]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/udev/udevd.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
udevd: event - fix event queue in daemenozied mode
[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 <fcntl.h>
30 #include <getopt.h>
31 #include <sys/file.h>
32 #include <sys/time.h>
33 #include <sys/prctl.h>
34 #include <sys/socket.h>
35 #include <sys/signalfd.h>
36 #include <sys/epoll.h>
37 #include <sys/mount.h>
38 #include <sys/wait.h>
39 #include <sys/stat.h>
40 #include <sys/ioctl.h>
41 #include <sys/inotify.h>
42
43 #include "sd-daemon.h"
44 #include "rtnl-util.h"
45 #include "cgroup-util.h"
46 #include "process-util.h"
47 #include "dev-setup.h"
48 #include "fileio.h"
49 #include "selinux-util.h"
50 #include "udev.h"
51 #include "udev-util.h"
52 #include "formats-util.h"
53 #include "hashmap.h"
54
55 static bool arg_debug = false;
56 static int arg_daemonize = false;
57 static int arg_resolve_names = 1;
58 static unsigned arg_children_max;
59 static int arg_exec_delay;
60 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
61 static usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
62
63 typedef struct Manager {
64 struct udev *udev;
65 Hashmap *workers;
66 struct udev_list_node events;
67 char *cgroup;
68 pid_t pid; /* the process that originally allocated the manager object */
69 sigset_t sigmask_orig;
70
71 struct udev_rules *rules;
72 struct udev_list properties;
73
74 struct udev_monitor *monitor;
75 struct udev_ctrl *ctrl;
76 struct udev_ctrl_connection *ctrl_conn_blocking;
77
78 int fd_ep;
79 int fd_ctrl;
80 int fd_uevent;
81 int fd_signal;
82 int fd_inotify;
83 int fd_worker;
84 int worker_watch[2];
85
86 bool stop_exec_queue:1;
87 bool reload:1;
88 bool exit:1;
89 } Manager;
90
91 enum event_state {
92 EVENT_UNDEF,
93 EVENT_QUEUED,
94 EVENT_RUNNING,
95 };
96
97 struct event {
98 struct udev_list_node node;
99 Manager *manager;
100 struct udev *udev;
101 struct udev_device *dev;
102 struct udev_device *dev_kernel;
103 struct worker *worker;
104 enum event_state state;
105 unsigned long long int delaying_seqnum;
106 unsigned long long int seqnum;
107 const char *devpath;
108 size_t devpath_len;
109 const char *devpath_old;
110 dev_t devnum;
111 int ifindex;
112 bool is_block;
113 usec_t start_usec;
114 bool warned;
115 };
116
117 static inline struct event *node_to_event(struct udev_list_node *node) {
118 return container_of(node, struct event, node);
119 }
120
121 static void event_queue_cleanup(Manager *manager, enum event_state type);
122
123 enum worker_state {
124 WORKER_UNDEF,
125 WORKER_RUNNING,
126 WORKER_IDLE,
127 WORKER_KILLED,
128 };
129
130 struct worker {
131 Manager *manager;
132 struct udev_list_node node;
133 int refcount;
134 pid_t pid;
135 struct udev_monitor *monitor;
136 enum worker_state state;
137 struct event *event;
138 };
139
140 /* passed from worker to main process */
141 struct worker_message {
142 };
143
144 static void event_free(struct event *event) {
145 int r;
146
147 if (!event)
148 return;
149
150 udev_list_node_remove(&event->node);
151 udev_device_unref(event->dev);
152 udev_device_unref(event->dev_kernel);
153
154 if (event->worker)
155 event->worker->event = NULL;
156
157 assert(event->manager);
158
159 if (udev_list_node_is_empty(&event->manager->events)) {
160 /* only clean up the queue from the process that created it */
161 if (event->manager->pid == getpid()) {
162 r = unlink("/run/udev/queue");
163 if (r < 0)
164 log_warning_errno(errno, "could not unlink /run/udev/queue: %m");
165 }
166 }
167
168 free(event);
169 }
170
171 static void worker_free(struct worker *worker) {
172 if (!worker)
173 return;
174
175 assert(worker->manager);
176
177 hashmap_remove(worker->manager->workers, UINT_TO_PTR(worker->pid));
178 udev_monitor_unref(worker->monitor);
179 event_free(worker->event);
180
181 free(worker);
182 }
183
184 static void manager_workers_free(Manager *manager) {
185 struct worker *worker;
186 Iterator i;
187
188 assert(manager);
189
190 HASHMAP_FOREACH(worker, manager->workers, i)
191 worker_free(worker);
192
193 manager->workers = hashmap_free(manager->workers);
194 }
195
196 static int worker_new(struct worker **ret, Manager *manager, struct udev_monitor *worker_monitor, pid_t pid) {
197 _cleanup_free_ struct worker *worker = NULL;
198 int r;
199
200 assert(ret);
201 assert(manager);
202 assert(worker_monitor);
203 assert(pid > 1);
204
205 worker = new0(struct worker, 1);
206 if (!worker)
207 return -ENOMEM;
208
209 worker->refcount = 1;
210 worker->manager = manager;
211 /* close monitor, but keep address around */
212 udev_monitor_disconnect(worker_monitor);
213 worker->monitor = udev_monitor_ref(worker_monitor);
214 worker->pid = pid;
215
216 r = hashmap_ensure_allocated(&manager->workers, NULL);
217 if (r < 0)
218 return r;
219
220 r = hashmap_put(manager->workers, UINT_TO_PTR(pid), worker);
221 if (r < 0)
222 return r;
223
224 *ret = worker;
225 worker = NULL;
226
227 return 0;
228 }
229
230 static int on_event_timeout(sd_event_source *s, uint64_t usec, void *userdata) {
231 struct event *event = userdata;
232
233 assert(event);
234 assert(event->worker);
235
236 kill_and_sigcont(event->worker->pid, SIGKILL);
237 event->worker->state = WORKER_KILLED;
238
239 log_error("seq %llu '%s' killed", udev_device_get_seqnum(event->dev), event->devpath);
240
241 return 1;
242 }
243
244 static int on_event_timeout_warning(sd_event_source *s, uint64_t usec, void *userdata) {
245 struct event *event = userdata;
246
247 assert(event);
248
249 log_warning("seq %llu '%s' is taking a long time", udev_device_get_seqnum(event->dev), event->devpath);
250
251 return 1;
252 }
253
254 static void worker_attach_event(struct worker *worker, struct event *event) {
255 assert(worker);
256 assert(event);
257 assert(!event->worker);
258 assert(!worker->event);
259
260 worker->state = WORKER_RUNNING;
261 worker->event = event;
262 event->state = EVENT_RUNNING;
263 event->start_usec = now(CLOCK_MONOTONIC);
264 event->warned = false;
265 event->worker = worker;
266 }
267
268 static void manager_free(Manager *manager) {
269 if (!manager)
270 return;
271
272 udev_builtin_exit(manager->udev);
273
274 udev_unref(manager->udev);
275 manager_workers_free(manager);
276 event_queue_cleanup(manager, EVENT_UNDEF);
277
278 udev_monitor_unref(manager->monitor);
279 udev_ctrl_unref(manager->ctrl);
280 udev_ctrl_connection_unref(manager->ctrl_conn_blocking);
281
282 udev_list_cleanup(&manager->properties);
283 udev_rules_unref(manager->rules);
284 free(manager->cgroup);
285
286 safe_close(manager->fd_ep);
287 safe_close(manager->fd_signal);
288 safe_close(manager->fd_inotify);
289 safe_close_pair(manager->worker_watch);
290
291 free(manager);
292 }
293
294 DEFINE_TRIVIAL_CLEANUP_FUNC(Manager*, manager_free);
295
296 static int worker_send_message(int fd) {
297 struct worker_message message = {};
298
299 return loop_write(fd, &message, sizeof(message), false);
300 }
301
302 static void worker_spawn(Manager *manager, struct event *event) {
303 struct udev *udev = event->udev;
304 _cleanup_udev_monitor_unref_ struct udev_monitor *worker_monitor = NULL;
305 pid_t pid;
306
307 /* listen for new events */
308 worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
309 if (worker_monitor == NULL)
310 return;
311 /* allow the main daemon netlink address to send devices to the worker */
312 udev_monitor_allow_unicast_sender(worker_monitor, manager->monitor);
313 udev_monitor_enable_receiving(worker_monitor);
314
315 pid = fork();
316 switch (pid) {
317 case 0: {
318 struct udev_device *dev = NULL;
319 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
320 int fd_monitor;
321 _cleanup_close_ int fd_signal = -1, fd_ep = -1;
322 struct epoll_event ep_signal = { .events = EPOLLIN };
323 struct epoll_event ep_monitor = { .events = EPOLLIN };
324 sigset_t mask;
325 int r = 0;
326
327 /* take initial device from queue */
328 dev = event->dev;
329 event->dev = NULL;
330
331 manager_workers_free(manager);
332 event_queue_cleanup(manager, EVENT_UNDEF);
333
334 manager->monitor = udev_monitor_unref(manager->monitor);
335 manager->ctrl_conn_blocking = udev_ctrl_connection_unref(manager->ctrl_conn_blocking);
336 manager->ctrl = udev_ctrl_unref(manager->ctrl);
337
338 manager->fd_ep = safe_close(manager->fd_ep);
339 manager->fd_signal = safe_close(manager->fd_signal);
340 manager->fd_inotify = safe_close(manager->fd_inotify);
341 manager->worker_watch[READ_END] = safe_close(manager->worker_watch[READ_END]);
342
343 sigfillset(&mask);
344 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
345 if (fd_signal < 0) {
346 r = log_error_errno(errno, "error creating signalfd %m");
347 goto out;
348 }
349 ep_signal.data.fd = fd_signal;
350
351 fd_monitor = udev_monitor_get_fd(worker_monitor);
352 ep_monitor.data.fd = fd_monitor;
353
354 fd_ep = epoll_create1(EPOLL_CLOEXEC);
355 if (fd_ep < 0) {
356 r = log_error_errno(errno, "error creating epoll fd: %m");
357 goto out;
358 }
359
360 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
361 epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
362 r = log_error_errno(errno, "fail to add fds to epoll: %m");
363 goto out;
364 }
365
366 /* request TERM signal if parent exits */
367 prctl(PR_SET_PDEATHSIG, SIGTERM);
368
369 /* reset OOM score, we only protect the main daemon */
370 write_string_file("/proc/self/oom_score_adj", "0");
371
372 for (;;) {
373 struct udev_event *udev_event;
374 int fd_lock = -1;
375
376 log_debug("seq %llu running", udev_device_get_seqnum(dev));
377 udev_event = udev_event_new(dev);
378 if (udev_event == NULL) {
379 r = -ENOMEM;
380 goto out;
381 }
382
383 /* needed for SIGCHLD/SIGTERM in spawn() */
384 udev_event->fd_signal = fd_signal;
385
386 if (arg_exec_delay > 0)
387 udev_event->exec_delay = arg_exec_delay;
388
389 /*
390 * Take a shared lock on the device node; this establishes
391 * a concept of device "ownership" to serialize device
392 * access. External processes holding an exclusive lock will
393 * cause udev to skip the event handling; in the case udev
394 * acquired the lock, the external process can block until
395 * udev has finished its event handling.
396 */
397 if (!streq_ptr(udev_device_get_action(dev), "remove") &&
398 streq_ptr("block", udev_device_get_subsystem(dev)) &&
399 !startswith(udev_device_get_sysname(dev), "dm-") &&
400 !startswith(udev_device_get_sysname(dev), "md")) {
401 struct udev_device *d = dev;
402
403 if (streq_ptr("partition", udev_device_get_devtype(d)))
404 d = udev_device_get_parent(d);
405
406 if (d) {
407 fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
408 if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
409 log_debug_errno(errno, "Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
410 fd_lock = safe_close(fd_lock);
411 r = -EAGAIN;
412 goto skip;
413 }
414 }
415 }
416
417 /* needed for renaming netifs */
418 udev_event->rtnl = rtnl;
419
420 /* apply rules, create node, symlinks */
421 udev_event_execute_rules(udev_event,
422 arg_event_timeout_usec, arg_event_timeout_warn_usec,
423 &manager->properties,
424 manager->rules,
425 &manager->sigmask_orig);
426
427 udev_event_execute_run(udev_event,
428 arg_event_timeout_usec, arg_event_timeout_warn_usec,
429 &manager->sigmask_orig);
430
431 if (udev_event->rtnl)
432 /* in case rtnl was initialized */
433 rtnl = sd_rtnl_ref(udev_event->rtnl);
434
435 /* apply/restore inotify watch */
436 if (udev_event->inotify_watch) {
437 udev_watch_begin(udev, dev);
438 udev_device_update_db(dev);
439 }
440
441 safe_close(fd_lock);
442
443 /* send processed event back to libudev listeners */
444 udev_monitor_send_device(worker_monitor, NULL, dev);
445
446 skip:
447 log_debug("seq %llu processed", udev_device_get_seqnum(dev));
448
449 /* send udevd the result of the event execution */
450 r = worker_send_message(manager->worker_watch[WRITE_END]);
451 if (r < 0)
452 log_error_errno(r, "failed to send result of seq %llu to main daemon: %m",
453 udev_device_get_seqnum(dev));
454
455 udev_device_unref(dev);
456 dev = NULL;
457
458 if (udev_event->sigterm) {
459 udev_event_unref(udev_event);
460 goto out;
461 }
462
463 udev_event_unref(udev_event);
464
465 /* wait for more device messages from main udevd, or term signal */
466 while (dev == NULL) {
467 struct epoll_event ev[4];
468 int fdcount;
469 int i;
470
471 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
472 if (fdcount < 0) {
473 if (errno == EINTR)
474 continue;
475 r = log_error_errno(errno, "failed to poll: %m");
476 goto out;
477 }
478
479 for (i = 0; i < fdcount; i++) {
480 if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
481 dev = udev_monitor_receive_device(worker_monitor);
482 break;
483 } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
484 struct signalfd_siginfo fdsi;
485 ssize_t size;
486
487 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
488 if (size != sizeof(struct signalfd_siginfo))
489 continue;
490 switch (fdsi.ssi_signo) {
491 case SIGTERM:
492 goto out;
493 }
494 }
495 }
496 }
497 }
498 out:
499 udev_device_unref(dev);
500 manager_free(manager);
501 log_close();
502 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
503 }
504 case -1:
505 event->state = EVENT_QUEUED;
506 log_error_errno(errno, "fork of child failed: %m");
507 break;
508 default:
509 {
510 struct worker *worker;
511 int r;
512
513 r = worker_new(&worker, manager, worker_monitor, pid);
514 if (r < 0)
515 return;
516
517 worker_attach_event(worker, event);
518
519 log_debug("seq %llu forked new worker ["PID_FMT"]", udev_device_get_seqnum(event->dev), pid);
520 break;
521 }
522 }
523 }
524
525 static void event_run(Manager *manager, struct event *event) {
526 struct worker *worker;
527 Iterator i;
528
529 assert(manager);
530 assert(event);
531
532 HASHMAP_FOREACH(worker, manager->workers, i) {
533 ssize_t count;
534
535 if (worker->state != WORKER_IDLE)
536 continue;
537
538 count = udev_monitor_send_device(manager->monitor, worker->monitor, event->dev);
539 if (count < 0) {
540 log_error_errno(errno, "worker ["PID_FMT"] did not accept message %zi (%m), kill it",
541 worker->pid, count);
542 kill(worker->pid, SIGKILL);
543 worker->state = WORKER_KILLED;
544 continue;
545 }
546 worker_attach_event(worker, event);
547 return;
548 }
549
550 if (hashmap_size(manager->workers) >= arg_children_max) {
551 if (arg_children_max > 1)
552 log_debug("maximum number (%i) of children reached", hashmap_size(manager->workers));
553 return;
554 }
555
556 /* start new worker and pass initial device */
557 worker_spawn(manager, event);
558 }
559
560 static int event_queue_insert(Manager *manager, struct udev_device *dev) {
561 struct event *event;
562 int r;
563
564 assert(manager);
565 assert(dev);
566
567 /* only one process can add events to the queue */
568 if (manager->pid == 0)
569 manager->pid = getpid();
570
571 assert(manager->pid == getpid());
572
573 event = new0(struct event, 1);
574 if (!event)
575 return -ENOMEM;
576
577 event->udev = udev_device_get_udev(dev);
578 event->manager = manager;
579 event->dev = dev;
580 event->dev_kernel = udev_device_shallow_clone(dev);
581 udev_device_copy_properties(event->dev_kernel, dev);
582 event->seqnum = udev_device_get_seqnum(dev);
583 event->devpath = udev_device_get_devpath(dev);
584 event->devpath_len = strlen(event->devpath);
585 event->devpath_old = udev_device_get_devpath_old(dev);
586 event->devnum = udev_device_get_devnum(dev);
587 event->is_block = streq("block", udev_device_get_subsystem(dev));
588 event->ifindex = udev_device_get_ifindex(dev);
589
590 log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
591 udev_device_get_action(dev), udev_device_get_subsystem(dev));
592
593 event->state = EVENT_QUEUED;
594
595 if (udev_list_node_is_empty(&manager->events)) {
596 r = touch("/run/udev/queue");
597 if (r < 0)
598 log_warning_errno(r, "could not touch /run/udev/queue: %m");
599 }
600
601 udev_list_node_append(&event->node, &manager->events);
602
603 return 0;
604 }
605
606 static void manager_kill_workers(Manager *manager) {
607 struct worker *worker;
608 Iterator i;
609
610 assert(manager);
611
612 HASHMAP_FOREACH(worker, manager->workers, i) {
613 if (worker->state == WORKER_KILLED)
614 continue;
615
616 worker->state = WORKER_KILLED;
617 kill(worker->pid, SIGTERM);
618 }
619 }
620
621 /* lookup event for identical, parent, child device */
622 static bool is_devpath_busy(Manager *manager, struct event *event) {
623 struct udev_list_node *loop;
624 size_t common;
625
626 /* check if queue contains events we depend on */
627 udev_list_node_foreach(loop, &manager->events) {
628 struct event *loop_event = node_to_event(loop);
629
630 /* we already found a later event, earlier can not block us, no need to check again */
631 if (loop_event->seqnum < event->delaying_seqnum)
632 continue;
633
634 /* event we checked earlier still exists, no need to check again */
635 if (loop_event->seqnum == event->delaying_seqnum)
636 return true;
637
638 /* found ourself, no later event can block us */
639 if (loop_event->seqnum >= event->seqnum)
640 break;
641
642 /* check major/minor */
643 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
644 return true;
645
646 /* check network device ifindex */
647 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
648 return true;
649
650 /* check our old name */
651 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
652 event->delaying_seqnum = loop_event->seqnum;
653 return true;
654 }
655
656 /* compare devpath */
657 common = MIN(loop_event->devpath_len, event->devpath_len);
658
659 /* one devpath is contained in the other? */
660 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
661 continue;
662
663 /* identical device event found */
664 if (loop_event->devpath_len == event->devpath_len) {
665 /* devices names might have changed/swapped in the meantime */
666 if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
667 continue;
668 if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
669 continue;
670 event->delaying_seqnum = loop_event->seqnum;
671 return true;
672 }
673
674 /* parent device event found */
675 if (event->devpath[common] == '/') {
676 event->delaying_seqnum = loop_event->seqnum;
677 return true;
678 }
679
680 /* child device event found */
681 if (loop_event->devpath[common] == '/') {
682 event->delaying_seqnum = loop_event->seqnum;
683 return true;
684 }
685
686 /* no matching device */
687 continue;
688 }
689
690 return false;
691 }
692
693 static void event_queue_start(Manager *manager) {
694 struct udev_list_node *loop;
695
696 assert(manager);
697
698 udev_list_node_foreach(loop, &manager->events) {
699 struct event *event = node_to_event(loop);
700
701 if (event->state != EVENT_QUEUED)
702 continue;
703
704 /* do not start event if parent or child event is still running */
705 if (is_devpath_busy(manager, event))
706 continue;
707
708 event_run(manager, event);
709 }
710 }
711
712 static void event_queue_cleanup(Manager *manager, enum event_state match_type) {
713 struct udev_list_node *loop, *tmp;
714
715 udev_list_node_foreach_safe(loop, tmp, &manager->events) {
716 struct event *event = node_to_event(loop);
717
718 if (match_type != EVENT_UNDEF && match_type != event->state)
719 continue;
720
721 event_free(event);
722 }
723 }
724
725 static int on_worker(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
726 Manager *manager = userdata;
727
728 assert(manager);
729
730 for (;;) {
731 struct worker_message msg;
732 struct iovec iovec = {
733 .iov_base = &msg,
734 .iov_len = sizeof(msg),
735 };
736 union {
737 struct cmsghdr cmsghdr;
738 uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
739 } control = {};
740 struct msghdr msghdr = {
741 .msg_iov = &iovec,
742 .msg_iovlen = 1,
743 .msg_control = &control,
744 .msg_controllen = sizeof(control),
745 };
746 struct cmsghdr *cmsg;
747 ssize_t size;
748 struct ucred *ucred = NULL;
749 struct worker *worker;
750
751 size = recvmsg(fd, &msghdr, MSG_DONTWAIT);
752 if (size < 0) {
753 if (errno == EINTR)
754 continue;
755 else if (errno == EAGAIN)
756 /* nothing more to read */
757 break;
758
759 return log_error_errno(errno, "failed to receive message: %m");
760 } else if (size != sizeof(struct worker_message)) {
761 log_warning_errno(EIO, "ignoring worker message with invalid size %zi bytes", size);
762 continue;
763 }
764
765 for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
766 if (cmsg->cmsg_level == SOL_SOCKET &&
767 cmsg->cmsg_type == SCM_CREDENTIALS &&
768 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
769 ucred = (struct ucred*) CMSG_DATA(cmsg);
770 }
771
772 if (!ucred || ucred->pid <= 0) {
773 log_warning_errno(EIO, "ignoring worker message without valid PID");
774 continue;
775 }
776
777 /* lookup worker who sent the signal */
778 worker = hashmap_get(manager->workers, UINT_TO_PTR(ucred->pid));
779 if (!worker) {
780 log_debug("worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid);
781 continue;
782 }
783
784 if (worker->state != WORKER_KILLED)
785 worker->state = WORKER_IDLE;
786
787 /* worker returned */
788 event_free(worker->event);
789 }
790
791 return 1;
792 }
793
794 static int on_uevent(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
795 Manager *manager = userdata;
796 struct udev_device *dev;
797 int r;
798
799 assert(manager);
800
801 dev = udev_monitor_receive_device(manager->monitor);
802 if (dev) {
803 udev_device_ensure_usec_initialized(dev, NULL);
804 r = event_queue_insert(manager, dev);
805 if (r < 0)
806 udev_device_unref(dev);
807 }
808
809 return 1;
810 }
811
812 /* receive the udevd message from userspace */
813 static int on_ctrl_msg(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
814 Manager *manager = userdata;
815 _cleanup_udev_ctrl_connection_unref_ struct udev_ctrl_connection *ctrl_conn = NULL;
816 _cleanup_udev_ctrl_msg_unref_ struct udev_ctrl_msg *ctrl_msg = NULL;
817 const char *str;
818 int i;
819
820 assert(manager);
821
822 ctrl_conn = udev_ctrl_get_connection(manager->ctrl);
823 if (!ctrl_conn)
824 return 1;
825
826 ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
827 if (!ctrl_msg)
828 return 1;
829
830 i = udev_ctrl_get_set_log_level(ctrl_msg);
831 if (i >= 0) {
832 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
833 log_set_max_level(i);
834 manager_kill_workers(manager);
835 }
836
837 if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
838 log_debug("udevd message (STOP_EXEC_QUEUE) received");
839 manager->stop_exec_queue = true;
840 }
841
842 if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
843 log_debug("udevd message (START_EXEC_QUEUE) received");
844 manager->stop_exec_queue = false;
845 }
846
847 if (udev_ctrl_get_reload(ctrl_msg) > 0) {
848 log_debug("udevd message (RELOAD) received");
849 manager->reload = true;
850 }
851
852 str = udev_ctrl_get_set_env(ctrl_msg);
853 if (str != NULL) {
854 _cleanup_free_ char *key = NULL;
855
856 key = strdup(str);
857 if (key) {
858 char *val;
859
860 val = strchr(key, '=');
861 if (val != NULL) {
862 val[0] = '\0';
863 val = &val[1];
864 if (val[0] == '\0') {
865 log_debug("udevd message (ENV) received, unset '%s'", key);
866 udev_list_entry_add(&manager->properties, key, NULL);
867 } else {
868 log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
869 udev_list_entry_add(&manager->properties, key, val);
870 }
871 } else
872 log_error("wrong key format '%s'", key);
873 }
874 manager_kill_workers(manager);
875 }
876
877 i = udev_ctrl_get_set_children_max(ctrl_msg);
878 if (i >= 0) {
879 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
880 arg_children_max = i;
881 }
882
883 if (udev_ctrl_get_ping(ctrl_msg) > 0)
884 log_debug("udevd message (SYNC) received");
885
886 if (udev_ctrl_get_exit(ctrl_msg) > 0) {
887 log_debug("udevd message (EXIT) received");
888 manager->exit = true;
889 /* keep reference to block the client until we exit
890 TODO: deal with several blocking exit requests */
891 manager->ctrl_conn_blocking = udev_ctrl_connection_ref(ctrl_conn);
892 }
893
894 return 1;
895 }
896
897 static int synthesize_change(struct udev_device *dev) {
898 char filename[UTIL_PATH_SIZE];
899 int r;
900
901 if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
902 streq_ptr("disk", udev_device_get_devtype(dev)) &&
903 !startswith(udev_device_get_sysname(dev), "dm-")) {
904 bool part_table_read = false;
905 bool has_partitions = false;
906 int fd;
907 struct udev *udev = udev_device_get_udev(dev);
908 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
909 struct udev_list_entry *item;
910
911 /*
912 * Try to re-read the partition table. This only succeeds if
913 * none of the devices is busy. The kernel returns 0 if no
914 * partition table is found, and we will not get an event for
915 * the disk.
916 */
917 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
918 if (fd >= 0) {
919 r = flock(fd, LOCK_EX|LOCK_NB);
920 if (r >= 0)
921 r = ioctl(fd, BLKRRPART, 0);
922
923 close(fd);
924 if (r >= 0)
925 part_table_read = true;
926 }
927
928 /* search for partitions */
929 e = udev_enumerate_new(udev);
930 if (!e)
931 return -ENOMEM;
932
933 r = udev_enumerate_add_match_parent(e, dev);
934 if (r < 0)
935 return r;
936
937 r = udev_enumerate_add_match_subsystem(e, "block");
938 if (r < 0)
939 return r;
940
941 r = udev_enumerate_scan_devices(e);
942 if (r < 0)
943 return r;
944
945 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
946 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
947
948 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
949 if (!d)
950 continue;
951
952 if (!streq_ptr("partition", udev_device_get_devtype(d)))
953 continue;
954
955 has_partitions = true;
956 break;
957 }
958
959 /*
960 * We have partitions and re-read the table, the kernel already sent
961 * out a "change" event for the disk, and "remove/add" for all
962 * partitions.
963 */
964 if (part_table_read && has_partitions)
965 return 0;
966
967 /*
968 * We have partitions but re-reading the partition table did not
969 * work, synthesize "change" for the disk and all partitions.
970 */
971 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
972 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
973 write_string_file(filename, "change");
974
975 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
976 _cleanup_udev_device_unref_ struct udev_device *d = NULL;
977
978 d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
979 if (!d)
980 continue;
981
982 if (!streq_ptr("partition", udev_device_get_devtype(d)))
983 continue;
984
985 log_debug("device %s closed, synthesising partition '%s' 'change'",
986 udev_device_get_devnode(dev), udev_device_get_devnode(d));
987 strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
988 write_string_file(filename, "change");
989 }
990
991 return 0;
992 }
993
994 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
995 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
996 write_string_file(filename, "change");
997
998 return 0;
999 }
1000
1001 static int on_inotify(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1002 Manager *manager = userdata;
1003 union inotify_event_buffer buffer;
1004 struct inotify_event *e;
1005 ssize_t l;
1006
1007 assert(manager);
1008
1009 l = read(fd, &buffer, sizeof(buffer));
1010 if (l < 0) {
1011 if (errno == EAGAIN || errno == EINTR)
1012 return 1;
1013
1014 return log_error_errno(errno, "Failed to read inotify fd: %m");
1015 }
1016
1017 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1018 _cleanup_udev_device_unref_ struct udev_device *dev = NULL;
1019
1020 dev = udev_watch_lookup(manager->udev, e->wd);
1021 if (!dev)
1022 continue;
1023
1024 log_debug("inotify event: %x for %s", e->mask, udev_device_get_devnode(dev));
1025 if (e->mask & IN_CLOSE_WRITE) {
1026 synthesize_change(dev);
1027
1028 /* settle might be waiting on us to determine the queue
1029 * state. If we just handled an inotify event, we might have
1030 * generated a "change" event, but we won't have queued up
1031 * the resultant uevent yet. Do that.
1032 */
1033 on_uevent(NULL, -1, 0, manager);
1034 } else if (e->mask & IN_IGNORED)
1035 udev_watch_end(manager->udev, dev);
1036 }
1037
1038 return 1;
1039 }
1040
1041 static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1042 Manager *manager = userdata;
1043
1044 assert(manager);
1045
1046 manager->exit = true;
1047
1048 return 1;
1049 }
1050
1051 static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1052 Manager *manager = userdata;
1053
1054 assert(manager);
1055
1056 manager->reload = true;
1057
1058 return 1;
1059 }
1060
1061 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1062 Manager *manager = userdata;
1063
1064 assert(manager);
1065
1066 for (;;) {
1067 pid_t pid;
1068 int status;
1069 struct worker *worker;
1070
1071 pid = waitpid(-1, &status, WNOHANG);
1072 if (pid <= 0)
1073 break;
1074
1075 worker = hashmap_get(manager->workers, UINT_TO_PTR(pid));
1076 if (!worker) {
1077 log_warning("worker ["PID_FMT"] is unknown, ignoring", pid);
1078 continue;
1079 }
1080
1081 if (WIFEXITED(status)) {
1082 if (WEXITSTATUS(status) == 0)
1083 log_debug("worker ["PID_FMT"] exited", pid);
1084 else
1085 log_warning("worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status));
1086 } else if (WIFSIGNALED(status)) {
1087 log_warning("worker ["PID_FMT"] terminated by signal %i (%s)", pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
1088 } else if (WIFSTOPPED(status)) {
1089 log_info("worker ["PID_FMT"] stopped", pid);
1090 continue;
1091 } else if (WIFCONTINUED(status)) {
1092 log_info("worker ["PID_FMT"] continued", pid);
1093 continue;
1094 } else
1095 log_warning("worker ["PID_FMT"] exit with status 0x%04x", pid, status);
1096
1097 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1098 if (worker->event) {
1099 log_error("worker ["PID_FMT"] failed while handling '%s'", pid, worker->event->devpath);
1100 /* delete state from disk */
1101 udev_device_delete_db(worker->event->dev);
1102 udev_device_tag_index(worker->event->dev, NULL, false);
1103 /* forward kernel event without amending it */
1104 udev_monitor_send_device(manager->monitor, NULL, worker->event->dev_kernel);
1105 }
1106 }
1107
1108 worker_free(worker);
1109 }
1110
1111 return 1;
1112 }
1113
1114 static int systemd_fds(int *rctrl, int *rnetlink) {
1115 int ctrl = -1, netlink = -1;
1116 int fd, n;
1117
1118 n = sd_listen_fds(true);
1119 if (n <= 0)
1120 return -1;
1121
1122 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
1123 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
1124 if (ctrl >= 0)
1125 return -1;
1126 ctrl = fd;
1127 continue;
1128 }
1129
1130 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
1131 if (netlink >= 0)
1132 return -1;
1133 netlink = fd;
1134 continue;
1135 }
1136
1137 return -1;
1138 }
1139
1140 if (ctrl < 0 || netlink < 0)
1141 return -1;
1142
1143 log_debug("ctrl=%i netlink=%i", ctrl, netlink);
1144 *rctrl = ctrl;
1145 *rnetlink = netlink;
1146 return 0;
1147 }
1148
1149 /*
1150 * read the kernel command line, in case we need to get into debug mode
1151 * udev.log-priority=<level> syslog priority
1152 * udev.children-max=<number of workers> events are fully serialized if set to 1
1153 * udev.exec-delay=<number of seconds> delay execution of every executed program
1154 * udev.event-timeout=<number of seconds> seconds to wait before terminating an event
1155 */
1156 static int parse_proc_cmdline_item(const char *key, const char *value) {
1157 int r;
1158
1159 assert(key);
1160
1161 if (!value)
1162 return 0;
1163
1164 if (startswith(key, "rd."))
1165 key += strlen("rd.");
1166
1167 if (startswith(key, "udev."))
1168 key += strlen("udev.");
1169 else
1170 return 0;
1171
1172 if (streq(key, "log-priority")) {
1173 int prio;
1174
1175 prio = util_log_priority(value);
1176 log_set_max_level(prio);
1177 } else if (streq(key, "children-max")) {
1178 r = safe_atou(value, &arg_children_max);
1179 if (r < 0)
1180 log_warning("invalid udev.children-max ignored: %s", value);
1181 } else if (streq(key, "exec-delay")) {
1182 r = safe_atoi(value, &arg_exec_delay);
1183 if (r < 0)
1184 log_warning("invalid udev.exec-delay ignored: %s", value);
1185 } else if (streq(key, "event-timeout")) {
1186 r = safe_atou64(value, &arg_event_timeout_usec);
1187 if (r < 0)
1188 log_warning("invalid udev.event-timeout ignored: %s", value);
1189 else {
1190 arg_event_timeout_usec *= USEC_PER_SEC;
1191 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1192 }
1193 }
1194
1195 return 0;
1196 }
1197
1198 static void help(void) {
1199 printf("%s [OPTIONS...]\n\n"
1200 "Manages devices.\n\n"
1201 " -h --help Print this message\n"
1202 " --version Print version of the program\n"
1203 " --daemon Detach and run in the background\n"
1204 " --debug Enable debug output\n"
1205 " --children-max=INT Set maximum number of workers\n"
1206 " --exec-delay=SECONDS Seconds to wait before executing RUN=\n"
1207 " --event-timeout=SECONDS Seconds to wait before terminating an event\n"
1208 " --resolve-names=early|late|never\n"
1209 " When to resolve users and groups\n"
1210 , program_invocation_short_name);
1211 }
1212
1213 static int parse_argv(int argc, char *argv[]) {
1214 static const struct option options[] = {
1215 { "daemon", no_argument, NULL, 'd' },
1216 { "debug", no_argument, NULL, 'D' },
1217 { "children-max", required_argument, NULL, 'c' },
1218 { "exec-delay", required_argument, NULL, 'e' },
1219 { "event-timeout", required_argument, NULL, 't' },
1220 { "resolve-names", required_argument, NULL, 'N' },
1221 { "help", no_argument, NULL, 'h' },
1222 { "version", no_argument, NULL, 'V' },
1223 {}
1224 };
1225
1226 int c;
1227
1228 assert(argc >= 0);
1229 assert(argv);
1230
1231 while ((c = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL)) >= 0) {
1232 int r;
1233
1234 switch (c) {
1235
1236 case 'd':
1237 arg_daemonize = true;
1238 break;
1239 case 'c':
1240 r = safe_atou(optarg, &arg_children_max);
1241 if (r < 0)
1242 log_warning("Invalid --children-max ignored: %s", optarg);
1243 break;
1244 case 'e':
1245 r = safe_atoi(optarg, &arg_exec_delay);
1246 if (r < 0)
1247 log_warning("Invalid --exec-delay ignored: %s", optarg);
1248 break;
1249 case 't':
1250 r = safe_atou64(optarg, &arg_event_timeout_usec);
1251 if (r < 0)
1252 log_warning("Invalid --event-timeout ignored: %s", optarg);
1253 else {
1254 arg_event_timeout_usec *= USEC_PER_SEC;
1255 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1256 }
1257 break;
1258 case 'D':
1259 arg_debug = true;
1260 break;
1261 case 'N':
1262 if (streq(optarg, "early")) {
1263 arg_resolve_names = 1;
1264 } else if (streq(optarg, "late")) {
1265 arg_resolve_names = 0;
1266 } else if (streq(optarg, "never")) {
1267 arg_resolve_names = -1;
1268 } else {
1269 log_error("resolve-names must be early, late or never");
1270 return 0;
1271 }
1272 break;
1273 case 'h':
1274 help();
1275 return 0;
1276 case 'V':
1277 printf("%s\n", VERSION);
1278 return 0;
1279 case '?':
1280 return -EINVAL;
1281 default:
1282 assert_not_reached("Unhandled option");
1283
1284 }
1285 }
1286
1287 return 1;
1288 }
1289
1290 static int manager_new(Manager **ret) {
1291 _cleanup_(manager_freep) Manager *manager = NULL;
1292 struct epoll_event ep_ctrl = { .events = EPOLLIN };
1293 struct epoll_event ep_inotify = { .events = EPOLLIN };
1294 struct epoll_event ep_signal = { .events = EPOLLIN };
1295 struct epoll_event ep_netlink = { .events = EPOLLIN };
1296 struct epoll_event ep_worker = { .events = EPOLLIN };
1297 sigset_t mask;
1298 int r, one = 1;
1299
1300 assert(ret);
1301
1302 manager = new0(Manager, 1);
1303 if (!manager)
1304 return log_oom();
1305
1306 manager->fd_ep = -1;
1307 manager->fd_ctrl = -1;
1308 manager->fd_uevent = -1;
1309 manager->fd_signal = -1;
1310 manager->fd_inotify = -1;
1311 manager->worker_watch[WRITE_END] = -1;
1312 manager->worker_watch[READ_END] = -1;
1313
1314 manager->udev = udev_new();
1315 if (!manager->udev)
1316 return log_error_errno(errno, "could not allocate udev context: %m");
1317
1318 udev_builtin_init(manager->udev);
1319
1320 manager->rules = udev_rules_new(manager->udev, arg_resolve_names);
1321 if (!manager->rules)
1322 return log_error_errno(ENOMEM, "error reading rules");
1323
1324 udev_list_node_init(&manager->events);
1325 udev_list_init(manager->udev, &manager->properties, true);
1326
1327 r = systemd_fds(&manager->fd_ctrl, &manager->fd_uevent);
1328 if (r >= 0) {
1329 /* get control and netlink socket from systemd */
1330 manager->ctrl = udev_ctrl_new_from_fd(manager->udev, manager->fd_ctrl);
1331 if (!manager->ctrl)
1332 return log_error_errno(EINVAL, "error taking over udev control socket");
1333
1334 manager->monitor = udev_monitor_new_from_netlink_fd(manager->udev, "kernel", manager->fd_uevent);
1335 if (!manager->monitor)
1336 return log_error_errno(EINVAL, "error taking over netlink socket");
1337
1338 /* get our own cgroup, we regularly kill everything udev has left behind */
1339 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &manager->cgroup);
1340 if (r < 0)
1341 log_warning_errno(r, "failed to get cgroup: %m");
1342 } else {
1343 /* open control and netlink socket */
1344 manager->ctrl = udev_ctrl_new(manager->udev);
1345 if (!manager->ctrl)
1346 return log_error_errno(EINVAL, "error initializing udev control socket");
1347
1348 manager->fd_ctrl = udev_ctrl_get_fd(manager->ctrl);
1349
1350 manager->monitor = udev_monitor_new_from_netlink(manager->udev, "kernel");
1351 if (!manager->monitor)
1352 return log_error_errno(EINVAL, "error initializing netlink socket");
1353
1354 manager->fd_uevent = udev_monitor_get_fd(manager->monitor);
1355
1356 (void) udev_monitor_set_receive_buffer_size(manager->monitor, 128 * 1024 * 1024);
1357 }
1358
1359 r = udev_monitor_enable_receiving(manager->monitor);
1360 if (r < 0)
1361 return log_error_errno(EINVAL, "error binding netlink socket");
1362
1363 r = udev_ctrl_enable_receiving(manager->ctrl);
1364 if (r < 0)
1365 return log_error_errno(EINVAL, "error binding udev control socket");
1366
1367 /* unnamed socket from workers to the main daemon */
1368 r = socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, manager->worker_watch);
1369 if (r < 0)
1370 return log_error_errno(errno, "error creating socketpair: %m");
1371
1372 manager->fd_worker = manager->worker_watch[READ_END];
1373
1374 r = setsockopt(manager->fd_worker, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
1375 if (r < 0)
1376 return log_error_errno(errno, "could not enable SO_PASSCRED: %m");
1377
1378 manager->fd_inotify = udev_watch_init(manager->udev);
1379 if (manager->fd_inotify < 0)
1380 return log_error_errno(ENOMEM, "error initializing inotify");
1381
1382 udev_watch_restore(manager->udev);
1383
1384 /* block and listen to all signals on signalfd */
1385 sigfillset(&mask);
1386 sigprocmask(SIG_SETMASK, &mask, &manager->sigmask_orig);
1387 manager->fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1388 if (manager->fd_signal < 0)
1389 return log_error_errno(errno, "error creating signalfd");
1390
1391 ep_ctrl.data.fd = manager->fd_ctrl;
1392 ep_inotify.data.fd = manager->fd_inotify;
1393 ep_signal.data.fd = manager->fd_signal;
1394 ep_netlink.data.fd = manager->fd_uevent;
1395 ep_worker.data.fd = manager->fd_worker;
1396
1397 manager->fd_ep = epoll_create1(EPOLL_CLOEXEC);
1398 if (manager->fd_ep < 0)
1399 return log_error_errno(errno, "error creating epoll fd: %m");
1400
1401 if (epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_ctrl, &ep_ctrl) < 0 ||
1402 epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_inotify, &ep_inotify) < 0 ||
1403 epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_signal, &ep_signal) < 0 ||
1404 epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_uevent, &ep_netlink) < 0 ||
1405 epoll_ctl(manager->fd_ep, EPOLL_CTL_ADD, manager->fd_worker, &ep_worker) < 0)
1406 return log_error_errno(errno, "fail to add fds to epoll: %m");
1407
1408 *ret = manager;
1409 manager = NULL;
1410
1411 return 1;
1412 }
1413
1414 int main(int argc, char *argv[]) {
1415 _cleanup_(manager_freep) Manager *manager = NULL;
1416 int r;
1417
1418 log_set_target(LOG_TARGET_AUTO);
1419 log_parse_environment();
1420 log_open();
1421
1422 r = parse_argv(argc, argv);
1423 if (r <= 0)
1424 goto exit;
1425
1426 r = parse_proc_cmdline(parse_proc_cmdline_item);
1427 if (r < 0)
1428 log_warning_errno(r, "failed to parse kernel command line, ignoring: %m");
1429
1430 if (arg_debug)
1431 log_set_max_level(LOG_DEBUG);
1432
1433 if (getuid() != 0) {
1434 r = log_error_errno(EPERM, "root privileges required");
1435 goto exit;
1436 }
1437
1438 if (arg_children_max == 0) {
1439 cpu_set_t cpu_set;
1440
1441 arg_children_max = 8;
1442
1443 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1444 arg_children_max += CPU_COUNT(&cpu_set) * 2;
1445 }
1446
1447 log_debug("set children_max to %u", arg_children_max);
1448 }
1449
1450 /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1451 if (arg_daemonize) {
1452 int fd;
1453
1454 fd = open("/dev/null", O_RDWR);
1455 if (fd < 0)
1456 log_error("cannot open /dev/null");
1457 else {
1458 if (write(STDOUT_FILENO, 0, 0) < 0)
1459 dup2(fd, STDOUT_FILENO);
1460 if (write(STDERR_FILENO, 0, 0) < 0)
1461 dup2(fd, STDERR_FILENO);
1462 if (fd > STDERR_FILENO)
1463 close(fd);
1464 }
1465 }
1466
1467 /* set umask before creating any file/directory */
1468 r = chdir("/");
1469 if (r < 0) {
1470 r = log_error_errno(errno, "could not change dir to /: %m");
1471 goto exit;
1472 }
1473
1474 umask(022);
1475
1476 r = mac_selinux_init("/dev");
1477 if (r < 0) {
1478 log_error_errno(r, "could not initialize labelling: %m");
1479 goto exit;
1480 }
1481
1482 r = mkdir("/run/udev", 0755);
1483 if (r < 0 && errno != EEXIST) {
1484 r = log_error_errno(errno, "could not create /run/udev: %m");
1485 goto exit;
1486 }
1487
1488 dev_setup(NULL, UID_INVALID, GID_INVALID);
1489
1490 r = manager_new(&manager);
1491 if (r < 0)
1492 goto exit;
1493
1494 log_info("starting version " VERSION);
1495
1496 r = udev_rules_apply_static_dev_perms(manager->rules);
1497 if (r < 0)
1498 log_error_errno(r, "failed to apply permissions on static device nodes: %m");
1499
1500 if (arg_daemonize) {
1501 pid_t pid;
1502
1503 pid = fork();
1504 switch (pid) {
1505 case 0:
1506 break;
1507 case -1:
1508 r = log_error_errno(errno, "fork of daemon failed: %m");
1509 goto exit;
1510 default:
1511 mac_selinux_finish();
1512 log_close();
1513 _exit(EXIT_SUCCESS);
1514 }
1515
1516 setsid();
1517
1518 write_string_file("/proc/self/oom_score_adj", "-1000");
1519 } else
1520 sd_notify(1, "READY=1");
1521
1522 for (;;) {
1523 static usec_t last_usec;
1524 struct epoll_event ev[8];
1525 int fdcount;
1526 int timeout;
1527 bool is_worker, is_signal, is_inotify, is_uevent, is_ctrl;
1528 int i;
1529
1530 if (manager->exit) {
1531 /* close sources of new events and discard buffered events */
1532 if (manager->fd_ctrl >= 0) {
1533 epoll_ctl(manager->fd_ep, EPOLL_CTL_DEL, manager->fd_ctrl, NULL);
1534 manager->fd_ctrl = safe_close(manager->fd_ctrl);
1535 }
1536
1537 if (manager->monitor) {
1538 epoll_ctl(manager->fd_ep, EPOLL_CTL_DEL, manager->fd_uevent, NULL);
1539 manager->monitor = udev_monitor_unref(manager->monitor);
1540 }
1541
1542 if (manager->fd_inotify >= 0) {
1543 epoll_ctl(manager->fd_ep, EPOLL_CTL_DEL, manager->fd_inotify, NULL);
1544 manager->fd_inotify = safe_close(manager->fd_inotify);
1545 }
1546
1547 /* discard queued events and kill workers */
1548 event_queue_cleanup(manager, EVENT_QUEUED);
1549 manager_kill_workers(manager);
1550
1551 /* exit after all has cleaned up */
1552 if (udev_list_node_is_empty(&manager->events) && hashmap_isempty(manager->workers))
1553 break;
1554
1555 /* timeout at exit for workers to finish */
1556 timeout = 30 * MSEC_PER_SEC;
1557 } else if (udev_list_node_is_empty(&manager->events) && hashmap_isempty(manager->workers)) {
1558 /* we are idle */
1559 timeout = -1;
1560
1561 /* cleanup possible left-over processes in our cgroup */
1562 if (manager->cgroup)
1563 cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, false, true, NULL);
1564 } else {
1565 /* kill idle or hanging workers */
1566 timeout = 3 * MSEC_PER_SEC;
1567 }
1568
1569 fdcount = epoll_wait(manager->fd_ep, ev, ELEMENTSOF(ev), timeout);
1570 if (fdcount < 0)
1571 continue;
1572
1573 if (fdcount == 0) {
1574 struct worker *worker;
1575 Iterator j;
1576
1577 /* timeout */
1578 if (manager->exit) {
1579 log_error("timeout, giving up waiting for workers to finish");
1580 break;
1581 }
1582
1583 /* kill idle workers */
1584 if (udev_list_node_is_empty(&manager->events)) {
1585 log_debug("cleanup idle workers");
1586 manager_kill_workers(manager);
1587 }
1588
1589 /* check for hanging events */
1590 HASHMAP_FOREACH(worker, manager->workers, j) {
1591 struct event *event = worker->event;
1592 usec_t ts;
1593
1594 if (worker->state != WORKER_RUNNING)
1595 continue;
1596
1597 assert(event);
1598
1599 ts = now(CLOCK_MONOTONIC);
1600
1601 if ((ts - event->start_usec) > arg_event_timeout_warn_usec) {
1602 if ((ts - event->start_usec) > arg_event_timeout_usec)
1603 on_event_timeout(NULL, 0, event);
1604 else if (!event->warned) {
1605 on_event_timeout_warning(NULL, 0, event);
1606 event->warned = true;
1607 }
1608 }
1609 }
1610
1611 }
1612
1613 is_worker = is_signal = is_inotify = is_uevent = is_ctrl = false;
1614 for (i = 0; i < fdcount; i++) {
1615 if (ev[i].data.fd == manager->fd_worker && ev[i].events & EPOLLIN)
1616 is_worker = true;
1617 else if (ev[i].data.fd == manager->fd_uevent && ev[i].events & EPOLLIN)
1618 is_uevent = true;
1619 else if (ev[i].data.fd == manager->fd_signal && ev[i].events & EPOLLIN)
1620 is_signal = true;
1621 else if (ev[i].data.fd == manager->fd_inotify && ev[i].events & EPOLLIN)
1622 is_inotify = true;
1623 else if (ev[i].data.fd == manager->fd_ctrl && ev[i].events & EPOLLIN)
1624 is_ctrl = true;
1625 }
1626
1627 /* check for changed config, every 3 seconds at most */
1628 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1629 if (udev_rules_check_timestamp(manager->rules))
1630 manager->reload = true;
1631 if (udev_builtin_validate(manager->udev))
1632 manager->reload = true;
1633
1634 last_usec = now(CLOCK_MONOTONIC);
1635 }
1636
1637 /* reload requested, HUP signal received, rules changed, builtin changed */
1638 if (manager->reload) {
1639 manager_kill_workers(manager);
1640 manager->rules = udev_rules_unref(manager->rules);
1641 udev_builtin_exit(manager->udev);
1642 manager->reload = false;
1643 }
1644
1645 /* event has finished */
1646 if (is_worker)
1647 on_worker(NULL, manager->fd_worker, 0, manager);
1648
1649 /* uevent from kernel */
1650 if (is_uevent)
1651 on_uevent(NULL, manager->fd_uevent, 0, manager);
1652
1653 /* start new events */
1654 if (!udev_list_node_is_empty(&manager->events) && !manager->exit && !manager->stop_exec_queue) {
1655 udev_builtin_init(manager->udev);
1656 if (!manager->rules)
1657 manager->rules = udev_rules_new(manager->udev, arg_resolve_names);
1658 if (manager->rules)
1659 event_queue_start(manager);
1660 }
1661
1662 if (is_signal) {
1663 struct signalfd_siginfo fdsi;
1664 ssize_t size;
1665
1666 size = read(manager->fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1667 if (size == sizeof(struct signalfd_siginfo)) {
1668 switch (fdsi.ssi_signo) {
1669 case SIGINT:
1670 case SIGTERM:
1671 on_sigterm(NULL, &fdsi, manager);
1672 break;
1673 case SIGHUP:
1674 on_sighup(NULL, &fdsi, manager);
1675 break;
1676 case SIGCHLD:
1677 on_sigchld(NULL, &fdsi, manager);
1678 break;
1679 }
1680 }
1681 }
1682
1683 /* we are shutting down, the events below are not handled anymore */
1684 if (manager->exit)
1685 continue;
1686
1687 /* device node watch */
1688 if (is_inotify)
1689 on_inotify(NULL, manager->fd_inotify, 0, manager);
1690
1691 /*
1692 * This needs to be after the inotify handling, to make sure,
1693 * that the ping is send back after the possibly generated
1694 * "change" events by the inotify device node watch.
1695 */
1696 if (is_ctrl)
1697 on_ctrl_msg(NULL, manager->fd_ctrl, 0, manager);
1698 }
1699
1700 exit:
1701 if (manager)
1702 udev_ctrl_cleanup(manager->ctrl);
1703 mac_selinux_finish();
1704 log_close();
1705 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
1706 }
Unnamed repository; edit this file 'description' to name the repository.