 turn_off:
2696 /* All children processed for now, turn off event source */
2697
2698 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF);
2699 if (r < 0)
2700 return log_error_errno(r, "Failed to disable SIGCHLD event source: %m");
2701
2702 return 0;
2703 }
2704
manager_start_special(Manager * m,const char * name,JobMode mode)2705 static void manager_start_special(Manager *m, const char *name, JobMode mode) {
2706 Job *job;
2707
2708 if (manager_add_job_by_name_and_warn(m, JOB_START, name, mode, NULL, &job) < 0)
2709 return;
2710
2711 const char *s = unit_status_string(job->unit, NULL);
2712
2713 log_info("Activating special unit %s...", s);
2714
2715 sd_notifyf(false,
2716 "STATUS=Activating special unit %s...", s);
2717 m->status_ready = false;
2718 }
2719
manager_handle_ctrl_alt_del(Manager * m)2720 static void manager_handle_ctrl_alt_del(Manager *m) {
2721 /* If the user presses C-A-D more than
2722 * 7 times within 2s, we reboot/shutdown immediately,
2723 * unless it was disabled in system.conf */
2724
2725 if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
2726 manager_start_special(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
2727 else
2728 emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1,
2729 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
2730 }
2731
manager_dispatch_signal_fd(sd_event_source * source,int fd,uint32_t revents,void * userdata)2732 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2733 Manager *m = userdata;
2734 ssize_t n;
2735 struct signalfd_siginfo sfsi;
2736 int r;
2737
2738 assert(m);
2739 assert(m->signal_fd == fd);
2740
2741 if (revents != EPOLLIN) {
2742 log_warning("Got unexpected events from signal file descriptor.");
2743 return 0;
2744 }
2745
2746 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
2747 if (n < 0) {
2748 if (ERRNO_IS_TRANSIENT(errno))
2749 return 0;
2750
2751 /* We return an error here, which will kill this handler,
2752 * to avoid a busy loop on read error. */
2753 return log_error_errno(errno, "Reading from signal fd failed: %m");
2754 }
2755 if (n != sizeof(sfsi)) {
2756 log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n);
2757 return 0;
2758 }
2759
2760 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
2761 (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
2762 ? LOG_DEBUG : LOG_INFO,
2763 &sfsi);
2764
2765 switch (sfsi.ssi_signo) {
2766
2767 case SIGCHLD:
2768 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
2769 if (r < 0)
2770 log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m");
2771
2772 break;
2773
2774 case SIGTERM:
2775 if (MANAGER_IS_SYSTEM(m)) {
2776 /* This is for compatibility with the original sysvinit */
2777 if (verify_run_space_and_log("Refusing to reexecute") < 0)
2778 break;
2779
2780 m->objective = MANAGER_REEXECUTE;
2781 break;
2782 }
2783
2784 _fallthrough_;
2785 case SIGINT:
2786 if (MANAGER_IS_SYSTEM(m))
2787 manager_handle_ctrl_alt_del(m);
2788 else
2789 manager_start_special(m, SPECIAL_EXIT_TARGET, JOB_REPLACE_IRREVERSIBLY);
2790 break;
2791
2792 case SIGWINCH:
2793 /* This is a nop on non-init */
2794 if (MANAGER_IS_SYSTEM(m))
2795 manager_start_special(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2796
2797 break;
2798
2799 case SIGPWR:
2800 /* This is a nop on non-init */
2801 if (MANAGER_IS_SYSTEM(m))
2802 manager_start_special(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2803
2804 break;
2805
2806 case SIGUSR1:
2807 if (manager_dbus_is_running(m, false)) {
2808 log_info("Trying to reconnect to bus...");
2809
2810 (void) bus_init_api(m);
2811
2812 if (MANAGER_IS_SYSTEM(m))
2813 (void) bus_init_system(m);
2814 } else
2815 manager_start_special(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2816
2817 break;
2818
2819 case SIGUSR2: {
2820 _cleanup_free_ char *dump = NULL;
2821
2822 r = manager_get_dump_string(m, &dump);
2823 if (r < 0) {
2824 log_warning_errno(errno, "Failed to acquire manager dump: %m");
2825 break;
2826 }
2827
2828 log_dump(LOG_INFO, dump);
2829 break;
2830 }
2831
2832 case SIGHUP:
2833 if (verify_run_space_and_log("Refusing to reload") < 0)
2834 break;
2835
2836 m->objective = MANAGER_RELOAD;
2837 break;
2838
2839 default: {
2840
2841 /* Starting SIGRTMIN+0 */
2842 static const struct {
2843 const char *target;
2844 JobMode mode;
2845 } target_table[] = {
2846 [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE },
2847 [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE },
2848 [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE },
2849 [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2850 [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY },
2851 [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY },
2852 [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY },
2853 };
2854
2855 /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2856 static const ManagerObjective objective_table[] = {
2857 [0] = MANAGER_HALT,
2858 [1] = MANAGER_POWEROFF,
2859 [2] = MANAGER_REBOOT,
2860 [3] = MANAGER_KEXEC,
2861 };
2862
2863 if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2864 (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2865 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2866 manager_start_special(m, target_table[idx].target, target_table[idx].mode);
2867 break;
2868 }
2869
2870 if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2871 (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) {
2872 m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13];
2873 break;
2874 }
2875
2876 switch (sfsi.ssi_signo - SIGRTMIN) {
2877
2878 case 20:
2879 manager_override_show_status(m, SHOW_STATUS_YES, "signal");
2880 break;
2881
2882 case 21:
2883 manager_override_show_status(m, SHOW_STATUS_NO, "signal");
2884 break;
2885
2886 case 22:
2887 manager_override_log_level(m, LOG_DEBUG);
2888 break;
2889
2890 case 23:
2891 manager_restore_original_log_level(m);
2892 break;
2893
2894 case 24:
2895 if (MANAGER_IS_USER(m)) {
2896 m->objective = MANAGER_EXIT;
2897 return 0;
2898 }
2899
2900 /* This is a nop on init */
2901 break;
2902
2903 case 25:
2904 m->objective = MANAGER_REEXECUTE;
2905 break;
2906
2907 case 26:
2908 case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2909 manager_restore_original_log_target(m);
2910 break;
2911
2912 case 27:
2913 manager_override_log_target(m, LOG_TARGET_CONSOLE);
2914 break;
2915
2916 case 28:
2917 manager_override_log_target(m, LOG_TARGET_KMSG);
2918 break;
2919
2920 default:
2921 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2922 }
2923 }}
2924
2925 return 0;
2926 }
2927
manager_dispatch_time_change_fd(sd_event_source * source,int fd,uint32_t revents,void * userdata)2928 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2929 Manager *m = userdata;
2930 Unit *u;
2931
2932 assert(m);
2933
2934 log_struct(LOG_DEBUG,
2935 "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR,
2936 LOG_MESSAGE("Time has been changed"));
2937
2938 /* Restart the watch */
2939 (void) manager_setup_time_change(m);
2940
2941 HASHMAP_FOREACH(u, m->units)
2942 if (UNIT_VTABLE(u)->time_change)
2943 UNIT_VTABLE(u)->time_change(u);
2944
2945 return 0;
2946 }
2947
manager_dispatch_timezone_change(sd_event_source * source,const struct inotify_event * e,void * userdata)2948 static int manager_dispatch_timezone_change(
2949 sd_event_source *source,
2950 const struct inotify_event *e,
2951 void *userdata) {
2952
2953 Manager *m = userdata;
2954 int changed;
2955 Unit *u;
2956
2957 assert(m);
2958
2959 log_debug("inotify event for /etc/localtime");
2960
2961 changed = manager_read_timezone_stat(m);
2962 if (changed <= 0)
2963 return changed;
2964
2965 /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */
2966 (void) manager_setup_timezone_change(m);
2967
2968 /* Read the new timezone */
2969 tzset();
2970
2971 log_debug("Timezone has been changed (now: %s).", tzname[daylight]);
2972
2973 HASHMAP_FOREACH(u, m->units)
2974 if (UNIT_VTABLE(u)->timezone_change)
2975 UNIT_VTABLE(u)->timezone_change(u);
2976
2977 return 0;
2978 }
2979
manager_dispatch_idle_pipe_fd(sd_event_source * source,int fd,uint32_t revents,void * userdata)2980 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2981 Manager *m = userdata;
2982
2983 assert(m);
2984 assert(m->idle_pipe[2] == fd);
2985
2986 /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's
2987 * now turn off any further console output if there's at least one service that needs console access, so that
2988 * from now on our own output should not spill into that service's output anymore. After all, we support
2989 * Type=idle only to beautify console output and it generally is set on services that want to own the console
2990 * exclusively without our interference. */
2991 m->no_console_output = m->n_on_console > 0;
2992
2993 /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer
2994 * by closing the pipes towards them, which is what they are waiting for. */
2995 manager_close_idle_pipe(m);
2996
2997 return 0;
2998 }
2999
manager_dispatch_jobs_in_progress(sd_event_source * source,usec_t usec,void * userdata)3000 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
3001 Manager *m = userdata;
3002 int r;
3003
3004 assert(m);
3005 assert(source);
3006
3007 manager_print_jobs_in_progress(m);
3008
3009 r = sd_event_source_set_time_relative(source, JOBS_IN_PROGRESS_PERIOD_USEC);
3010 if (r < 0)
3011 return r;
3012
3013 return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
3014 }
3015
manager_loop(Manager * m)3016 int manager_loop(Manager *m) {
3017 RateLimit rl = { .interval = 1*USEC_PER_SEC, .burst = 50000 };
3018 int r;
3019
3020 assert(m);
3021 assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */
3022
3023 manager_check_finished(m);
3024
3025 /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */
3026 r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON);
3027 if (r < 0)
3028 return log_error_errno(r, "Failed to enable SIGCHLD event source: %m");
3029
3030 while (m->objective == MANAGER_OK) {
3031
3032 (void) watchdog_ping();
3033
3034 if (!ratelimit_below(&rl)) {
3035 /* Yay, something is going seriously wrong, pause a little */
3036 log_warning("Looping too fast. Throttling execution a little.");
3037 sleep(1);
3038 }
3039
3040 if (manager_dispatch_load_queue(m) > 0)
3041 continue;
3042
3043 if (manager_dispatch_gc_job_queue(m) > 0)
3044 continue;
3045
3046 if (manager_dispatch_gc_unit_queue(m) > 0)
3047 continue;
3048
3049 if (manager_dispatch_cleanup_queue(m) > 0)
3050 continue;
3051
3052 if (manager_dispatch_cgroup_realize_queue(m) > 0)
3053 continue;
3054
3055 if (manager_dispatch_start_when_upheld_queue(m) > 0)
3056 continue;
3057
3058 if (manager_dispatch_stop_when_bound_queue(m) > 0)
3059 continue;
3060
3061 if (manager_dispatch_stop_when_unneeded_queue(m) > 0)
3062 continue;
3063
3064 if (manager_dispatch_dbus_queue(m) > 0)
3065 continue;
3066
3067 /* Sleep for watchdog runtime wait time */
3068 r = sd_event_run(m->event, watchdog_runtime_wait());
3069 if (r < 0)
3070 return log_error_errno(r, "Failed to run event loop: %m");
3071 }
3072
3073 return m->objective;
3074 }
3075
manager_load_unit_from_dbus_path(Manager * m,const char * s,sd_bus_error * e,Unit ** _u)3076 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
3077 _cleanup_free_ char *n = NULL;
3078 sd_id128_t invocation_id;
3079 Unit *u;
3080 int r;
3081
3082 assert(m);
3083 assert(s);
3084 assert(_u);
3085
3086 r = unit_name_from_dbus_path(s, &n);
3087 if (r < 0)
3088 return r;
3089
3090 /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
3091 * as invocation ID. */
3092 r = sd_id128_from_string(n, &invocation_id);
3093 if (r >= 0) {
3094 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
3095 if (u) {
3096 *_u = u;
3097 return 0;
3098 }
3099
3100 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID,
3101 "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.",
3102 SD_ID128_FORMAT_VAL(invocation_id));
3103 }
3104
3105 /* If this didn't work, we check if this is a unit name */
3106 if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
3107 _cleanup_free_ char *nn = NULL;
3108
3109 nn = cescape(n);
3110 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS,
3111 "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn));
3112 }
3113
3114 r = manager_load_unit(m, n, NULL, e, &u);
3115 if (r < 0)
3116 return r;
3117
3118 *_u = u;
3119 return 0;
3120 }
3121
manager_get_job_from_dbus_path(Manager * m,const char * s,Job ** _j)3122 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
3123 const char *p;
3124 unsigned id;
3125 Job *j;
3126 int r;
3127
3128 assert(m);
3129 assert(s);
3130 assert(_j);
3131
3132 p = startswith(s, "/org/freedesktop/systemd1/job/");
3133 if (!p)
3134 return -EINVAL;
3135
3136 r = safe_atou(p, &id);
3137 if (r < 0)
3138 return r;
3139
3140 j = manager_get_job(m, id);
3141 if (!j)
3142 return -ENOENT;
3143
3144 *_j = j;
3145
3146 return 0;
3147 }
3148
manager_send_unit_audit(Manager * m,Unit * u,int type,bool success)3149 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
3150
3151 #if HAVE_AUDIT
3152 _cleanup_free_ char *p = NULL;
3153 const char *msg;
3154 int audit_fd, r;
3155
3156 if (!MANAGER_IS_SYSTEM(m))
3157 return;
3158
3159 audit_fd = get_audit_fd();
3160 if (audit_fd < 0)
3161 return;
3162
3163 /* Don't generate audit events if the service was already
3164 * started and we're just deserializing */
3165 if (MANAGER_IS_RELOADING(m))
3166 return;
3167
3168 if (u->type != UNIT_SERVICE)
3169 return;
3170
3171 r = unit_name_to_prefix_and_instance(u->id, &p);
3172 if (r < 0) {
3173 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
3174 return;
3175 }
3176
3177 msg = strjoina("unit=", p);
3178 if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
3179 if (errno == EPERM)
3180 /* We aren't allowed to send audit messages?
3181 * Then let's not retry again. */
3182 close_audit_fd();
3183 else
3184 log_warning_errno(errno, "Failed to send audit message: %m");
3185 }
3186 #endif
3187
3188 }
3189
manager_send_unit_plymouth(Manager * m,Unit * u)3190 void manager_send_unit_plymouth(Manager *m, Unit *u) {
3191 static const union sockaddr_union sa = PLYMOUTH_SOCKET;
3192 _cleanup_free_ char *message = NULL;
3193 _cleanup_close_ int fd = -1;
3194 int n = 0;
3195
3196 /* Don't generate plymouth events if the service was already
3197 * started and we're just deserializing */
3198 if (MANAGER_IS_RELOADING(m))
3199 return;
3200
3201 if (!MANAGER_IS_SYSTEM(m))
3202 return;
3203
3204 if (detect_container() > 0)
3205 return;
3206
3207 if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP))
3208 return;
3209
3210 /* We set SOCK_NONBLOCK here so that we rather drop the
3211 * message then wait for plymouth */
3212 fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
3213 if (fd < 0) {
3214 log_error_errno(errno, "socket() failed: %m");
3215 return;
3216 }
3217
3218 if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
3219 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3220 log_error_errno(errno, "connect() failed: %m");
3221 return;
3222 }
3223
3224 if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0)
3225 return (void) log_oom();
3226
3227 errno = 0;
3228 if (write(fd, message, n + 1) != n + 1)
3229 if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno))
3230 log_error_errno(errno, "Failed to write Plymouth message: %m");
3231 }
3232
manager_get_watchdog(Manager * m,WatchdogType t)3233 usec_t manager_get_watchdog(Manager *m, WatchdogType t) {
3234 assert(m);
3235
3236 if (MANAGER_IS_USER(m))
3237 return USEC_INFINITY;
3238
3239 if (timestamp_is_set(m->watchdog_overridden[t]))
3240 return m->watchdog_overridden[t];
3241
3242 return m->watchdog[t];
3243 }
3244
manager_set_watchdog(Manager * m,WatchdogType t,usec_t timeout)3245 void manager_set_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3246
3247 assert(m);
3248
3249 if (MANAGER_IS_USER(m))
3250 return;
3251
3252 if (m->watchdog[t] == timeout)
3253 return;
3254
3255 if (t == WATCHDOG_RUNTIME) {
3256 if (!timestamp_is_set(m->watchdog_overridden[WATCHDOG_RUNTIME]))
3257 (void) watchdog_setup(timeout);
3258 } else if (t == WATCHDOG_PRETIMEOUT)
3259 if (m->watchdog_overridden[WATCHDOG_PRETIMEOUT] == USEC_INFINITY)
3260 (void) watchdog_setup_pretimeout(timeout);
3261
3262 m->watchdog[t] = timeout;
3263 }
3264
manager_override_watchdog(Manager * m,WatchdogType t,usec_t timeout)3265 void manager_override_watchdog(Manager *m, WatchdogType t, usec_t timeout) {
3266
3267 assert(m);
3268
3269 if (MANAGER_IS_USER(m))
3270 return;
3271
3272 if (m->watchdog_overridden[t] == timeout)
3273 return;
3274
3275 if (t == WATCHDOG_RUNTIME) {
3276 usec_t usec = timestamp_is_set(timeout) ? timeout : m->watchdog[t];
3277
3278 (void) watchdog_setup(usec);
3279 } else if (t == WATCHDOG_PRETIMEOUT)
3280 (void) watchdog_setup_pretimeout(timeout);
3281
3282 m->watchdog_overridden[t] = timeout;
3283 }
3284
manager_set_watchdog_pretimeout_governor(Manager * m,const char * governor)3285 int manager_set_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3286 _cleanup_free_ char *p = NULL;
3287 int r;
3288
3289 assert(m);
3290
3291 if (MANAGER_IS_USER(m))
3292 return 0;
3293
3294 if (streq_ptr(m->watchdog_pretimeout_governor, governor))
3295 return 0;
3296
3297 p = strdup(governor);
3298 if (!p)
3299 return -ENOMEM;
3300
3301 r = watchdog_setup_pretimeout_governor(governor);
3302 if (r < 0)
3303 return r;
3304
3305 return free_and_replace(m->watchdog_pretimeout_governor, p);
3306 }
3307
manager_override_watchdog_pretimeout_governor(Manager * m,const char * governor)3308 int manager_override_watchdog_pretimeout_governor(Manager *m, const char *governor) {
3309 _cleanup_free_ char *p = NULL;
3310 int r;
3311
3312 assert(m);
3313
3314 if (MANAGER_IS_USER(m))
3315 return 0;
3316
3317 if (streq_ptr(m->watchdog_pretimeout_governor_overridden, governor))
3318 return 0;
3319
3320 p = strdup(governor);
3321 if (!p)
3322 return -ENOMEM;
3323
3324 r = watchdog_setup_pretimeout_governor(governor);
3325 if (r < 0)
3326 return r;
3327
3328 return free_and_replace(m->watchdog_pretimeout_governor_overridden, p);
3329 }
3330
manager_reload(Manager * m)3331 int manager_reload(Manager *m) {
3332 _unused_ _cleanup_(manager_reloading_stopp) Manager *reloading = NULL;
3333 _cleanup_fdset_free_ FDSet *fds = NULL;
3334 _cleanup_fclose_ FILE *f = NULL;
3335 int r;
3336
3337 assert(m);
3338
3339 r = manager_open_serialization(m, &f);
3340 if (r < 0)
3341 return log_error_errno(r, "Failed to create serialization file: %m");
3342
3343 fds = fdset_new();
3344 if (!fds)
3345 return log_oom();
3346
3347 /* We are officially in reload mode from here on. */
3348 reloading = manager_reloading_start(m);
3349
3350 r = manager_serialize(m, f, fds, false);
3351 if (r < 0)
3352 return r;
3353
3354 if (fseeko(f, 0, SEEK_SET) < 0)
3355 return log_error_errno(errno, "Failed to seek to beginning of serialization: %m");
3356
3357 /* �� This is the point of no return, from here on there is no way back. �� */
3358 reloading = NULL;
3359
3360 bus_manager_send_reloading(m, true);
3361
3362 /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users
3363 * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need
3364 * it. */
3365
3366 manager_clear_jobs_and_units(m);
3367 lookup_paths_flush_generator(&m->lookup_paths);
3368 lookup_paths_free(&m->lookup_paths);
3369 exec_runtime_vacuum(m);
3370 dynamic_user_vacuum(m, false);
3371 m->uid_refs = hashmap_free(m->uid_refs);
3372 m->gid_refs = hashmap_free(m->gid_refs);
3373
3374 r = lookup_paths_init_or_warn(&m->lookup_paths, m->unit_file_scope, 0, NULL);
3375 if (r < 0)
3376 return r;
3377
3378 (void) manager_run_environment_generators(m);
3379 (void) manager_run_generators(m);
3380
3381 lookup_paths_log(&m->lookup_paths);
3382
3383 /* We flushed out generated files, for which we don't watch mtime, so we should flush the old map. */
3384 manager_free_unit_name_maps(m);
3385
3386 /* First, enumerate what we can from kernel and suchlike */
3387 manager_enumerate_perpetual(m);
3388 manager_enumerate(m);
3389
3390 /* Second, deserialize our stored data */
3391 r = manager_deserialize(m, f, fds);
3392 if (r < 0)
3393 log_warning_errno(r, "Deserialization failed, proceeding anyway: %m");
3394
3395 /* We don't need the serialization anymore */
3396 f = safe_fclose(f);
3397
3398 /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */
3399 (void) manager_setup_notify(m);
3400 (void) manager_setup_cgroups_agent(m);
3401 (void) manager_setup_user_lookup_fd(m);
3402
3403 /* Third, fire things up! */
3404 manager_coldplug(m);
3405
3406 /* Clean up runtime objects no longer referenced */
3407 manager_vacuum(m);
3408
3409 /* Clean up deserialized tracked clients */
3410 m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
3411
3412 /* Consider the reload process complete now. */
3413 assert(m->n_reloading > 0);
3414 m->n_reloading--;
3415
3416 /* On manager reloading, device tag data should exists, thus, we should honor the results of device
3417 * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So,
3418 * let's always set the flag here for safety. */
3419 m->honor_device_enumeration = true;
3420
3421 manager_ready(m);
3422
3423 m->send_reloading_done = true;
3424 return 0;
3425 }
3426
manager_reset_failed(Manager * m)3427 void manager_reset_failed(Manager *m) {
3428 Unit *u;
3429
3430 assert(m);
3431
3432 HASHMAP_FOREACH(u, m->units)
3433 unit_reset_failed(u);
3434 }
3435
manager_unit_inactive_or_pending(Manager * m,const char * name)3436 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
3437 Unit *u;
3438
3439 assert(m);
3440 assert(name);
3441
3442 /* Returns true if the unit is inactive or going down */
3443 u = manager_get_unit(m, name);
3444 if (!u)
3445 return true;
3446
3447 return unit_inactive_or_pending(u);
3448 }
3449
log_taint_string(Manager * m)3450 static void log_taint_string(Manager *m) {
3451 _cleanup_free_ char *taint = NULL;
3452
3453 assert(m);
3454
3455 if (MANAGER_IS_USER(m) || m->taint_logged)
3456 return;
3457
3458 m->taint_logged = true; /* only check for taint once */
3459
3460 taint = manager_taint_string(m);
3461 if (isempty(taint))
3462 return;
3463
3464 log_struct(LOG_NOTICE,
3465 LOG_MESSAGE("System is tainted: %s", taint),
3466 "TAINT=%s", taint,
3467 "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR);
3468 }
3469
manager_notify_finished(Manager * m)3470 static void manager_notify_finished(Manager *m) {
3471 usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
3472
3473 if (MANAGER_IS_TEST_RUN(m))
3474 return;
3475
3476 if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
3477 char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")]
3478 = {};
3479 char *p = buf;
3480 size_t size = sizeof buf;
3481
3482 /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and
3483 * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered
3484 * negative values. */
3485
3486 firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic;
3487 loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3488 userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3489 total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic;
3490
3491 if (firmware_usec > 0)
3492 size = strpcpyf(&p, size, "%s (firmware) + ", FORMAT_TIMESPAN(firmware_usec, USEC_PER_MSEC));
3493 if (loader_usec > 0)
3494 size = strpcpyf(&p, size, "%s (loader) + ", FORMAT_TIMESPAN(loader_usec, USEC_PER_MSEC));
3495
3496 if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) {
3497
3498 /* The initrd case on bare-metal */
3499 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3500 initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic;
3501
3502 log_struct(LOG_INFO,
3503 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3504 "KERNEL_USEC="USEC_FMT, kernel_usec,
3505 "INITRD_USEC="USEC_FMT, initrd_usec,
3506 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3507 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.",
3508 buf,
3509 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3510 FORMAT_TIMESPAN(initrd_usec, USEC_PER_MSEC),
3511 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3512 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3513 } else {
3514 /* The initrd-less case on bare-metal */
3515
3516 kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic;
3517 initrd_usec = 0;
3518
3519 log_struct(LOG_INFO,
3520 "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR,
3521 "KERNEL_USEC="USEC_FMT, kernel_usec,
3522 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3523 LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.",
3524 buf,
3525 FORMAT_TIMESPAN(kernel_usec, USEC_PER_MSEC),
3526 FORMAT_TIMESPAN(userspace_usec, USEC_PER_MSEC),
3527 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3528 }
3529 } else {
3530 /* The container and --user case */
3531 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
3532 total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic;
3533
3534 log_struct(LOG_INFO,
3535 "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR,
3536 "USERSPACE_USEC="USEC_FMT, userspace_usec,
3537 LOG_MESSAGE("Startup finished in %s.",
3538 FORMAT_TIMESPAN(total_usec, USEC_PER_MSEC)));
3539 }
3540
3541 bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
3542
3543 log_taint_string(m);
3544 }
3545
user_manager_send_ready(Manager * m)3546 static void user_manager_send_ready(Manager *m) {
3547 int r;
3548
3549 assert(m);
3550
3551 /* We send READY=1 on reaching basic.target only when running in --user mode. */
3552 if (!MANAGER_IS_USER(m) || m->ready_sent)
3553 return;
3554
3555 r = sd_notify(false,
3556 "READY=1\n"
3557 "STATUS=Reached " SPECIAL_BASIC_TARGET ".");
3558 if (r < 0)
3559 log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
3560
3561 m->ready_sent = true;
3562 m->status_ready = false;
3563 }
3564
manager_send_ready(Manager * m)3565 static void manager_send_ready(Manager *m) {
3566 int r;
3567
3568 if (m->ready_sent && m->status_ready)
3569 /* Skip the notification if nothing changed. */
3570 return;
3571
3572 r = sd_notify(false,
3573 "READY=1\n"
3574 "STATUS=Ready.");
3575 if (r < 0)
3576 log_full_errno(m->ready_sent ? LOG_DEBUG : LOG_WARNING, r,
3577 "Failed to send readiness notification, ignoring: %m");
3578
3579 m->ready_sent = m->status_ready = true;
3580 }
3581
manager_check_basic_target(Manager * m)3582 static void manager_check_basic_target(Manager *m) {
3583 Unit *u;
3584
3585 assert(m);
3586
3587 /* Small shortcut */
3588 if (m->ready_sent && m->taint_logged)
3589 return;
3590
3591 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3592 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3593 return;
3594
3595 /* For user managers, send out READY=1 as soon as we reach basic.target */
3596 user_manager_send_ready(m);
3597
3598 /* Log the taint string as soon as we reach basic.target */
3599 log_taint_string(m);
3600 }
3601
manager_check_finished(Manager * m)3602 void manager_check_finished(Manager *m) {
3603 assert(m);
3604
3605 if (MANAGER_IS_RELOADING(m))
3606 return;
3607
3608 /* Verify that we have entered the event loop already, and not left it again. */
3609 if (!MANAGER_IS_RUNNING(m))
3610 return;
3611
3612 manager_check_basic_target(m);
3613
3614 if (hashmap_size(m->jobs) > 0) {
3615 if (m->jobs_in_progress_event_source)
3616 /* Ignore any failure, this is only for feedback */
3617 (void) sd_event_source_set_time(m->jobs_in_progress_event_source,
3618 manager_watch_jobs_next_time(m));
3619 return;
3620 }
3621
3622 /* The jobs hashmap tends to grow a lot during boot, and then it's not reused until shutdown. Let's
3623 kill the hashmap if it is relatively large. */
3624 if (hashmap_buckets(m->jobs) > hashmap_size(m->units) / 10)
3625 m->jobs = hashmap_free(m->jobs);
3626
3627 manager_send_ready(m);
3628
3629 /* Notify Type=idle units that we are done now */
3630 manager_close_idle_pipe(m);
3631
3632 if (MANAGER_IS_FINISHED(m))
3633 return;
3634
3635 manager_flip_auto_status(m, false, "boot finished");
3636
3637 /* Turn off confirm spawn now */
3638 m->confirm_spawn = NULL;
3639
3640 /* No need to update ask password status when we're going non-interactive */
3641 manager_close_ask_password(m);
3642
3643 /* This is no longer the first boot */
3644 manager_set_first_boot(m, false);
3645
3646 dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH);
3647
3648 manager_notify_finished(m);
3649
3650 manager_invalidate_startup_units(m);
3651 }
3652
generator_path_any(const char * const * paths)3653 static bool generator_path_any(const char* const* paths) {
3654 bool found = false;
3655
3656 /* Optimize by skipping the whole process by not creating output directories
3657 * if no generators are found. */
3658 STRV_FOREACH(path, paths)
3659 if (access(*path, F_OK) == 0)
3660 found = true;
3661 else if (errno != ENOENT)
3662 log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3663
3664 return found;
3665 }
3666
manager_run_environment_generators(Manager * m)3667 static int manager_run_environment_generators(Manager *m) {
3668 char **tmp = NULL; /* this is only used in the forked process, no cleanup here */
3669 _cleanup_strv_free_ char **paths = NULL;
3670 void* args[] = {
3671 [STDOUT_GENERATE] = &tmp,
3672 [STDOUT_COLLECT] = &tmp,
3673 [STDOUT_CONSUME] = &m->transient_environment,
3674 };
3675 int r;
3676
3677 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS))
3678 return 0;
3679
3680 paths = env_generator_binary_paths(MANAGER_IS_SYSTEM(m));
3681 if (!paths)
3682 return log_oom();
3683
3684 if (!generator_path_any((const char* const*) paths))
3685 return 0;
3686
3687 RUN_WITH_UMASK(0022)
3688 r = execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, gather_environment,
3689 args, NULL, m->transient_environment,
3690 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3691 return r;
3692 }
3693
build_generator_environment(Manager * m,char *** ret)3694 static int build_generator_environment(Manager *m, char ***ret) {
3695 _cleanup_strv_free_ char **nl = NULL;
3696 Virtualization v;
3697 int r;
3698
3699 assert(m);
3700 assert(ret);
3701
3702 /* Generators oftentimes want to know some basic facts about the environment they run in, in order to
3703 * adjust generated units to that. Let's pass down some bits of information that are easy for us to
3704 * determine (but a bit harder for generator scripts to determine), as environment variables. */
3705
3706 nl = strv_copy(m->transient_environment);
3707 if (!nl)
3708 return -ENOMEM;
3709
3710 r = strv_env_assign(&nl, "SYSTEMD_SCOPE", MANAGER_IS_SYSTEM(m) ? "system" : "user");
3711 if (r < 0)
3712 return r;
3713
3714 if (MANAGER_IS_SYSTEM(m)) {
3715 /* Note that $SYSTEMD_IN_INITRD may be used to override the initrd detection in much of our
3716 * codebase. This is hence more than purely informational. It will shortcut detection of the
3717 * initrd state if generators invoke our own tools. But that's OK, as it would come to the
3718 * same results (hopefully). */
3719 r = strv_env_assign(&nl, "SYSTEMD_IN_INITRD", one_zero(in_initrd()));
3720 if (r < 0)
3721 return r;
3722
3723 if (m->first_boot >= 0) {
3724 r = strv_env_assign(&nl, "SYSTEMD_FIRST_BOOT", one_zero(m->first_boot));
3725 if (r < 0)
3726 return r;
3727 }
3728 }
3729
3730 v = detect_virtualization();
3731 if (v < 0)
3732 log_debug_errno(v, "Failed to detect virtualization, ignoring: %m");
3733 else if (v > 0) {
3734 const char *s;
3735
3736 s = strjoina(VIRTUALIZATION_IS_VM(v) ? "vm:" :
3737 VIRTUALIZATION_IS_CONTAINER(v) ? "container:" : ":",
3738 virtualization_to_string(v));
3739
3740 r = strv_env_assign(&nl, "SYSTEMD_VIRTUALIZATION", s);
3741 if (r < 0)
3742 return r;
3743 }
3744
3745 r = strv_env_assign(&nl, "SYSTEMD_ARCHITECTURE", architecture_to_string(uname_architecture()));
3746 if (r < 0)
3747 return r;
3748
3749 *ret = TAKE_PTR(nl);
3750 return 0;
3751 }
3752
manager_run_generators(Manager * m)3753 static int manager_run_generators(Manager *m) {
3754 _cleanup_strv_free_ char **paths = NULL, **ge = NULL;
3755 int r;
3756
3757 assert(m);
3758
3759 if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS))
3760 return 0;
3761
3762 paths = generator_binary_paths(m->unit_file_scope);
3763 if (!paths)
3764 return log_oom();
3765
3766 if (!generator_path_any((const char* const*) paths))
3767 return 0;
3768
3769 r = lookup_paths_mkdir_generator(&m->lookup_paths);
3770 if (r < 0) {
3771 log_error_errno(r, "Failed to create generator directories: %m");
3772 goto finish;
3773 }
3774
3775 const char *argv[] = {
3776 NULL, /* Leave this empty, execute_directory() will fill something in */
3777 m->lookup_paths.generator,
3778 m->lookup_paths.generator_early,
3779 m->lookup_paths.generator_late,
3780 NULL,
3781 };
3782
3783 r = build_generator_environment(m, &ge);
3784 if (r < 0) {
3785 log_error_errno(r, "Failed to build generator environment: %m");
3786 goto finish;
3787 }
3788
3789 RUN_WITH_UMASK(0022)
3790 (void) execute_directories(
3791 (const char* const*) paths,
3792 DEFAULT_TIMEOUT_USEC,
3793 /* callbacks= */ NULL, /* callback_args= */ NULL,
3794 (char**) argv,
3795 ge,
3796 EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS | EXEC_DIR_SET_SYSTEMD_EXEC_PID);
3797
3798 r = 0;
3799
3800 finish:
3801 lookup_paths_trim_generator(&m->lookup_paths);
3802 return r;
3803 }
3804
manager_transient_environment_add(Manager * m,char ** plus)3805 int manager_transient_environment_add(Manager *m, char **plus) {
3806 char **a;
3807
3808 assert(m);
3809
3810 if (strv_isempty(plus))
3811 return 0;
3812
3813 a = strv_env_merge(m->transient_environment, plus);
3814 if (!a)
3815 return log_oom();
3816
3817 sanitize_environment(a);
3818
3819 return strv_free_and_replace(m->transient_environment, a);
3820 }
3821
manager_client_environment_modify(Manager * m,char ** minus,char ** plus)3822 int manager_client_environment_modify(
3823 Manager *m,
3824 char **minus,
3825 char **plus) {
3826
3827 char **a = NULL, **b = NULL, **l;
3828
3829 assert(m);
3830
3831 if (strv_isempty(minus) && strv_isempty(plus))
3832 return 0;
3833
3834 l = m->client_environment;
3835
3836 if (!strv_isempty(minus)) {
3837 a = strv_env_delete(l, 1, minus);
3838 if (!a)
3839 return -ENOMEM;
3840
3841 l = a;
3842 }
3843
3844 if (!strv_isempty(plus)) {
3845 b = strv_env_merge(l, plus);
3846 if (!b) {
3847 strv_free(a);
3848 return -ENOMEM;
3849 }
3850
3851 l = b;
3852 }
3853
3854 if (m->client_environment != l)
3855 strv_free(m->client_environment);
3856
3857 if (a != l)
3858 strv_free(a);
3859 if (b != l)
3860 strv_free(b);
3861
3862 m->client_environment = sanitize_environment(l);
3863 return 0;
3864 }
3865
manager_get_effective_environment(Manager * m,char *** ret)3866 int manager_get_effective_environment(Manager *m, char ***ret) {
3867 char **l;
3868
3869 assert(m);
3870 assert(ret);
3871
3872 l = strv_env_merge(m->transient_environment, m->client_environment);
3873 if (!l)
3874 return -ENOMEM;
3875
3876 *ret = l;
3877 return 0;
3878 }
3879
manager_set_default_rlimits(Manager * m,struct rlimit ** default_rlimit)3880 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3881 assert(m);
3882
3883 for (unsigned i = 0; i < _RLIMIT_MAX; i++) {
3884 m->rlimit[i] = mfree(m->rlimit[i]);
3885
3886 if (!default_rlimit[i])
3887 continue;
3888
3889 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3890 if (!m->rlimit[i])
3891 return log_oom();
3892 }
3893
3894 return 0;
3895 }
3896
manager_recheck_dbus(Manager * m)3897 void manager_recheck_dbus(Manager *m) {
3898 assert(m);
3899
3900 /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all
3901 * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the
3902 * connection of the API bus). That's because the system bus after all runs as service of the system instance,
3903 * while in the user instance we can assume it's already there. */
3904
3905 if (MANAGER_IS_RELOADING(m))
3906 return; /* don't check while we are reloading… */
3907
3908 if (manager_dbus_is_running(m, false)) {
3909 (void) bus_init_api(m);
3910
3911 if (MANAGER_IS_SYSTEM(m))
3912 (void) bus_init_system(m);
3913 } else {
3914 (void) bus_done_api(m);
3915
3916 if (MANAGER_IS_SYSTEM(m))
3917 (void) bus_done_system(m);
3918 }
3919 }
3920
manager_journal_is_running(Manager * m)3921 static bool manager_journal_is_running(Manager *m) {
3922 Unit *u;
3923
3924 assert(m);
3925
3926 if (MANAGER_IS_TEST_RUN(m))
3927 return false;
3928
3929 /* If we are the user manager we can safely assume that the journal is up */
3930 if (!MANAGER_IS_SYSTEM(m))
3931 return true;
3932
3933 /* Check that the socket is not only up, but in RUNNING state */
3934 u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3935 if (!u)
3936 return false;
3937 if (SOCKET(u)->state != SOCKET_RUNNING)
3938 return false;
3939
3940 /* Similar, check if the daemon itself is fully up, too */
3941 u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3942 if (!u)
3943 return false;
3944 if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING))
3945 return false;
3946
3947 return true;
3948 }
3949
disable_printk_ratelimit(void)3950 void disable_printk_ratelimit(void) {
3951 /* Disable kernel's printk ratelimit.
3952 *
3953 * Logging to /dev/kmsg is most useful during early boot and shutdown, where normal logging
3954 * mechanisms are not available. The semantics of this sysctl are such that any kernel command-line
3955 * setting takes precedence. */
3956 int r;
3957
3958 r = sysctl_write("kernel/printk_devkmsg", "on");
3959 if (r < 0)
3960 log_debug_errno(r, "Failed to set sysctl kernel.printk_devkmsg=on: %m");
3961 }
3962
manager_recheck_journal(Manager * m)3963 void manager_recheck_journal(Manager *m) {
3964
3965 assert(m);
3966
3967 /* Don't bother with this unless we are in the special situation of being PID 1 */
3968 if (getpid_cached() != 1)
3969 return;
3970
3971 /* Don't check this while we are reloading, things might still change */
3972 if (MANAGER_IS_RELOADING(m))
3973 return;
3974
3975 /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the
3976 * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger
3977 * an activation ourselves we can't fulfill. */
3978 log_set_prohibit_ipc(!manager_journal_is_running(m));
3979 log_open();
3980 }
3981
manager_get_show_status(Manager * m)3982 static ShowStatus manager_get_show_status(Manager *m) {
3983 assert(m);
3984
3985 if (MANAGER_IS_USER(m))
3986 return _SHOW_STATUS_INVALID;
3987
3988 if (m->show_status_overridden != _SHOW_STATUS_INVALID)
3989 return m->show_status_overridden;
3990
3991 return m->show_status;
3992 }
3993
manager_get_show_status_on(Manager * m)3994 bool manager_get_show_status_on(Manager *m) {
3995 assert(m);
3996
3997 return show_status_on(manager_get_show_status(m));
3998 }
3999
set_show_status_marker(bool b)4000 static void set_show_status_marker(bool b) {
4001 if (b)
4002 (void) touch("/run/systemd/show-status");
4003 else
4004 (void) unlink("/run/systemd/show-status");
4005 }
4006
manager_set_show_status(Manager * m,ShowStatus mode,const char * reason)4007 void manager_set_show_status(Manager *m, ShowStatus mode, const char *reason) {
4008 assert(m);
4009 assert(reason);
4010 assert(mode >= 0 && mode < _SHOW_STATUS_MAX);
4011
4012 if (MANAGER_IS_USER(m))
4013 return;
4014
4015 if (mode == m->show_status)
4016 return;
4017
4018 if (m->show_status_overridden == _SHOW_STATUS_INVALID) {
4019 bool enabled;
4020
4021 enabled = show_status_on(mode);
4022 log_debug("%s (%s) showing of status (%s).",
4023 enabled ? "Enabling" : "Disabling",
4024 strna(show_status_to_string(mode)),
4025 reason);
4026
4027 set_show_status_marker(enabled);
4028 }
4029
4030 m->show_status = mode;
4031 }
4032
manager_override_show_status(Manager * m,ShowStatus mode,const char * reason)4033 void manager_override_show_status(Manager *m, ShowStatus mode, const char *reason) {
4034 assert(m);
4035 assert(mode < _SHOW_STATUS_MAX);
4036
4037 if (MANAGER_IS_USER(m))
4038 return;
4039
4040 if (mode == m->show_status_overridden)
4041 return;
4042
4043 m->show_status_overridden = mode;
4044
4045 if (mode == _SHOW_STATUS_INVALID)
4046 mode = m->show_status;
4047
4048 log_debug("%s (%s) showing of status (%s).",
4049 m->show_status_overridden != _SHOW_STATUS_INVALID ? "Overriding" : "Restoring",
4050 strna(show_status_to_string(mode)),
4051 reason);
4052
4053 set_show_status_marker(show_status_on(mode));
4054 }
4055
manager_get_confirm_spawn(Manager * m)4056 const char *manager_get_confirm_spawn(Manager *m) {
4057 static int last_errno = 0;
4058 struct stat st;
4059 int r;
4060
4061 assert(m);
4062
4063 /* Here's the deal: we want to test the validity of the console but don't want
4064 * PID1 to go through the whole console process which might block. But we also
4065 * want to warn the user only once if something is wrong with the console so we
4066 * cannot do the sanity checks after spawning our children. So here we simply do
4067 * really basic tests to hopefully trap common errors.
4068 *
4069 * If the console suddenly disappear at the time our children will really it
4070 * then they will simply fail to acquire it and a positive answer will be
4071 * assumed. New children will fall back to /dev/console though.
4072 *
4073 * Note: TTYs are devices that can come and go any time, and frequently aren't
4074 * available yet during early boot (consider a USB rs232 dongle...). If for any
4075 * reason the configured console is not ready, we fall back to the default
4076 * console. */
4077
4078 if (!m->confirm_spawn || path_equal(m->confirm_spawn, "/dev/console"))
4079 return m->confirm_spawn;
4080
4081 if (stat(m->confirm_spawn, &st) < 0) {
4082 r = -errno;
4083 goto fail;
4084 }
4085
4086 if (!S_ISCHR(st.st_mode)) {
4087 r = -ENOTTY;
4088 goto fail;
4089 }
4090
4091 last_errno = 0;
4092 return m->confirm_spawn;
4093
4094 fail:
4095 if (last_errno != r)
4096 last_errno = log_warning_errno(r, "Failed to open %s, using default console: %m", m->confirm_spawn);
4097
4098 return "/dev/console";
4099 }
4100
manager_set_first_boot(Manager * m,bool b)4101 void manager_set_first_boot(Manager *m, bool b) {
4102 assert(m);
4103
4104 if (!MANAGER_IS_SYSTEM(m))
4105 return;
4106
4107 if (m->first_boot != (int) b) {
4108 if (b)
4109 (void) touch("/run/systemd/first-boot");
4110 else
4111 (void) unlink("/run/systemd/first-boot");
4112 }
4113
4114 m->first_boot = b;
4115 }
4116
manager_disable_confirm_spawn(void)4117 void manager_disable_confirm_spawn(void) {
4118 (void) touch("/run/systemd/confirm_spawn_disabled");
4119 }
4120
manager_is_confirm_spawn_disabled(Manager * m)4121 bool manager_is_confirm_spawn_disabled(Manager *m) {
4122 if (!m->confirm_spawn)
4123 return true;
4124
4125 return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
4126 }
4127
manager_should_show_status(Manager * m,StatusType type)4128 static bool manager_should_show_status(Manager *m, StatusType type) {
4129 assert(m);
4130
4131 if (!MANAGER_IS_SYSTEM(m))
4132 return false;
4133
4134 if (m->no_console_output)
4135 return false;
4136
4137 if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
4138 return false;
4139
4140 /* If we cannot find out the status properly, just proceed. */
4141 if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
4142 return false;
4143
4144 if (type == STATUS_TYPE_NOTICE && m->show_status != SHOW_STATUS_NO)
4145 return true;
4146
4147 return manager_get_show_status_on(m);
4148 }
4149
manager_status_printf(Manager * m,StatusType type,const char * status,const char * format,...)4150 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
4151 va_list ap;
4152
4153 /* If m is NULL, assume we're after shutdown and let the messages through. */
4154
4155 if (m && !manager_should_show_status(m, type))
4156 return;
4157
4158 /* XXX We should totally drop the check for ephemeral here
4159 * and thus effectively make 'Type=idle' pointless. */
4160 if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
4161 return;
4162
4163 va_start(ap, format);
4164 status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap);
4165 va_end(ap);
4166 }
4167
manager_get_units_requiring_mounts_for(Manager * m,const char * path)4168 Set* manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
4169 assert(m);
4170 assert(path);
4171
4172 if (path_equal(path, "/"))
4173 path = "";
4174
4175 return hashmap_get(m->units_requiring_mounts_for, path);
4176 }
4177
manager_update_failed_units(Manager * m,Unit * u,bool failed)4178 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
4179 unsigned size;
4180 int r;
4181
4182 assert(m);
4183 assert(u->manager == m);
4184
4185 size = set_size(m->failed_units);
4186
4187 if (failed) {
4188 r = set_ensure_put(&m->failed_units, NULL, u);
4189 if (r < 0)
4190 return log_oom();
4191 } else
4192 (void) set_remove(m->failed_units, u);
4193
4194 if (set_size(m->failed_units) != size)
4195 bus_manager_send_change_signal(m);
4196
4197 return 0;
4198 }
4199
manager_state(Manager * m)4200 ManagerState manager_state(Manager *m) {
4201 Unit *u;
4202
4203 assert(m);
4204
4205 /* Is the special shutdown target active or queued? If so, we are in shutdown state */
4206 u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
4207 if (u && unit_active_or_pending(u))
4208 return MANAGER_STOPPING;
4209
4210 /* Did we ever finish booting? If not then we are still starting up */
4211 if (!MANAGER_IS_FINISHED(m)) {
4212
4213 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
4214 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
4215 return MANAGER_INITIALIZING;
4216
4217 return MANAGER_STARTING;
4218 }
4219
4220 if (MANAGER_IS_SYSTEM(m)) {
4221 /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
4222 u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
4223 if (u && unit_active_or_pending(u))
4224 return MANAGER_MAINTENANCE;
4225
4226 u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
4227 if (u && unit_active_or_pending(u))
4228 return MANAGER_MAINTENANCE;
4229 }
4230
4231 /* Are there any failed units? If so, we are in degraded mode */
4232 if (set_size(m->failed_units) > 0)
4233 return MANAGER_DEGRADED;
4234
4235 return MANAGER_RUNNING;
4236 }
4237
manager_unref_uid_internal(Hashmap * uid_refs,uid_t uid,bool destroy_now,int (* _clean_ipc)(uid_t uid))4238 static void manager_unref_uid_internal(
4239 Hashmap *uid_refs,
4240 uid_t uid,
4241 bool destroy_now,
4242 int (*_clean_ipc)(uid_t uid)) {
4243
4244 uint32_t c, n;
4245
4246 assert(uid_is_valid(uid));
4247 assert(_clean_ipc);
4248
4249 /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
4250 * that uid_t and gid_t are actually defined the same way, with the same validity rules.
4251 *
4252 * We store a hashmap where the key is the UID/GID and the value is a 32bit reference counter, whose highest
4253 * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
4254 * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
4255 * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
4256
4257 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4258 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4259
4260 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4261 return;
4262
4263 c = PTR_TO_UINT32(hashmap_get(uid_refs, UID_TO_PTR(uid)));
4264
4265 n = c & ~DESTROY_IPC_FLAG;
4266 assert(n > 0);
4267 n--;
4268
4269 if (destroy_now && n == 0) {
4270 hashmap_remove(uid_refs, UID_TO_PTR(uid));
4271
4272 if (c & DESTROY_IPC_FLAG) {
4273 log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
4274 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4275 uid);
4276 (void) _clean_ipc(uid);
4277 }
4278 } else {
4279 c = n | (c & DESTROY_IPC_FLAG);
4280 assert_se(hashmap_update(uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
4281 }
4282 }
4283
manager_unref_uid(Manager * m,uid_t uid,bool destroy_now)4284 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
4285 manager_unref_uid_internal(m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
4286 }
4287
manager_unref_gid(Manager * m,gid_t gid,bool destroy_now)4288 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
4289 manager_unref_uid_internal(m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
4290 }
4291
manager_ref_uid_internal(Hashmap ** uid_refs,uid_t uid,bool clean_ipc)4292 static int manager_ref_uid_internal(
4293 Hashmap **uid_refs,
4294 uid_t uid,
4295 bool clean_ipc) {
4296
4297 uint32_t c, n;
4298 int r;
4299
4300 assert(uid_refs);
4301 assert(uid_is_valid(uid));
4302
4303 /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
4304 * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
4305
4306 assert_cc(sizeof(uid_t) == sizeof(gid_t));
4307 assert_cc(UID_INVALID == (uid_t) GID_INVALID);
4308
4309 if (uid == 0) /* We don't keep track of root, and will never destroy it */
4310 return 0;
4311
4312 r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
4313 if (r < 0)
4314 return r;
4315
4316 c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
4317
4318 n = c & ~DESTROY_IPC_FLAG;
4319 n++;
4320
4321 if (n & DESTROY_IPC_FLAG) /* check for overflow */
4322 return -EOVERFLOW;
4323
4324 c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
4325
4326 return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
4327 }
4328
manager_ref_uid(Manager * m,uid_t uid,bool clean_ipc)4329 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
4330 return manager_ref_uid_internal(&m->uid_refs, uid, clean_ipc);
4331 }
4332
manager_ref_gid(Manager * m,gid_t gid,bool clean_ipc)4333 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
4334 return manager_ref_uid_internal(&m->gid_refs, (uid_t) gid, clean_ipc);
4335 }
4336
manager_vacuum_uid_refs_internal(Hashmap * uid_refs,int (* _clean_ipc)(uid_t uid))4337 static void manager_vacuum_uid_refs_internal(
4338 Hashmap *uid_refs,
4339 int (*_clean_ipc)(uid_t uid)) {
4340
4341 void *p, *k;
4342
4343 assert(_clean_ipc);
4344
4345 HASHMAP_FOREACH_KEY(p, k, uid_refs) {
4346 uint32_t c, n;
4347 uid_t uid;
4348
4349 uid = PTR_TO_UID(k);
4350 c = PTR_TO_UINT32(p);
4351
4352 n = c & ~DESTROY_IPC_FLAG;
4353 if (n > 0)
4354 continue;
4355
4356 if (c & DESTROY_IPC_FLAG) {
4357 log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
4358 _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
4359 uid);
4360 (void) _clean_ipc(uid);
4361 }
4362
4363 assert_se(hashmap_remove(uid_refs, k) == p);
4364 }
4365 }
4366
manager_vacuum_uid_refs(Manager * m)4367 static void manager_vacuum_uid_refs(Manager *m) {
4368 manager_vacuum_uid_refs_internal(m->uid_refs, clean_ipc_by_uid);
4369 }
4370
manager_vacuum_gid_refs(Manager * m)4371 static void manager_vacuum_gid_refs(Manager *m) {
4372 manager_vacuum_uid_refs_internal(m->gid_refs, clean_ipc_by_gid);
4373 }
4374
manager_vacuum(Manager * m)4375 static void manager_vacuum(Manager *m) {
4376 assert(m);
4377
4378 /* Release any dynamic users no longer referenced */
4379 dynamic_user_vacuum(m, true);
4380
4381 /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
4382 manager_vacuum_uid_refs(m);
4383 manager_vacuum_gid_refs(m);
4384
4385 /* Release any runtimes no longer referenced */
4386 exec_runtime_vacuum(m);
4387 }
4388
manager_dispatch_user_lookup_fd(sd_event_source * source,int fd,uint32_t revents,void * userdata)4389 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
4390 struct buffer {
4391 uid_t uid;
4392 gid_t gid;
4393 char unit_name[UNIT_NAME_MAX+1];
4394 } _packed_ buffer;
4395
4396 Manager *m = userdata;
4397 ssize_t l;
4398 size_t n;
4399 Unit *u;
4400
4401 assert_se(source);
4402 assert_se(m);
4403
4404 /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
4405 * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
4406 * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
4407
4408 l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
4409 if (l < 0) {
4410 if (ERRNO_IS_TRANSIENT(errno))
4411 return 0;
4412
4413 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
4414 }
4415
4416 if ((size_t) l <= offsetof(struct buffer, unit_name)) {
4417 log_warning("Received too short user lookup message, ignoring.");
4418 return 0;
4419 }
4420
4421 if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
4422 log_warning("Received too long user lookup message, ignoring.");
4423 return 0;
4424 }
4425
4426 if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
4427 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
4428 return 0;
4429 }
4430
4431 n = (size_t) l - offsetof(struct buffer, unit_name);
4432 if (memchr(buffer.unit_name, 0, n)) {
4433 log_warning("Received lookup message with embedded NUL character, ignoring.");
4434 return 0;
4435 }
4436
4437 buffer.unit_name[n] = 0;
4438 u = manager_get_unit(m, buffer.unit_name);
4439 if (!u) {
4440 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
4441 return 0;
4442 }
4443
4444 log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
4445
4446 unit_notify_user_lookup(u, buffer.uid, buffer.gid);
4447 return 0;
4448 }
4449
short_uid_range(const char * path)4450 static int short_uid_range(const char *path) {
4451 _cleanup_free_ UidRange *p = NULL;
4452 size_t n = 0;
4453 int r;
4454
4455 assert(path);
4456
4457 /* Taint systemd if we the UID range assigned to this environment doesn't at least cover 0…65534,
4458 * i.e. from root to nobody. */
4459
4460 r = uid_range_load_userns(&p, &n, path);
4461 if (ERRNO_IS_NOT_SUPPORTED(r))
4462 return false;
4463 if (r < 0)
4464 return log_debug_errno(r, "Failed to load %s: %m", path);
4465
4466 return !uid_range_covers(p, n, 0, 65535);
4467 }
4468
manager_taint_string(const Manager * m)4469 char* manager_taint_string(const Manager *m) {
4470 /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". Only things that are detected at
4471 * runtime should be tagged here. For stuff that is known during compilation, emit a warning in the
4472 * configuration phase. */
4473
4474 assert(m);
4475
4476 const char* stage[12] = {};
4477 size_t n = 0;
4478
4479 if (m->taint_usr)
4480 stage[n++] = "split-usr";
4481
4482 _cleanup_free_ char *usrbin = NULL;
4483 if (readlink_malloc("/bin", &usrbin) < 0 || !PATH_IN_SET(usrbin, "usr/bin", "/usr/bin"))
4484 stage[n++] = "unmerged-usr";
4485
4486 if (access("/proc/cgroups", F_OK) < 0)
4487 stage[n++] = "cgroups-missing";
4488
4489 if (cg_all_unified() == 0)
4490 stage[n++] = "cgroupsv1";
4491
4492 if (clock_is_localtime(NULL) > 0)
4493 stage[n++] = "local-hwclock";
4494
4495 _cleanup_free_ char *destination = NULL;
4496 if (readlink_malloc("/var/run", &destination) < 0 ||
4497 !PATH_IN_SET(destination, "../run", "/run"))
4498 stage[n++] = "var-run-bad";
4499
4500 _cleanup_free_ char *overflowuid = NULL, *overflowgid = NULL;
4501 if (read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid) >= 0 &&
4502 !streq(overflowuid, "65534"))
4503 stage[n++] = "overflowuid-not-65534";
4504 if (read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid) >= 0 &&
4505 !streq(overflowgid, "65534"))
4506 stage[n++] = "overflowgid-not-65534";
4507
4508 struct utsname uts;
4509 assert_se(uname(&uts) >= 0);
4510 if (strverscmp_improved(uts.release, KERNEL_BASELINE_VERSION) < 0)
4511 stage[n++] = "old-kernel";
4512
4513 if (short_uid_range("/proc/self/uid_map") > 0)
4514 stage[n++] = "short-uid-range";
4515 if (short_uid_range("/proc/self/gid_map") > 0)
4516 stage[n++] = "short-gid-range";
4517
4518 assert(n < ELEMENTSOF(stage) - 1); /* One extra for NULL terminator */
4519
4520 return strv_join((char**) stage, ":");
4521 }
4522
manager_ref_console(Manager * m)4523 void manager_ref_console(Manager *m) {
4524 assert(m);
4525
4526 m->n_on_console++;
4527 }
4528
manager_unref_console(Manager * m)4529 void manager_unref_console(Manager *m) {
4530
4531 assert(m->n_on_console > 0);
4532 m->n_on_console--;
4533
4534 if (m->n_on_console == 0)
4535 m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */
4536 }
4537
manager_override_log_level(Manager * m,int level)4538 void manager_override_log_level(Manager *m, int level) {
4539 _cleanup_free_ char *s = NULL;
4540 assert(m);
4541
4542 if (!m->log_level_overridden) {
4543 m->original_log_level = log_get_max_level();
4544 m->log_level_overridden = true;
4545 }
4546
4547 (void) log_level_to_string_alloc(level, &s);
4548 log_info("Setting log level to %s.", strna(s));
4549
4550 log_set_max_level(level);
4551 }
4552
manager_restore_original_log_level(Manager * m)4553 void manager_restore_original_log_level(Manager *m) {
4554 _cleanup_free_ char *s = NULL;
4555 assert(m);
4556
4557 if (!m->log_level_overridden)
4558 return;
4559
4560 (void) log_level_to_string_alloc(m->original_log_level, &s);
4561 log_info("Restoring log level to original (%s).", strna(s));
4562
4563 log_set_max_level(m->original_log_level);
4564 m->log_level_overridden = false;
4565 }
4566
manager_override_log_target(Manager * m,LogTarget target)4567 void manager_override_log_target(Manager *m, LogTarget target) {
4568 assert(m);
4569
4570 if (!m->log_target_overridden) {
4571 m->original_log_target = log_get_target();
4572 m->log_target_overridden = true;
4573 }
4574
4575 log_info("Setting log target to %s.", log_target_to_string(target));
4576 log_set_target(target);
4577 }
4578
manager_restore_original_log_target(Manager * m)4579 void manager_restore_original_log_target(Manager *m) {
4580 assert(m);
4581
4582 if (!m->log_target_overridden)
4583 return;
4584
4585 log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target));
4586
4587 log_set_target(m->original_log_target);
4588 m->log_target_overridden = false;
4589 }
4590
manager_timestamp_initrd_mangle(ManagerTimestamp s)4591 ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) {
4592 if (in_initrd() &&
4593 s >= MANAGER_TIMESTAMP_SECURITY_START &&
4594 s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)
4595 return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START;
4596 return s;
4597 }
4598
4599 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
4600 [MANAGER_INITIALIZING] = "initializing",
4601 [MANAGER_STARTING] = "starting",
4602 [MANAGER_RUNNING] = "running",
4603 [MANAGER_DEGRADED] = "degraded",
4604 [MANAGER_MAINTENANCE] = "maintenance",
4605 [MANAGER_STOPPING] = "stopping",
4606 };
4607
4608 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);
4609
4610 static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = {
4611 [MANAGER_TIMESTAMP_FIRMWARE] = "firmware",
4612 [MANAGER_TIMESTAMP_LOADER] = "loader",
4613 [MANAGER_TIMESTAMP_KERNEL] = "kernel",
4614 [MANAGER_TIMESTAMP_INITRD] = "initrd",
4615 [MANAGER_TIMESTAMP_USERSPACE] = "userspace",
4616 [MANAGER_TIMESTAMP_FINISH] = "finish",
4617 [MANAGER_TIMESTAMP_SECURITY_START] = "security-start",
4618 [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish",
4619 [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start",
4620 [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish",
4621 [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start",
4622 [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish",
4623 [MANAGER_TIMESTAMP_UNITS_LOAD] = "units-load",
4624 [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start",
4625 [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish",
4626 [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start",
4627 [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish",
4628 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start",
4629 [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish",
4630 };
4631
4632 DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp);
4633
4634 static const char* const oom_policy_table[_OOM_POLICY_MAX] = {
4635 [OOM_CONTINUE] = "continue",
4636 [OOM_STOP] = "stop",
4637 [OOM_KILL] = "kill",
4638 };
4639
4640 DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy);
4641