]>
Commit | Line | Data |
---|---|---|
1 | /* SPDX-License-Identifier: LGPL-2.1+ */ | |
2 | ||
3 | #include <errno.h> | |
4 | #include <fcntl.h> | |
5 | #include <linux/kd.h> | |
6 | #include <signal.h> | |
7 | #include <string.h> | |
8 | #include <sys/epoll.h> | |
9 | #include <sys/inotify.h> | |
10 | #include <sys/ioctl.h> | |
11 | #include <sys/reboot.h> | |
12 | #include <sys/timerfd.h> | |
13 | #include <sys/wait.h> | |
14 | #include <unistd.h> | |
15 | ||
16 | #if HAVE_AUDIT | |
17 | #include <libaudit.h> | |
18 | #endif | |
19 | ||
20 | #include "sd-daemon.h" | |
21 | #include "sd-messages.h" | |
22 | #include "sd-path.h" | |
23 | ||
24 | #include "all-units.h" | |
25 | #include "alloc-util.h" | |
26 | #include "audit-fd.h" | |
27 | #include "boot-timestamps.h" | |
28 | #include "bus-common-errors.h" | |
29 | #include "bus-error.h" | |
30 | #include "bus-kernel.h" | |
31 | #include "bus-util.h" | |
32 | #include "clean-ipc.h" | |
33 | #include "clock-util.h" | |
34 | #include "dbus-job.h" | |
35 | #include "dbus-manager.h" | |
36 | #include "dbus-unit.h" | |
37 | #include "dbus.h" | |
38 | #include "dirent-util.h" | |
39 | #include "env-util.h" | |
40 | #include "escape.h" | |
41 | #include "exec-util.h" | |
42 | #include "execute.h" | |
43 | #include "exit-status.h" | |
44 | #include "fd-util.h" | |
45 | #include "fileio.h" | |
46 | #include "fs-util.h" | |
47 | #include "hashmap.h" | |
48 | #include "io-util.h" | |
49 | #include "label.h" | |
50 | #include "locale-setup.h" | |
51 | #include "log.h" | |
52 | #include "macro.h" | |
53 | #include "manager.h" | |
54 | #include "memory-util.h" | |
55 | #include "missing.h" | |
56 | #include "mkdir.h" | |
57 | #include "parse-util.h" | |
58 | #include "path-lookup.h" | |
59 | #include "path-util.h" | |
60 | #include "plymouth-util.h" | |
61 | #include "process-util.h" | |
62 | #include "ratelimit.h" | |
63 | #include "rlimit-util.h" | |
64 | #include "rm-rf.h" | |
65 | #include "serialize.h" | |
66 | #include "signal-util.h" | |
67 | #include "socket-util.h" | |
68 | #include "special.h" | |
69 | #include "stat-util.h" | |
70 | #include "string-table.h" | |
71 | #include "string-util.h" | |
72 | #include "strv.h" | |
73 | #include "strxcpyx.h" | |
74 | #include "syslog-util.h" | |
75 | #include "terminal-util.h" | |
76 | #include "time-util.h" | |
77 | #include "transaction.h" | |
78 | #include "umask-util.h" | |
79 | #include "unit-name.h" | |
80 | #include "user-util.h" | |
81 | #include "virt.h" | |
82 | #include "watchdog.h" | |
83 | ||
84 | #define NOTIFY_RCVBUF_SIZE (8*1024*1024) | |
85 | #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024) | |
86 | ||
87 | /* Initial delay and the interval for printing status messages about running jobs */ | |
88 | #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC) | |
89 | #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3) | |
90 | #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3 | |
91 | ||
92 | /* If there are more than 1K bus messages queue across our API and direct busses, then let's not add more on top until | |
93 | * the queue gets more empty. */ | |
94 | #define MANAGER_BUS_BUSY_THRESHOLD 1024LU | |
95 | ||
96 | /* How many units and jobs to process of the bus queue before returning to the event loop. */ | |
97 | #define MANAGER_BUS_MESSAGE_BUDGET 100U | |
98 | ||
99 | static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
100 | static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
101 | static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
102 | static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
103 | static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
104 | static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata); | |
105 | static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata); | |
106 | static int manager_dispatch_run_queue(sd_event_source *source, void *userdata); | |
107 | static int manager_dispatch_sigchld(sd_event_source *source, void *userdata); | |
108 | static int manager_dispatch_timezone_change(sd_event_source *source, const struct inotify_event *event, void *userdata); | |
109 | static int manager_run_environment_generators(Manager *m); | |
110 | static int manager_run_generators(Manager *m); | |
111 | ||
112 | static void manager_watch_jobs_in_progress(Manager *m) { | |
113 | usec_t next; | |
114 | int r; | |
115 | ||
116 | assert(m); | |
117 | ||
118 | /* We do not want to show the cylon animation if the user | |
119 | * needs to confirm service executions otherwise confirmation | |
120 | * messages will be screwed by the cylon animation. */ | |
121 | if (!manager_is_confirm_spawn_disabled(m)) | |
122 | return; | |
123 | ||
124 | if (m->jobs_in_progress_event_source) | |
125 | return; | |
126 | ||
127 | next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC; | |
128 | r = sd_event_add_time( | |
129 | m->event, | |
130 | &m->jobs_in_progress_event_source, | |
131 | CLOCK_MONOTONIC, | |
132 | next, 0, | |
133 | manager_dispatch_jobs_in_progress, m); | |
134 | if (r < 0) | |
135 | return; | |
136 | ||
137 | (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress"); | |
138 | } | |
139 | ||
140 | #define CYLON_BUFFER_EXTRA (2*STRLEN(ANSI_RED) + STRLEN(ANSI_HIGHLIGHT_RED) + 2*STRLEN(ANSI_NORMAL)) | |
141 | ||
142 | static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) { | |
143 | char *p = buffer; | |
144 | ||
145 | assert(buflen >= CYLON_BUFFER_EXTRA + width + 1); | |
146 | assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */ | |
147 | ||
148 | if (pos > 1) { | |
149 | if (pos > 2) | |
150 | p = mempset(p, ' ', pos-2); | |
151 | if (log_get_show_color()) | |
152 | p = stpcpy(p, ANSI_RED); | |
153 | *p++ = '*'; | |
154 | } | |
155 | ||
156 | if (pos > 0 && pos <= width) { | |
157 | if (log_get_show_color()) | |
158 | p = stpcpy(p, ANSI_HIGHLIGHT_RED); | |
159 | *p++ = '*'; | |
160 | } | |
161 | ||
162 | if (log_get_show_color()) | |
163 | p = stpcpy(p, ANSI_NORMAL); | |
164 | ||
165 | if (pos < width) { | |
166 | if (log_get_show_color()) | |
167 | p = stpcpy(p, ANSI_RED); | |
168 | *p++ = '*'; | |
169 | if (pos < width-1) | |
170 | p = mempset(p, ' ', width-1-pos); | |
171 | if (log_get_show_color()) | |
172 | strcpy(p, ANSI_NORMAL); | |
173 | } | |
174 | } | |
175 | ||
176 | void manager_flip_auto_status(Manager *m, bool enable) { | |
177 | assert(m); | |
178 | ||
179 | if (enable) { | |
180 | if (m->show_status == SHOW_STATUS_AUTO) | |
181 | manager_set_show_status(m, SHOW_STATUS_TEMPORARY); | |
182 | } else { | |
183 | if (m->show_status == SHOW_STATUS_TEMPORARY) | |
184 | manager_set_show_status(m, SHOW_STATUS_AUTO); | |
185 | } | |
186 | } | |
187 | ||
188 | static void manager_print_jobs_in_progress(Manager *m) { | |
189 | _cleanup_free_ char *job_of_n = NULL; | |
190 | Iterator i; | |
191 | Job *j; | |
192 | unsigned counter = 0, print_nr; | |
193 | char cylon[6 + CYLON_BUFFER_EXTRA + 1]; | |
194 | unsigned cylon_pos; | |
195 | char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit"; | |
196 | uint64_t x; | |
197 | ||
198 | assert(m); | |
199 | assert(m->n_running_jobs > 0); | |
200 | ||
201 | manager_flip_auto_status(m, true); | |
202 | ||
203 | print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs; | |
204 | ||
205 | HASHMAP_FOREACH(j, m->jobs, i) | |
206 | if (j->state == JOB_RUNNING && counter++ == print_nr) | |
207 | break; | |
208 | ||
209 | /* m->n_running_jobs must be consistent with the contents of m->jobs, | |
210 | * so the above loop must have succeeded in finding j. */ | |
211 | assert(counter == print_nr + 1); | |
212 | assert(j); | |
213 | ||
214 | cylon_pos = m->jobs_in_progress_iteration % 14; | |
215 | if (cylon_pos >= 8) | |
216 | cylon_pos = 14 - cylon_pos; | |
217 | draw_cylon(cylon, sizeof(cylon), 6, cylon_pos); | |
218 | ||
219 | m->jobs_in_progress_iteration++; | |
220 | ||
221 | if (m->n_running_jobs > 1) { | |
222 | if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0) | |
223 | job_of_n = NULL; | |
224 | } | |
225 | ||
226 | format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC); | |
227 | if (job_get_timeout(j, &x) > 0) | |
228 | format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC); | |
229 | ||
230 | manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon, | |
231 | "%sA %s job is running for %s (%s / %s)", | |
232 | strempty(job_of_n), | |
233 | job_type_to_string(j->type), | |
234 | unit_description(j->unit), | |
235 | time, limit); | |
236 | } | |
237 | ||
238 | static int have_ask_password(void) { | |
239 | _cleanup_closedir_ DIR *dir; | |
240 | struct dirent *de; | |
241 | ||
242 | dir = opendir("/run/systemd/ask-password"); | |
243 | if (!dir) { | |
244 | if (errno == ENOENT) | |
245 | return false; | |
246 | else | |
247 | return -errno; | |
248 | } | |
249 | ||
250 | FOREACH_DIRENT_ALL(de, dir, return -errno) { | |
251 | if (startswith(de->d_name, "ask.")) | |
252 | return true; | |
253 | } | |
254 | return false; | |
255 | } | |
256 | ||
257 | static int manager_dispatch_ask_password_fd(sd_event_source *source, | |
258 | int fd, uint32_t revents, void *userdata) { | |
259 | Manager *m = userdata; | |
260 | ||
261 | assert(m); | |
262 | ||
263 | (void) flush_fd(fd); | |
264 | ||
265 | m->have_ask_password = have_ask_password(); | |
266 | if (m->have_ask_password < 0) | |
267 | /* Log error but continue. Negative have_ask_password | |
268 | * is treated as unknown status. */ | |
269 | log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m"); | |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
274 | static void manager_close_ask_password(Manager *m) { | |
275 | assert(m); | |
276 | ||
277 | m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source); | |
278 | m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd); | |
279 | m->have_ask_password = -EINVAL; | |
280 | } | |
281 | ||
282 | static int manager_check_ask_password(Manager *m) { | |
283 | int r; | |
284 | ||
285 | assert(m); | |
286 | ||
287 | if (!m->ask_password_event_source) { | |
288 | assert(m->ask_password_inotify_fd < 0); | |
289 | ||
290 | mkdir_p_label("/run/systemd/ask-password", 0755); | |
291 | ||
292 | m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); | |
293 | if (m->ask_password_inotify_fd < 0) | |
294 | return log_error_errno(errno, "Failed to create inotify object: %m"); | |
295 | ||
296 | if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) { | |
297 | log_error_errno(errno, "Failed to watch \"/run/systemd/ask-password\": %m"); | |
298 | manager_close_ask_password(m); | |
299 | return -errno; | |
300 | } | |
301 | ||
302 | r = sd_event_add_io(m->event, &m->ask_password_event_source, | |
303 | m->ask_password_inotify_fd, EPOLLIN, | |
304 | manager_dispatch_ask_password_fd, m); | |
305 | if (r < 0) { | |
306 | log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m"); | |
307 | manager_close_ask_password(m); | |
308 | return -errno; | |
309 | } | |
310 | ||
311 | (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password"); | |
312 | ||
313 | /* Queries might have been added meanwhile... */ | |
314 | manager_dispatch_ask_password_fd(m->ask_password_event_source, | |
315 | m->ask_password_inotify_fd, EPOLLIN, m); | |
316 | } | |
317 | ||
318 | return m->have_ask_password; | |
319 | } | |
320 | ||
321 | static int manager_watch_idle_pipe(Manager *m) { | |
322 | int r; | |
323 | ||
324 | assert(m); | |
325 | ||
326 | if (m->idle_pipe_event_source) | |
327 | return 0; | |
328 | ||
329 | if (m->idle_pipe[2] < 0) | |
330 | return 0; | |
331 | ||
332 | r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m); | |
333 | if (r < 0) | |
334 | return log_error_errno(r, "Failed to watch idle pipe: %m"); | |
335 | ||
336 | (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe"); | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
341 | static void manager_close_idle_pipe(Manager *m) { | |
342 | assert(m); | |
343 | ||
344 | m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source); | |
345 | ||
346 | safe_close_pair(m->idle_pipe); | |
347 | safe_close_pair(m->idle_pipe + 2); | |
348 | } | |
349 | ||
350 | static int manager_setup_time_change(Manager *m) { | |
351 | int r; | |
352 | ||
353 | assert(m); | |
354 | ||
355 | if (MANAGER_IS_TEST_RUN(m)) | |
356 | return 0; | |
357 | ||
358 | m->time_change_event_source = sd_event_source_unref(m->time_change_event_source); | |
359 | m->time_change_fd = safe_close(m->time_change_fd); | |
360 | ||
361 | m->time_change_fd = time_change_fd(); | |
362 | if (m->time_change_fd < 0) | |
363 | return log_error_errno(m->time_change_fd, "Failed to create timer change timer fd: %m"); | |
364 | ||
365 | r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m); | |
366 | if (r < 0) | |
367 | return log_error_errno(r, "Failed to create time change event source: %m"); | |
368 | ||
369 | /* Schedule this slightly earlier than the .timer event sources */ | |
370 | r = sd_event_source_set_priority(m->time_change_event_source, SD_EVENT_PRIORITY_NORMAL-1); | |
371 | if (r < 0) | |
372 | return log_error_errno(r, "Failed to set priority of time change event sources: %m"); | |
373 | ||
374 | (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change"); | |
375 | ||
376 | log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd."); | |
377 | ||
378 | return 0; | |
379 | } | |
380 | ||
381 | static int manager_read_timezone_stat(Manager *m) { | |
382 | struct stat st; | |
383 | bool changed; | |
384 | ||
385 | assert(m); | |
386 | ||
387 | /* Read the current stat() data of /etc/localtime so that we detect changes */ | |
388 | if (lstat("/etc/localtime", &st) < 0) { | |
389 | log_debug_errno(errno, "Failed to stat /etc/localtime, ignoring: %m"); | |
390 | changed = m->etc_localtime_accessible; | |
391 | m->etc_localtime_accessible = false; | |
392 | } else { | |
393 | usec_t k; | |
394 | ||
395 | k = timespec_load(&st.st_mtim); | |
396 | changed = !m->etc_localtime_accessible || k != m->etc_localtime_mtime; | |
397 | ||
398 | m->etc_localtime_mtime = k; | |
399 | m->etc_localtime_accessible = true; | |
400 | } | |
401 | ||
402 | return changed; | |
403 | } | |
404 | ||
405 | static int manager_setup_timezone_change(Manager *m) { | |
406 | _cleanup_(sd_event_source_unrefp) sd_event_source *new_event = NULL; | |
407 | int r; | |
408 | ||
409 | assert(m); | |
410 | ||
411 | if (MANAGER_IS_TEST_RUN(m)) | |
412 | return 0; | |
413 | ||
414 | /* We watch /etc/localtime for three events: change of the link count (which might mean removal from /etc even | |
415 | * though another link might be kept), renames, and file close operations after writing. Note we don't bother | |
416 | * with IN_DELETE_SELF, as that would just report when the inode is removed entirely, i.e. after the link count | |
417 | * went to zero and all fds to it are closed. | |
418 | * | |
419 | * Note that we never follow symlinks here. This is a simplification, but should cover almost all cases | |
420 | * correctly. | |
421 | * | |
422 | * Note that we create the new event source first here, before releasing the old one. This should optimize | |
423 | * behaviour as this way sd-event can reuse the old watch in case the inode didn't change. */ | |
424 | ||
425 | r = sd_event_add_inotify(m->event, &new_event, "/etc/localtime", | |
426 | IN_ATTRIB|IN_MOVE_SELF|IN_CLOSE_WRITE|IN_DONT_FOLLOW, manager_dispatch_timezone_change, m); | |
427 | if (r == -ENOENT) { | |
428 | /* If the file doesn't exist yet, subscribe to /etc instead, and wait until it is created either by | |
429 | * O_CREATE or by rename() */ | |
430 | ||
431 | log_debug_errno(r, "/etc/localtime doesn't exist yet, watching /etc instead."); | |
432 | r = sd_event_add_inotify(m->event, &new_event, "/etc", | |
433 | IN_CREATE|IN_MOVED_TO|IN_ONLYDIR, manager_dispatch_timezone_change, m); | |
434 | } | |
435 | if (r < 0) | |
436 | return log_error_errno(r, "Failed to create timezone change event source: %m"); | |
437 | ||
438 | /* Schedule this slightly earlier than the .timer event sources */ | |
439 | r = sd_event_source_set_priority(new_event, SD_EVENT_PRIORITY_NORMAL-1); | |
440 | if (r < 0) | |
441 | return log_error_errno(r, "Failed to set priority of timezone change event sources: %m"); | |
442 | ||
443 | sd_event_source_unref(m->timezone_change_event_source); | |
444 | m->timezone_change_event_source = TAKE_PTR(new_event); | |
445 | ||
446 | return 0; | |
447 | } | |
448 | ||
449 | static int enable_special_signals(Manager *m) { | |
450 | _cleanup_close_ int fd = -1; | |
451 | ||
452 | assert(m); | |
453 | ||
454 | if (MANAGER_IS_TEST_RUN(m)) | |
455 | return 0; | |
456 | ||
457 | /* Enable that we get SIGINT on control-alt-del. In containers | |
458 | * this will fail with EPERM (older) or EINVAL (newer), so | |
459 | * ignore that. */ | |
460 | if (reboot(RB_DISABLE_CAD) < 0 && !IN_SET(errno, EPERM, EINVAL)) | |
461 | log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m"); | |
462 | ||
463 | fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC); | |
464 | if (fd < 0) { | |
465 | /* Support systems without virtual console */ | |
466 | if (fd != -ENOENT) | |
467 | log_warning_errno(errno, "Failed to open /dev/tty0: %m"); | |
468 | } else { | |
469 | /* Enable that we get SIGWINCH on kbrequest */ | |
470 | if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0) | |
471 | log_warning_errno(errno, "Failed to enable kbrequest handling: %m"); | |
472 | } | |
473 | ||
474 | return 0; | |
475 | } | |
476 | ||
477 | #define RTSIG_IF_AVAILABLE(signum) (signum <= SIGRTMAX ? signum : -1) | |
478 | ||
479 | static int manager_setup_signals(Manager *m) { | |
480 | struct sigaction sa = { | |
481 | .sa_handler = SIG_DFL, | |
482 | .sa_flags = SA_NOCLDSTOP|SA_RESTART, | |
483 | }; | |
484 | sigset_t mask; | |
485 | int r; | |
486 | ||
487 | assert(m); | |
488 | ||
489 | assert_se(sigaction(SIGCHLD, &sa, NULL) == 0); | |
490 | ||
491 | /* We make liberal use of realtime signals here. On | |
492 | * Linux/glibc we have 30 of them (with the exception of Linux | |
493 | * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30 | |
494 | * (aka SIGRTMAX). */ | |
495 | ||
496 | assert_se(sigemptyset(&mask) == 0); | |
497 | sigset_add_many(&mask, | |
498 | SIGCHLD, /* Child died */ | |
499 | SIGTERM, /* Reexecute daemon */ | |
500 | SIGHUP, /* Reload configuration */ | |
501 | SIGUSR1, /* systemd/upstart: reconnect to D-Bus */ | |
502 | SIGUSR2, /* systemd: dump status */ | |
503 | SIGINT, /* Kernel sends us this on control-alt-del */ | |
504 | SIGWINCH, /* Kernel sends us this on kbrequest (alt-arrowup) */ | |
505 | SIGPWR, /* Some kernel drivers and upsd send us this on power failure */ | |
506 | ||
507 | SIGRTMIN+0, /* systemd: start default.target */ | |
508 | SIGRTMIN+1, /* systemd: isolate rescue.target */ | |
509 | SIGRTMIN+2, /* systemd: isolate emergency.target */ | |
510 | SIGRTMIN+3, /* systemd: start halt.target */ | |
511 | SIGRTMIN+4, /* systemd: start poweroff.target */ | |
512 | SIGRTMIN+5, /* systemd: start reboot.target */ | |
513 | SIGRTMIN+6, /* systemd: start kexec.target */ | |
514 | ||
515 | /* ... space for more special targets ... */ | |
516 | ||
517 | SIGRTMIN+13, /* systemd: Immediate halt */ | |
518 | SIGRTMIN+14, /* systemd: Immediate poweroff */ | |
519 | SIGRTMIN+15, /* systemd: Immediate reboot */ | |
520 | SIGRTMIN+16, /* systemd: Immediate kexec */ | |
521 | ||
522 | /* ... space for more immediate system state changes ... */ | |
523 | ||
524 | SIGRTMIN+20, /* systemd: enable status messages */ | |
525 | SIGRTMIN+21, /* systemd: disable status messages */ | |
526 | SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */ | |
527 | SIGRTMIN+23, /* systemd: set log level to LOG_INFO */ | |
528 | SIGRTMIN+24, /* systemd: Immediate exit (--user only) */ | |
529 | ||
530 | /* .. one free signal here ... */ | |
531 | ||
532 | /* Apparently Linux on hppa had fewer RT signals until v3.18, | |
533 | * SIGRTMAX was SIGRTMIN+25, and then SIGRTMIN was lowered, | |
534 | * see commit v3.17-7614-g1f25df2eff. | |
535 | * | |
536 | * We cannot unconditionally make use of those signals here, | |
537 | * so let's use a runtime check. Since these commands are | |
538 | * accessible by different means and only really a safety | |
539 | * net, the missing functionality on hppa shouldn't matter. | |
540 | */ | |
541 | ||
542 | RTSIG_IF_AVAILABLE(SIGRTMIN+26), /* systemd: set log target to journal-or-kmsg */ | |
543 | RTSIG_IF_AVAILABLE(SIGRTMIN+27), /* systemd: set log target to console */ | |
544 | RTSIG_IF_AVAILABLE(SIGRTMIN+28), /* systemd: set log target to kmsg */ | |
545 | RTSIG_IF_AVAILABLE(SIGRTMIN+29), /* systemd: set log target to syslog-or-kmsg (obsolete) */ | |
546 | ||
547 | /* ... one free signal here SIGRTMIN+30 ... */ | |
548 | -1); | |
549 | assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); | |
550 | ||
551 | m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC); | |
552 | if (m->signal_fd < 0) | |
553 | return -errno; | |
554 | ||
555 | r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m); | |
556 | if (r < 0) | |
557 | return r; | |
558 | ||
559 | (void) sd_event_source_set_description(m->signal_event_source, "manager-signal"); | |
560 | ||
561 | /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the | |
562 | * notify processing can still figure out to which process/service a message belongs, before we reap the | |
563 | * process. Also, process this before handling cgroup notifications, so that we always collect child exit | |
564 | * status information before detecting that there's no process in a cgroup. */ | |
565 | r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6); | |
566 | if (r < 0) | |
567 | return r; | |
568 | ||
569 | if (MANAGER_IS_SYSTEM(m)) | |
570 | return enable_special_signals(m); | |
571 | ||
572 | return 0; | |
573 | } | |
574 | ||
575 | static char** sanitize_environment(char **l) { | |
576 | ||
577 | /* Let's remove some environment variables that we need ourselves to communicate with our clients */ | |
578 | strv_env_unset_many( | |
579 | l, | |
580 | "EXIT_CODE", | |
581 | "EXIT_STATUS", | |
582 | "INVOCATION_ID", | |
583 | "JOURNAL_STREAM", | |
584 | "LISTEN_FDNAMES", | |
585 | "LISTEN_FDS", | |
586 | "LISTEN_PID", | |
587 | "MAINPID", | |
588 | "MANAGERPID", | |
589 | "NOTIFY_SOCKET", | |
590 | "PIDFILE", | |
591 | "REMOTE_ADDR", | |
592 | "REMOTE_PORT", | |
593 | "SERVICE_RESULT", | |
594 | "WATCHDOG_PID", | |
595 | "WATCHDOG_USEC", | |
596 | NULL); | |
597 | ||
598 | /* Let's order the environment alphabetically, just to make it pretty */ | |
599 | strv_sort(l); | |
600 | ||
601 | return l; | |
602 | } | |
603 | ||
604 | int manager_default_environment(Manager *m) { | |
605 | assert(m); | |
606 | ||
607 | m->transient_environment = strv_free(m->transient_environment); | |
608 | ||
609 | if (MANAGER_IS_SYSTEM(m)) { | |
610 | /* The system manager always starts with a clean | |
611 | * environment for its children. It does not import | |
612 | * the kernel's or the parents' exported variables. | |
613 | * | |
614 | * The initial passed environment is untouched to keep | |
615 | * /proc/self/environ valid; it is used for tagging | |
616 | * the init process inside containers. */ | |
617 | m->transient_environment = strv_new("PATH=" DEFAULT_PATH); | |
618 | ||
619 | /* Import locale variables LC_*= from configuration */ | |
620 | (void) locale_setup(&m->transient_environment); | |
621 | } else | |
622 | /* The user manager passes its own environment | |
623 | * along to its children. */ | |
624 | m->transient_environment = strv_copy(environ); | |
625 | ||
626 | if (!m->transient_environment) | |
627 | return log_oom(); | |
628 | ||
629 | sanitize_environment(m->transient_environment); | |
630 | ||
631 | return 0; | |
632 | } | |
633 | ||
634 | static int manager_setup_prefix(Manager *m) { | |
635 | struct table_entry { | |
636 | uint64_t type; | |
637 | const char *suffix; | |
638 | }; | |
639 | ||
640 | static const struct table_entry paths_system[_EXEC_DIRECTORY_TYPE_MAX] = { | |
641 | [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_SYSTEM_RUNTIME, NULL }, | |
642 | [EXEC_DIRECTORY_STATE] = { SD_PATH_SYSTEM_STATE_PRIVATE, NULL }, | |
643 | [EXEC_DIRECTORY_CACHE] = { SD_PATH_SYSTEM_STATE_CACHE, NULL }, | |
644 | [EXEC_DIRECTORY_LOGS] = { SD_PATH_SYSTEM_STATE_LOGS, NULL }, | |
645 | [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_SYSTEM_CONFIGURATION, NULL }, | |
646 | }; | |
647 | ||
648 | static const struct table_entry paths_user[_EXEC_DIRECTORY_TYPE_MAX] = { | |
649 | [EXEC_DIRECTORY_RUNTIME] = { SD_PATH_USER_RUNTIME, NULL }, | |
650 | [EXEC_DIRECTORY_STATE] = { SD_PATH_USER_CONFIGURATION, NULL }, | |
651 | [EXEC_DIRECTORY_CACHE] = { SD_PATH_USER_STATE_CACHE, NULL }, | |
652 | [EXEC_DIRECTORY_LOGS] = { SD_PATH_USER_CONFIGURATION, "log" }, | |
653 | [EXEC_DIRECTORY_CONFIGURATION] = { SD_PATH_USER_CONFIGURATION, NULL }, | |
654 | }; | |
655 | ||
656 | const struct table_entry *p; | |
657 | ExecDirectoryType i; | |
658 | int r; | |
659 | ||
660 | assert(m); | |
661 | ||
662 | if (MANAGER_IS_SYSTEM(m)) | |
663 | p = paths_system; | |
664 | else | |
665 | p = paths_user; | |
666 | ||
667 | for (i = 0; i < _EXEC_DIRECTORY_TYPE_MAX; i++) { | |
668 | r = sd_path_home(p[i].type, p[i].suffix, &m->prefix[i]); | |
669 | if (r < 0) | |
670 | return r; | |
671 | } | |
672 | ||
673 | return 0; | |
674 | } | |
675 | ||
676 | static int manager_setup_run_queue(Manager *m) { | |
677 | int r; | |
678 | ||
679 | assert(m); | |
680 | assert(!m->run_queue_event_source); | |
681 | ||
682 | r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m); | |
683 | if (r < 0) | |
684 | return r; | |
685 | ||
686 | r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE); | |
687 | if (r < 0) | |
688 | return r; | |
689 | ||
690 | r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF); | |
691 | if (r < 0) | |
692 | return r; | |
693 | ||
694 | (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue"); | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
699 | static int manager_setup_sigchld_event_source(Manager *m) { | |
700 | int r; | |
701 | ||
702 | assert(m); | |
703 | assert(!m->sigchld_event_source); | |
704 | ||
705 | r = sd_event_add_defer(m->event, &m->sigchld_event_source, manager_dispatch_sigchld, m); | |
706 | if (r < 0) | |
707 | return r; | |
708 | ||
709 | r = sd_event_source_set_priority(m->sigchld_event_source, SD_EVENT_PRIORITY_NORMAL-7); | |
710 | if (r < 0) | |
711 | return r; | |
712 | ||
713 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF); | |
714 | if (r < 0) | |
715 | return r; | |
716 | ||
717 | (void) sd_event_source_set_description(m->sigchld_event_source, "manager-sigchld"); | |
718 | ||
719 | return 0; | |
720 | } | |
721 | ||
722 | int manager_new(UnitFileScope scope, ManagerTestRunFlags test_run_flags, Manager **_m) { | |
723 | _cleanup_(manager_freep) Manager *m = NULL; | |
724 | int r; | |
725 | ||
726 | assert(_m); | |
727 | assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER)); | |
728 | ||
729 | m = new(Manager, 1); | |
730 | if (!m) | |
731 | return -ENOMEM; | |
732 | ||
733 | *m = (Manager) { | |
734 | .unit_file_scope = scope, | |
735 | .objective = _MANAGER_OBJECTIVE_INVALID, | |
736 | ||
737 | .default_timer_accuracy_usec = USEC_PER_MINUTE, | |
738 | .default_memory_accounting = MEMORY_ACCOUNTING_DEFAULT, | |
739 | .default_tasks_accounting = true, | |
740 | .default_tasks_max = UINT64_MAX, | |
741 | .default_timeout_start_usec = DEFAULT_TIMEOUT_USEC, | |
742 | .default_timeout_stop_usec = DEFAULT_TIMEOUT_USEC, | |
743 | .default_restart_usec = DEFAULT_RESTART_USEC, | |
744 | ||
745 | .original_log_level = -1, | |
746 | .original_log_target = _LOG_TARGET_INVALID, | |
747 | ||
748 | .notify_fd = -1, | |
749 | .cgroups_agent_fd = -1, | |
750 | .signal_fd = -1, | |
751 | .time_change_fd = -1, | |
752 | .user_lookup_fds = { -1, -1 }, | |
753 | .private_listen_fd = -1, | |
754 | .dev_autofs_fd = -1, | |
755 | .cgroup_inotify_fd = -1, | |
756 | .pin_cgroupfs_fd = -1, | |
757 | .ask_password_inotify_fd = -1, | |
758 | .idle_pipe = { -1, -1, -1, -1}, | |
759 | ||
760 | /* start as id #1, so that we can leave #0 around as "null-like" value */ | |
761 | .current_job_id = 1, | |
762 | ||
763 | .have_ask_password = -EINVAL, /* we don't know */ | |
764 | .first_boot = -1, | |
765 | .test_run_flags = test_run_flags, | |
766 | ||
767 | .default_oom_policy = OOM_STOP, | |
768 | }; | |
769 | ||
770 | #if ENABLE_EFI | |
771 | if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) | |
772 | boot_timestamps(m->timestamps + MANAGER_TIMESTAMP_USERSPACE, | |
773 | m->timestamps + MANAGER_TIMESTAMP_FIRMWARE, | |
774 | m->timestamps + MANAGER_TIMESTAMP_LOADER); | |
775 | #endif | |
776 | ||
777 | /* Prepare log fields we can use for structured logging */ | |
778 | if (MANAGER_IS_SYSTEM(m)) { | |
779 | m->unit_log_field = "UNIT="; | |
780 | m->unit_log_format_string = "UNIT=%s"; | |
781 | ||
782 | m->invocation_log_field = "INVOCATION_ID="; | |
783 | m->invocation_log_format_string = "INVOCATION_ID=%s"; | |
784 | } else { | |
785 | m->unit_log_field = "USER_UNIT="; | |
786 | m->unit_log_format_string = "USER_UNIT=%s"; | |
787 | ||
788 | m->invocation_log_field = "USER_INVOCATION_ID="; | |
789 | m->invocation_log_format_string = "USER_INVOCATION_ID=%s"; | |
790 | } | |
791 | ||
792 | /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */ | |
793 | RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7); | |
794 | ||
795 | r = manager_default_environment(m); | |
796 | if (r < 0) | |
797 | return r; | |
798 | ||
799 | r = hashmap_ensure_allocated(&m->units, &string_hash_ops); | |
800 | if (r < 0) | |
801 | return r; | |
802 | ||
803 | r = hashmap_ensure_allocated(&m->jobs, NULL); | |
804 | if (r < 0) | |
805 | return r; | |
806 | ||
807 | r = hashmap_ensure_allocated(&m->cgroup_unit, &path_hash_ops); | |
808 | if (r < 0) | |
809 | return r; | |
810 | ||
811 | r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops); | |
812 | if (r < 0) | |
813 | return r; | |
814 | ||
815 | r = manager_setup_prefix(m); | |
816 | if (r < 0) | |
817 | return r; | |
818 | ||
819 | r = sd_event_default(&m->event); | |
820 | if (r < 0) | |
821 | return r; | |
822 | ||
823 | r = manager_setup_run_queue(m); | |
824 | if (r < 0) | |
825 | return r; | |
826 | ||
827 | if (test_run_flags == MANAGER_TEST_RUN_MINIMAL) { | |
828 | m->cgroup_root = strdup(""); | |
829 | if (!m->cgroup_root) | |
830 | return -ENOMEM; | |
831 | } else { | |
832 | r = manager_setup_signals(m); | |
833 | if (r < 0) | |
834 | return r; | |
835 | ||
836 | r = manager_setup_cgroup(m); | |
837 | if (r < 0) | |
838 | return r; | |
839 | ||
840 | r = manager_setup_time_change(m); | |
841 | if (r < 0) | |
842 | return r; | |
843 | ||
844 | r = manager_read_timezone_stat(m); | |
845 | if (r < 0) | |
846 | return r; | |
847 | ||
848 | (void) manager_setup_timezone_change(m); | |
849 | ||
850 | r = manager_setup_sigchld_event_source(m); | |
851 | if (r < 0) | |
852 | return r; | |
853 | } | |
854 | ||
855 | if (MANAGER_IS_SYSTEM(m) && test_run_flags == 0) { | |
856 | r = mkdir_label("/run/systemd/units", 0755); | |
857 | if (r < 0 && r != -EEXIST) | |
858 | return r; | |
859 | } | |
860 | ||
861 | m->taint_usr = | |
862 | !in_initrd() && | |
863 | dir_is_empty("/usr") > 0; | |
864 | ||
865 | /* Note that we do not set up the notify fd here. We do that after deserialization, | |
866 | * since they might have gotten serialized across the reexec. */ | |
867 | ||
868 | *_m = TAKE_PTR(m); | |
869 | ||
870 | return 0; | |
871 | } | |
872 | ||
873 | static int manager_setup_notify(Manager *m) { | |
874 | int r; | |
875 | ||
876 | if (MANAGER_IS_TEST_RUN(m)) | |
877 | return 0; | |
878 | ||
879 | if (m->notify_fd < 0) { | |
880 | _cleanup_close_ int fd = -1; | |
881 | union sockaddr_union sa = {}; | |
882 | int salen; | |
883 | ||
884 | /* First free all secondary fields */ | |
885 | m->notify_socket = mfree(m->notify_socket); | |
886 | m->notify_event_source = sd_event_source_unref(m->notify_event_source); | |
887 | ||
888 | fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
889 | if (fd < 0) | |
890 | return log_error_errno(errno, "Failed to allocate notification socket: %m"); | |
891 | ||
892 | fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE); | |
893 | ||
894 | m->notify_socket = strappend(m->prefix[EXEC_DIRECTORY_RUNTIME], "/systemd/notify"); | |
895 | if (!m->notify_socket) | |
896 | return log_oom(); | |
897 | ||
898 | salen = sockaddr_un_set_path(&sa.un, m->notify_socket); | |
899 | if (salen < 0) | |
900 | return log_error_errno(salen, "Notify socket '%s' not valid for AF_UNIX socket address, refusing.", m->notify_socket); | |
901 | ||
902 | (void) mkdir_parents_label(m->notify_socket, 0755); | |
903 | (void) sockaddr_un_unlink(&sa.un); | |
904 | ||
905 | r = bind(fd, &sa.sa, salen); | |
906 | if (r < 0) | |
907 | return log_error_errno(errno, "bind(%s) failed: %m", m->notify_socket); | |
908 | ||
909 | r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true); | |
910 | if (r < 0) | |
911 | return log_error_errno(r, "SO_PASSCRED failed: %m"); | |
912 | ||
913 | m->notify_fd = TAKE_FD(fd); | |
914 | ||
915 | log_debug("Using notification socket %s", m->notify_socket); | |
916 | } | |
917 | ||
918 | if (!m->notify_event_source) { | |
919 | r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m); | |
920 | if (r < 0) | |
921 | return log_error_errno(r, "Failed to allocate notify event source: %m"); | |
922 | ||
923 | /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which | |
924 | * service an exit message belongs. */ | |
925 | r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-8); | |
926 | if (r < 0) | |
927 | return log_error_errno(r, "Failed to set priority of notify event source: %m"); | |
928 | ||
929 | (void) sd_event_source_set_description(m->notify_event_source, "manager-notify"); | |
930 | } | |
931 | ||
932 | return 0; | |
933 | } | |
934 | ||
935 | static int manager_setup_cgroups_agent(Manager *m) { | |
936 | ||
937 | static const union sockaddr_union sa = { | |
938 | .un.sun_family = AF_UNIX, | |
939 | .un.sun_path = "/run/systemd/cgroups-agent", | |
940 | }; | |
941 | int r; | |
942 | ||
943 | /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering | |
944 | * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and | |
945 | * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on | |
946 | * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number | |
947 | * of D-Bus connections may be queued until the kernel will start dropping further incoming connections, | |
948 | * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX | |
949 | * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and | |
950 | * we thus won't lose messages. | |
951 | * | |
952 | * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen | |
953 | * to it. The system instance hence listens on this special socket, but the user instances listen on the system | |
954 | * bus for these messages. */ | |
955 | ||
956 | if (MANAGER_IS_TEST_RUN(m)) | |
957 | return 0; | |
958 | ||
959 | if (!MANAGER_IS_SYSTEM(m)) | |
960 | return 0; | |
961 | ||
962 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
963 | if (r < 0) | |
964 | return log_error_errno(r, "Failed to determine whether unified cgroups hierarchy is used: %m"); | |
965 | if (r > 0) /* We don't need this anymore on the unified hierarchy */ | |
966 | return 0; | |
967 | ||
968 | if (m->cgroups_agent_fd < 0) { | |
969 | _cleanup_close_ int fd = -1; | |
970 | ||
971 | /* First free all secondary fields */ | |
972 | m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source); | |
973 | ||
974 | fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
975 | if (fd < 0) | |
976 | return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m"); | |
977 | ||
978 | fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE); | |
979 | ||
980 | (void) sockaddr_un_unlink(&sa.un); | |
981 | ||
982 | /* Only allow root to connect to this socket */ | |
983 | RUN_WITH_UMASK(0077) | |
984 | r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)); | |
985 | if (r < 0) | |
986 | return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path); | |
987 | ||
988 | m->cgroups_agent_fd = TAKE_FD(fd); | |
989 | } | |
990 | ||
991 | if (!m->cgroups_agent_event_source) { | |
992 | r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m); | |
993 | if (r < 0) | |
994 | return log_error_errno(r, "Failed to allocate cgroups agent event source: %m"); | |
995 | ||
996 | /* Process cgroups notifications early. Note that when the agent notification is received | |
997 | * we'll just enqueue the unit in the cgroup empty queue, hence pick a high priority than | |
998 | * that. Also see handling of cgroup inotify for the unified cgroup stuff. */ | |
999 | r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
1000 | if (r < 0) | |
1001 | return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m"); | |
1002 | ||
1003 | (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent"); | |
1004 | } | |
1005 | ||
1006 | return 0; | |
1007 | } | |
1008 | ||
1009 | static int manager_setup_user_lookup_fd(Manager *m) { | |
1010 | int r; | |
1011 | ||
1012 | assert(m); | |
1013 | ||
1014 | /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID | |
1015 | * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation, | |
1016 | * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked | |
1017 | * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects | |
1018 | * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes, | |
1019 | * hence we establish this communication channel so that forked off processes can pass their UID/GID | |
1020 | * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple | |
1021 | * datagram, along with their unit name, so that we can share one communication socket pair among all units for | |
1022 | * this purpose. | |
1023 | * | |
1024 | * You might wonder why we need a communication channel for this that is independent of the usual notification | |
1025 | * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET | |
1026 | * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user | |
1027 | * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available. | |
1028 | * | |
1029 | * Note that this function is called under two circumstances: when we first initialize (in which case we | |
1030 | * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload | |
1031 | * (in which case the socket pair already exists but we still need to allocate the event source for it). */ | |
1032 | ||
1033 | if (m->user_lookup_fds[0] < 0) { | |
1034 | ||
1035 | /* Free all secondary fields */ | |
1036 | safe_close_pair(m->user_lookup_fds); | |
1037 | m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source); | |
1038 | ||
1039 | if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0) | |
1040 | return log_error_errno(errno, "Failed to allocate user lookup socket: %m"); | |
1041 | ||
1042 | (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE); | |
1043 | } | |
1044 | ||
1045 | if (!m->user_lookup_event_source) { | |
1046 | r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m); | |
1047 | if (r < 0) | |
1048 | return log_error_errno(errno, "Failed to allocate user lookup event source: %m"); | |
1049 | ||
1050 | /* Process even earlier than the notify event source, so that we always know first about valid UID/GID | |
1051 | * resolutions */ | |
1052 | r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-11); | |
1053 | if (r < 0) | |
1054 | return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m"); | |
1055 | ||
1056 | (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup"); | |
1057 | } | |
1058 | ||
1059 | return 0; | |
1060 | } | |
1061 | ||
1062 | static unsigned manager_dispatch_cleanup_queue(Manager *m) { | |
1063 | Unit *u; | |
1064 | unsigned n = 0; | |
1065 | ||
1066 | assert(m); | |
1067 | ||
1068 | while ((u = m->cleanup_queue)) { | |
1069 | assert(u->in_cleanup_queue); | |
1070 | ||
1071 | unit_free(u); | |
1072 | n++; | |
1073 | } | |
1074 | ||
1075 | return n; | |
1076 | } | |
1077 | ||
1078 | enum { | |
1079 | GC_OFFSET_IN_PATH, /* This one is on the path we were traveling */ | |
1080 | GC_OFFSET_UNSURE, /* No clue */ | |
1081 | GC_OFFSET_GOOD, /* We still need this unit */ | |
1082 | GC_OFFSET_BAD, /* We don't need this unit anymore */ | |
1083 | _GC_OFFSET_MAX | |
1084 | }; | |
1085 | ||
1086 | static void unit_gc_mark_good(Unit *u, unsigned gc_marker) { | |
1087 | Unit *other; | |
1088 | Iterator i; | |
1089 | void *v; | |
1090 | ||
1091 | u->gc_marker = gc_marker + GC_OFFSET_GOOD; | |
1092 | ||
1093 | /* Recursively mark referenced units as GOOD as well */ | |
1094 | HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REFERENCES], i) | |
1095 | if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE) | |
1096 | unit_gc_mark_good(other, gc_marker); | |
1097 | } | |
1098 | ||
1099 | static void unit_gc_sweep(Unit *u, unsigned gc_marker) { | |
1100 | Unit *other; | |
1101 | bool is_bad; | |
1102 | Iterator i; | |
1103 | void *v; | |
1104 | ||
1105 | assert(u); | |
1106 | ||
1107 | if (IN_SET(u->gc_marker - gc_marker, | |
1108 | GC_OFFSET_GOOD, GC_OFFSET_BAD, GC_OFFSET_UNSURE, GC_OFFSET_IN_PATH)) | |
1109 | return; | |
1110 | ||
1111 | if (u->in_cleanup_queue) | |
1112 | goto bad; | |
1113 | ||
1114 | if (!unit_may_gc(u)) | |
1115 | goto good; | |
1116 | ||
1117 | u->gc_marker = gc_marker + GC_OFFSET_IN_PATH; | |
1118 | ||
1119 | is_bad = true; | |
1120 | ||
1121 | HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REFERENCED_BY], i) { | |
1122 | unit_gc_sweep(other, gc_marker); | |
1123 | ||
1124 | if (other->gc_marker == gc_marker + GC_OFFSET_GOOD) | |
1125 | goto good; | |
1126 | ||
1127 | if (other->gc_marker != gc_marker + GC_OFFSET_BAD) | |
1128 | is_bad = false; | |
1129 | } | |
1130 | ||
1131 | if (u->refs_by_target) { | |
1132 | const UnitRef *ref; | |
1133 | ||
1134 | LIST_FOREACH(refs_by_target, ref, u->refs_by_target) { | |
1135 | unit_gc_sweep(ref->source, gc_marker); | |
1136 | ||
1137 | if (ref->source->gc_marker == gc_marker + GC_OFFSET_GOOD) | |
1138 | goto good; | |
1139 | ||
1140 | if (ref->source->gc_marker != gc_marker + GC_OFFSET_BAD) | |
1141 | is_bad = false; | |
1142 | } | |
1143 | } | |
1144 | ||
1145 | if (is_bad) | |
1146 | goto bad; | |
1147 | ||
1148 | /* We were unable to find anything out about this entry, so | |
1149 | * let's investigate it later */ | |
1150 | u->gc_marker = gc_marker + GC_OFFSET_UNSURE; | |
1151 | unit_add_to_gc_queue(u); | |
1152 | return; | |
1153 | ||
1154 | bad: | |
1155 | /* We definitely know that this one is not useful anymore, so | |
1156 | * let's mark it for deletion */ | |
1157 | u->gc_marker = gc_marker + GC_OFFSET_BAD; | |
1158 | unit_add_to_cleanup_queue(u); | |
1159 | return; | |
1160 | ||
1161 | good: | |
1162 | unit_gc_mark_good(u, gc_marker); | |
1163 | } | |
1164 | ||
1165 | static unsigned manager_dispatch_gc_unit_queue(Manager *m) { | |
1166 | unsigned n = 0, gc_marker; | |
1167 | Unit *u; | |
1168 | ||
1169 | assert(m); | |
1170 | ||
1171 | /* log_debug("Running GC..."); */ | |
1172 | ||
1173 | m->gc_marker += _GC_OFFSET_MAX; | |
1174 | if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX) | |
1175 | m->gc_marker = 1; | |
1176 | ||
1177 | gc_marker = m->gc_marker; | |
1178 | ||
1179 | while ((u = m->gc_unit_queue)) { | |
1180 | assert(u->in_gc_queue); | |
1181 | ||
1182 | unit_gc_sweep(u, gc_marker); | |
1183 | ||
1184 | LIST_REMOVE(gc_queue, m->gc_unit_queue, u); | |
1185 | u->in_gc_queue = false; | |
1186 | ||
1187 | n++; | |
1188 | ||
1189 | if (IN_SET(u->gc_marker - gc_marker, | |
1190 | GC_OFFSET_BAD, GC_OFFSET_UNSURE)) { | |
1191 | if (u->id) | |
1192 | log_unit_debug(u, "Collecting."); | |
1193 | u->gc_marker = gc_marker + GC_OFFSET_BAD; | |
1194 | unit_add_to_cleanup_queue(u); | |
1195 | } | |
1196 | } | |
1197 | ||
1198 | return n; | |
1199 | } | |
1200 | ||
1201 | static unsigned manager_dispatch_gc_job_queue(Manager *m) { | |
1202 | unsigned n = 0; | |
1203 | Job *j; | |
1204 | ||
1205 | assert(m); | |
1206 | ||
1207 | while ((j = m->gc_job_queue)) { | |
1208 | assert(j->in_gc_queue); | |
1209 | ||
1210 | LIST_REMOVE(gc_queue, m->gc_job_queue, j); | |
1211 | j->in_gc_queue = false; | |
1212 | ||
1213 | n++; | |
1214 | ||
1215 | if (!job_may_gc(j)) | |
1216 | continue; | |
1217 | ||
1218 | log_unit_debug(j->unit, "Collecting job."); | |
1219 | (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false); | |
1220 | } | |
1221 | ||
1222 | return n; | |
1223 | } | |
1224 | ||
1225 | static unsigned manager_dispatch_stop_when_unneeded_queue(Manager *m) { | |
1226 | unsigned n = 0; | |
1227 | Unit *u; | |
1228 | int r; | |
1229 | ||
1230 | assert(m); | |
1231 | ||
1232 | while ((u = m->stop_when_unneeded_queue)) { | |
1233 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1234 | assert(m->stop_when_unneeded_queue); | |
1235 | ||
1236 | assert(u->in_stop_when_unneeded_queue); | |
1237 | LIST_REMOVE(stop_when_unneeded_queue, m->stop_when_unneeded_queue, u); | |
1238 | u->in_stop_when_unneeded_queue = false; | |
1239 | ||
1240 | n++; | |
1241 | ||
1242 | if (!unit_is_unneeded(u)) | |
1243 | continue; | |
1244 | ||
1245 | log_unit_debug(u, "Unit is not needed anymore."); | |
1246 | ||
1247 | /* If stopping a unit fails continuously we might enter a stop loop here, hence stop acting on the | |
1248 | * service being unnecessary after a while. */ | |
1249 | ||
1250 | if (!ratelimit_below(&u->auto_stop_ratelimit)) { | |
1251 | log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently."); | |
1252 | continue; | |
1253 | } | |
1254 | ||
1255 | /* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */ | |
1256 | r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL); | |
1257 | if (r < 0) | |
1258 | log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r)); | |
1259 | } | |
1260 | ||
1261 | return n; | |
1262 | } | |
1263 | ||
1264 | static void manager_clear_jobs_and_units(Manager *m) { | |
1265 | Unit *u; | |
1266 | ||
1267 | assert(m); | |
1268 | ||
1269 | while ((u = hashmap_first(m->units))) | |
1270 | unit_free(u); | |
1271 | ||
1272 | manager_dispatch_cleanup_queue(m); | |
1273 | ||
1274 | assert(!m->load_queue); | |
1275 | assert(!m->run_queue); | |
1276 | assert(!m->dbus_unit_queue); | |
1277 | assert(!m->dbus_job_queue); | |
1278 | assert(!m->cleanup_queue); | |
1279 | assert(!m->gc_unit_queue); | |
1280 | assert(!m->gc_job_queue); | |
1281 | assert(!m->stop_when_unneeded_queue); | |
1282 | ||
1283 | assert(hashmap_isempty(m->jobs)); | |
1284 | assert(hashmap_isempty(m->units)); | |
1285 | ||
1286 | m->n_on_console = 0; | |
1287 | m->n_running_jobs = 0; | |
1288 | m->n_installed_jobs = 0; | |
1289 | m->n_failed_jobs = 0; | |
1290 | } | |
1291 | ||
1292 | Manager* manager_free(Manager *m) { | |
1293 | ExecDirectoryType dt; | |
1294 | UnitType c; | |
1295 | ||
1296 | if (!m) | |
1297 | return NULL; | |
1298 | ||
1299 | manager_clear_jobs_and_units(m); | |
1300 | ||
1301 | for (c = 0; c < _UNIT_TYPE_MAX; c++) | |
1302 | if (unit_vtable[c]->shutdown) | |
1303 | unit_vtable[c]->shutdown(m); | |
1304 | ||
1305 | /* Keep the cgroup hierarchy in place except when we know we are going down for good */ | |
1306 | manager_shutdown_cgroup(m, IN_SET(m->objective, MANAGER_EXIT, MANAGER_REBOOT, MANAGER_POWEROFF, MANAGER_HALT, MANAGER_KEXEC)); | |
1307 | ||
1308 | lookup_paths_flush_generator(&m->lookup_paths); | |
1309 | ||
1310 | bus_done(m); | |
1311 | ||
1312 | exec_runtime_vacuum(m); | |
1313 | hashmap_free(m->exec_runtime_by_id); | |
1314 | ||
1315 | dynamic_user_vacuum(m, false); | |
1316 | hashmap_free(m->dynamic_users); | |
1317 | ||
1318 | hashmap_free(m->units); | |
1319 | hashmap_free(m->units_by_invocation_id); | |
1320 | hashmap_free(m->jobs); | |
1321 | hashmap_free(m->watch_pids); | |
1322 | hashmap_free(m->watch_bus); | |
1323 | ||
1324 | set_free(m->startup_units); | |
1325 | set_free(m->failed_units); | |
1326 | ||
1327 | sd_event_source_unref(m->signal_event_source); | |
1328 | sd_event_source_unref(m->sigchld_event_source); | |
1329 | sd_event_source_unref(m->notify_event_source); | |
1330 | sd_event_source_unref(m->cgroups_agent_event_source); | |
1331 | sd_event_source_unref(m->time_change_event_source); | |
1332 | sd_event_source_unref(m->timezone_change_event_source); | |
1333 | sd_event_source_unref(m->jobs_in_progress_event_source); | |
1334 | sd_event_source_unref(m->run_queue_event_source); | |
1335 | sd_event_source_unref(m->user_lookup_event_source); | |
1336 | sd_event_source_unref(m->sync_bus_names_event_source); | |
1337 | ||
1338 | safe_close(m->signal_fd); | |
1339 | safe_close(m->notify_fd); | |
1340 | safe_close(m->cgroups_agent_fd); | |
1341 | safe_close(m->time_change_fd); | |
1342 | safe_close_pair(m->user_lookup_fds); | |
1343 | ||
1344 | manager_close_ask_password(m); | |
1345 | ||
1346 | manager_close_idle_pipe(m); | |
1347 | ||
1348 | sd_event_unref(m->event); | |
1349 | ||
1350 | free(m->notify_socket); | |
1351 | ||
1352 | lookup_paths_free(&m->lookup_paths); | |
1353 | strv_free(m->transient_environment); | |
1354 | strv_free(m->client_environment); | |
1355 | ||
1356 | hashmap_free(m->cgroup_unit); | |
1357 | set_free_free(m->unit_path_cache); | |
1358 | ||
1359 | free(m->switch_root); | |
1360 | free(m->switch_root_init); | |
1361 | ||
1362 | rlimit_free_all(m->rlimit); | |
1363 | ||
1364 | assert(hashmap_isempty(m->units_requiring_mounts_for)); | |
1365 | hashmap_free(m->units_requiring_mounts_for); | |
1366 | ||
1367 | hashmap_free(m->uid_refs); | |
1368 | hashmap_free(m->gid_refs); | |
1369 | ||
1370 | for (dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) | |
1371 | m->prefix[dt] = mfree(m->prefix[dt]); | |
1372 | ||
1373 | return mfree(m); | |
1374 | } | |
1375 | ||
1376 | static void manager_enumerate_perpetual(Manager *m) { | |
1377 | UnitType c; | |
1378 | ||
1379 | assert(m); | |
1380 | ||
1381 | /* Let's ask every type to load all units from disk/kernel that it might know */ | |
1382 | for (c = 0; c < _UNIT_TYPE_MAX; c++) { | |
1383 | if (!unit_type_supported(c)) { | |
1384 | log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c)); | |
1385 | continue; | |
1386 | } | |
1387 | ||
1388 | if (unit_vtable[c]->enumerate_perpetual) | |
1389 | unit_vtable[c]->enumerate_perpetual(m); | |
1390 | } | |
1391 | } | |
1392 | ||
1393 | static void manager_enumerate(Manager *m) { | |
1394 | UnitType c; | |
1395 | ||
1396 | assert(m); | |
1397 | ||
1398 | /* Let's ask every type to load all units from disk/kernel that it might know */ | |
1399 | for (c = 0; c < _UNIT_TYPE_MAX; c++) { | |
1400 | if (!unit_type_supported(c)) { | |
1401 | log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c)); | |
1402 | continue; | |
1403 | } | |
1404 | ||
1405 | if (unit_vtable[c]->enumerate) | |
1406 | unit_vtable[c]->enumerate(m); | |
1407 | } | |
1408 | ||
1409 | manager_dispatch_load_queue(m); | |
1410 | } | |
1411 | ||
1412 | static void manager_coldplug(Manager *m) { | |
1413 | Iterator i; | |
1414 | Unit *u; | |
1415 | char *k; | |
1416 | int r; | |
1417 | ||
1418 | assert(m); | |
1419 | ||
1420 | log_debug("Invoking unit coldplug() handlers…"); | |
1421 | ||
1422 | /* Let's place the units back into their deserialized state */ | |
1423 | HASHMAP_FOREACH_KEY(u, k, m->units, i) { | |
1424 | ||
1425 | /* ignore aliases */ | |
1426 | if (u->id != k) | |
1427 | continue; | |
1428 | ||
1429 | r = unit_coldplug(u); | |
1430 | if (r < 0) | |
1431 | log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id); | |
1432 | } | |
1433 | } | |
1434 | ||
1435 | static void manager_catchup(Manager *m) { | |
1436 | Iterator i; | |
1437 | Unit *u; | |
1438 | char *k; | |
1439 | ||
1440 | assert(m); | |
1441 | ||
1442 | log_debug("Invoking unit catchup() handlers…"); | |
1443 | ||
1444 | /* Let's catch up on any state changes that happened while we were reloading/reexecing */ | |
1445 | HASHMAP_FOREACH_KEY(u, k, m->units, i) { | |
1446 | ||
1447 | /* ignore aliases */ | |
1448 | if (u->id != k) | |
1449 | continue; | |
1450 | ||
1451 | unit_catchup(u); | |
1452 | } | |
1453 | } | |
1454 | ||
1455 | static void manager_build_unit_path_cache(Manager *m) { | |
1456 | char **i; | |
1457 | int r; | |
1458 | ||
1459 | assert(m); | |
1460 | ||
1461 | set_free_free(m->unit_path_cache); | |
1462 | ||
1463 | m->unit_path_cache = set_new(&path_hash_ops); | |
1464 | if (!m->unit_path_cache) { | |
1465 | r = -ENOMEM; | |
1466 | goto fail; | |
1467 | } | |
1468 | ||
1469 | /* This simply builds a list of files we know exist, so that | |
1470 | * we don't always have to go to disk */ | |
1471 | ||
1472 | STRV_FOREACH(i, m->lookup_paths.search_path) { | |
1473 | _cleanup_closedir_ DIR *d = NULL; | |
1474 | struct dirent *de; | |
1475 | ||
1476 | d = opendir(*i); | |
1477 | if (!d) { | |
1478 | if (errno != ENOENT) | |
1479 | log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i); | |
1480 | continue; | |
1481 | } | |
1482 | ||
1483 | FOREACH_DIRENT(de, d, r = -errno; goto fail) { | |
1484 | char *p; | |
1485 | ||
1486 | p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name); | |
1487 | if (!p) { | |
1488 | r = -ENOMEM; | |
1489 | goto fail; | |
1490 | } | |
1491 | ||
1492 | r = set_consume(m->unit_path_cache, p); | |
1493 | if (r < 0) | |
1494 | goto fail; | |
1495 | } | |
1496 | } | |
1497 | ||
1498 | return; | |
1499 | ||
1500 | fail: | |
1501 | log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m"); | |
1502 | m->unit_path_cache = set_free_free(m->unit_path_cache); | |
1503 | } | |
1504 | ||
1505 | static void manager_distribute_fds(Manager *m, FDSet *fds) { | |
1506 | Iterator i; | |
1507 | Unit *u; | |
1508 | ||
1509 | assert(m); | |
1510 | ||
1511 | HASHMAP_FOREACH(u, m->units, i) { | |
1512 | ||
1513 | if (fdset_size(fds) <= 0) | |
1514 | break; | |
1515 | ||
1516 | if (!UNIT_VTABLE(u)->distribute_fds) | |
1517 | continue; | |
1518 | ||
1519 | UNIT_VTABLE(u)->distribute_fds(u, fds); | |
1520 | } | |
1521 | } | |
1522 | ||
1523 | static bool manager_dbus_is_running(Manager *m, bool deserialized) { | |
1524 | Unit *u; | |
1525 | ||
1526 | assert(m); | |
1527 | ||
1528 | /* This checks whether the dbus instance we are supposed to expose our APIs on is up. We check both the socket | |
1529 | * and the service unit. If the 'deserialized' parameter is true we'll check the deserialized state of the unit | |
1530 | * rather than the current one. */ | |
1531 | ||
1532 | if (MANAGER_IS_TEST_RUN(m)) | |
1533 | return false; | |
1534 | ||
1535 | u = manager_get_unit(m, SPECIAL_DBUS_SOCKET); | |
1536 | if (!u) | |
1537 | return false; | |
1538 | if ((deserialized ? SOCKET(u)->deserialized_state : SOCKET(u)->state) != SOCKET_RUNNING) | |
1539 | return false; | |
1540 | ||
1541 | u = manager_get_unit(m, SPECIAL_DBUS_SERVICE); | |
1542 | if (!u) | |
1543 | return false; | |
1544 | if (!IN_SET((deserialized ? SERVICE(u)->deserialized_state : SERVICE(u)->state), SERVICE_RUNNING, SERVICE_RELOAD)) | |
1545 | return false; | |
1546 | ||
1547 | return true; | |
1548 | } | |
1549 | ||
1550 | static void manager_setup_bus(Manager *m) { | |
1551 | assert(m); | |
1552 | ||
1553 | /* Let's set up our private bus connection now, unconditionally */ | |
1554 | (void) bus_init_private(m); | |
1555 | ||
1556 | /* If we are in --user mode also connect to the system bus now */ | |
1557 | if (MANAGER_IS_USER(m)) | |
1558 | (void) bus_init_system(m); | |
1559 | ||
1560 | /* Let's connect to the bus now, but only if the unit is supposed to be up */ | |
1561 | if (manager_dbus_is_running(m, MANAGER_IS_RELOADING(m))) { | |
1562 | (void) bus_init_api(m); | |
1563 | ||
1564 | if (MANAGER_IS_SYSTEM(m)) | |
1565 | (void) bus_init_system(m); | |
1566 | } | |
1567 | } | |
1568 | ||
1569 | static void manager_preset_all(Manager *m) { | |
1570 | int r; | |
1571 | ||
1572 | assert(m); | |
1573 | ||
1574 | if (m->first_boot <= 0) | |
1575 | return; | |
1576 | ||
1577 | if (!MANAGER_IS_SYSTEM(m)) | |
1578 | return; | |
1579 | ||
1580 | if (MANAGER_IS_TEST_RUN(m)) | |
1581 | return; | |
1582 | ||
1583 | /* If this is the first boot, and we are in the host system, then preset everything */ | |
1584 | r = unit_file_preset_all(UNIT_FILE_SYSTEM, 0, NULL, UNIT_FILE_PRESET_ENABLE_ONLY, NULL, 0); | |
1585 | if (r < 0) | |
1586 | log_full_errno(r == -EEXIST ? LOG_NOTICE : LOG_WARNING, r, | |
1587 | "Failed to populate /etc with preset unit settings, ignoring: %m"); | |
1588 | else | |
1589 | log_info("Populated /etc with preset unit settings."); | |
1590 | } | |
1591 | ||
1592 | static void manager_vacuum(Manager *m) { | |
1593 | assert(m); | |
1594 | ||
1595 | /* Release any dynamic users no longer referenced */ | |
1596 | dynamic_user_vacuum(m, true); | |
1597 | ||
1598 | /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */ | |
1599 | manager_vacuum_uid_refs(m); | |
1600 | manager_vacuum_gid_refs(m); | |
1601 | ||
1602 | /* Release any runtimes no longer referenced */ | |
1603 | exec_runtime_vacuum(m); | |
1604 | } | |
1605 | ||
1606 | static void manager_ready(Manager *m) { | |
1607 | assert(m); | |
1608 | ||
1609 | /* After having loaded everything, do the final round of catching up with what might have changed */ | |
1610 | ||
1611 | m->objective = MANAGER_OK; /* Tell everyone we are up now */ | |
1612 | ||
1613 | /* It might be safe to log to the journal now and connect to dbus */ | |
1614 | manager_recheck_journal(m); | |
1615 | manager_recheck_dbus(m); | |
1616 | ||
1617 | /* Sync current state of bus names with our set of listening units */ | |
1618 | (void) manager_enqueue_sync_bus_names(m); | |
1619 | ||
1620 | /* Let's finally catch up with any changes that took place while we were reloading/reexecing */ | |
1621 | manager_catchup(m); | |
1622 | ||
1623 | m->honor_device_enumeration = true; | |
1624 | } | |
1625 | ||
1626 | static Manager* manager_reloading_start(Manager *m) { | |
1627 | m->n_reloading++; | |
1628 | return m; | |
1629 | } | |
1630 | static void manager_reloading_stopp(Manager **m) { | |
1631 | if (*m) { | |
1632 | assert((*m)->n_reloading > 0); | |
1633 | (*m)->n_reloading--; | |
1634 | } | |
1635 | } | |
1636 | ||
1637 | int manager_startup(Manager *m, FILE *serialization, FDSet *fds) { | |
1638 | int r; | |
1639 | ||
1640 | assert(m); | |
1641 | ||
1642 | /* If we are running in test mode, we still want to run the generators, | |
1643 | * but we should not touch the real generator directories. */ | |
1644 | r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, | |
1645 | MANAGER_IS_TEST_RUN(m) ? LOOKUP_PATHS_TEMPORARY_GENERATED : 0, | |
1646 | NULL); | |
1647 | if (r < 0) | |
1648 | return log_error_errno(r, "Failed to initialize path lookup table: %m"); | |
1649 | ||
1650 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_START)); | |
1651 | r = manager_run_environment_generators(m); | |
1652 | if (r >= 0) | |
1653 | r = manager_run_generators(m); | |
1654 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_GENERATORS_FINISH)); | |
1655 | if (r < 0) | |
1656 | return r; | |
1657 | ||
1658 | manager_preset_all(m); | |
1659 | ||
1660 | r = lookup_paths_reduce(&m->lookup_paths); | |
1661 | if (r < 0) | |
1662 | log_warning_errno(r, "Failed ot reduce unit file paths, ignoring: %m"); | |
1663 | ||
1664 | manager_build_unit_path_cache(m); | |
1665 | ||
1666 | { | |
1667 | /* This block is (optionally) done with the reloading counter bumped */ | |
1668 | _cleanup_(manager_reloading_stopp) Manager *reloading = NULL; | |
1669 | ||
1670 | /* If we will deserialize make sure that during enumeration this is already known, so we increase the | |
1671 | * counter here already */ | |
1672 | if (serialization) | |
1673 | reloading = manager_reloading_start(m); | |
1674 | ||
1675 | /* First, enumerate what we can from all config files */ | |
1676 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_START)); | |
1677 | manager_enumerate_perpetual(m); | |
1678 | manager_enumerate(m); | |
1679 | dual_timestamp_get(m->timestamps + manager_timestamp_initrd_mangle(MANAGER_TIMESTAMP_UNITS_LOAD_FINISH)); | |
1680 | ||
1681 | /* Second, deserialize if there is something to deserialize */ | |
1682 | if (serialization) { | |
1683 | r = manager_deserialize(m, serialization, fds); | |
1684 | if (r < 0) | |
1685 | return log_error_errno(r, "Deserialization failed: %m"); | |
1686 | } | |
1687 | ||
1688 | /* Any fds left? Find some unit which wants them. This is useful to allow container managers to pass | |
1689 | * some file descriptors to us pre-initialized. This enables socket-based activation of entire | |
1690 | * containers. */ | |
1691 | manager_distribute_fds(m, fds); | |
1692 | ||
1693 | /* We might have deserialized the notify fd, but if we didn't then let's create the bus now */ | |
1694 | r = manager_setup_notify(m); | |
1695 | if (r < 0) | |
1696 | /* No sense to continue without notifications, our children would fail anyway. */ | |
1697 | return r; | |
1698 | ||
1699 | r = manager_setup_cgroups_agent(m); | |
1700 | if (r < 0) | |
1701 | /* Likewise, no sense to continue without empty cgroup notifications. */ | |
1702 | return r; | |
1703 | ||
1704 | r = manager_setup_user_lookup_fd(m); | |
1705 | if (r < 0) | |
1706 | /* This shouldn't fail, except if things are really broken. */ | |
1707 | return r; | |
1708 | ||
1709 | /* Connect to the bus if we are good for it */ | |
1710 | manager_setup_bus(m); | |
1711 | ||
1712 | /* Now that we are connected to all possible busses, let's deserialize who is tracking us. */ | |
1713 | r = bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed); | |
1714 | if (r < 0) | |
1715 | log_warning_errno(r, "Failed to deserialized tracked clients, ignoring: %m"); | |
1716 | m->deserialized_subscribed = strv_free(m->deserialized_subscribed); | |
1717 | ||
1718 | /* Third, fire things up! */ | |
1719 | manager_coldplug(m); | |
1720 | ||
1721 | /* Clean up runtime objects */ | |
1722 | manager_vacuum(m); | |
1723 | ||
1724 | if (serialization) | |
1725 | /* Let's wait for the UnitNew/JobNew messages being sent, before we notify that the | |
1726 | * reload is finished */ | |
1727 | m->send_reloading_done = true; | |
1728 | } | |
1729 | ||
1730 | manager_ready(m); | |
1731 | ||
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | int manager_add_job( | |
1736 | Manager *m, | |
1737 | JobType type, | |
1738 | Unit *unit, | |
1739 | JobMode mode, | |
1740 | Set *affected_jobs, | |
1741 | sd_bus_error *error, | |
1742 | Job **ret) { | |
1743 | ||
1744 | Transaction *tr; | |
1745 | int r; | |
1746 | ||
1747 | assert(m); | |
1748 | assert(type < _JOB_TYPE_MAX); | |
1749 | assert(unit); | |
1750 | assert(mode < _JOB_MODE_MAX); | |
1751 | ||
1752 | if (mode == JOB_ISOLATE && type != JOB_START) | |
1753 | return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start."); | |
1754 | ||
1755 | if (mode == JOB_ISOLATE && !unit->allow_isolate) | |
1756 | return sd_bus_error_setf(error, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated."); | |
1757 | ||
1758 | log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode)); | |
1759 | ||
1760 | type = job_type_collapse(type, unit); | |
1761 | ||
1762 | tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY); | |
1763 | if (!tr) | |
1764 | return -ENOMEM; | |
1765 | ||
1766 | r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false, | |
1767 | IN_SET(mode, JOB_IGNORE_DEPENDENCIES, JOB_IGNORE_REQUIREMENTS), | |
1768 | mode == JOB_IGNORE_DEPENDENCIES, error); | |
1769 | if (r < 0) | |
1770 | goto tr_abort; | |
1771 | ||
1772 | if (mode == JOB_ISOLATE) { | |
1773 | r = transaction_add_isolate_jobs(tr, m); | |
1774 | if (r < 0) | |
1775 | goto tr_abort; | |
1776 | } | |
1777 | ||
1778 | r = transaction_activate(tr, m, mode, affected_jobs, error); | |
1779 | if (r < 0) | |
1780 | goto tr_abort; | |
1781 | ||
1782 | log_unit_debug(unit, | |
1783 | "Enqueued job %s/%s as %u", unit->id, | |
1784 | job_type_to_string(type), (unsigned) tr->anchor_job->id); | |
1785 | ||
1786 | if (ret) | |
1787 | *ret = tr->anchor_job; | |
1788 | ||
1789 | transaction_free(tr); | |
1790 | return 0; | |
1791 | ||
1792 | tr_abort: | |
1793 | transaction_abort(tr); | |
1794 | transaction_free(tr); | |
1795 | return r; | |
1796 | } | |
1797 | ||
1798 | int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, sd_bus_error *e, Job **ret) { | |
1799 | Unit *unit = NULL; /* just to appease gcc, initialization is not really necessary */ | |
1800 | int r; | |
1801 | ||
1802 | assert(m); | |
1803 | assert(type < _JOB_TYPE_MAX); | |
1804 | assert(name); | |
1805 | assert(mode < _JOB_MODE_MAX); | |
1806 | ||
1807 | r = manager_load_unit(m, name, NULL, NULL, &unit); | |
1808 | if (r < 0) | |
1809 | return r; | |
1810 | assert(unit); | |
1811 | ||
1812 | return manager_add_job(m, type, unit, mode, affected_jobs, e, ret); | |
1813 | } | |
1814 | ||
1815 | int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Set *affected_jobs, Job **ret) { | |
1816 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1817 | int r; | |
1818 | ||
1819 | assert(m); | |
1820 | assert(type < _JOB_TYPE_MAX); | |
1821 | assert(name); | |
1822 | assert(mode < _JOB_MODE_MAX); | |
1823 | ||
1824 | r = manager_add_job_by_name(m, type, name, mode, affected_jobs, &error, ret); | |
1825 | if (r < 0) | |
1826 | return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r)); | |
1827 | ||
1828 | return r; | |
1829 | } | |
1830 | ||
1831 | int manager_propagate_reload(Manager *m, Unit *unit, JobMode mode, sd_bus_error *e) { | |
1832 | int r; | |
1833 | Transaction *tr; | |
1834 | ||
1835 | assert(m); | |
1836 | assert(unit); | |
1837 | assert(mode < _JOB_MODE_MAX); | |
1838 | assert(mode != JOB_ISOLATE); /* Isolate is only valid for start */ | |
1839 | ||
1840 | tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY); | |
1841 | if (!tr) | |
1842 | return -ENOMEM; | |
1843 | ||
1844 | /* We need an anchor job */ | |
1845 | r = transaction_add_job_and_dependencies(tr, JOB_NOP, unit, NULL, false, false, true, true, e); | |
1846 | if (r < 0) | |
1847 | goto tr_abort; | |
1848 | ||
1849 | /* Failure in adding individual dependencies is ignored, so this always succeeds. */ | |
1850 | transaction_add_propagate_reload_jobs(tr, unit, tr->anchor_job, mode == JOB_IGNORE_DEPENDENCIES, e); | |
1851 | ||
1852 | r = transaction_activate(tr, m, mode, NULL, e); | |
1853 | if (r < 0) | |
1854 | goto tr_abort; | |
1855 | ||
1856 | transaction_free(tr); | |
1857 | return 0; | |
1858 | ||
1859 | tr_abort: | |
1860 | transaction_abort(tr); | |
1861 | transaction_free(tr); | |
1862 | return r; | |
1863 | } | |
1864 | ||
1865 | Job *manager_get_job(Manager *m, uint32_t id) { | |
1866 | assert(m); | |
1867 | ||
1868 | return hashmap_get(m->jobs, UINT32_TO_PTR(id)); | |
1869 | } | |
1870 | ||
1871 | Unit *manager_get_unit(Manager *m, const char *name) { | |
1872 | assert(m); | |
1873 | assert(name); | |
1874 | ||
1875 | return hashmap_get(m->units, name); | |
1876 | } | |
1877 | ||
1878 | static int manager_dispatch_target_deps_queue(Manager *m) { | |
1879 | Unit *u; | |
1880 | unsigned k; | |
1881 | int r = 0; | |
1882 | ||
1883 | static const UnitDependency deps[] = { | |
1884 | UNIT_REQUIRED_BY, | |
1885 | UNIT_REQUISITE_OF, | |
1886 | UNIT_WANTED_BY, | |
1887 | UNIT_BOUND_BY | |
1888 | }; | |
1889 | ||
1890 | assert(m); | |
1891 | ||
1892 | while ((u = m->target_deps_queue)) { | |
1893 | assert(u->in_target_deps_queue); | |
1894 | ||
1895 | LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u); | |
1896 | u->in_target_deps_queue = false; | |
1897 | ||
1898 | for (k = 0; k < ELEMENTSOF(deps); k++) { | |
1899 | Unit *target; | |
1900 | Iterator i; | |
1901 | void *v; | |
1902 | ||
1903 | HASHMAP_FOREACH_KEY(v, target, u->dependencies[deps[k]], i) { | |
1904 | r = unit_add_default_target_dependency(u, target); | |
1905 | if (r < 0) | |
1906 | return r; | |
1907 | } | |
1908 | } | |
1909 | } | |
1910 | ||
1911 | return r; | |
1912 | } | |
1913 | ||
1914 | unsigned manager_dispatch_load_queue(Manager *m) { | |
1915 | Unit *u; | |
1916 | unsigned n = 0; | |
1917 | ||
1918 | assert(m); | |
1919 | ||
1920 | /* Make sure we are not run recursively */ | |
1921 | if (m->dispatching_load_queue) | |
1922 | return 0; | |
1923 | ||
1924 | m->dispatching_load_queue = true; | |
1925 | ||
1926 | /* Dispatches the load queue. Takes a unit from the queue and | |
1927 | * tries to load its data until the queue is empty */ | |
1928 | ||
1929 | while ((u = m->load_queue)) { | |
1930 | assert(u->in_load_queue); | |
1931 | ||
1932 | unit_load(u); | |
1933 | n++; | |
1934 | } | |
1935 | ||
1936 | m->dispatching_load_queue = false; | |
1937 | ||
1938 | /* Dispatch the units waiting for their target dependencies to be added now, as all targets that we know about | |
1939 | * should be loaded and have aliases resolved */ | |
1940 | (void) manager_dispatch_target_deps_queue(m); | |
1941 | ||
1942 | return n; | |
1943 | } | |
1944 | ||
1945 | int manager_load_unit_prepare( | |
1946 | Manager *m, | |
1947 | const char *name, | |
1948 | const char *path, | |
1949 | sd_bus_error *e, | |
1950 | Unit **_ret) { | |
1951 | ||
1952 | _cleanup_(unit_freep) Unit *cleanup_ret = NULL; | |
1953 | Unit *ret; | |
1954 | UnitType t; | |
1955 | int r; | |
1956 | ||
1957 | assert(m); | |
1958 | assert(name || path); | |
1959 | assert(_ret); | |
1960 | ||
1961 | /* This will prepare the unit for loading, but not actually | |
1962 | * load anything from disk. */ | |
1963 | ||
1964 | if (path && !is_path(path)) | |
1965 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path); | |
1966 | ||
1967 | if (!name) | |
1968 | name = basename(path); | |
1969 | ||
1970 | t = unit_name_to_type(name); | |
1971 | ||
1972 | if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) { | |
1973 | if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) | |
1974 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name); | |
1975 | ||
1976 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name); | |
1977 | } | |
1978 | ||
1979 | ret = manager_get_unit(m, name); | |
1980 | if (ret) { | |
1981 | *_ret = ret; | |
1982 | return 1; | |
1983 | } | |
1984 | ||
1985 | ret = cleanup_ret = unit_new(m, unit_vtable[t]->object_size); | |
1986 | if (!ret) | |
1987 | return -ENOMEM; | |
1988 | ||
1989 | if (path) { | |
1990 | ret->fragment_path = strdup(path); | |
1991 | if (!ret->fragment_path) | |
1992 | return -ENOMEM; | |
1993 | } | |
1994 | ||
1995 | r = unit_add_name(ret, name); | |
1996 | if (r < 0) | |
1997 | return r; | |
1998 | ||
1999 | unit_add_to_load_queue(ret); | |
2000 | unit_add_to_dbus_queue(ret); | |
2001 | unit_add_to_gc_queue(ret); | |
2002 | ||
2003 | *_ret = ret; | |
2004 | cleanup_ret = NULL; | |
2005 | ||
2006 | return 0; | |
2007 | } | |
2008 | ||
2009 | int manager_load_unit( | |
2010 | Manager *m, | |
2011 | const char *name, | |
2012 | const char *path, | |
2013 | sd_bus_error *e, | |
2014 | Unit **_ret) { | |
2015 | ||
2016 | int r; | |
2017 | ||
2018 | assert(m); | |
2019 | assert(_ret); | |
2020 | ||
2021 | /* This will load the service information files, but not actually | |
2022 | * start any services or anything. */ | |
2023 | ||
2024 | r = manager_load_unit_prepare(m, name, path, e, _ret); | |
2025 | if (r != 0) | |
2026 | return r; | |
2027 | ||
2028 | manager_dispatch_load_queue(m); | |
2029 | ||
2030 | *_ret = unit_follow_merge(*_ret); | |
2031 | return 0; | |
2032 | } | |
2033 | ||
2034 | int manager_load_startable_unit_or_warn( | |
2035 | Manager *m, | |
2036 | const char *name, | |
2037 | const char *path, | |
2038 | Unit **ret) { | |
2039 | ||
2040 | /* Load a unit, make sure it loaded fully and is not masked. */ | |
2041 | ||
2042 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2043 | Unit *unit; | |
2044 | int r; | |
2045 | ||
2046 | r = manager_load_unit(m, name, path, &error, &unit); | |
2047 | if (r < 0) | |
2048 | return log_error_errno(r, "Failed to load %s %s: %s", | |
2049 | name ? "unit" : "unit file", name ?: path, | |
2050 | bus_error_message(&error, r)); | |
2051 | ||
2052 | r = bus_unit_validate_load_state(unit, &error); | |
2053 | if (r < 0) | |
2054 | return log_error_errno(r, "%s", bus_error_message(&error, r)); | |
2055 | ||
2056 | *ret = unit; | |
2057 | return 0; | |
2058 | } | |
2059 | ||
2060 | void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) { | |
2061 | Iterator i; | |
2062 | Job *j; | |
2063 | ||
2064 | assert(s); | |
2065 | assert(f); | |
2066 | ||
2067 | HASHMAP_FOREACH(j, s->jobs, i) | |
2068 | job_dump(j, f, prefix); | |
2069 | } | |
2070 | ||
2071 | void manager_dump_units(Manager *s, FILE *f, const char *prefix) { | |
2072 | Iterator i; | |
2073 | Unit *u; | |
2074 | const char *t; | |
2075 | ||
2076 | assert(s); | |
2077 | assert(f); | |
2078 | ||
2079 | HASHMAP_FOREACH_KEY(u, t, s->units, i) | |
2080 | if (u->id == t) | |
2081 | unit_dump(u, f, prefix); | |
2082 | } | |
2083 | ||
2084 | void manager_dump(Manager *m, FILE *f, const char *prefix) { | |
2085 | ManagerTimestamp q; | |
2086 | ||
2087 | assert(m); | |
2088 | assert(f); | |
2089 | ||
2090 | for (q = 0; q < _MANAGER_TIMESTAMP_MAX; q++) { | |
2091 | char buf[FORMAT_TIMESTAMP_MAX]; | |
2092 | ||
2093 | if (dual_timestamp_is_set(m->timestamps + q)) | |
2094 | fprintf(f, "%sTimestamp %s: %s\n", | |
2095 | strempty(prefix), | |
2096 | manager_timestamp_to_string(q), | |
2097 | format_timestamp(buf, sizeof(buf), m->timestamps[q].realtime)); | |
2098 | } | |
2099 | ||
2100 | manager_dump_units(m, f, prefix); | |
2101 | manager_dump_jobs(m, f, prefix); | |
2102 | } | |
2103 | ||
2104 | int manager_get_dump_string(Manager *m, char **ret) { | |
2105 | _cleanup_free_ char *dump = NULL; | |
2106 | _cleanup_fclose_ FILE *f = NULL; | |
2107 | size_t size; | |
2108 | int r; | |
2109 | ||
2110 | assert(m); | |
2111 | assert(ret); | |
2112 | ||
2113 | f = open_memstream_unlocked(&dump, &size); | |
2114 | if (!f) | |
2115 | return -errno; | |
2116 | ||
2117 | manager_dump(m, f, NULL); | |
2118 | ||
2119 | r = fflush_and_check(f); | |
2120 | if (r < 0) | |
2121 | return r; | |
2122 | ||
2123 | f = safe_fclose(f); | |
2124 | ||
2125 | *ret = TAKE_PTR(dump); | |
2126 | ||
2127 | return 0; | |
2128 | } | |
2129 | ||
2130 | void manager_clear_jobs(Manager *m) { | |
2131 | Job *j; | |
2132 | ||
2133 | assert(m); | |
2134 | ||
2135 | while ((j = hashmap_first(m->jobs))) | |
2136 | /* No need to recurse. We're cancelling all jobs. */ | |
2137 | job_finish_and_invalidate(j, JOB_CANCELED, false, false); | |
2138 | } | |
2139 | ||
2140 | void manager_unwatch_pid(Manager *m, pid_t pid) { | |
2141 | assert(m); | |
2142 | ||
2143 | /* First let's drop the unit keyed as "pid". */ | |
2144 | (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid)); | |
2145 | ||
2146 | /* Then, let's also drop the array keyed by -pid. */ | |
2147 | free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid))); | |
2148 | } | |
2149 | ||
2150 | static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) { | |
2151 | Manager *m = userdata; | |
2152 | Job *j; | |
2153 | ||
2154 | assert(source); | |
2155 | assert(m); | |
2156 | ||
2157 | while ((j = m->run_queue)) { | |
2158 | assert(j->installed); | |
2159 | assert(j->in_run_queue); | |
2160 | ||
2161 | (void) job_run_and_invalidate(j); | |
2162 | } | |
2163 | ||
2164 | if (m->n_running_jobs > 0) | |
2165 | manager_watch_jobs_in_progress(m); | |
2166 | ||
2167 | if (m->n_on_console > 0) | |
2168 | manager_watch_idle_pipe(m); | |
2169 | ||
2170 | return 1; | |
2171 | } | |
2172 | ||
2173 | static unsigned manager_dispatch_dbus_queue(Manager *m) { | |
2174 | unsigned n = 0, budget; | |
2175 | Unit *u; | |
2176 | Job *j; | |
2177 | ||
2178 | assert(m); | |
2179 | ||
2180 | /* When we are reloading, let's not wait with generating signals, since we need to exit the manager as quickly | |
2181 | * as we can. There's no point in throttling generation of signals in that case. */ | |
2182 | if (MANAGER_IS_RELOADING(m) || m->send_reloading_done || m->pending_reload_message) | |
2183 | budget = (unsigned) -1; /* infinite budget in this case */ | |
2184 | else { | |
2185 | /* Anything to do at all? */ | |
2186 | if (!m->dbus_unit_queue && !m->dbus_job_queue) | |
2187 | return 0; | |
2188 | ||
2189 | /* Do we have overly many messages queued at the moment? If so, let's not enqueue more on top, let's | |
2190 | * sit this cycle out, and process things in a later cycle when the queues got a bit emptier. */ | |
2191 | if (manager_bus_n_queued_write(m) > MANAGER_BUS_BUSY_THRESHOLD) | |
2192 | return 0; | |
2193 | ||
2194 | /* Only process a certain number of units/jobs per event loop iteration. Even if the bus queue wasn't | |
2195 | * overly full before this call we shouldn't increase it in size too wildly in one step, and we | |
2196 | * shouldn't monopolize CPU time with generating these messages. Note the difference in counting of | |
2197 | * this "budget" and the "threshold" above: the "budget" is decreased only once per generated message, | |
2198 | * regardless how many busses/direct connections it is enqueued on, while the "threshold" is applied to | |
2199 | * each queued instance of bus message, i.e. if the same message is enqueued to five busses/direct | |
2200 | * connections it will be counted five times. This difference in counting ("references" | |
2201 | * vs. "instances") is primarily a result of the fact that it's easier to implement it this way, | |
2202 | * however it also reflects the thinking that the "threshold" should put a limit on used queue memory, | |
2203 | * i.e. space, while the "budget" should put a limit on time. Also note that the "threshold" is | |
2204 | * currently chosen much higher than the "budget". */ | |
2205 | budget = MANAGER_BUS_MESSAGE_BUDGET; | |
2206 | } | |
2207 | ||
2208 | while (budget != 0 && (u = m->dbus_unit_queue)) { | |
2209 | ||
2210 | assert(u->in_dbus_queue); | |
2211 | ||
2212 | bus_unit_send_change_signal(u); | |
2213 | n++; | |
2214 | ||
2215 | if (budget != (unsigned) -1) | |
2216 | budget--; | |
2217 | } | |
2218 | ||
2219 | while (budget != 0 && (j = m->dbus_job_queue)) { | |
2220 | assert(j->in_dbus_queue); | |
2221 | ||
2222 | bus_job_send_change_signal(j); | |
2223 | n++; | |
2224 | ||
2225 | if (budget != (unsigned) -1) | |
2226 | budget--; | |
2227 | } | |
2228 | ||
2229 | if (m->send_reloading_done) { | |
2230 | m->send_reloading_done = false; | |
2231 | bus_manager_send_reloading(m, false); | |
2232 | n++; | |
2233 | } | |
2234 | ||
2235 | if (m->pending_reload_message) { | |
2236 | bus_send_pending_reload_message(m); | |
2237 | n++; | |
2238 | } | |
2239 | ||
2240 | return n; | |
2241 | } | |
2242 | ||
2243 | static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { | |
2244 | Manager *m = userdata; | |
2245 | char buf[PATH_MAX]; | |
2246 | ssize_t n; | |
2247 | ||
2248 | n = recv(fd, buf, sizeof(buf), 0); | |
2249 | if (n < 0) | |
2250 | return log_error_errno(errno, "Failed to read cgroups agent message: %m"); | |
2251 | if (n == 0) { | |
2252 | log_error("Got zero-length cgroups agent message, ignoring."); | |
2253 | return 0; | |
2254 | } | |
2255 | if ((size_t) n >= sizeof(buf)) { | |
2256 | log_error("Got overly long cgroups agent message, ignoring."); | |
2257 | return 0; | |
2258 | } | |
2259 | ||
2260 | if (memchr(buf, 0, n)) { | |
2261 | log_error("Got cgroups agent message with embedded NUL byte, ignoring."); | |
2262 | return 0; | |
2263 | } | |
2264 | buf[n] = 0; | |
2265 | ||
2266 | manager_notify_cgroup_empty(m, buf); | |
2267 | (void) bus_forward_agent_released(m, buf); | |
2268 | ||
2269 | return 0; | |
2270 | } | |
2271 | ||
2272 | static void manager_invoke_notify_message( | |
2273 | Manager *m, | |
2274 | Unit *u, | |
2275 | const struct ucred *ucred, | |
2276 | const char *buf, | |
2277 | FDSet *fds) { | |
2278 | ||
2279 | assert(m); | |
2280 | assert(u); | |
2281 | assert(ucred); | |
2282 | assert(buf); | |
2283 | ||
2284 | if (u->notifygen == m->notifygen) /* Already invoked on this same unit in this same iteration? */ | |
2285 | return; | |
2286 | u->notifygen = m->notifygen; | |
2287 | ||
2288 | if (UNIT_VTABLE(u)->notify_message) { | |
2289 | _cleanup_strv_free_ char **tags = NULL; | |
2290 | ||
2291 | tags = strv_split(buf, NEWLINE); | |
2292 | if (!tags) { | |
2293 | log_oom(); | |
2294 | return; | |
2295 | } | |
2296 | ||
2297 | UNIT_VTABLE(u)->notify_message(u, ucred, tags, fds); | |
2298 | ||
2299 | } else if (DEBUG_LOGGING) { | |
2300 | _cleanup_free_ char *x = NULL, *y = NULL; | |
2301 | ||
2302 | x = ellipsize(buf, 20, 90); | |
2303 | if (x) | |
2304 | y = cescape(x); | |
2305 | ||
2306 | log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y)); | |
2307 | } | |
2308 | } | |
2309 | ||
2310 | static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { | |
2311 | ||
2312 | _cleanup_fdset_free_ FDSet *fds = NULL; | |
2313 | Manager *m = userdata; | |
2314 | char buf[NOTIFY_BUFFER_MAX+1]; | |
2315 | struct iovec iovec = { | |
2316 | .iov_base = buf, | |
2317 | .iov_len = sizeof(buf)-1, | |
2318 | }; | |
2319 | union { | |
2320 | struct cmsghdr cmsghdr; | |
2321 | uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) + | |
2322 | CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)]; | |
2323 | } control = {}; | |
2324 | struct msghdr msghdr = { | |
2325 | .msg_iov = &iovec, | |
2326 | .msg_iovlen = 1, | |
2327 | .msg_control = &control, | |
2328 | .msg_controllen = sizeof(control), | |
2329 | }; | |
2330 | ||
2331 | struct cmsghdr *cmsg; | |
2332 | struct ucred *ucred = NULL; | |
2333 | _cleanup_free_ Unit **array_copy = NULL; | |
2334 | Unit *u1, *u2, **array; | |
2335 | int r, *fd_array = NULL; | |
2336 | size_t n_fds = 0; | |
2337 | bool found = false; | |
2338 | ssize_t n; | |
2339 | ||
2340 | assert(m); | |
2341 | assert(m->notify_fd == fd); | |
2342 | ||
2343 | if (revents != EPOLLIN) { | |
2344 | log_warning("Got unexpected poll event for notify fd."); | |
2345 | return 0; | |
2346 | } | |
2347 | ||
2348 | n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC); | |
2349 | if (n < 0) { | |
2350 | if (IN_SET(errno, EAGAIN, EINTR)) | |
2351 | return 0; /* Spurious wakeup, try again */ | |
2352 | ||
2353 | /* If this is any other, real error, then let's stop processing this socket. This of course means we | |
2354 | * won't take notification messages anymore, but that's still better than busy looping around this: | |
2355 | * being woken up over and over again but being unable to actually read the message off the socket. */ | |
2356 | return log_error_errno(errno, "Failed to receive notification message: %m"); | |
2357 | } | |
2358 | ||
2359 | CMSG_FOREACH(cmsg, &msghdr) { | |
2360 | if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { | |
2361 | ||
2362 | fd_array = (int*) CMSG_DATA(cmsg); | |
2363 | n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int); | |
2364 | ||
2365 | } else if (cmsg->cmsg_level == SOL_SOCKET && | |
2366 | cmsg->cmsg_type == SCM_CREDENTIALS && | |
2367 | cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) { | |
2368 | ||
2369 | ucred = (struct ucred*) CMSG_DATA(cmsg); | |
2370 | } | |
2371 | } | |
2372 | ||
2373 | if (n_fds > 0) { | |
2374 | assert(fd_array); | |
2375 | ||
2376 | r = fdset_new_array(&fds, fd_array, n_fds); | |
2377 | if (r < 0) { | |
2378 | close_many(fd_array, n_fds); | |
2379 | log_oom(); | |
2380 | return 0; | |
2381 | } | |
2382 | } | |
2383 | ||
2384 | if (!ucred || !pid_is_valid(ucred->pid)) { | |
2385 | log_warning("Received notify message without valid credentials. Ignoring."); | |
2386 | return 0; | |
2387 | } | |
2388 | ||
2389 | if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) { | |
2390 | log_warning("Received notify message exceeded maximum size. Ignoring."); | |
2391 | return 0; | |
2392 | } | |
2393 | ||
2394 | /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one | |
2395 | * trailing NUL byte in the message, but don't expect it. */ | |
2396 | if (n > 1 && memchr(buf, 0, n-1)) { | |
2397 | log_warning("Received notify message with embedded NUL bytes. Ignoring."); | |
2398 | return 0; | |
2399 | } | |
2400 | ||
2401 | /* Make sure it's NUL-terminated. */ | |
2402 | buf[n] = 0; | |
2403 | ||
2404 | /* Increase the generation counter used for filtering out duplicate unit invocations. */ | |
2405 | m->notifygen++; | |
2406 | ||
2407 | /* Notify every unit that might be interested, which might be multiple. */ | |
2408 | u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid); | |
2409 | u2 = hashmap_get(m->watch_pids, PID_TO_PTR(ucred->pid)); | |
2410 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-ucred->pid)); | |
2411 | if (array) { | |
2412 | size_t k = 0; | |
2413 | ||
2414 | while (array[k]) | |
2415 | k++; | |
2416 | ||
2417 | array_copy = newdup(Unit*, array, k+1); | |
2418 | if (!array_copy) | |
2419 | log_oom(); | |
2420 | } | |
2421 | /* And now invoke the per-unit callbacks. Note that manager_invoke_notify_message() will handle duplicate units | |
2422 | * make sure we only invoke each unit's handler once. */ | |
2423 | if (u1) { | |
2424 | manager_invoke_notify_message(m, u1, ucred, buf, fds); | |
2425 | found = true; | |
2426 | } | |
2427 | if (u2) { | |
2428 | manager_invoke_notify_message(m, u2, ucred, buf, fds); | |
2429 | found = true; | |
2430 | } | |
2431 | if (array_copy) | |
2432 | for (size_t i = 0; array_copy[i]; i++) { | |
2433 | manager_invoke_notify_message(m, array_copy[i], ucred, buf, fds); | |
2434 | found = true; | |
2435 | } | |
2436 | ||
2437 | if (!found) | |
2438 | log_warning("Cannot find unit for notify message of PID "PID_FMT", ignoring.", ucred->pid); | |
2439 | ||
2440 | if (fdset_size(fds) > 0) | |
2441 | log_warning("Got extra auxiliary fds with notification message, closing them."); | |
2442 | ||
2443 | return 0; | |
2444 | } | |
2445 | ||
2446 | static void manager_invoke_sigchld_event( | |
2447 | Manager *m, | |
2448 | Unit *u, | |
2449 | const siginfo_t *si) { | |
2450 | ||
2451 | assert(m); | |
2452 | assert(u); | |
2453 | assert(si); | |
2454 | ||
2455 | /* Already invoked the handler of this unit in this iteration? Then don't process this again */ | |
2456 | if (u->sigchldgen == m->sigchldgen) | |
2457 | return; | |
2458 | u->sigchldgen = m->sigchldgen; | |
2459 | ||
2460 | log_unit_debug(u, "Child "PID_FMT" belongs to %s.", si->si_pid, u->id); | |
2461 | unit_unwatch_pid(u, si->si_pid); | |
2462 | ||
2463 | if (UNIT_VTABLE(u)->sigchld_event) | |
2464 | UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status); | |
2465 | } | |
2466 | ||
2467 | static int manager_dispatch_sigchld(sd_event_source *source, void *userdata) { | |
2468 | Manager *m = userdata; | |
2469 | siginfo_t si = {}; | |
2470 | int r; | |
2471 | ||
2472 | assert(source); | |
2473 | assert(m); | |
2474 | ||
2475 | /* First we call waitd() for a PID and do not reap the zombie. That way we can still access /proc/$PID for it | |
2476 | * while it is a zombie. */ | |
2477 | ||
2478 | if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) { | |
2479 | ||
2480 | if (errno != ECHILD) | |
2481 | log_error_errno(errno, "Failed to peek for child with waitid(), ignoring: %m"); | |
2482 | ||
2483 | goto turn_off; | |
2484 | } | |
2485 | ||
2486 | if (si.si_pid <= 0) | |
2487 | goto turn_off; | |
2488 | ||
2489 | if (IN_SET(si.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) { | |
2490 | _cleanup_free_ Unit **array_copy = NULL; | |
2491 | _cleanup_free_ char *name = NULL; | |
2492 | Unit *u1, *u2, **array; | |
2493 | ||
2494 | (void) get_process_comm(si.si_pid, &name); | |
2495 | ||
2496 | log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)", | |
2497 | si.si_pid, strna(name), | |
2498 | sigchld_code_to_string(si.si_code), | |
2499 | si.si_status, | |
2500 | strna(si.si_code == CLD_EXITED | |
2501 | ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL) | |
2502 | : signal_to_string(si.si_status))); | |
2503 | ||
2504 | /* Increase the generation counter used for filtering out duplicate unit invocations */ | |
2505 | m->sigchldgen++; | |
2506 | ||
2507 | /* And now figure out the unit this belongs to, it might be multiple... */ | |
2508 | u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid); | |
2509 | u2 = hashmap_get(m->watch_pids, PID_TO_PTR(si.si_pid)); | |
2510 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-si.si_pid)); | |
2511 | if (array) { | |
2512 | size_t n = 0; | |
2513 | ||
2514 | /* Cound how many entries the array has */ | |
2515 | while (array[n]) | |
2516 | n++; | |
2517 | ||
2518 | /* Make a copy of the array so that we don't trip up on the array changing beneath us */ | |
2519 | array_copy = newdup(Unit*, array, n+1); | |
2520 | if (!array_copy) | |
2521 | log_oom(); | |
2522 | } | |
2523 | ||
2524 | /* Finally, execute them all. Note that u1, u2 and the array might contain duplicates, but | |
2525 | * that's fine, manager_invoke_sigchld_event() will ensure we only invoke the handlers once for | |
2526 | * each iteration. */ | |
2527 | if (u1) | |
2528 | manager_invoke_sigchld_event(m, u1, &si); | |
2529 | if (u2) | |
2530 | manager_invoke_sigchld_event(m, u2, &si); | |
2531 | if (array_copy) | |
2532 | for (size_t i = 0; array_copy[i]; i++) | |
2533 | manager_invoke_sigchld_event(m, array_copy[i], &si); | |
2534 | } | |
2535 | ||
2536 | /* And now, we actually reap the zombie. */ | |
2537 | if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) { | |
2538 | log_error_errno(errno, "Failed to dequeue child, ignoring: %m"); | |
2539 | return 0; | |
2540 | } | |
2541 | ||
2542 | return 0; | |
2543 | ||
2544 | turn_off: | |
2545 | /* All children processed for now, turn off event source */ | |
2546 | ||
2547 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_OFF); | |
2548 | if (r < 0) | |
2549 | return log_error_errno(r, "Failed to disable SIGCHLD event source: %m"); | |
2550 | ||
2551 | return 0; | |
2552 | } | |
2553 | ||
2554 | static void manager_start_target(Manager *m, const char *name, JobMode mode) { | |
2555 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2556 | int r; | |
2557 | ||
2558 | log_debug("Activating special unit %s", name); | |
2559 | ||
2560 | r = manager_add_job_by_name(m, JOB_START, name, mode, NULL, &error, NULL); | |
2561 | if (r < 0) | |
2562 | log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r)); | |
2563 | } | |
2564 | ||
2565 | static void manager_handle_ctrl_alt_del(Manager *m) { | |
2566 | /* If the user presses C-A-D more than | |
2567 | * 7 times within 2s, we reboot/shutdown immediately, | |
2568 | * unless it was disabled in system.conf */ | |
2569 | ||
2570 | if (ratelimit_below(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE) | |
2571 | manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY); | |
2572 | else | |
2573 | emergency_action(m, m->cad_burst_action, EMERGENCY_ACTION_WARN, NULL, -1, | |
2574 | "Ctrl-Alt-Del was pressed more than 7 times within 2s"); | |
2575 | } | |
2576 | ||
2577 | static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { | |
2578 | Manager *m = userdata; | |
2579 | ssize_t n; | |
2580 | struct signalfd_siginfo sfsi; | |
2581 | int r; | |
2582 | ||
2583 | assert(m); | |
2584 | assert(m->signal_fd == fd); | |
2585 | ||
2586 | if (revents != EPOLLIN) { | |
2587 | log_warning("Got unexpected events from signal file descriptor."); | |
2588 | return 0; | |
2589 | } | |
2590 | ||
2591 | n = read(m->signal_fd, &sfsi, sizeof(sfsi)); | |
2592 | if (n != sizeof(sfsi)) { | |
2593 | if (n >= 0) { | |
2594 | log_warning("Truncated read from signal fd (%zu bytes), ignoring!", n); | |
2595 | return 0; | |
2596 | } | |
2597 | ||
2598 | if (IN_SET(errno, EINTR, EAGAIN)) | |
2599 | return 0; | |
2600 | ||
2601 | /* We return an error here, which will kill this handler, | |
2602 | * to avoid a busy loop on read error. */ | |
2603 | return log_error_errno(errno, "Reading from signal fd failed: %m"); | |
2604 | } | |
2605 | ||
2606 | log_received_signal(sfsi.ssi_signo == SIGCHLD || | |
2607 | (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m)) | |
2608 | ? LOG_DEBUG : LOG_INFO, | |
2609 | &sfsi); | |
2610 | ||
2611 | switch (sfsi.ssi_signo) { | |
2612 | ||
2613 | case SIGCHLD: | |
2614 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON); | |
2615 | if (r < 0) | |
2616 | log_warning_errno(r, "Failed to enable SIGCHLD event source, ignoring: %m"); | |
2617 | ||
2618 | break; | |
2619 | ||
2620 | case SIGTERM: | |
2621 | if (MANAGER_IS_SYSTEM(m)) { | |
2622 | /* This is for compatibility with the original sysvinit */ | |
2623 | if (verify_run_space_and_log("Refusing to reexecute") < 0) | |
2624 | break; | |
2625 | ||
2626 | m->objective = MANAGER_REEXECUTE; | |
2627 | break; | |
2628 | } | |
2629 | ||
2630 | _fallthrough_; | |
2631 | case SIGINT: | |
2632 | if (MANAGER_IS_SYSTEM(m)) | |
2633 | manager_handle_ctrl_alt_del(m); | |
2634 | else | |
2635 | manager_start_target(m, SPECIAL_EXIT_TARGET, | |
2636 | JOB_REPLACE_IRREVERSIBLY); | |
2637 | break; | |
2638 | ||
2639 | case SIGWINCH: | |
2640 | /* This is a nop on non-init */ | |
2641 | if (MANAGER_IS_SYSTEM(m)) | |
2642 | manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE); | |
2643 | ||
2644 | break; | |
2645 | ||
2646 | case SIGPWR: | |
2647 | /* This is a nop on non-init */ | |
2648 | if (MANAGER_IS_SYSTEM(m)) | |
2649 | manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE); | |
2650 | ||
2651 | break; | |
2652 | ||
2653 | case SIGUSR1: | |
2654 | if (manager_dbus_is_running(m, false)) { | |
2655 | log_info("Trying to reconnect to bus..."); | |
2656 | ||
2657 | (void) bus_init_api(m); | |
2658 | ||
2659 | if (MANAGER_IS_SYSTEM(m)) | |
2660 | (void) bus_init_system(m); | |
2661 | } else { | |
2662 | log_info("Starting D-Bus service..."); | |
2663 | manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE); | |
2664 | } | |
2665 | ||
2666 | break; | |
2667 | ||
2668 | case SIGUSR2: { | |
2669 | _cleanup_free_ char *dump = NULL; | |
2670 | ||
2671 | r = manager_get_dump_string(m, &dump); | |
2672 | if (r < 0) { | |
2673 | log_warning_errno(errno, "Failed to acquire manager dump: %m"); | |
2674 | break; | |
2675 | } | |
2676 | ||
2677 | log_dump(LOG_INFO, dump); | |
2678 | break; | |
2679 | } | |
2680 | ||
2681 | case SIGHUP: | |
2682 | if (verify_run_space_and_log("Refusing to reload") < 0) | |
2683 | break; | |
2684 | ||
2685 | m->objective = MANAGER_RELOAD; | |
2686 | break; | |
2687 | ||
2688 | default: { | |
2689 | ||
2690 | /* Starting SIGRTMIN+0 */ | |
2691 | static const struct { | |
2692 | const char *target; | |
2693 | JobMode mode; | |
2694 | } target_table[] = { | |
2695 | [0] = { SPECIAL_DEFAULT_TARGET, JOB_ISOLATE }, | |
2696 | [1] = { SPECIAL_RESCUE_TARGET, JOB_ISOLATE }, | |
2697 | [2] = { SPECIAL_EMERGENCY_TARGET, JOB_ISOLATE }, | |
2698 | [3] = { SPECIAL_HALT_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2699 | [4] = { SPECIAL_POWEROFF_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2700 | [5] = { SPECIAL_REBOOT_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2701 | [6] = { SPECIAL_KEXEC_TARGET, JOB_REPLACE_IRREVERSIBLY }, | |
2702 | }; | |
2703 | ||
2704 | /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */ | |
2705 | static const ManagerObjective objective_table[] = { | |
2706 | [0] = MANAGER_HALT, | |
2707 | [1] = MANAGER_POWEROFF, | |
2708 | [2] = MANAGER_REBOOT, | |
2709 | [3] = MANAGER_KEXEC, | |
2710 | }; | |
2711 | ||
2712 | if ((int) sfsi.ssi_signo >= SIGRTMIN+0 && | |
2713 | (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) { | |
2714 | int idx = (int) sfsi.ssi_signo - SIGRTMIN; | |
2715 | manager_start_target(m, target_table[idx].target, | |
2716 | target_table[idx].mode); | |
2717 | break; | |
2718 | } | |
2719 | ||
2720 | if ((int) sfsi.ssi_signo >= SIGRTMIN+13 && | |
2721 | (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(objective_table)) { | |
2722 | m->objective = objective_table[sfsi.ssi_signo - SIGRTMIN - 13]; | |
2723 | break; | |
2724 | } | |
2725 | ||
2726 | switch (sfsi.ssi_signo - SIGRTMIN) { | |
2727 | ||
2728 | case 20: | |
2729 | manager_set_show_status(m, SHOW_STATUS_YES); | |
2730 | break; | |
2731 | ||
2732 | case 21: | |
2733 | manager_set_show_status(m, SHOW_STATUS_NO); | |
2734 | break; | |
2735 | ||
2736 | case 22: | |
2737 | manager_override_log_level(m, LOG_DEBUG); | |
2738 | break; | |
2739 | ||
2740 | case 23: | |
2741 | manager_restore_original_log_level(m); | |
2742 | break; | |
2743 | ||
2744 | case 24: | |
2745 | if (MANAGER_IS_USER(m)) { | |
2746 | m->objective = MANAGER_EXIT; | |
2747 | return 0; | |
2748 | } | |
2749 | ||
2750 | /* This is a nop on init */ | |
2751 | break; | |
2752 | ||
2753 | case 26: | |
2754 | case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */ | |
2755 | manager_restore_original_log_target(m); | |
2756 | break; | |
2757 | ||
2758 | case 27: | |
2759 | manager_override_log_target(m, LOG_TARGET_CONSOLE); | |
2760 | break; | |
2761 | ||
2762 | case 28: | |
2763 | manager_override_log_target(m, LOG_TARGET_KMSG); | |
2764 | break; | |
2765 | ||
2766 | default: | |
2767 | log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo)); | |
2768 | } | |
2769 | }} | |
2770 | ||
2771 | return 0; | |
2772 | } | |
2773 | ||
2774 | static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { | |
2775 | Manager *m = userdata; | |
2776 | Iterator i; | |
2777 | Unit *u; | |
2778 | ||
2779 | assert(m); | |
2780 | assert(m->time_change_fd == fd); | |
2781 | ||
2782 | log_struct(LOG_DEBUG, | |
2783 | "MESSAGE_ID=" SD_MESSAGE_TIME_CHANGE_STR, | |
2784 | LOG_MESSAGE("Time has been changed")); | |
2785 | ||
2786 | /* Restart the watch */ | |
2787 | (void) manager_setup_time_change(m); | |
2788 | ||
2789 | HASHMAP_FOREACH(u, m->units, i) | |
2790 | if (UNIT_VTABLE(u)->time_change) | |
2791 | UNIT_VTABLE(u)->time_change(u); | |
2792 | ||
2793 | return 0; | |
2794 | } | |
2795 | ||
2796 | static int manager_dispatch_timezone_change( | |
2797 | sd_event_source *source, | |
2798 | const struct inotify_event *e, | |
2799 | void *userdata) { | |
2800 | ||
2801 | Manager *m = userdata; | |
2802 | int changed; | |
2803 | Iterator i; | |
2804 | Unit *u; | |
2805 | ||
2806 | assert(m); | |
2807 | ||
2808 | log_debug("inotify event for /etc/localtime"); | |
2809 | ||
2810 | changed = manager_read_timezone_stat(m); | |
2811 | if (changed <= 0) | |
2812 | return changed; | |
2813 | ||
2814 | /* Something changed, restart the watch, to ensure we watch the new /etc/localtime if it changed */ | |
2815 | (void) manager_setup_timezone_change(m); | |
2816 | ||
2817 | /* Read the new timezone */ | |
2818 | tzset(); | |
2819 | ||
2820 | log_debug("Timezone has been changed (now: %s).", tzname[daylight]); | |
2821 | ||
2822 | HASHMAP_FOREACH(u, m->units, i) | |
2823 | if (UNIT_VTABLE(u)->timezone_change) | |
2824 | UNIT_VTABLE(u)->timezone_change(u); | |
2825 | ||
2826 | return 0; | |
2827 | } | |
2828 | ||
2829 | static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { | |
2830 | Manager *m = userdata; | |
2831 | ||
2832 | assert(m); | |
2833 | assert(m->idle_pipe[2] == fd); | |
2834 | ||
2835 | /* There's at least one Type=idle child that just gave up on us waiting for the boot process to complete. Let's | |
2836 | * now turn off any further console output if there's at least one service that needs console access, so that | |
2837 | * from now on our own output should not spill into that service's output anymore. After all, we support | |
2838 | * Type=idle only to beautify console output and it generally is set on services that want to own the console | |
2839 | * exclusively without our interference. */ | |
2840 | m->no_console_output = m->n_on_console > 0; | |
2841 | ||
2842 | /* Acknowledge the child's request, and let all all other children know too that they shouldn't wait any longer | |
2843 | * by closing the pipes towards them, which is what they are waiting for. */ | |
2844 | manager_close_idle_pipe(m); | |
2845 | ||
2846 | return 0; | |
2847 | } | |
2848 | ||
2849 | static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) { | |
2850 | Manager *m = userdata; | |
2851 | int r; | |
2852 | uint64_t next; | |
2853 | ||
2854 | assert(m); | |
2855 | assert(source); | |
2856 | ||
2857 | manager_print_jobs_in_progress(m); | |
2858 | ||
2859 | next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC; | |
2860 | r = sd_event_source_set_time(source, next); | |
2861 | if (r < 0) | |
2862 | return r; | |
2863 | ||
2864 | return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT); | |
2865 | } | |
2866 | ||
2867 | int manager_loop(Manager *m) { | |
2868 | int r; | |
2869 | ||
2870 | RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000); | |
2871 | ||
2872 | assert(m); | |
2873 | assert(m->objective == MANAGER_OK); /* Ensure manager_startup() has been called */ | |
2874 | ||
2875 | /* Release the path cache */ | |
2876 | m->unit_path_cache = set_free_free(m->unit_path_cache); | |
2877 | ||
2878 | manager_check_finished(m); | |
2879 | ||
2880 | /* There might still be some zombies hanging around from before we were exec()'ed. Let's reap them. */ | |
2881 | r = sd_event_source_set_enabled(m->sigchld_event_source, SD_EVENT_ON); | |
2882 | if (r < 0) | |
2883 | return log_error_errno(r, "Failed to enable SIGCHLD event source: %m"); | |
2884 | ||
2885 | while (m->objective == MANAGER_OK) { | |
2886 | usec_t wait_usec; | |
2887 | ||
2888 | if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) | |
2889 | watchdog_ping(); | |
2890 | ||
2891 | if (!ratelimit_below(&rl)) { | |
2892 | /* Yay, something is going seriously wrong, pause a little */ | |
2893 | log_warning("Looping too fast. Throttling execution a little."); | |
2894 | sleep(1); | |
2895 | } | |
2896 | ||
2897 | if (manager_dispatch_load_queue(m) > 0) | |
2898 | continue; | |
2899 | ||
2900 | if (manager_dispatch_gc_job_queue(m) > 0) | |
2901 | continue; | |
2902 | ||
2903 | if (manager_dispatch_gc_unit_queue(m) > 0) | |
2904 | continue; | |
2905 | ||
2906 | if (manager_dispatch_cleanup_queue(m) > 0) | |
2907 | continue; | |
2908 | ||
2909 | if (manager_dispatch_cgroup_realize_queue(m) > 0) | |
2910 | continue; | |
2911 | ||
2912 | if (manager_dispatch_stop_when_unneeded_queue(m) > 0) | |
2913 | continue; | |
2914 | ||
2915 | if (manager_dispatch_dbus_queue(m) > 0) | |
2916 | continue; | |
2917 | ||
2918 | /* Sleep for half the watchdog time */ | |
2919 | if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) { | |
2920 | wait_usec = m->runtime_watchdog / 2; | |
2921 | if (wait_usec <= 0) | |
2922 | wait_usec = 1; | |
2923 | } else | |
2924 | wait_usec = USEC_INFINITY; | |
2925 | ||
2926 | r = sd_event_run(m->event, wait_usec); | |
2927 | if (r < 0) | |
2928 | return log_error_errno(r, "Failed to run event loop: %m"); | |
2929 | } | |
2930 | ||
2931 | return m->objective; | |
2932 | } | |
2933 | ||
2934 | int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) { | |
2935 | _cleanup_free_ char *n = NULL; | |
2936 | sd_id128_t invocation_id; | |
2937 | Unit *u; | |
2938 | int r; | |
2939 | ||
2940 | assert(m); | |
2941 | assert(s); | |
2942 | assert(_u); | |
2943 | ||
2944 | r = unit_name_from_dbus_path(s, &n); | |
2945 | if (r < 0) | |
2946 | return r; | |
2947 | ||
2948 | /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it | |
2949 | * as invocation ID. */ | |
2950 | r = sd_id128_from_string(n, &invocation_id); | |
2951 | if (r >= 0) { | |
2952 | u = hashmap_get(m->units_by_invocation_id, &invocation_id); | |
2953 | if (u) { | |
2954 | *_u = u; | |
2955 | return 0; | |
2956 | } | |
2957 | ||
2958 | return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID, | |
2959 | "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.", | |
2960 | SD_ID128_FORMAT_VAL(invocation_id)); | |
2961 | } | |
2962 | ||
2963 | /* If this didn't work, we check if this is a unit name */ | |
2964 | if (!unit_name_is_valid(n, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) { | |
2965 | _cleanup_free_ char *nn = NULL; | |
2966 | ||
2967 | nn = cescape(n); | |
2968 | return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, | |
2969 | "Unit name %s is neither a valid invocation ID nor unit name.", strnull(nn)); | |
2970 | } | |
2971 | ||
2972 | r = manager_load_unit(m, n, NULL, e, &u); | |
2973 | if (r < 0) | |
2974 | return r; | |
2975 | ||
2976 | *_u = u; | |
2977 | return 0; | |
2978 | } | |
2979 | ||
2980 | int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) { | |
2981 | const char *p; | |
2982 | unsigned id; | |
2983 | Job *j; | |
2984 | int r; | |
2985 | ||
2986 | assert(m); | |
2987 | assert(s); | |
2988 | assert(_j); | |
2989 | ||
2990 | p = startswith(s, "/org/freedesktop/systemd1/job/"); | |
2991 | if (!p) | |
2992 | return -EINVAL; | |
2993 | ||
2994 | r = safe_atou(p, &id); | |
2995 | if (r < 0) | |
2996 | return r; | |
2997 | ||
2998 | j = manager_get_job(m, id); | |
2999 | if (!j) | |
3000 | return -ENOENT; | |
3001 | ||
3002 | *_j = j; | |
3003 | ||
3004 | return 0; | |
3005 | } | |
3006 | ||
3007 | void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) { | |
3008 | ||
3009 | #if HAVE_AUDIT | |
3010 | _cleanup_free_ char *p = NULL; | |
3011 | const char *msg; | |
3012 | int audit_fd, r; | |
3013 | ||
3014 | if (!MANAGER_IS_SYSTEM(m)) | |
3015 | return; | |
3016 | ||
3017 | audit_fd = get_audit_fd(); | |
3018 | if (audit_fd < 0) | |
3019 | return; | |
3020 | ||
3021 | /* Don't generate audit events if the service was already | |
3022 | * started and we're just deserializing */ | |
3023 | if (MANAGER_IS_RELOADING(m)) | |
3024 | return; | |
3025 | ||
3026 | if (u->type != UNIT_SERVICE) | |
3027 | return; | |
3028 | ||
3029 | r = unit_name_to_prefix_and_instance(u->id, &p); | |
3030 | if (r < 0) { | |
3031 | log_error_errno(r, "Failed to extract prefix and instance of unit name: %m"); | |
3032 | return; | |
3033 | } | |
3034 | ||
3035 | msg = strjoina("unit=", p); | |
3036 | if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) { | |
3037 | if (errno == EPERM) | |
3038 | /* We aren't allowed to send audit messages? | |
3039 | * Then let's not retry again. */ | |
3040 | close_audit_fd(); | |
3041 | else | |
3042 | log_warning_errno(errno, "Failed to send audit message: %m"); | |
3043 | } | |
3044 | #endif | |
3045 | ||
3046 | } | |
3047 | ||
3048 | void manager_send_unit_plymouth(Manager *m, Unit *u) { | |
3049 | static const union sockaddr_union sa = PLYMOUTH_SOCKET; | |
3050 | _cleanup_free_ char *message = NULL; | |
3051 | _cleanup_close_ int fd = -1; | |
3052 | int n = 0; | |
3053 | ||
3054 | /* Don't generate plymouth events if the service was already | |
3055 | * started and we're just deserializing */ | |
3056 | if (MANAGER_IS_RELOADING(m)) | |
3057 | return; | |
3058 | ||
3059 | if (!MANAGER_IS_SYSTEM(m)) | |
3060 | return; | |
3061 | ||
3062 | if (detect_container() > 0) | |
3063 | return; | |
3064 | ||
3065 | if (!IN_SET(u->type, UNIT_SERVICE, UNIT_MOUNT, UNIT_SWAP)) | |
3066 | return; | |
3067 | ||
3068 | /* We set SOCK_NONBLOCK here so that we rather drop the | |
3069 | * message then wait for plymouth */ | |
3070 | fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
3071 | if (fd < 0) { | |
3072 | log_error_errno(errno, "socket() failed: %m"); | |
3073 | return; | |
3074 | } | |
3075 | ||
3076 | if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) { | |
3077 | if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno)) | |
3078 | log_error_errno(errno, "connect() failed: %m"); | |
3079 | return; | |
3080 | } | |
3081 | ||
3082 | if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) { | |
3083 | log_oom(); | |
3084 | return; | |
3085 | } | |
3086 | ||
3087 | errno = 0; | |
3088 | if (write(fd, message, n + 1) != n + 1) | |
3089 | if (!IN_SET(errno, EAGAIN, ENOENT) && !ERRNO_IS_DISCONNECT(errno)) | |
3090 | log_error_errno(errno, "Failed to write Plymouth message: %m"); | |
3091 | } | |
3092 | ||
3093 | int manager_open_serialization(Manager *m, FILE **_f) { | |
3094 | int fd; | |
3095 | FILE *f; | |
3096 | ||
3097 | assert(_f); | |
3098 | ||
3099 | fd = open_serialization_fd("systemd-state"); | |
3100 | if (fd < 0) | |
3101 | return fd; | |
3102 | ||
3103 | f = fdopen(fd, "w+"); | |
3104 | if (!f) { | |
3105 | safe_close(fd); | |
3106 | return -errno; | |
3107 | } | |
3108 | ||
3109 | *_f = f; | |
3110 | return 0; | |
3111 | } | |
3112 | ||
3113 | static bool manager_timestamp_shall_serialize(ManagerTimestamp t) { | |
3114 | ||
3115 | if (!in_initrd()) | |
3116 | return true; | |
3117 | ||
3118 | /* The following timestamps only apply to the host system, hence only serialize them there */ | |
3119 | return !IN_SET(t, | |
3120 | MANAGER_TIMESTAMP_USERSPACE, MANAGER_TIMESTAMP_FINISH, | |
3121 | MANAGER_TIMESTAMP_SECURITY_START, MANAGER_TIMESTAMP_SECURITY_FINISH, | |
3122 | MANAGER_TIMESTAMP_GENERATORS_START, MANAGER_TIMESTAMP_GENERATORS_FINISH, | |
3123 | MANAGER_TIMESTAMP_UNITS_LOAD_START, MANAGER_TIMESTAMP_UNITS_LOAD_FINISH); | |
3124 | } | |
3125 | ||
3126 | int manager_serialize( | |
3127 | Manager *m, | |
3128 | FILE *f, | |
3129 | FDSet *fds, | |
3130 | bool switching_root) { | |
3131 | ||
3132 | ManagerTimestamp q; | |
3133 | const char *t; | |
3134 | Iterator i; | |
3135 | Unit *u; | |
3136 | int r; | |
3137 | ||
3138 | assert(m); | |
3139 | assert(f); | |
3140 | assert(fds); | |
3141 | ||
3142 | _cleanup_(manager_reloading_stopp) _unused_ Manager *reloading = manager_reloading_start(m); | |
3143 | ||
3144 | (void) serialize_item_format(f, "current-job-id", "%" PRIu32, m->current_job_id); | |
3145 | (void) serialize_item_format(f, "n-installed-jobs", "%u", m->n_installed_jobs); | |
3146 | (void) serialize_item_format(f, "n-failed-jobs", "%u", m->n_failed_jobs); | |
3147 | (void) serialize_bool(f, "taint-usr", m->taint_usr); | |
3148 | (void) serialize_bool(f, "ready-sent", m->ready_sent); | |
3149 | (void) serialize_bool(f, "taint-logged", m->taint_logged); | |
3150 | (void) serialize_bool(f, "service-watchdogs", m->service_watchdogs); | |
3151 | ||
3152 | /* After switching root, udevd has not been started yet. So, enumeration results should not be emitted. */ | |
3153 | (void) serialize_bool(f, "honor-device-enumeration", !switching_root); | |
3154 | ||
3155 | t = show_status_to_string(m->show_status); | |
3156 | if (t) | |
3157 | (void) serialize_item(f, "show-status", t); | |
3158 | ||
3159 | if (m->log_level_overridden) | |
3160 | (void) serialize_item_format(f, "log-level-override", "%i", log_get_max_level()); | |
3161 | if (m->log_target_overridden) | |
3162 | (void) serialize_item(f, "log-target-override", log_target_to_string(log_get_target())); | |
3163 | ||
3164 | for (q = 0; q < _MANAGER_TIMESTAMP_MAX; q++) { | |
3165 | _cleanup_free_ char *joined = NULL; | |
3166 | ||
3167 | if (!manager_timestamp_shall_serialize(q)) | |
3168 | continue; | |
3169 | ||
3170 | joined = strjoin(manager_timestamp_to_string(q), "-timestamp"); | |
3171 | if (!joined) | |
3172 | return log_oom(); | |
3173 | ||
3174 | (void) serialize_dual_timestamp(f, joined, m->timestamps + q); | |
3175 | } | |
3176 | ||
3177 | if (!switching_root) | |
3178 | (void) serialize_strv(f, "env", m->client_environment); | |
3179 | ||
3180 | if (m->notify_fd >= 0) { | |
3181 | r = serialize_fd(f, fds, "notify-fd", m->notify_fd); | |
3182 | if (r < 0) | |
3183 | return r; | |
3184 | ||
3185 | (void) serialize_item(f, "notify-socket", m->notify_socket); | |
3186 | } | |
3187 | ||
3188 | if (m->cgroups_agent_fd >= 0) { | |
3189 | r = serialize_fd(f, fds, "cgroups-agent-fd", m->cgroups_agent_fd); | |
3190 | if (r < 0) | |
3191 | return r; | |
3192 | } | |
3193 | ||
3194 | if (m->user_lookup_fds[0] >= 0) { | |
3195 | int copy0, copy1; | |
3196 | ||
3197 | copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]); | |
3198 | if (copy0 < 0) | |
3199 | return log_error_errno(copy0, "Failed to add user lookup fd to serialization: %m"); | |
3200 | ||
3201 | copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]); | |
3202 | if (copy1 < 0) | |
3203 | return log_error_errno(copy1, "Failed to add user lookup fd to serialization: %m"); | |
3204 | ||
3205 | (void) serialize_item_format(f, "user-lookup", "%i %i", copy0, copy1); | |
3206 | } | |
3207 | ||
3208 | bus_track_serialize(m->subscribed, f, "subscribed"); | |
3209 | ||
3210 | r = dynamic_user_serialize(m, f, fds); | |
3211 | if (r < 0) | |
3212 | return r; | |
3213 | ||
3214 | manager_serialize_uid_refs(m, f); | |
3215 | manager_serialize_gid_refs(m, f); | |
3216 | ||
3217 | r = exec_runtime_serialize(m, f, fds); | |
3218 | if (r < 0) | |
3219 | return r; | |
3220 | ||
3221 | (void) fputc('\n', f); | |
3222 | ||
3223 | HASHMAP_FOREACH_KEY(u, t, m->units, i) { | |
3224 | if (u->id != t) | |
3225 | continue; | |
3226 | ||
3227 | /* Start marker */ | |
3228 | fputs(u->id, f); | |
3229 | fputc('\n', f); | |
3230 | ||
3231 | r = unit_serialize(u, f, fds, !switching_root); | |
3232 | if (r < 0) | |
3233 | return r; | |
3234 | } | |
3235 | ||
3236 | r = fflush_and_check(f); | |
3237 | if (r < 0) | |
3238 | return log_error_errno(r, "Failed to flush serialization: %m"); | |
3239 | ||
3240 | r = bus_fdset_add_all(m, fds); | |
3241 | if (r < 0) | |
3242 | return log_error_errno(r, "Failed to add bus sockets to serialization: %m"); | |
3243 | ||
3244 | return 0; | |
3245 | } | |
3246 | ||
3247 | static int manager_deserialize_one_unit(Manager *m, const char *name, FILE *f, FDSet *fds) { | |
3248 | Unit *u; | |
3249 | int r; | |
3250 | ||
3251 | r = manager_load_unit(m, name, NULL, NULL, &u); | |
3252 | if (r < 0) { | |
3253 | if (r == -ENOMEM) | |
3254 | return r; | |
3255 | return log_notice_errno(r, "Failed to load unit \"%s\", skipping deserialization: %m", name); | |
3256 | } | |
3257 | ||
3258 | r = unit_deserialize(u, f, fds); | |
3259 | if (r < 0) { | |
3260 | if (r == -ENOMEM) | |
3261 | return r; | |
3262 | return log_notice_errno(r, "Failed to deserialize unit \"%s\", skipping: %m", name); | |
3263 | } | |
3264 | ||
3265 | return 0; | |
3266 | } | |
3267 | ||
3268 | static int manager_deserialize_units(Manager *m, FILE *f, FDSet *fds) { | |
3269 | const char *unit_name; | |
3270 | int r; | |
3271 | ||
3272 | for (;;) { | |
3273 | _cleanup_free_ char *line = NULL; | |
3274 | /* Start marker */ | |
3275 | r = read_line(f, LONG_LINE_MAX, &line); | |
3276 | if (r < 0) | |
3277 | return log_error_errno(r, "Failed to read serialization line: %m"); | |
3278 | if (r == 0) | |
3279 | break; | |
3280 | ||
3281 | unit_name = strstrip(line); | |
3282 | ||
3283 | r = manager_deserialize_one_unit(m, unit_name, f, fds); | |
3284 | if (r == -ENOMEM) | |
3285 | return r; | |
3286 | if (r < 0) { | |
3287 | r = unit_deserialize_skip(f); | |
3288 | if (r < 0) | |
3289 | return r; | |
3290 | } | |
3291 | } | |
3292 | ||
3293 | return 0; | |
3294 | } | |
3295 | ||
3296 | int manager_deserialize(Manager *m, FILE *f, FDSet *fds) { | |
3297 | int r = 0; | |
3298 | ||
3299 | assert(m); | |
3300 | assert(f); | |
3301 | ||
3302 | log_debug("Deserializing state..."); | |
3303 | ||
3304 | /* If we are not in reload mode yet, enter it now. Not that this is recursive, a caller might already have | |
3305 | * increased it to non-zero, which is why we just increase it by one here and down again at the end of this | |
3306 | * call. */ | |
3307 | _cleanup_(manager_reloading_stopp) _unused_ Manager *reloading = manager_reloading_start(m); | |
3308 | ||
3309 | for (;;) { | |
3310 | _cleanup_free_ char *line = NULL; | |
3311 | const char *val, *l; | |
3312 | ||
3313 | r = read_line(f, LONG_LINE_MAX, &line); | |
3314 | if (r < 0) | |
3315 | return log_error_errno(r, "Failed to read serialization line: %m"); | |
3316 | if (r == 0) | |
3317 | break; | |
3318 | ||
3319 | l = strstrip(line); | |
3320 | if (isempty(l)) /* end marker */ | |
3321 | break; | |
3322 | ||
3323 | if ((val = startswith(l, "current-job-id="))) { | |
3324 | uint32_t id; | |
3325 | ||
3326 | if (safe_atou32(val, &id) < 0) | |
3327 | log_notice("Failed to parse current job id value '%s', ignoring.", val); | |
3328 | else | |
3329 | m->current_job_id = MAX(m->current_job_id, id); | |
3330 | ||
3331 | } else if ((val = startswith(l, "n-installed-jobs="))) { | |
3332 | uint32_t n; | |
3333 | ||
3334 | if (safe_atou32(val, &n) < 0) | |
3335 | log_notice("Failed to parse installed jobs counter '%s', ignoring.", val); | |
3336 | else | |
3337 | m->n_installed_jobs += n; | |
3338 | ||
3339 | } else if ((val = startswith(l, "n-failed-jobs="))) { | |
3340 | uint32_t n; | |
3341 | ||
3342 | if (safe_atou32(val, &n) < 0) | |
3343 | log_notice("Failed to parse failed jobs counter '%s', ignoring.", val); | |
3344 | else | |
3345 | m->n_failed_jobs += n; | |
3346 | ||
3347 | } else if ((val = startswith(l, "taint-usr="))) { | |
3348 | int b; | |
3349 | ||
3350 | b = parse_boolean(val); | |
3351 | if (b < 0) | |
3352 | log_notice("Failed to parse taint /usr flag '%s', ignoring.", val); | |
3353 | else | |
3354 | m->taint_usr = m->taint_usr || b; | |
3355 | ||
3356 | } else if ((val = startswith(l, "ready-sent="))) { | |
3357 | int b; | |
3358 | ||
3359 | b = parse_boolean(val); | |
3360 | if (b < 0) | |
3361 | log_notice("Failed to parse ready-sent flag '%s', ignoring.", val); | |
3362 | else | |
3363 | m->ready_sent = m->ready_sent || b; | |
3364 | ||
3365 | } else if ((val = startswith(l, "taint-logged="))) { | |
3366 | int b; | |
3367 | ||
3368 | b = parse_boolean(val); | |
3369 | if (b < 0) | |
3370 | log_notice("Failed to parse taint-logged flag '%s', ignoring.", val); | |
3371 | else | |
3372 | m->taint_logged = m->taint_logged || b; | |
3373 | ||
3374 | } else if ((val = startswith(l, "service-watchdogs="))) { | |
3375 | int b; | |
3376 | ||
3377 | b = parse_boolean(val); | |
3378 | if (b < 0) | |
3379 | log_notice("Failed to parse service-watchdogs flag '%s', ignoring.", val); | |
3380 | else | |
3381 | m->service_watchdogs = b; | |
3382 | ||
3383 | } else if ((val = startswith(l, "honor-device-enumeration="))) { | |
3384 | int b; | |
3385 | ||
3386 | b = parse_boolean(val); | |
3387 | if (b < 0) | |
3388 | log_notice("Failed to parse honor-device-enumeration flag '%s', ignoring.", val); | |
3389 | else | |
3390 | m->honor_device_enumeration = b; | |
3391 | ||
3392 | } else if ((val = startswith(l, "show-status="))) { | |
3393 | ShowStatus s; | |
3394 | ||
3395 | s = show_status_from_string(val); | |
3396 | if (s < 0) | |
3397 | log_notice("Failed to parse show-status flag '%s', ignoring.", val); | |
3398 | else | |
3399 | manager_set_show_status(m, s); | |
3400 | ||
3401 | } else if ((val = startswith(l, "log-level-override="))) { | |
3402 | int level; | |
3403 | ||
3404 | level = log_level_from_string(val); | |
3405 | if (level < 0) | |
3406 | log_notice("Failed to parse log-level-override value '%s', ignoring.", val); | |
3407 | else | |
3408 | manager_override_log_level(m, level); | |
3409 | ||
3410 | } else if ((val = startswith(l, "log-target-override="))) { | |
3411 | LogTarget target; | |
3412 | ||
3413 | target = log_target_from_string(val); | |
3414 | if (target < 0) | |
3415 | log_notice("Failed to parse log-target-override value '%s', ignoring.", val); | |
3416 | else | |
3417 | manager_override_log_target(m, target); | |
3418 | ||
3419 | } else if (startswith(l, "env=")) { | |
3420 | r = deserialize_environment(l + 4, &m->client_environment); | |
3421 | if (r < 0) | |
3422 | log_notice_errno(r, "Failed to parse environment entry: \"%s\", ignoring: %m", l); | |
3423 | ||
3424 | } else if ((val = startswith(l, "notify-fd="))) { | |
3425 | int fd; | |
3426 | ||
3427 | if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd)) | |
3428 | log_notice("Failed to parse notify fd, ignoring: \"%s\"", val); | |
3429 | else { | |
3430 | m->notify_event_source = sd_event_source_unref(m->notify_event_source); | |
3431 | safe_close(m->notify_fd); | |
3432 | m->notify_fd = fdset_remove(fds, fd); | |
3433 | } | |
3434 | ||
3435 | } else if ((val = startswith(l, "notify-socket="))) { | |
3436 | r = free_and_strdup(&m->notify_socket, val); | |
3437 | if (r < 0) | |
3438 | return r; | |
3439 | ||
3440 | } else if ((val = startswith(l, "cgroups-agent-fd="))) { | |
3441 | int fd; | |
3442 | ||
3443 | if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd)) | |
3444 | log_notice("Failed to parse cgroups agent fd, ignoring.: %s", val); | |
3445 | else { | |
3446 | m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source); | |
3447 | safe_close(m->cgroups_agent_fd); | |
3448 | m->cgroups_agent_fd = fdset_remove(fds, fd); | |
3449 | } | |
3450 | ||
3451 | } else if ((val = startswith(l, "user-lookup="))) { | |
3452 | int fd0, fd1; | |
3453 | ||
3454 | if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1)) | |
3455 | log_notice("Failed to parse user lookup fd, ignoring: %s", val); | |
3456 | else { | |
3457 | m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source); | |
3458 | safe_close_pair(m->user_lookup_fds); | |
3459 | m->user_lookup_fds[0] = fdset_remove(fds, fd0); | |
3460 | m->user_lookup_fds[1] = fdset_remove(fds, fd1); | |
3461 | } | |
3462 | ||
3463 | } else if ((val = startswith(l, "dynamic-user="))) | |
3464 | dynamic_user_deserialize_one(m, val, fds); | |
3465 | else if ((val = startswith(l, "destroy-ipc-uid="))) | |
3466 | manager_deserialize_uid_refs_one(m, val); | |
3467 | else if ((val = startswith(l, "destroy-ipc-gid="))) | |
3468 | manager_deserialize_gid_refs_one(m, val); | |
3469 | else if ((val = startswith(l, "exec-runtime="))) | |
3470 | exec_runtime_deserialize_one(m, val, fds); | |
3471 | else if ((val = startswith(l, "subscribed="))) { | |
3472 | ||
3473 | if (strv_extend(&m->deserialized_subscribed, val) < 0) | |
3474 | return -ENOMEM; | |
3475 | ||
3476 | } else { | |
3477 | ManagerTimestamp q; | |
3478 | ||
3479 | for (q = 0; q < _MANAGER_TIMESTAMP_MAX; q++) { | |
3480 | val = startswith(l, manager_timestamp_to_string(q)); | |
3481 | if (!val) | |
3482 | continue; | |
3483 | ||
3484 | val = startswith(val, "-timestamp="); | |
3485 | if (val) | |
3486 | break; | |
3487 | } | |
3488 | ||
3489 | if (q < _MANAGER_TIMESTAMP_MAX) /* found it */ | |
3490 | (void) deserialize_dual_timestamp(val, m->timestamps + q); | |
3491 | else if (!startswith(l, "kdbus-fd=")) /* ignore kdbus */ | |
3492 | log_notice("Unknown serialization item '%s', ignoring.", l); | |
3493 | } | |
3494 | } | |
3495 | ||
3496 | return manager_deserialize_units(m, f, fds); | |
3497 | } | |
3498 | ||
3499 | int manager_reload(Manager *m) { | |
3500 | _cleanup_(manager_reloading_stopp) Manager *reloading = NULL; | |
3501 | _cleanup_fdset_free_ FDSet *fds = NULL; | |
3502 | _cleanup_fclose_ FILE *f = NULL; | |
3503 | int r; | |
3504 | ||
3505 | assert(m); | |
3506 | ||
3507 | r = manager_open_serialization(m, &f); | |
3508 | if (r < 0) | |
3509 | return log_error_errno(r, "Failed to create serialization file: %m"); | |
3510 | ||
3511 | fds = fdset_new(); | |
3512 | if (!fds) | |
3513 | return log_oom(); | |
3514 | ||
3515 | /* We are officially in reload mode from here on. */ | |
3516 | reloading = manager_reloading_start(m); | |
3517 | ||
3518 | r = manager_serialize(m, f, fds, false); | |
3519 | if (r < 0) | |
3520 | return r; | |
3521 | ||
3522 | if (fseeko(f, 0, SEEK_SET) < 0) | |
3523 | return log_error_errno(errno, "Failed to seek to beginning of serialization: %m"); | |
3524 | ||
3525 | /* 💀 This is the point of no return, from here on there is no way back. 💀 */ | |
3526 | reloading = NULL; | |
3527 | ||
3528 | bus_manager_send_reloading(m, true); | |
3529 | ||
3530 | /* Start by flushing out all jobs and units, all generated units, all runtime environments, all dynamic users | |
3531 | * and everything else that is worth flushing out. We'll get it all back from the serialization — if we need | |
3532 | * it.*/ | |
3533 | ||
3534 | manager_clear_jobs_and_units(m); | |
3535 | lookup_paths_flush_generator(&m->lookup_paths); | |
3536 | lookup_paths_free(&m->lookup_paths); | |
3537 | exec_runtime_vacuum(m); | |
3538 | dynamic_user_vacuum(m, false); | |
3539 | m->uid_refs = hashmap_free(m->uid_refs); | |
3540 | m->gid_refs = hashmap_free(m->gid_refs); | |
3541 | ||
3542 | r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL); | |
3543 | if (r < 0) | |
3544 | log_warning_errno(r, "Failed to initialize path lookup table, ignoring: %m"); | |
3545 | ||
3546 | (void) manager_run_environment_generators(m); | |
3547 | (void) manager_run_generators(m); | |
3548 | ||
3549 | r = lookup_paths_reduce(&m->lookup_paths); | |
3550 | if (r < 0) | |
3551 | log_warning_errno(r, "Failed ot reduce unit file paths, ignoring: %m"); | |
3552 | ||
3553 | manager_build_unit_path_cache(m); | |
3554 | ||
3555 | /* First, enumerate what we can from kernel and suchlike */ | |
3556 | manager_enumerate_perpetual(m); | |
3557 | manager_enumerate(m); | |
3558 | ||
3559 | /* Second, deserialize our stored data */ | |
3560 | r = manager_deserialize(m, f, fds); | |
3561 | if (r < 0) | |
3562 | log_warning_errno(r, "Deserialization failed, proceeding anyway: %m"); | |
3563 | ||
3564 | /* We don't need the serialization anymore */ | |
3565 | f = safe_fclose(f); | |
3566 | ||
3567 | /* Re-register notify_fd as event source, and set up other sockets/communication channels we might need */ | |
3568 | (void) manager_setup_notify(m); | |
3569 | (void) manager_setup_cgroups_agent(m); | |
3570 | (void) manager_setup_user_lookup_fd(m); | |
3571 | ||
3572 | /* Third, fire things up! */ | |
3573 | manager_coldplug(m); | |
3574 | ||
3575 | /* Clean up runtime objects no longer referenced */ | |
3576 | manager_vacuum(m); | |
3577 | ||
3578 | /* Consider the reload process complete now. */ | |
3579 | assert(m->n_reloading > 0); | |
3580 | m->n_reloading--; | |
3581 | ||
3582 | /* On manager reloading, device tag data should exists, thus, we should honor the results of device | |
3583 | * enumeration. The flag should be always set correctly by the serialized data, but it may fail. So, | |
3584 | * let's always set the flag here for safety. */ | |
3585 | m->honor_device_enumeration = true; | |
3586 | ||
3587 | manager_ready(m); | |
3588 | ||
3589 | m->send_reloading_done = true; | |
3590 | return 0; | |
3591 | } | |
3592 | ||
3593 | void manager_reset_failed(Manager *m) { | |
3594 | Unit *u; | |
3595 | Iterator i; | |
3596 | ||
3597 | assert(m); | |
3598 | ||
3599 | HASHMAP_FOREACH(u, m->units, i) | |
3600 | unit_reset_failed(u); | |
3601 | } | |
3602 | ||
3603 | bool manager_unit_inactive_or_pending(Manager *m, const char *name) { | |
3604 | Unit *u; | |
3605 | ||
3606 | assert(m); | |
3607 | assert(name); | |
3608 | ||
3609 | /* Returns true if the unit is inactive or going down */ | |
3610 | u = manager_get_unit(m, name); | |
3611 | if (!u) | |
3612 | return true; | |
3613 | ||
3614 | return unit_inactive_or_pending(u); | |
3615 | } | |
3616 | ||
3617 | static void log_taint_string(Manager *m) { | |
3618 | _cleanup_free_ char *taint = NULL; | |
3619 | ||
3620 | assert(m); | |
3621 | ||
3622 | if (MANAGER_IS_USER(m) || m->taint_logged) | |
3623 | return; | |
3624 | ||
3625 | m->taint_logged = true; /* only check for taint once */ | |
3626 | ||
3627 | taint = manager_taint_string(m); | |
3628 | if (isempty(taint)) | |
3629 | return; | |
3630 | ||
3631 | log_struct(LOG_NOTICE, | |
3632 | LOG_MESSAGE("System is tainted: %s", taint), | |
3633 | "TAINT=%s", taint, | |
3634 | "MESSAGE_ID=" SD_MESSAGE_TAINTED_STR); | |
3635 | } | |
3636 | ||
3637 | static void manager_notify_finished(Manager *m) { | |
3638 | char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX]; | |
3639 | usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec; | |
3640 | ||
3641 | if (MANAGER_IS_TEST_RUN(m)) | |
3642 | return; | |
3643 | ||
3644 | if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) { | |
3645 | char ts[FORMAT_TIMESPAN_MAX]; | |
3646 | char buf[FORMAT_TIMESPAN_MAX + STRLEN(" (firmware) + ") + FORMAT_TIMESPAN_MAX + STRLEN(" (loader) + ")] | |
3647 | = {}; | |
3648 | char *p = buf; | |
3649 | size_t size = sizeof buf; | |
3650 | ||
3651 | /* Note that MANAGER_TIMESTAMP_KERNEL's monotonic value is always at 0, and | |
3652 | * MANAGER_TIMESTAMP_FIRMWARE's and MANAGER_TIMESTAMP_LOADER's monotonic value should be considered | |
3653 | * negative values. */ | |
3654 | ||
3655 | firmware_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic - m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic; | |
3656 | loader_usec = m->timestamps[MANAGER_TIMESTAMP_LOADER].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; | |
3657 | userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic; | |
3658 | total_usec = m->timestamps[MANAGER_TIMESTAMP_FIRMWARE].monotonic + m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic; | |
3659 | ||
3660 | if (firmware_usec > 0) | |
3661 | size = strpcpyf(&p, size, "%s (firmware) + ", format_timespan(ts, sizeof(ts), firmware_usec, USEC_PER_MSEC)); | |
3662 | if (loader_usec > 0) | |
3663 | size = strpcpyf(&p, size, "%s (loader) + ", format_timespan(ts, sizeof(ts), loader_usec, USEC_PER_MSEC)); | |
3664 | ||
3665 | if (dual_timestamp_is_set(&m->timestamps[MANAGER_TIMESTAMP_INITRD])) { | |
3666 | ||
3667 | /* The initrd case on bare-metal*/ | |
3668 | kernel_usec = m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; | |
3669 | initrd_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_INITRD].monotonic; | |
3670 | ||
3671 | log_struct(LOG_INFO, | |
3672 | "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR, | |
3673 | "KERNEL_USEC="USEC_FMT, kernel_usec, | |
3674 | "INITRD_USEC="USEC_FMT, initrd_usec, | |
3675 | "USERSPACE_USEC="USEC_FMT, userspace_usec, | |
3676 | LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (initrd) + %s (userspace) = %s.", | |
3677 | buf, | |
3678 | format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC), | |
3679 | format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC), | |
3680 | format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC), | |
3681 | format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC))); | |
3682 | } else { | |
3683 | /* The initrd-less case on bare-metal*/ | |
3684 | ||
3685 | kernel_usec = m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic - m->timestamps[MANAGER_TIMESTAMP_KERNEL].monotonic; | |
3686 | initrd_usec = 0; | |
3687 | ||
3688 | log_struct(LOG_INFO, | |
3689 | "MESSAGE_ID=" SD_MESSAGE_STARTUP_FINISHED_STR, | |
3690 | "KERNEL_USEC="USEC_FMT, kernel_usec, | |
3691 | "USERSPACE_USEC="USEC_FMT, userspace_usec, | |
3692 | LOG_MESSAGE("Startup finished in %s%s (kernel) + %s (userspace) = %s.", | |
3693 | buf, | |
3694 | format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC), | |
3695 | format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC), | |
3696 | format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC))); | |
3697 | } | |
3698 | } else { | |
3699 | /* The container and --user case */ | |
3700 | firmware_usec = loader_usec = initrd_usec = kernel_usec = 0; | |
3701 | total_usec = userspace_usec = m->timestamps[MANAGER_TIMESTAMP_FINISH].monotonic - m->timestamps[MANAGER_TIMESTAMP_USERSPACE].monotonic; | |
3702 | ||
3703 | log_struct(LOG_INFO, | |
3704 | "MESSAGE_ID=" SD_MESSAGE_USER_STARTUP_FINISHED_STR, | |
3705 | "USERSPACE_USEC="USEC_FMT, userspace_usec, | |
3706 | LOG_MESSAGE("Startup finished in %s.", | |
3707 | format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC))); | |
3708 | } | |
3709 | ||
3710 | bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec); | |
3711 | ||
3712 | sd_notifyf(false, | |
3713 | m->ready_sent ? "STATUS=Startup finished in %s." | |
3714 | : "READY=1\n" | |
3715 | "STATUS=Startup finished in %s.", | |
3716 | format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)); | |
3717 | m->ready_sent = true; | |
3718 | ||
3719 | log_taint_string(m); | |
3720 | } | |
3721 | ||
3722 | static void manager_send_ready(Manager *m) { | |
3723 | assert(m); | |
3724 | ||
3725 | /* We send READY=1 on reaching basic.target only when running in --user mode. */ | |
3726 | if (!MANAGER_IS_USER(m) || m->ready_sent) | |
3727 | return; | |
3728 | ||
3729 | m->ready_sent = true; | |
3730 | ||
3731 | sd_notifyf(false, | |
3732 | "READY=1\n" | |
3733 | "STATUS=Reached " SPECIAL_BASIC_TARGET "."); | |
3734 | } | |
3735 | ||
3736 | static void manager_check_basic_target(Manager *m) { | |
3737 | Unit *u; | |
3738 | ||
3739 | assert(m); | |
3740 | ||
3741 | /* Small shortcut */ | |
3742 | if (m->ready_sent && m->taint_logged) | |
3743 | return; | |
3744 | ||
3745 | u = manager_get_unit(m, SPECIAL_BASIC_TARGET); | |
3746 | if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) | |
3747 | return; | |
3748 | ||
3749 | /* For user managers, send out READY=1 as soon as we reach basic.target */ | |
3750 | manager_send_ready(m); | |
3751 | ||
3752 | /* Log the taint string as soon as we reach basic.target */ | |
3753 | log_taint_string(m); | |
3754 | } | |
3755 | ||
3756 | void manager_check_finished(Manager *m) { | |
3757 | assert(m); | |
3758 | ||
3759 | if (MANAGER_IS_RELOADING(m)) | |
3760 | return; | |
3761 | ||
3762 | /* Verify that we have entered the event loop already, and not left it again. */ | |
3763 | if (!MANAGER_IS_RUNNING(m)) | |
3764 | return; | |
3765 | ||
3766 | manager_check_basic_target(m); | |
3767 | ||
3768 | if (hashmap_size(m->jobs) > 0) { | |
3769 | if (m->jobs_in_progress_event_source) | |
3770 | /* Ignore any failure, this is only for feedback */ | |
3771 | (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC); | |
3772 | ||
3773 | return; | |
3774 | } | |
3775 | ||
3776 | manager_flip_auto_status(m, false); | |
3777 | ||
3778 | /* Notify Type=idle units that we are done now */ | |
3779 | manager_close_idle_pipe(m); | |
3780 | ||
3781 | /* Turn off confirm spawn now */ | |
3782 | m->confirm_spawn = NULL; | |
3783 | ||
3784 | /* No need to update ask password status when we're going non-interactive */ | |
3785 | manager_close_ask_password(m); | |
3786 | ||
3787 | /* This is no longer the first boot */ | |
3788 | manager_set_first_boot(m, false); | |
3789 | ||
3790 | if (MANAGER_IS_FINISHED(m)) | |
3791 | return; | |
3792 | ||
3793 | dual_timestamp_get(m->timestamps + MANAGER_TIMESTAMP_FINISH); | |
3794 | ||
3795 | manager_notify_finished(m); | |
3796 | ||
3797 | manager_invalidate_startup_units(m); | |
3798 | } | |
3799 | ||
3800 | static bool generator_path_any(const char* const* paths) { | |
3801 | char **path; | |
3802 | bool found = false; | |
3803 | ||
3804 | /* Optimize by skipping the whole process by not creating output directories | |
3805 | * if no generators are found. */ | |
3806 | STRV_FOREACH(path, (char**) paths) | |
3807 | if (access(*path, F_OK) == 0) | |
3808 | found = true; | |
3809 | else if (errno != ENOENT) | |
3810 | log_warning_errno(errno, "Failed to open generator directory %s: %m", *path); | |
3811 | ||
3812 | return found; | |
3813 | } | |
3814 | ||
3815 | static const char *const system_env_generator_binary_paths[] = { | |
3816 | "/run/systemd/system-environment-generators", | |
3817 | "/etc/systemd/system-environment-generators", | |
3818 | "/usr/local/lib/systemd/system-environment-generators", | |
3819 | SYSTEM_ENV_GENERATOR_PATH, | |
3820 | NULL | |
3821 | }; | |
3822 | ||
3823 | static const char *const user_env_generator_binary_paths[] = { | |
3824 | "/run/systemd/user-environment-generators", | |
3825 | "/etc/systemd/user-environment-generators", | |
3826 | "/usr/local/lib/systemd/user-environment-generators", | |
3827 | USER_ENV_GENERATOR_PATH, | |
3828 | NULL | |
3829 | }; | |
3830 | ||
3831 | static int manager_run_environment_generators(Manager *m) { | |
3832 | char **tmp = NULL; /* this is only used in the forked process, no cleanup here */ | |
3833 | const char *const *paths; | |
3834 | void* args[] = { | |
3835 | [STDOUT_GENERATE] = &tmp, | |
3836 | [STDOUT_COLLECT] = &tmp, | |
3837 | [STDOUT_CONSUME] = &m->transient_environment, | |
3838 | }; | |
3839 | int r; | |
3840 | ||
3841 | if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_ENV_GENERATORS)) | |
3842 | return 0; | |
3843 | ||
3844 | paths = MANAGER_IS_SYSTEM(m) ? system_env_generator_binary_paths : user_env_generator_binary_paths; | |
3845 | ||
3846 | if (!generator_path_any(paths)) | |
3847 | return 0; | |
3848 | ||
3849 | RUN_WITH_UMASK(0022) | |
3850 | r = execute_directories(paths, DEFAULT_TIMEOUT_USEC, gather_environment, | |
3851 | args, NULL, m->transient_environment, EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS); | |
3852 | return r; | |
3853 | } | |
3854 | ||
3855 | static int manager_run_generators(Manager *m) { | |
3856 | _cleanup_strv_free_ char **paths = NULL; | |
3857 | const char *argv[5]; | |
3858 | int r; | |
3859 | ||
3860 | assert(m); | |
3861 | ||
3862 | if (MANAGER_IS_TEST_RUN(m) && !(m->test_run_flags & MANAGER_TEST_RUN_GENERATORS)) | |
3863 | return 0; | |
3864 | ||
3865 | paths = generator_binary_paths(m->unit_file_scope); | |
3866 | if (!paths) | |
3867 | return log_oom(); | |
3868 | ||
3869 | if (!generator_path_any((const char* const*) paths)) | |
3870 | return 0; | |
3871 | ||
3872 | r = lookup_paths_mkdir_generator(&m->lookup_paths); | |
3873 | if (r < 0) { | |
3874 | log_error_errno(r, "Failed to create generator directories: %m"); | |
3875 | goto finish; | |
3876 | } | |
3877 | ||
3878 | argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */ | |
3879 | argv[1] = m->lookup_paths.generator; | |
3880 | argv[2] = m->lookup_paths.generator_early; | |
3881 | argv[3] = m->lookup_paths.generator_late; | |
3882 | argv[4] = NULL; | |
3883 | ||
3884 | RUN_WITH_UMASK(0022) | |
3885 | (void) execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, NULL, NULL, | |
3886 | (char**) argv, m->transient_environment, EXEC_DIR_PARALLEL | EXEC_DIR_IGNORE_ERRORS); | |
3887 | ||
3888 | r = 0; | |
3889 | ||
3890 | finish: | |
3891 | lookup_paths_trim_generator(&m->lookup_paths); | |
3892 | return r; | |
3893 | } | |
3894 | ||
3895 | int manager_transient_environment_add(Manager *m, char **plus) { | |
3896 | char **a; | |
3897 | ||
3898 | assert(m); | |
3899 | ||
3900 | if (strv_isempty(plus)) | |
3901 | return 0; | |
3902 | ||
3903 | a = strv_env_merge(2, m->transient_environment, plus); | |
3904 | if (!a) | |
3905 | return log_oom(); | |
3906 | ||
3907 | sanitize_environment(a); | |
3908 | ||
3909 | return strv_free_and_replace(m->transient_environment, a); | |
3910 | } | |
3911 | ||
3912 | int manager_client_environment_modify( | |
3913 | Manager *m, | |
3914 | char **minus, | |
3915 | char **plus) { | |
3916 | ||
3917 | char **a = NULL, **b = NULL, **l; | |
3918 | ||
3919 | assert(m); | |
3920 | ||
3921 | if (strv_isempty(minus) && strv_isempty(plus)) | |
3922 | return 0; | |
3923 | ||
3924 | l = m->client_environment; | |
3925 | ||
3926 | if (!strv_isempty(minus)) { | |
3927 | a = strv_env_delete(l, 1, minus); | |
3928 | if (!a) | |
3929 | return -ENOMEM; | |
3930 | ||
3931 | l = a; | |
3932 | } | |
3933 | ||
3934 | if (!strv_isempty(plus)) { | |
3935 | b = strv_env_merge(2, l, plus); | |
3936 | if (!b) { | |
3937 | strv_free(a); | |
3938 | return -ENOMEM; | |
3939 | } | |
3940 | ||
3941 | l = b; | |
3942 | } | |
3943 | ||
3944 | if (m->client_environment != l) | |
3945 | strv_free(m->client_environment); | |
3946 | ||
3947 | if (a != l) | |
3948 | strv_free(a); | |
3949 | if (b != l) | |
3950 | strv_free(b); | |
3951 | ||
3952 | m->client_environment = sanitize_environment(l); | |
3953 | return 0; | |
3954 | } | |
3955 | ||
3956 | int manager_get_effective_environment(Manager *m, char ***ret) { | |
3957 | char **l; | |
3958 | ||
3959 | assert(m); | |
3960 | assert(ret); | |
3961 | ||
3962 | l = strv_env_merge(2, m->transient_environment, m->client_environment); | |
3963 | if (!l) | |
3964 | return -ENOMEM; | |
3965 | ||
3966 | *ret = l; | |
3967 | return 0; | |
3968 | } | |
3969 | ||
3970 | int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) { | |
3971 | int i; | |
3972 | ||
3973 | assert(m); | |
3974 | ||
3975 | for (i = 0; i < _RLIMIT_MAX; i++) { | |
3976 | m->rlimit[i] = mfree(m->rlimit[i]); | |
3977 | ||
3978 | if (!default_rlimit[i]) | |
3979 | continue; | |
3980 | ||
3981 | m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1); | |
3982 | if (!m->rlimit[i]) | |
3983 | return log_oom(); | |
3984 | } | |
3985 | ||
3986 | return 0; | |
3987 | } | |
3988 | ||
3989 | void manager_recheck_dbus(Manager *m) { | |
3990 | assert(m); | |
3991 | ||
3992 | /* Connects to the bus if the dbus service and socket are running. If we are running in user mode this is all | |
3993 | * it does. In system mode we'll also connect to the system bus (which will most likely just reuse the | |
3994 | * connection of the API bus). That's because the system bus after all runs as service of the system instance, | |
3995 | * while in the user instance we can assume it's already there. */ | |
3996 | ||
3997 | if (MANAGER_IS_RELOADING(m)) | |
3998 | return; /* don't check while we are reloading… */ | |
3999 | ||
4000 | if (manager_dbus_is_running(m, false)) { | |
4001 | (void) bus_init_api(m); | |
4002 | ||
4003 | if (MANAGER_IS_SYSTEM(m)) | |
4004 | (void) bus_init_system(m); | |
4005 | } else { | |
4006 | (void) bus_done_api(m); | |
4007 | ||
4008 | if (MANAGER_IS_SYSTEM(m)) | |
4009 | (void) bus_done_system(m); | |
4010 | } | |
4011 | } | |
4012 | ||
4013 | static bool manager_journal_is_running(Manager *m) { | |
4014 | Unit *u; | |
4015 | ||
4016 | assert(m); | |
4017 | ||
4018 | if (MANAGER_IS_TEST_RUN(m)) | |
4019 | return false; | |
4020 | ||
4021 | /* If we are the user manager we can safely assume that the journal is up */ | |
4022 | if (!MANAGER_IS_SYSTEM(m)) | |
4023 | return true; | |
4024 | ||
4025 | /* Check that the socket is not only up, but in RUNNING state */ | |
4026 | u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET); | |
4027 | if (!u) | |
4028 | return false; | |
4029 | if (SOCKET(u)->state != SOCKET_RUNNING) | |
4030 | return false; | |
4031 | ||
4032 | /* Similar, check if the daemon itself is fully up, too */ | |
4033 | u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE); | |
4034 | if (!u) | |
4035 | return false; | |
4036 | if (!IN_SET(SERVICE(u)->state, SERVICE_RELOAD, SERVICE_RUNNING)) | |
4037 | return false; | |
4038 | ||
4039 | return true; | |
4040 | } | |
4041 | ||
4042 | void manager_recheck_journal(Manager *m) { | |
4043 | ||
4044 | assert(m); | |
4045 | ||
4046 | /* Don't bother with this unless we are in the special situation of being PID 1 */ | |
4047 | if (getpid_cached() != 1) | |
4048 | return; | |
4049 | ||
4050 | /* Don't check this while we are reloading, things might still change */ | |
4051 | if (MANAGER_IS_RELOADING(m)) | |
4052 | return; | |
4053 | ||
4054 | /* The journal is fully and entirely up? If so, let's permit logging to it, if that's configured. If the | |
4055 | * journal is down, don't ever log to it, otherwise we might end up deadlocking ourselves as we might trigger | |
4056 | * an activation ourselves we can't fulfill. */ | |
4057 | log_set_prohibit_ipc(!manager_journal_is_running(m)); | |
4058 | log_open(); | |
4059 | } | |
4060 | ||
4061 | void manager_set_show_status(Manager *m, ShowStatus mode) { | |
4062 | assert(m); | |
4063 | assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY)); | |
4064 | ||
4065 | if (!MANAGER_IS_SYSTEM(m)) | |
4066 | return; | |
4067 | ||
4068 | if (m->show_status != mode) | |
4069 | log_debug("%s showing of status.", | |
4070 | mode == SHOW_STATUS_NO ? "Disabling" : "Enabling"); | |
4071 | m->show_status = mode; | |
4072 | ||
4073 | if (IN_SET(mode, SHOW_STATUS_TEMPORARY, SHOW_STATUS_YES)) | |
4074 | (void) touch("/run/systemd/show-status"); | |
4075 | else | |
4076 | (void) unlink("/run/systemd/show-status"); | |
4077 | } | |
4078 | ||
4079 | static bool manager_get_show_status(Manager *m, StatusType type) { | |
4080 | assert(m); | |
4081 | ||
4082 | if (!MANAGER_IS_SYSTEM(m)) | |
4083 | return false; | |
4084 | ||
4085 | if (m->no_console_output) | |
4086 | return false; | |
4087 | ||
4088 | if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING)) | |
4089 | return false; | |
4090 | ||
4091 | /* If we cannot find out the status properly, just proceed. */ | |
4092 | if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0) | |
4093 | return false; | |
4094 | ||
4095 | return IN_SET(m->show_status, SHOW_STATUS_TEMPORARY, SHOW_STATUS_YES); | |
4096 | } | |
4097 | ||
4098 | const char *manager_get_confirm_spawn(Manager *m) { | |
4099 | static int last_errno = 0; | |
4100 | const char *vc = m->confirm_spawn; | |
4101 | struct stat st; | |
4102 | int r; | |
4103 | ||
4104 | /* Here's the deal: we want to test the validity of the console but don't want | |
4105 | * PID1 to go through the whole console process which might block. But we also | |
4106 | * want to warn the user only once if something is wrong with the console so we | |
4107 | * cannot do the sanity checks after spawning our children. So here we simply do | |
4108 | * really basic tests to hopefully trap common errors. | |
4109 | * | |
4110 | * If the console suddenly disappear at the time our children will really it | |
4111 | * then they will simply fail to acquire it and a positive answer will be | |
4112 | * assumed. New children will fallback to /dev/console though. | |
4113 | * | |
4114 | * Note: TTYs are devices that can come and go any time, and frequently aren't | |
4115 | * available yet during early boot (consider a USB rs232 dongle...). If for any | |
4116 | * reason the configured console is not ready, we fallback to the default | |
4117 | * console. */ | |
4118 | ||
4119 | if (!vc || path_equal(vc, "/dev/console")) | |
4120 | return vc; | |
4121 | ||
4122 | r = stat(vc, &st); | |
4123 | if (r < 0) | |
4124 | goto fail; | |
4125 | ||
4126 | if (!S_ISCHR(st.st_mode)) { | |
4127 | errno = ENOTTY; | |
4128 | goto fail; | |
4129 | } | |
4130 | ||
4131 | last_errno = 0; | |
4132 | return vc; | |
4133 | fail: | |
4134 | if (last_errno != errno) { | |
4135 | last_errno = errno; | |
4136 | log_warning_errno(errno, "Failed to open %s: %m, using default console", vc); | |
4137 | } | |
4138 | return "/dev/console"; | |
4139 | } | |
4140 | ||
4141 | void manager_set_first_boot(Manager *m, bool b) { | |
4142 | assert(m); | |
4143 | ||
4144 | if (!MANAGER_IS_SYSTEM(m)) | |
4145 | return; | |
4146 | ||
4147 | if (m->first_boot != (int) b) { | |
4148 | if (b) | |
4149 | (void) touch("/run/systemd/first-boot"); | |
4150 | else | |
4151 | (void) unlink("/run/systemd/first-boot"); | |
4152 | } | |
4153 | ||
4154 | m->first_boot = b; | |
4155 | } | |
4156 | ||
4157 | void manager_disable_confirm_spawn(void) { | |
4158 | (void) touch("/run/systemd/confirm_spawn_disabled"); | |
4159 | } | |
4160 | ||
4161 | bool manager_is_confirm_spawn_disabled(Manager *m) { | |
4162 | if (!m->confirm_spawn) | |
4163 | return true; | |
4164 | ||
4165 | return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0; | |
4166 | } | |
4167 | ||
4168 | void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) { | |
4169 | va_list ap; | |
4170 | ||
4171 | /* If m is NULL, assume we're after shutdown and let the messages through. */ | |
4172 | ||
4173 | if (m && !manager_get_show_status(m, type)) | |
4174 | return; | |
4175 | ||
4176 | /* XXX We should totally drop the check for ephemeral here | |
4177 | * and thus effectively make 'Type=idle' pointless. */ | |
4178 | if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0) | |
4179 | return; | |
4180 | ||
4181 | va_start(ap, format); | |
4182 | status_vprintf(status, SHOW_STATUS_ELLIPSIZE|(type == STATUS_TYPE_EPHEMERAL ? SHOW_STATUS_EPHEMERAL : 0), format, ap); | |
4183 | va_end(ap); | |
4184 | } | |
4185 | ||
4186 | Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) { | |
4187 | char p[strlen(path)+1]; | |
4188 | ||
4189 | assert(m); | |
4190 | assert(path); | |
4191 | ||
4192 | strcpy(p, path); | |
4193 | path_simplify(p, false); | |
4194 | ||
4195 | return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p); | |
4196 | } | |
4197 | ||
4198 | int manager_update_failed_units(Manager *m, Unit *u, bool failed) { | |
4199 | unsigned size; | |
4200 | int r; | |
4201 | ||
4202 | assert(m); | |
4203 | assert(u->manager == m); | |
4204 | ||
4205 | size = set_size(m->failed_units); | |
4206 | ||
4207 | if (failed) { | |
4208 | r = set_ensure_allocated(&m->failed_units, NULL); | |
4209 | if (r < 0) | |
4210 | return log_oom(); | |
4211 | ||
4212 | if (set_put(m->failed_units, u) < 0) | |
4213 | return log_oom(); | |
4214 | } else | |
4215 | (void) set_remove(m->failed_units, u); | |
4216 | ||
4217 | if (set_size(m->failed_units) != size) | |
4218 | bus_manager_send_change_signal(m); | |
4219 | ||
4220 | return 0; | |
4221 | } | |
4222 | ||
4223 | ManagerState manager_state(Manager *m) { | |
4224 | Unit *u; | |
4225 | ||
4226 | assert(m); | |
4227 | ||
4228 | /* Did we ever finish booting? If not then we are still starting up */ | |
4229 | if (!MANAGER_IS_FINISHED(m)) { | |
4230 | ||
4231 | u = manager_get_unit(m, SPECIAL_BASIC_TARGET); | |
4232 | if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) | |
4233 | return MANAGER_INITIALIZING; | |
4234 | ||
4235 | return MANAGER_STARTING; | |
4236 | } | |
4237 | ||
4238 | /* Is the special shutdown target active or queued? If so, we are in shutdown state */ | |
4239 | u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET); | |
4240 | if (u && unit_active_or_pending(u)) | |
4241 | return MANAGER_STOPPING; | |
4242 | ||
4243 | if (MANAGER_IS_SYSTEM(m)) { | |
4244 | /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */ | |
4245 | u = manager_get_unit(m, SPECIAL_RESCUE_TARGET); | |
4246 | if (u && unit_active_or_pending(u)) | |
4247 | return MANAGER_MAINTENANCE; | |
4248 | ||
4249 | u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET); | |
4250 | if (u && unit_active_or_pending(u)) | |
4251 | return MANAGER_MAINTENANCE; | |
4252 | } | |
4253 | ||
4254 | /* Are there any failed units? If so, we are in degraded mode */ | |
4255 | if (set_size(m->failed_units) > 0) | |
4256 | return MANAGER_DEGRADED; | |
4257 | ||
4258 | return MANAGER_RUNNING; | |
4259 | } | |
4260 | ||
4261 | #define DESTROY_IPC_FLAG (UINT32_C(1) << 31) | |
4262 | ||
4263 | static void manager_unref_uid_internal( | |
4264 | Manager *m, | |
4265 | Hashmap **uid_refs, | |
4266 | uid_t uid, | |
4267 | bool destroy_now, | |
4268 | int (*_clean_ipc)(uid_t uid)) { | |
4269 | ||
4270 | uint32_t c, n; | |
4271 | ||
4272 | assert(m); | |
4273 | assert(uid_refs); | |
4274 | assert(uid_is_valid(uid)); | |
4275 | assert(_clean_ipc); | |
4276 | ||
4277 | /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption | |
4278 | * that uid_t and gid_t are actually defined the same way, with the same validity rules. | |
4279 | * | |
4280 | * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest | |
4281 | * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID | |
4282 | * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added | |
4283 | * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */ | |
4284 | ||
4285 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); | |
4286 | assert_cc(UID_INVALID == (uid_t) GID_INVALID); | |
4287 | ||
4288 | if (uid == 0) /* We don't keep track of root, and will never destroy it */ | |
4289 | return; | |
4290 | ||
4291 | c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid))); | |
4292 | ||
4293 | n = c & ~DESTROY_IPC_FLAG; | |
4294 | assert(n > 0); | |
4295 | n--; | |
4296 | ||
4297 | if (destroy_now && n == 0) { | |
4298 | hashmap_remove(*uid_refs, UID_TO_PTR(uid)); | |
4299 | ||
4300 | if (c & DESTROY_IPC_FLAG) { | |
4301 | log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.", | |
4302 | _clean_ipc == clean_ipc_by_uid ? "UID" : "GID", | |
4303 | uid); | |
4304 | (void) _clean_ipc(uid); | |
4305 | } | |
4306 | } else { | |
4307 | c = n | (c & DESTROY_IPC_FLAG); | |
4308 | assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0); | |
4309 | } | |
4310 | } | |
4311 | ||
4312 | void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) { | |
4313 | manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid); | |
4314 | } | |
4315 | ||
4316 | void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) { | |
4317 | manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid); | |
4318 | } | |
4319 | ||
4320 | static int manager_ref_uid_internal( | |
4321 | Manager *m, | |
4322 | Hashmap **uid_refs, | |
4323 | uid_t uid, | |
4324 | bool clean_ipc) { | |
4325 | ||
4326 | uint32_t c, n; | |
4327 | int r; | |
4328 | ||
4329 | assert(m); | |
4330 | assert(uid_refs); | |
4331 | assert(uid_is_valid(uid)); | |
4332 | ||
4333 | /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption | |
4334 | * that uid_t and gid_t are actually defined the same way, with the same validity rules. */ | |
4335 | ||
4336 | assert_cc(sizeof(uid_t) == sizeof(gid_t)); | |
4337 | assert_cc(UID_INVALID == (uid_t) GID_INVALID); | |
4338 | ||
4339 | if (uid == 0) /* We don't keep track of root, and will never destroy it */ | |
4340 | return 0; | |
4341 | ||
4342 | r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops); | |
4343 | if (r < 0) | |
4344 | return r; | |
4345 | ||
4346 | c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid))); | |
4347 | ||
4348 | n = c & ~DESTROY_IPC_FLAG; | |
4349 | n++; | |
4350 | ||
4351 | if (n & DESTROY_IPC_FLAG) /* check for overflow */ | |
4352 | return -EOVERFLOW; | |
4353 | ||
4354 | c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0); | |
4355 | ||
4356 | return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)); | |
4357 | } | |
4358 | ||
4359 | int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) { | |
4360 | return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc); | |
4361 | } | |
4362 | ||
4363 | int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) { | |
4364 | return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc); | |
4365 | } | |
4366 | ||
4367 | static void manager_vacuum_uid_refs_internal( | |
4368 | Manager *m, | |
4369 | Hashmap **uid_refs, | |
4370 | int (*_clean_ipc)(uid_t uid)) { | |
4371 | ||
4372 | Iterator i; | |
4373 | void *p, *k; | |
4374 | ||
4375 | assert(m); | |
4376 | assert(uid_refs); | |
4377 | assert(_clean_ipc); | |
4378 | ||
4379 | HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) { | |
4380 | uint32_t c, n; | |
4381 | uid_t uid; | |
4382 | ||
4383 | uid = PTR_TO_UID(k); | |
4384 | c = PTR_TO_UINT32(p); | |
4385 | ||
4386 | n = c & ~DESTROY_IPC_FLAG; | |
4387 | if (n > 0) | |
4388 | continue; | |
4389 | ||
4390 | if (c & DESTROY_IPC_FLAG) { | |
4391 | log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.", | |
4392 | _clean_ipc == clean_ipc_by_uid ? "UID" : "GID", | |
4393 | uid); | |
4394 | (void) _clean_ipc(uid); | |
4395 | } | |
4396 | ||
4397 | assert_se(hashmap_remove(*uid_refs, k) == p); | |
4398 | } | |
4399 | } | |
4400 | ||
4401 | void manager_vacuum_uid_refs(Manager *m) { | |
4402 | manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid); | |
4403 | } | |
4404 | ||
4405 | void manager_vacuum_gid_refs(Manager *m) { | |
4406 | manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid); | |
4407 | } | |
4408 | ||
4409 | static void manager_serialize_uid_refs_internal( | |
4410 | Manager *m, | |
4411 | FILE *f, | |
4412 | Hashmap **uid_refs, | |
4413 | const char *field_name) { | |
4414 | ||
4415 | Iterator i; | |
4416 | void *p, *k; | |
4417 | ||
4418 | assert(m); | |
4419 | assert(f); | |
4420 | assert(uid_refs); | |
4421 | assert(field_name); | |
4422 | ||
4423 | /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter | |
4424 | * of it is better rebuild after a reload/reexec. */ | |
4425 | ||
4426 | HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) { | |
4427 | uint32_t c; | |
4428 | uid_t uid; | |
4429 | ||
4430 | uid = PTR_TO_UID(k); | |
4431 | c = PTR_TO_UINT32(p); | |
4432 | ||
4433 | if (!(c & DESTROY_IPC_FLAG)) | |
4434 | continue; | |
4435 | ||
4436 | (void) serialize_item_format(f, field_name, UID_FMT, uid); | |
4437 | } | |
4438 | } | |
4439 | ||
4440 | void manager_serialize_uid_refs(Manager *m, FILE *f) { | |
4441 | manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid"); | |
4442 | } | |
4443 | ||
4444 | void manager_serialize_gid_refs(Manager *m, FILE *f) { | |
4445 | manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid"); | |
4446 | } | |
4447 | ||
4448 | static void manager_deserialize_uid_refs_one_internal( | |
4449 | Manager *m, | |
4450 | Hashmap** uid_refs, | |
4451 | const char *value) { | |
4452 | ||
4453 | uid_t uid; | |
4454 | uint32_t c; | |
4455 | int r; | |
4456 | ||
4457 | assert(m); | |
4458 | assert(uid_refs); | |
4459 | assert(value); | |
4460 | ||
4461 | r = parse_uid(value, &uid); | |
4462 | if (r < 0 || uid == 0) { | |
4463 | log_debug("Unable to parse UID reference serialization"); | |
4464 | return; | |
4465 | } | |
4466 | ||
4467 | r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops); | |
4468 | if (r < 0) { | |
4469 | log_oom(); | |
4470 | return; | |
4471 | } | |
4472 | ||
4473 | c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid))); | |
4474 | if (c & DESTROY_IPC_FLAG) | |
4475 | return; | |
4476 | ||
4477 | c |= DESTROY_IPC_FLAG; | |
4478 | ||
4479 | r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)); | |
4480 | if (r < 0) { | |
4481 | log_debug_errno(r, "Failed to add UID reference entry: %m"); | |
4482 | return; | |
4483 | } | |
4484 | } | |
4485 | ||
4486 | void manager_deserialize_uid_refs_one(Manager *m, const char *value) { | |
4487 | manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value); | |
4488 | } | |
4489 | ||
4490 | void manager_deserialize_gid_refs_one(Manager *m, const char *value) { | |
4491 | manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value); | |
4492 | } | |
4493 | ||
4494 | int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) { | |
4495 | struct buffer { | |
4496 | uid_t uid; | |
4497 | gid_t gid; | |
4498 | char unit_name[UNIT_NAME_MAX+1]; | |
4499 | } _packed_ buffer; | |
4500 | ||
4501 | Manager *m = userdata; | |
4502 | ssize_t l; | |
4503 | size_t n; | |
4504 | Unit *u; | |
4505 | ||
4506 | assert_se(source); | |
4507 | assert_se(m); | |
4508 | ||
4509 | /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID | |
4510 | * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the | |
4511 | * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */ | |
4512 | ||
4513 | l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT); | |
4514 | if (l < 0) { | |
4515 | if (IN_SET(errno, EINTR, EAGAIN)) | |
4516 | return 0; | |
4517 | ||
4518 | return log_error_errno(errno, "Failed to read from user lookup fd: %m"); | |
4519 | } | |
4520 | ||
4521 | if ((size_t) l <= offsetof(struct buffer, unit_name)) { | |
4522 | log_warning("Received too short user lookup message, ignoring."); | |
4523 | return 0; | |
4524 | } | |
4525 | ||
4526 | if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) { | |
4527 | log_warning("Received too long user lookup message, ignoring."); | |
4528 | return 0; | |
4529 | } | |
4530 | ||
4531 | if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) { | |
4532 | log_warning("Got user lookup message with invalid UID/GID pair, ignoring."); | |
4533 | return 0; | |
4534 | } | |
4535 | ||
4536 | n = (size_t) l - offsetof(struct buffer, unit_name); | |
4537 | if (memchr(buffer.unit_name, 0, n)) { | |
4538 | log_warning("Received lookup message with embedded NUL character, ignoring."); | |
4539 | return 0; | |
4540 | } | |
4541 | ||
4542 | buffer.unit_name[n] = 0; | |
4543 | u = manager_get_unit(m, buffer.unit_name); | |
4544 | if (!u) { | |
4545 | log_debug("Got user lookup message but unit doesn't exist, ignoring."); | |
4546 | return 0; | |
4547 | } | |
4548 | ||
4549 | log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid); | |
4550 | ||
4551 | unit_notify_user_lookup(u, buffer.uid, buffer.gid); | |
4552 | return 0; | |
4553 | } | |
4554 | ||
4555 | char *manager_taint_string(Manager *m) { | |
4556 | _cleanup_free_ char *destination = NULL, *overflowuid = NULL, *overflowgid = NULL; | |
4557 | char *buf, *e; | |
4558 | int r; | |
4559 | ||
4560 | /* Returns a "taint string", e.g. "local-hwclock:var-run-bad". | |
4561 | * Only things that are detected at runtime should be tagged | |
4562 | * here. For stuff that is set during compilation, emit a warning | |
4563 | * in the configuration phase. */ | |
4564 | ||
4565 | assert(m); | |
4566 | ||
4567 | buf = new(char, sizeof("split-usr:" | |
4568 | "cgroups-missing:" | |
4569 | "local-hwclock:" | |
4570 | "var-run-bad:" | |
4571 | "overflowuid-not-65534:" | |
4572 | "overflowgid-not-65534:")); | |
4573 | if (!buf) | |
4574 | return NULL; | |
4575 | ||
4576 | e = buf; | |
4577 | buf[0] = 0; | |
4578 | ||
4579 | if (m->taint_usr) | |
4580 | e = stpcpy(e, "split-usr:"); | |
4581 | ||
4582 | if (access("/proc/cgroups", F_OK) < 0) | |
4583 | e = stpcpy(e, "cgroups-missing:"); | |
4584 | ||
4585 | if (clock_is_localtime(NULL) > 0) | |
4586 | e = stpcpy(e, "local-hwclock:"); | |
4587 | ||
4588 | r = readlink_malloc("/var/run", &destination); | |
4589 | if (r < 0 || !PATH_IN_SET(destination, "../run", "/run")) | |
4590 | e = stpcpy(e, "var-run-bad:"); | |
4591 | ||
4592 | r = read_one_line_file("/proc/sys/kernel/overflowuid", &overflowuid); | |
4593 | if (r >= 0 && !streq(overflowuid, "65534")) | |
4594 | e = stpcpy(e, "overflowuid-not-65534:"); | |
4595 | ||
4596 | r = read_one_line_file("/proc/sys/kernel/overflowgid", &overflowgid); | |
4597 | if (r >= 0 && !streq(overflowgid, "65534")) | |
4598 | e = stpcpy(e, "overflowgid-not-65534:"); | |
4599 | ||
4600 | /* remove the last ':' */ | |
4601 | if (e != buf) | |
4602 | e[-1] = 0; | |
4603 | ||
4604 | return buf; | |
4605 | } | |
4606 | ||
4607 | void manager_ref_console(Manager *m) { | |
4608 | assert(m); | |
4609 | ||
4610 | m->n_on_console++; | |
4611 | } | |
4612 | ||
4613 | void manager_unref_console(Manager *m) { | |
4614 | ||
4615 | assert(m->n_on_console > 0); | |
4616 | m->n_on_console--; | |
4617 | ||
4618 | if (m->n_on_console == 0) | |
4619 | m->no_console_output = false; /* unset no_console_output flag, since the console is definitely free now */ | |
4620 | } | |
4621 | ||
4622 | void manager_override_log_level(Manager *m, int level) { | |
4623 | _cleanup_free_ char *s = NULL; | |
4624 | assert(m); | |
4625 | ||
4626 | if (!m->log_level_overridden) { | |
4627 | m->original_log_level = log_get_max_level(); | |
4628 | m->log_level_overridden = true; | |
4629 | } | |
4630 | ||
4631 | (void) log_level_to_string_alloc(level, &s); | |
4632 | log_info("Setting log level to %s.", strna(s)); | |
4633 | ||
4634 | log_set_max_level(level); | |
4635 | } | |
4636 | ||
4637 | void manager_restore_original_log_level(Manager *m) { | |
4638 | _cleanup_free_ char *s = NULL; | |
4639 | assert(m); | |
4640 | ||
4641 | if (!m->log_level_overridden) | |
4642 | return; | |
4643 | ||
4644 | (void) log_level_to_string_alloc(m->original_log_level, &s); | |
4645 | log_info("Restoring log level to original (%s).", strna(s)); | |
4646 | ||
4647 | log_set_max_level(m->original_log_level); | |
4648 | m->log_level_overridden = false; | |
4649 | } | |
4650 | ||
4651 | void manager_override_log_target(Manager *m, LogTarget target) { | |
4652 | assert(m); | |
4653 | ||
4654 | if (!m->log_target_overridden) { | |
4655 | m->original_log_target = log_get_target(); | |
4656 | m->log_target_overridden = true; | |
4657 | } | |
4658 | ||
4659 | log_info("Setting log target to %s.", log_target_to_string(target)); | |
4660 | log_set_target(target); | |
4661 | } | |
4662 | ||
4663 | void manager_restore_original_log_target(Manager *m) { | |
4664 | assert(m); | |
4665 | ||
4666 | if (!m->log_target_overridden) | |
4667 | return; | |
4668 | ||
4669 | log_info("Restoring log target to original %s.", log_target_to_string(m->original_log_target)); | |
4670 | ||
4671 | log_set_target(m->original_log_target); | |
4672 | m->log_target_overridden = false; | |
4673 | } | |
4674 | ||
4675 | ManagerTimestamp manager_timestamp_initrd_mangle(ManagerTimestamp s) { | |
4676 | if (in_initrd() && | |
4677 | s >= MANAGER_TIMESTAMP_SECURITY_START && | |
4678 | s <= MANAGER_TIMESTAMP_UNITS_LOAD_FINISH) | |
4679 | return s - MANAGER_TIMESTAMP_SECURITY_START + MANAGER_TIMESTAMP_INITRD_SECURITY_START; | |
4680 | return s; | |
4681 | } | |
4682 | ||
4683 | static const char *const manager_state_table[_MANAGER_STATE_MAX] = { | |
4684 | [MANAGER_INITIALIZING] = "initializing", | |
4685 | [MANAGER_STARTING] = "starting", | |
4686 | [MANAGER_RUNNING] = "running", | |
4687 | [MANAGER_DEGRADED] = "degraded", | |
4688 | [MANAGER_MAINTENANCE] = "maintenance", | |
4689 | [MANAGER_STOPPING] = "stopping", | |
4690 | }; | |
4691 | ||
4692 | DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState); | |
4693 | ||
4694 | static const char *const manager_timestamp_table[_MANAGER_TIMESTAMP_MAX] = { | |
4695 | [MANAGER_TIMESTAMP_FIRMWARE] = "firmware", | |
4696 | [MANAGER_TIMESTAMP_LOADER] = "loader", | |
4697 | [MANAGER_TIMESTAMP_KERNEL] = "kernel", | |
4698 | [MANAGER_TIMESTAMP_INITRD] = "initrd", | |
4699 | [MANAGER_TIMESTAMP_USERSPACE] = "userspace", | |
4700 | [MANAGER_TIMESTAMP_FINISH] = "finish", | |
4701 | [MANAGER_TIMESTAMP_SECURITY_START] = "security-start", | |
4702 | [MANAGER_TIMESTAMP_SECURITY_FINISH] = "security-finish", | |
4703 | [MANAGER_TIMESTAMP_GENERATORS_START] = "generators-start", | |
4704 | [MANAGER_TIMESTAMP_GENERATORS_FINISH] = "generators-finish", | |
4705 | [MANAGER_TIMESTAMP_UNITS_LOAD_START] = "units-load-start", | |
4706 | [MANAGER_TIMESTAMP_UNITS_LOAD_FINISH] = "units-load-finish", | |
4707 | [MANAGER_TIMESTAMP_INITRD_SECURITY_START] = "initrd-security-start", | |
4708 | [MANAGER_TIMESTAMP_INITRD_SECURITY_FINISH] = "initrd-security-finish", | |
4709 | [MANAGER_TIMESTAMP_INITRD_GENERATORS_START] = "initrd-generators-start", | |
4710 | [MANAGER_TIMESTAMP_INITRD_GENERATORS_FINISH] = "initrd-generators-finish", | |
4711 | [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_START] = "initrd-units-load-start", | |
4712 | [MANAGER_TIMESTAMP_INITRD_UNITS_LOAD_FINISH] = "initrd-units-load-finish", | |
4713 | }; | |
4714 | ||
4715 | DEFINE_STRING_TABLE_LOOKUP(manager_timestamp, ManagerTimestamp); | |
4716 | ||
4717 | static const char* const oom_policy_table[_OOM_POLICY_MAX] = { | |
4718 | [OOM_CONTINUE] = "continue", | |
4719 | [OOM_STOP] = "stop", | |
4720 | [OOM_KILL] = "kill", | |
4721 | }; | |
4722 | ||
4723 | DEFINE_STRING_TABLE_LOOKUP(oom_policy, OOMPolicy); |