]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/udev/udevd.c
udev: use initialization instead of zeroing in one place
[thirdparty/systemd.git] / src / udev / udevd.c
1 /*
2 * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org>
3 * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
4 * Copyright (C) 2009 Canonical Ltd.
5 * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
6 *
7 * This program is free software: you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation, either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program. If not, see <http://www.gnu.org/licenses/>.
19 */
20
21 #include <stddef.h>
22 #include <signal.h>
23 #include <unistd.h>
24 #include <errno.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <stdbool.h>
28 #include <string.h>
29 #include <ctype.h>
30 #include <fcntl.h>
31 #include <time.h>
32 #include <getopt.h>
33 #include <dirent.h>
34 #include <sys/time.h>
35 #include <sys/prctl.h>
36 #include <sys/socket.h>
37 #include <sys/un.h>
38 #include <sys/signalfd.h>
39 #include <sys/epoll.h>
40 #include <sys/poll.h>
41 #include <sys/wait.h>
42 #include <sys/stat.h>
43 #include <sys/ioctl.h>
44 #include <sys/inotify.h>
45 #include <sys/utsname.h>
46
47 #include "udev.h"
48 #include "sd-daemon.h"
49 #include "cgroup-util.h"
50 #include "dev-setup.h"
51 #include "fileio.h"
52
53 static bool debug;
54
55 void udev_main_log(struct udev *udev, int priority,
56 const char *file, int line, const char *fn,
57 const char *format, va_list args)
58 {
59 log_metav(priority, file, line, fn, format, args);
60 }
61
62 static struct udev_rules *rules;
63 static struct udev_queue_export *udev_queue_export;
64 static struct udev_ctrl *udev_ctrl;
65 static struct udev_monitor *monitor;
66 static int worker_watch[2] = { -1, -1 };
67 static int fd_signal = -1;
68 static int fd_ep = -1;
69 static int fd_inotify = -1;
70 static bool stop_exec_queue;
71 static bool reload;
72 static int children;
73 static int children_max;
74 static int exec_delay;
75 static sigset_t sigmask_orig;
76 static UDEV_LIST(event_list);
77 static UDEV_LIST(worker_list);
78 char *udev_cgroup;
79 static bool udev_exit;
80
81 enum event_state {
82 EVENT_UNDEF,
83 EVENT_QUEUED,
84 EVENT_RUNNING,
85 };
86
87 struct event {
88 struct udev_list_node node;
89 struct udev *udev;
90 struct udev_device *dev;
91 enum event_state state;
92 int exitcode;
93 unsigned long long int delaying_seqnum;
94 unsigned long long int seqnum;
95 const char *devpath;
96 size_t devpath_len;
97 const char *devpath_old;
98 dev_t devnum;
99 int ifindex;
100 bool is_block;
101 #ifdef HAVE_FIRMWARE
102 bool nodelay;
103 #endif
104 };
105
106 static inline struct event *node_to_event(struct udev_list_node *node)
107 {
108 return container_of(node, struct event, node);
109 }
110
111 static void event_queue_cleanup(struct udev *udev, enum event_state type);
112
113 enum worker_state {
114 WORKER_UNDEF,
115 WORKER_RUNNING,
116 WORKER_IDLE,
117 WORKER_KILLED,
118 };
119
120 struct worker {
121 struct udev_list_node node;
122 struct udev *udev;
123 int refcount;
124 pid_t pid;
125 struct udev_monitor *monitor;
126 enum worker_state state;
127 struct event *event;
128 usec_t event_start_usec;
129 };
130
131 /* passed from worker to main process */
132 struct worker_message {
133 pid_t pid;
134 int exitcode;
135 };
136
137 static inline struct worker *node_to_worker(struct udev_list_node *node)
138 {
139 return container_of(node, struct worker, node);
140 }
141
142 static void event_queue_delete(struct event *event, bool export)
143 {
144 udev_list_node_remove(&event->node);
145
146 if (export) {
147 udev_queue_export_device_finished(udev_queue_export, event->dev);
148 log_debug("seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode);
149 }
150 udev_device_unref(event->dev);
151 free(event);
152 }
153
154 static struct worker *worker_ref(struct worker *worker)
155 {
156 worker->refcount++;
157 return worker;
158 }
159
160 static void worker_cleanup(struct worker *worker)
161 {
162 udev_list_node_remove(&worker->node);
163 udev_monitor_unref(worker->monitor);
164 children--;
165 free(worker);
166 }
167
168 static void worker_unref(struct worker *worker)
169 {
170 worker->refcount--;
171 if (worker->refcount > 0)
172 return;
173 log_debug("worker [%u] cleaned up\n", worker->pid);
174 worker_cleanup(worker);
175 }
176
177 static void worker_list_cleanup(struct udev *udev)
178 {
179 struct udev_list_node *loop, *tmp;
180
181 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
182 struct worker *worker = node_to_worker(loop);
183
184 worker_cleanup(worker);
185 }
186 }
187
188 static void worker_new(struct event *event)
189 {
190 struct udev *udev = event->udev;
191 struct worker *worker;
192 struct udev_monitor *worker_monitor;
193 pid_t pid;
194
195 /* listen for new events */
196 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
197 if (worker_monitor == NULL)
198 return;
199 /* allow the main daemon netlink address to send devices to the worker */
200 udev_monitor_allow_unicast_sender(worker_monitor, monitor);
201 udev_monitor_enable_receiving(worker_monitor);
202
203 worker = calloc(1, sizeof(struct worker));
204 if (worker == NULL) {
205 udev_monitor_unref(worker_monitor);
206 return;
207 }
208 /* worker + event reference */
209 worker->refcount = 2;
210 worker->udev = udev;
211
212 pid = fork();
213 switch (pid) {
214 case 0: {
215 struct udev_device *dev = NULL;
216 int fd_monitor;
217 struct epoll_event ep_signal, ep_monitor;
218 sigset_t mask;
219 int rc = EXIT_SUCCESS;
220
221 /* take initial device from queue */
222 dev = event->dev;
223 event->dev = NULL;
224
225 free(worker);
226 worker_list_cleanup(udev);
227 event_queue_cleanup(udev, EVENT_UNDEF);
228 udev_queue_export_unref(udev_queue_export);
229 udev_monitor_unref(monitor);
230 udev_ctrl_unref(udev_ctrl);
231 close(fd_signal);
232 close(fd_ep);
233 close(worker_watch[READ_END]);
234
235 sigfillset(&mask);
236 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
237 if (fd_signal < 0) {
238 log_error("error creating signalfd %m\n");
239 rc = 2;
240 goto out;
241 }
242
243 fd_ep = epoll_create1(EPOLL_CLOEXEC);
244 if (fd_ep < 0) {
245 log_error("error creating epoll fd: %m\n");
246 rc = 3;
247 goto out;
248 }
249
250 memset(&ep_signal, 0, sizeof(struct epoll_event));
251 ep_signal.events = EPOLLIN;
252 ep_signal.data.fd = fd_signal;
253
254 fd_monitor = udev_monitor_get_fd(worker_monitor);
255 memset(&ep_monitor, 0, sizeof(struct epoll_event));
256 ep_monitor.events = EPOLLIN;
257 ep_monitor.data.fd = fd_monitor;
258
259 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
260 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
261 log_error("fail to add fds to epoll: %m\n");
262 rc = 4;
263 goto out;
264 }
265
266 /* request TERM signal if parent exits */
267 prctl(PR_SET_PDEATHSIG, SIGTERM);
268
269 /* reset OOM score, we only protect the main daemon */
270 write_string_file("/proc/self/oom_score_adj", "0");
271
272 for (;;) {
273 struct udev_event *udev_event;
274 struct worker_message msg;
275 int err;
276
277 log_debug("seq %llu running\n", udev_device_get_seqnum(dev));
278 udev_event = udev_event_new(dev);
279 if (udev_event == NULL) {
280 rc = 5;
281 goto out;
282 }
283
284 /* needed for SIGCHLD/SIGTERM in spawn() */
285 udev_event->fd_signal = fd_signal;
286
287 if (exec_delay > 0)
288 udev_event->exec_delay = exec_delay;
289
290 /* apply rules, create node, symlinks */
291 err = udev_event_execute_rules(udev_event, rules, &sigmask_orig);
292
293 if (err == 0)
294 udev_event_execute_run(udev_event, &sigmask_orig);
295
296 /* apply/restore inotify watch */
297 if (err == 0 && udev_event->inotify_watch) {
298 udev_watch_begin(udev, dev);
299 udev_device_update_db(dev);
300 }
301
302 /* send processed event back to libudev listeners */
303 udev_monitor_send_device(worker_monitor, NULL, dev);
304
305 /* send udevd the result of the event execution */
306 memset(&msg, 0, sizeof(struct worker_message));
307 if (err != 0)
308 msg.exitcode = err;
309 msg.pid = getpid();
310 send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
311
312 log_debug("seq %llu processed with %i\n", udev_device_get_seqnum(dev), err);
313
314 udev_device_unref(dev);
315 dev = NULL;
316
317 if (udev_event->sigterm) {
318 udev_event_unref(udev_event);
319 goto out;
320 }
321
322 udev_event_unref(udev_event);
323
324 /* wait for more device messages from main udevd, or term signal */
325 while (dev == NULL) {
326 struct epoll_event ev[4];
327 int fdcount;
328 int i;
329
330 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
331 if (fdcount < 0) {
332 if (errno == EINTR)
333 continue;
334 log_error("failed to poll: %m\n");
335 goto out;
336 }
337
338 for (i = 0; i < fdcount; i++) {
339 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
340 dev = udev_monitor_receive_device(worker_monitor);
341 break;
342 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
343 struct signalfd_siginfo fdsi;
344 ssize_t size;
345
346 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
347 if (size != sizeof(struct signalfd_siginfo))
348 continue;
349 switch (fdsi.ssi_signo) {
350 case SIGTERM:
351 goto out;
352 }
353 }
354 }
355 }
356 }
357 out:
358 udev_device_unref(dev);
359 if (fd_signal >= 0)
360 close(fd_signal);
361 if (fd_ep >= 0)
362 close(fd_ep);
363 close(fd_inotify);
364 close(worker_watch[WRITE_END]);
365 udev_rules_unref(rules);
366 udev_builtin_exit(udev);
367 udev_monitor_unref(worker_monitor);
368 udev_unref(udev);
369 log_close();
370 exit(rc);
371 }
372 case -1:
373 udev_monitor_unref(worker_monitor);
374 event->state = EVENT_QUEUED;
375 free(worker);
376 log_error("fork of child failed: %m\n");
377 break;
378 default:
379 /* close monitor, but keep address around */
380 udev_monitor_disconnect(worker_monitor);
381 worker->monitor = worker_monitor;
382 worker->pid = pid;
383 worker->state = WORKER_RUNNING;
384 worker->event_start_usec = now(CLOCK_MONOTONIC);
385 worker->event = event;
386 event->state = EVENT_RUNNING;
387 udev_list_node_append(&worker->node, &worker_list);
388 children++;
389 log_debug("seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid);
390 break;
391 }
392 }
393
394 static void event_run(struct event *event)
395 {
396 struct udev_list_node *loop;
397
398 udev_list_node_foreach(loop, &worker_list) {
399 struct worker *worker = node_to_worker(loop);
400 ssize_t count;
401
402 if (worker->state != WORKER_IDLE)
403 continue;
404
405 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
406 if (count < 0) {
407 log_error("worker [%u] did not accept message %zi (%m), kill it\n", worker->pid, count);
408 kill(worker->pid, SIGKILL);
409 worker->state = WORKER_KILLED;
410 continue;
411 }
412 worker_ref(worker);
413 worker->event = event;
414 worker->state = WORKER_RUNNING;
415 worker->event_start_usec = now(CLOCK_MONOTONIC);
416 event->state = EVENT_RUNNING;
417 return;
418 }
419
420 if (children >= children_max) {
421 if (children_max > 1)
422 log_debug("maximum number (%i) of children reached\n", children);
423 return;
424 }
425
426 /* start new worker and pass initial device */
427 worker_new(event);
428 }
429
430 static int event_queue_insert(struct udev_device *dev)
431 {
432 struct event *event;
433
434 event = calloc(1, sizeof(struct event));
435 if (event == NULL)
436 return -1;
437
438 event->udev = udev_device_get_udev(dev);
439 event->dev = dev;
440 event->seqnum = udev_device_get_seqnum(dev);
441 event->devpath = udev_device_get_devpath(dev);
442 event->devpath_len = strlen(event->devpath);
443 event->devpath_old = udev_device_get_devpath_old(dev);
444 event->devnum = udev_device_get_devnum(dev);
445 event->is_block = streq("block", udev_device_get_subsystem(dev));
446 event->ifindex = udev_device_get_ifindex(dev);
447 #ifdef HAVE_FIRMWARE
448 if (streq(udev_device_get_subsystem(dev), "firmware"))
449 event->nodelay = true;
450 #endif
451
452 udev_queue_export_device_queued(udev_queue_export, dev);
453 log_debug("seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev),
454 udev_device_get_action(dev), udev_device_get_subsystem(dev));
455
456 event->state = EVENT_QUEUED;
457 udev_list_node_append(&event->node, &event_list);
458 return 0;
459 }
460
461 static void worker_kill(struct udev *udev)
462 {
463 struct udev_list_node *loop;
464
465 udev_list_node_foreach(loop, &worker_list) {
466 struct worker *worker = node_to_worker(loop);
467
468 if (worker->state == WORKER_KILLED)
469 continue;
470
471 worker->state = WORKER_KILLED;
472 kill(worker->pid, SIGTERM);
473 }
474 }
475
476 /* lookup event for identical, parent, child device */
477 static bool is_devpath_busy(struct event *event)
478 {
479 struct udev_list_node *loop;
480 size_t common;
481
482 /* check if queue contains events we depend on */
483 udev_list_node_foreach(loop, &event_list) {
484 struct event *loop_event = node_to_event(loop);
485
486 /* we already found a later event, earlier can not block us, no need to check again */
487 if (loop_event->seqnum < event->delaying_seqnum)
488 continue;
489
490 /* event we checked earlier still exists, no need to check again */
491 if (loop_event->seqnum == event->delaying_seqnum)
492 return true;
493
494 /* found ourself, no later event can block us */
495 if (loop_event->seqnum >= event->seqnum)
496 break;
497
498 /* check major/minor */
499 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
500 return true;
501
502 /* check network device ifindex */
503 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
504 return true;
505
506 /* check our old name */
507 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
508 event->delaying_seqnum = loop_event->seqnum;
509 return true;
510 }
511
512 /* compare devpath */
513 common = MIN(loop_event->devpath_len, event->devpath_len);
514
515 /* one devpath is contained in the other? */
516 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
517 continue;
518
519 /* identical device event found */
520 if (loop_event->devpath_len == event->devpath_len) {
521 /* devices names might have changed/swapped in the meantime */
522 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
523 continue;
524 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
525 continue;
526 event->delaying_seqnum = loop_event->seqnum;
527 return true;
528 }
529
530 #ifdef HAVE_FIRMWARE
531 /* allow to bypass the dependency tracking */
532 if (event->nodelay)
533 continue;
534 #endif
535
536 /* parent device event found */
537 if (event->devpath[common] == '/') {
538 event->delaying_seqnum = loop_event->seqnum;
539 return true;
540 }
541
542 /* child device event found */
543 if (loop_event->devpath[common] == '/') {
544 event->delaying_seqnum = loop_event->seqnum;
545 return true;
546 }
547
548 /* no matching device */
549 continue;
550 }
551
552 return false;
553 }
554
555 static void event_queue_start(struct udev *udev)
556 {
557 struct udev_list_node *loop;
558
559 udev_list_node_foreach(loop, &event_list) {
560 struct event *event = node_to_event(loop);
561
562 if (event->state != EVENT_QUEUED)
563 continue;
564
565 /* do not start event if parent or child event is still running */
566 if (is_devpath_busy(event))
567 continue;
568
569 event_run(event);
570 }
571 }
572
573 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
574 {
575 struct udev_list_node *loop, *tmp;
576
577 udev_list_node_foreach_safe(loop, tmp, &event_list) {
578 struct event *event = node_to_event(loop);
579
580 if (match_type != EVENT_UNDEF && match_type != event->state)
581 continue;
582
583 event_queue_delete(event, false);
584 }
585 }
586
587 static void worker_returned(int fd_worker)
588 {
589 for (;;) {
590 struct worker_message msg;
591 ssize_t size;
592 struct udev_list_node *loop;
593
594 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
595 if (size != sizeof(struct worker_message))
596 break;
597
598 /* lookup worker who sent the signal */
599 udev_list_node_foreach(loop, &worker_list) {
600 struct worker *worker = node_to_worker(loop);
601
602 if (worker->pid != msg.pid)
603 continue;
604
605 /* worker returned */
606 if (worker->event) {
607 worker->event->exitcode = msg.exitcode;
608 event_queue_delete(worker->event, true);
609 worker->event = NULL;
610 }
611 if (worker->state != WORKER_KILLED)
612 worker->state = WORKER_IDLE;
613 worker_unref(worker);
614 break;
615 }
616 }
617 }
618
619 /* receive the udevd message from userspace */
620 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
621 {
622 struct udev *udev = udev_ctrl_get_udev(uctrl);
623 struct udev_ctrl_connection *ctrl_conn;
624 struct udev_ctrl_msg *ctrl_msg = NULL;
625 const char *str;
626 int i;
627
628 ctrl_conn = udev_ctrl_get_connection(uctrl);
629 if (ctrl_conn == NULL)
630 goto out;
631
632 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
633 if (ctrl_msg == NULL)
634 goto out;
635
636 i = udev_ctrl_get_set_log_level(ctrl_msg);
637 if (i >= 0) {
638 log_debug("udevd message (SET_LOG_PRIORITY) received, log_priority=%i\n", i);
639 log_set_max_level(i);
640 udev_set_log_priority(udev, i);
641 worker_kill(udev);
642 }
643
644 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
645 log_debug("udevd message (STOP_EXEC_QUEUE) received\n");
646 stop_exec_queue = true;
647 }
648
649 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
650 log_debug("udevd message (START_EXEC_QUEUE) received\n");
651 stop_exec_queue = false;
652 }
653
654 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
655 log_debug("udevd message (RELOAD) received\n");
656 reload = true;
657 }
658
659 str = udev_ctrl_get_set_env(ctrl_msg);
660 if (str != NULL) {
661 char *key;
662
663 key = strdup(str);
664 if (key != NULL) {
665 char *val;
666
667 val = strchr(key, '=');
668 if (val != NULL) {
669 val[0] = '\0';
670 val = &val[1];
671 if (val[0] == '\0') {
672 log_debug("udevd message (ENV) received, unset '%s'\n", key);
673 udev_add_property(udev, key, NULL);
674 } else {
675 log_debug("udevd message (ENV) received, set '%s=%s'\n", key, val);
676 udev_add_property(udev, key, val);
677 }
678 } else {
679 log_error("wrong key format '%s'\n", key);
680 }
681 free(key);
682 }
683 worker_kill(udev);
684 }
685
686 i = udev_ctrl_get_set_children_max(ctrl_msg);
687 if (i >= 0) {
688 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i\n", i);
689 children_max = i;
690 }
691
692 if (udev_ctrl_get_ping(ctrl_msg) > 0)
693 log_debug("udevd message (SYNC) received\n");
694
695 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
696 log_debug("udevd message (EXIT) received\n");
697 udev_exit = true;
698 /* keep reference to block the client until we exit */
699 udev_ctrl_connection_ref(ctrl_conn);
700 }
701 out:
702 udev_ctrl_msg_unref(ctrl_msg);
703 return udev_ctrl_connection_unref(ctrl_conn);
704 }
705
706 /* read inotify messages */
707 static int handle_inotify(struct udev *udev)
708 {
709 int nbytes, pos;
710 char *buf;
711 struct inotify_event *ev;
712
713 if ((ioctl(fd_inotify, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
714 return 0;
715
716 buf = malloc(nbytes);
717 if (buf == NULL) {
718 log_error("error getting buffer for inotify\n");
719 return -1;
720 }
721
722 nbytes = read(fd_inotify, buf, nbytes);
723
724 for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
725 struct udev_device *dev;
726
727 ev = (struct inotify_event *)(buf + pos);
728 dev = udev_watch_lookup(udev, ev->wd);
729 if (dev != NULL) {
730 log_debug("inotify event: %x for %s\n", ev->mask, udev_device_get_devnode(dev));
731 if (ev->mask & IN_CLOSE_WRITE) {
732 char filename[UTIL_PATH_SIZE];
733 int fd;
734
735 log_debug("device %s closed, synthesising 'change'\n", udev_device_get_devnode(dev));
736 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
737 fd = open(filename, O_WRONLY);
738 if (fd >= 0) {
739 if (write(fd, "change", 6) < 0)
740 log_debug("error writing uevent: %m\n");
741 close(fd);
742 }
743 }
744 if (ev->mask & IN_IGNORED)
745 udev_watch_end(udev, dev);
746
747 udev_device_unref(dev);
748 }
749
750 }
751
752 free(buf);
753 return 0;
754 }
755
756 static void handle_signal(struct udev *udev, int signo)
757 {
758 switch (signo) {
759 case SIGINT:
760 case SIGTERM:
761 udev_exit = true;
762 break;
763 case SIGCHLD:
764 for (;;) {
765 pid_t pid;
766 int status;
767 struct udev_list_node *loop, *tmp;
768
769 pid = waitpid(-1, &status, WNOHANG);
770 if (pid <= 0)
771 break;
772
773 udev_list_node_foreach_safe(loop, tmp, &worker_list) {
774 struct worker *worker = node_to_worker(loop);
775
776 if (worker->pid != pid)
777 continue;
778 log_debug("worker [%u] exit\n", pid);
779
780 if (WIFEXITED(status)) {
781 if (WEXITSTATUS(status) != 0)
782 log_error("worker [%u] exit with return code %i\n",
783 pid, WEXITSTATUS(status));
784 } else if (WIFSIGNALED(status)) {
785 log_error("worker [%u] terminated by signal %i (%s)\n",
786 pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
787 } else if (WIFSTOPPED(status)) {
788 log_error("worker [%u] stopped\n", pid);
789 } else if (WIFCONTINUED(status)) {
790 log_error("worker [%u] continued\n", pid);
791 } else {
792 log_error("worker [%u] exit with status 0x%04x\n", pid, status);
793 }
794
795 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
796 if (worker->event) {
797 log_error("worker [%u] failed while handling '%s'\n",
798 pid, worker->event->devpath);
799 worker->event->exitcode = -32;
800 event_queue_delete(worker->event, true);
801 /* drop reference taken for state 'running' */
802 worker_unref(worker);
803 }
804 }
805 worker_unref(worker);
806 break;
807 }
808 }
809 break;
810 case SIGHUP:
811 reload = true;
812 break;
813 }
814 }
815
816 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
817 {
818 int ctrl = -1, netlink = -1;
819 int fd, n;
820
821 n = sd_listen_fds(true);
822 if (n <= 0)
823 return -1;
824
825 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
826 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
827 if (ctrl >= 0)
828 return -1;
829 ctrl = fd;
830 continue;
831 }
832
833 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
834 if (netlink >= 0)
835 return -1;
836 netlink = fd;
837 continue;
838 }
839
840 return -1;
841 }
842
843 if (ctrl < 0 || netlink < 0)
844 return -1;
845
846 log_debug("ctrl=%i netlink=%i\n", ctrl, netlink);
847 *rctrl = ctrl;
848 *rnetlink = netlink;
849 return 0;
850 }
851
852 /*
853 * read the kernel commandline, in case we need to get into debug mode
854 * udev.log-priority=<level> syslog priority
855 * udev.children-max=<number of workers> events are fully serialized if set to 1
856 * udev.exec-delay=<number of seconds> delay execution of every executed program
857 */
858 static void kernel_cmdline_options(struct udev *udev)
859 {
860 char *line, *w, *state;
861 size_t l;
862
863 if (read_one_line_file("/proc/cmdline", &line) < 0)
864 return;
865
866 FOREACH_WORD_QUOTED(w, l, line, state) {
867 char *s, *opt;
868
869 s = strndup(w, l);
870 if (!s)
871 break;
872
873 /* accept the same options for the initrd, prefixed with "rd." */
874 if (in_initrd() && startswith(s, "rd."))
875 opt = s + 3;
876 else
877 opt = s;
878
879 if (startswith(opt, "udev.log-priority=")) {
880 int prio;
881
882 prio = util_log_priority(opt + 18);
883 log_set_max_level(prio);
884 udev_set_log_priority(udev, prio);
885 } else if (startswith(opt, "udev.children-max=")) {
886 children_max = strtoul(opt + 18, NULL, 0);
887 } else if (startswith(opt, "udev.exec-delay=")) {
888 exec_delay = strtoul(opt + 16, NULL, 0);
889 }
890
891 free(s);
892 }
893
894 free(line);
895 }
896
897 int main(int argc, char *argv[])
898 {
899 struct udev *udev;
900 sigset_t mask;
901 int daemonize = false;
902 int resolve_names = 1;
903 static const struct option options[] = {
904 { "daemon", no_argument, NULL, 'd' },
905 { "debug", no_argument, NULL, 'D' },
906 { "children-max", required_argument, NULL, 'c' },
907 { "exec-delay", required_argument, NULL, 'e' },
908 { "resolve-names", required_argument, NULL, 'N' },
909 { "help", no_argument, NULL, 'h' },
910 { "version", no_argument, NULL, 'V' },
911 {}
912 };
913 int fd_ctrl = -1;
914 int fd_netlink = -1;
915 int fd_worker = -1;
916 struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
917 struct udev_ctrl_connection *ctrl_conn = NULL;
918 int rc = 1;
919
920 udev = udev_new();
921 if (udev == NULL)
922 goto exit;
923
924 log_set_target(LOG_TARGET_AUTO);
925 log_parse_environment();
926 log_open();
927
928 udev_set_log_fn(udev, udev_main_log);
929 log_set_max_level(udev_get_log_priority(udev));
930
931 log_debug("version %s\n", VERSION);
932 label_init("/dev");
933
934 for (;;) {
935 int option;
936
937 option = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL);
938 if (option == -1)
939 break;
940
941 switch (option) {
942 case 'd':
943 daemonize = true;
944 break;
945 case 'c':
946 children_max = strtoul(optarg, NULL, 0);
947 break;
948 case 'e':
949 exec_delay = strtoul(optarg, NULL, 0);
950 break;
951 case 'D':
952 debug = true;
953 log_set_max_level(LOG_DEBUG);
954 udev_set_log_priority(udev, LOG_DEBUG);
955 break;
956 case 'N':
957 if (streq(optarg, "early")) {
958 resolve_names = 1;
959 } else if (streq(optarg, "late")) {
960 resolve_names = 0;
961 } else if (streq(optarg, "never")) {
962 resolve_names = -1;
963 } else {
964 fprintf(stderr, "resolve-names must be early, late or never\n");
965 log_error("resolve-names must be early, late or never\n");
966 goto exit;
967 }
968 break;
969 case 'h':
970 printf("Usage: udevd OPTIONS\n"
971 " --daemon\n"
972 " --debug\n"
973 " --children-max=<maximum number of workers>\n"
974 " --exec-delay=<seconds to wait before executing RUN=>\n"
975 " --resolve-names=early|late|never\n"
976 " --version\n"
977 " --help\n"
978 "\n");
979 goto exit;
980 case 'V':
981 printf("%s\n", VERSION);
982 goto exit;
983 default:
984 goto exit;
985 }
986 }
987
988 kernel_cmdline_options(udev);
989
990 if (getuid() != 0) {
991 fprintf(stderr, "root privileges required\n");
992 log_error("root privileges required\n");
993 goto exit;
994 }
995
996 /* set umask before creating any file/directory */
997 chdir("/");
998 umask(022);
999
1000 mkdir("/run/udev", 0755);
1001
1002 dev_setup(NULL);
1003
1004 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1005 if (daemonize) {
1006 int fd;
1007
1008 fd = open("/dev/null", O_RDWR);
1009 if (fd >= 0) {
1010 if (write(STDOUT_FILENO, 0, 0) < 0)
1011 dup2(fd, STDOUT_FILENO);
1012 if (write(STDERR_FILENO, 0, 0) < 0)
1013 dup2(fd, STDERR_FILENO);
1014 if (fd > STDERR_FILENO)
1015 close(fd);
1016 } else {
1017 fprintf(stderr, "cannot open /dev/null\n");
1018 log_error("cannot open /dev/null\n");
1019 }
1020 }
1021
1022 if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1023 /* get control and netlink socket from systemd */
1024 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1025 if (udev_ctrl == NULL) {
1026 log_error("error taking over udev control socket");
1027 rc = 1;
1028 goto exit;
1029 }
1030
1031 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1032 if (monitor == NULL) {
1033 log_error("error taking over netlink socket\n");
1034 rc = 3;
1035 goto exit;
1036 }
1037
1038 /* get our own cgroup, we regularly kill everything udev has left behind */
1039 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1040 udev_cgroup = NULL;
1041 } else {
1042 /* open control and netlink socket */
1043 udev_ctrl = udev_ctrl_new(udev);
1044 if (udev_ctrl == NULL) {
1045 fprintf(stderr, "error initializing udev control socket");
1046 log_error("error initializing udev control socket");
1047 rc = 1;
1048 goto exit;
1049 }
1050 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1051
1052 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1053 if (monitor == NULL) {
1054 fprintf(stderr, "error initializing netlink socket\n");
1055 log_error("error initializing netlink socket\n");
1056 rc = 3;
1057 goto exit;
1058 }
1059 fd_netlink = udev_monitor_get_fd(monitor);
1060 }
1061
1062 if (udev_monitor_enable_receiving(monitor) < 0) {
1063 fprintf(stderr, "error binding netlink socket\n");
1064 log_error("error binding netlink socket\n");
1065 rc = 3;
1066 goto exit;
1067 }
1068
1069 if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1070 fprintf(stderr, "error binding udev control socket\n");
1071 log_error("error binding udev control socket\n");
1072 rc = 1;
1073 goto exit;
1074 }
1075
1076 udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
1077
1078 /* create queue file before signalling 'ready', to make sure we block 'settle' */
1079 udev_queue_export = udev_queue_export_new(udev);
1080 if (udev_queue_export == NULL) {
1081 log_error("error creating queue file\n");
1082 goto exit;
1083 }
1084
1085 if (daemonize) {
1086 pid_t pid;
1087
1088 pid = fork();
1089 switch (pid) {
1090 case 0:
1091 break;
1092 case -1:
1093 log_error("fork of daemon failed: %m\n");
1094 rc = 4;
1095 goto exit;
1096 default:
1097 rc = EXIT_SUCCESS;
1098 goto exit_daemonize;
1099 }
1100
1101 setsid();
1102
1103 write_string_file("/proc/self/oom_score_adj", "-1000");
1104 } else {
1105 sd_notify(1, "READY=1");
1106 }
1107
1108 print_kmsg("starting version " VERSION "\n");
1109
1110 if (!debug) {
1111 int fd;
1112
1113 fd = open("/dev/null", O_RDWR);
1114 if (fd >= 0) {
1115 dup2(fd, STDIN_FILENO);
1116 dup2(fd, STDOUT_FILENO);
1117 dup2(fd, STDERR_FILENO);
1118 close(fd);
1119 }
1120 }
1121
1122 fd_inotify = udev_watch_init(udev);
1123 if (fd_inotify < 0) {
1124 fprintf(stderr, "error initializing inotify\n");
1125 log_error("error initializing inotify\n");
1126 rc = 4;
1127 goto exit;
1128 }
1129 udev_watch_restore(udev);
1130
1131 /* block and listen to all signals on signalfd */
1132 sigfillset(&mask);
1133 sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1134 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1135 if (fd_signal < 0) {
1136 fprintf(stderr, "error creating signalfd\n");
1137 log_error("error creating signalfd\n");
1138 rc = 5;
1139 goto exit;
1140 }
1141
1142 /* unnamed socket from workers to the main daemon */
1143 if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1144 fprintf(stderr, "error creating socketpair\n");
1145 log_error("error creating socketpair\n");
1146 rc = 6;
1147 goto exit;
1148 }
1149 fd_worker = worker_watch[READ_END];
1150
1151 udev_builtin_init(udev);
1152
1153 rules = udev_rules_new(udev, resolve_names);
1154 if (rules == NULL) {
1155 log_error("error reading rules\n");
1156 goto exit;
1157 }
1158
1159 memset(&ep_ctrl, 0, sizeof(struct epoll_event));
1160 ep_ctrl.events = EPOLLIN;
1161 ep_ctrl.data.fd = fd_ctrl;
1162
1163 memset(&ep_inotify, 0, sizeof(struct epoll_event));
1164 ep_inotify.events = EPOLLIN;
1165 ep_inotify.data.fd = fd_inotify;
1166
1167 memset(&ep_signal, 0, sizeof(struct epoll_event));
1168 ep_signal.events = EPOLLIN;
1169 ep_signal.data.fd = fd_signal;
1170
1171 memset(&ep_netlink, 0, sizeof(struct epoll_event));
1172 ep_netlink.events = EPOLLIN;
1173 ep_netlink.data.fd = fd_netlink;
1174
1175 memset(&ep_worker, 0, sizeof(struct epoll_event));
1176 ep_worker.events = EPOLLIN;
1177 ep_worker.data.fd = fd_worker;
1178
1179 fd_ep = epoll_create1(EPOLL_CLOEXEC);
1180 if (fd_ep < 0) {
1181 log_error("error creating epoll fd: %m\n");
1182 goto exit;
1183 }
1184 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1185 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1186 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1187 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1188 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1189 log_error("fail to add fds to epoll: %m\n");
1190 goto exit;
1191 }
1192
1193 if (children_max <= 0) {
1194 cpu_set_t cpu_set;
1195
1196 children_max = 8;
1197
1198 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1199 children_max += CPU_COUNT(&cpu_set) * 2;
1200 }
1201 }
1202 log_debug("set children_max to %u\n", children_max);
1203
1204 rc = udev_rules_apply_static_dev_perms(rules);
1205 if (rc < 0)
1206 log_error("failed to apply permissions on static device nodes - %s\n", strerror(-rc));
1207
1208 udev_list_node_init(&event_list);
1209 udev_list_node_init(&worker_list);
1210
1211 for (;;) {
1212 static usec_t last_usec;
1213 struct epoll_event ev[8];
1214 int fdcount;
1215 int timeout;
1216 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1217 int i;
1218
1219 if (udev_exit) {
1220 /* close sources of new events and discard buffered events */
1221 if (fd_ctrl >= 0) {
1222 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1223 fd_ctrl = -1;
1224 }
1225 if (monitor != NULL) {
1226 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1227 udev_monitor_unref(monitor);
1228 monitor = NULL;
1229 }
1230 if (fd_inotify >= 0) {
1231 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1232 close(fd_inotify);
1233 fd_inotify = -1;
1234 }
1235
1236 /* discard queued events and kill workers */
1237 event_queue_cleanup(udev, EVENT_QUEUED);
1238 worker_kill(udev);
1239
1240 /* exit after all has cleaned up */
1241 if (udev_list_node_is_empty(&event_list) && udev_list_node_is_empty(&worker_list))
1242 break;
1243
1244 /* timeout at exit for workers to finish */
1245 timeout = 30 * 1000;
1246 } else if (udev_list_node_is_empty(&event_list) && !children) {
1247 /* we are idle */
1248 timeout = -1;
1249
1250 /* cleanup possible left-over processes in our cgroup */
1251 if (udev_cgroup)
1252 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1253 } else {
1254 /* kill idle or hanging workers */
1255 timeout = 3 * 1000;
1256 }
1257 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1258 if (fdcount < 0)
1259 continue;
1260
1261 if (fdcount == 0) {
1262 struct udev_list_node *loop;
1263
1264 /* timeout */
1265 if (udev_exit) {
1266 log_error("timeout, giving up waiting for workers to finish\n");
1267 break;
1268 }
1269
1270 /* kill idle workers */
1271 if (udev_list_node_is_empty(&event_list)) {
1272 log_debug("cleanup idle workers\n");
1273 worker_kill(udev);
1274 }
1275
1276 /* check for hanging events */
1277 udev_list_node_foreach(loop, &worker_list) {
1278 struct worker *worker = node_to_worker(loop);
1279
1280 if (worker->state != WORKER_RUNNING)
1281 continue;
1282
1283 if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > 30 * 1000 * 1000) {
1284 log_error("worker [%u] %s timeout; kill it\n", worker->pid,
1285 worker->event ? worker->event->devpath : "<idle>");
1286 kill(worker->pid, SIGKILL);
1287 worker->state = WORKER_KILLED;
1288 /* drop reference taken for state 'running' */
1289 worker_unref(worker);
1290 if (worker->event) {
1291 log_error("seq %llu '%s' killed\n",
1292 udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1293 worker->event->exitcode = -64;
1294 event_queue_delete(worker->event, true);
1295 worker->event = NULL;
1296 }
1297 }
1298 }
1299
1300 }
1301
1302 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1303 for (i = 0; i < fdcount; i++) {
1304 if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1305 is_worker = true;
1306 else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1307 is_netlink = true;
1308 else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1309 is_signal = true;
1310 else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1311 is_inotify = true;
1312 else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1313 is_ctrl = true;
1314 }
1315
1316 /* check for changed config, every 3 seconds at most */
1317 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * 1000 * 1000) {
1318 if (udev_rules_check_timestamp(rules))
1319 reload = true;
1320 if (udev_builtin_validate(udev))
1321 reload = true;
1322
1323 last_usec = now(CLOCK_MONOTONIC);
1324 }
1325
1326 /* reload requested, HUP signal received, rules changed, builtin changed */
1327 if (reload) {
1328 worker_kill(udev);
1329 rules = udev_rules_unref(rules);
1330 udev_builtin_exit(udev);
1331 reload = false;
1332 }
1333
1334 /* event has finished */
1335 if (is_worker)
1336 worker_returned(fd_worker);
1337
1338 if (is_netlink) {
1339 struct udev_device *dev;
1340
1341 dev = udev_monitor_receive_device(monitor);
1342 if (dev != NULL) {
1343 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1344 if (event_queue_insert(dev) < 0)
1345 udev_device_unref(dev);
1346 }
1347 }
1348
1349 /* start new events */
1350 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1351 udev_builtin_init(udev);
1352 if (rules == NULL)
1353 rules = udev_rules_new(udev, resolve_names);
1354 if (rules != NULL)
1355 event_queue_start(udev);
1356 }
1357
1358 if (is_signal) {
1359 struct signalfd_siginfo fdsi;
1360 ssize_t size;
1361
1362 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1363 if (size == sizeof(struct signalfd_siginfo))
1364 handle_signal(udev, fdsi.ssi_signo);
1365 }
1366
1367 /* we are shutting down, the events below are not handled anymore */
1368 if (udev_exit)
1369 continue;
1370
1371 /* device node watch */
1372 if (is_inotify)
1373 handle_inotify(udev);
1374
1375 /*
1376 * This needs to be after the inotify handling, to make sure,
1377 * that the ping is send back after the possibly generated
1378 * "change" events by the inotify device node watch.
1379 *
1380 * A single time we may receive a client connection which we need to
1381 * keep open to block the client. It will be closed right before we
1382 * exit.
1383 */
1384 if (is_ctrl)
1385 ctrl_conn = handle_ctrl_msg(udev_ctrl);
1386 }
1387
1388 rc = EXIT_SUCCESS;
1389 exit:
1390 udev_queue_export_cleanup(udev_queue_export);
1391 udev_ctrl_cleanup(udev_ctrl);
1392 exit_daemonize:
1393 if (fd_ep >= 0)
1394 close(fd_ep);
1395 worker_list_cleanup(udev);
1396 event_queue_cleanup(udev, EVENT_UNDEF);
1397 udev_rules_unref(rules);
1398 udev_builtin_exit(udev);
1399 if (fd_signal >= 0)
1400 close(fd_signal);
1401 if (worker_watch[READ_END] >= 0)
1402 close(worker_watch[READ_END]);
1403 if (worker_watch[WRITE_END] >= 0)
1404 close(worker_watch[WRITE_END]);
1405 udev_monitor_unref(monitor);
1406 udev_queue_export_unref(udev_queue_export);
1407 udev_ctrl_connection_unref(ctrl_conn);
1408 udev_ctrl_unref(udev_ctrl);
1409 label_finish();
1410 udev_unref(udev);
1411 log_close();
1412 return rc;
1413 }