chiark / gitweb /
udevd: create queue file before daemonizing to reliably block 'settle'
[elogind.git] / udev / udevd.c
1 /*
2  * Copyright (C) 2004-2011 Kay Sievers <kay.sievers@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
50 static bool debug;
51
52 static void log_fn(struct udev *udev, int priority,
53                    const char *file, int line, const char *fn,
54                    const char *format, va_list args)
55 {
56         if (debug) {
57                 char buf[1024];
58                 struct timeval tv;
59                 struct timezone tz;
60
61                 vsnprintf(buf, sizeof(buf), format, args);
62                 gettimeofday(&tv, &tz);
63                 fprintf(stderr, "%llu.%06u [%u] %s: %s",
64                         (unsigned long long) tv.tv_sec, (unsigned int) tv.tv_usec,
65                         (int) getpid(), fn, buf);
66         } else {
67                 vsyslog(priority, format, args);
68         }
69 }
70
71 static struct udev_rules *rules;
72 static struct udev_queue_export *udev_queue_export;
73 static struct udev_ctrl *udev_ctrl;
74 static struct udev_monitor *monitor;
75 static int worker_watch[2] = { -1, -1 };
76 static int fd_signal = -1;
77 static int fd_ep = -1;
78 static int fd_inotify = -1;
79 static bool stop_exec_queue;
80 static bool reload_config;
81 static int children;
82 static int children_max;
83 static int exec_delay;
84 static sigset_t sigmask_orig;
85 static UDEV_LIST(event_list);
86 static UDEV_LIST(worker_list);
87 static bool udev_exit;
88
89 enum event_state {
90         EVENT_UNDEF,
91         EVENT_QUEUED,
92         EVENT_RUNNING,
93 };
94
95 struct event {
96         struct udev_list_node node;
97         struct udev *udev;
98         struct udev_device *dev;
99         enum event_state state;
100         int exitcode;
101         unsigned long long int delaying_seqnum;
102         unsigned long long int seqnum;
103         const char *devpath;
104         size_t devpath_len;
105         const char *devpath_old;
106         dev_t devnum;
107         bool is_block;
108         int ifindex;
109 };
110
111 static struct event *node_to_event(struct udev_list_node *node)
112 {
113         char *event;
114
115         event = (char *)node;
116         event -= offsetof(struct event, node);
117         return (struct event *)event;
118 }
119
120 static void event_queue_cleanup(struct udev *udev, enum event_state type);
121
122 enum worker_state {
123         WORKER_UNDEF,
124         WORKER_RUNNING,
125         WORKER_IDLE,
126         WORKER_KILLED,
127 };
128
129 struct worker {
130         struct udev_list_node node;
131         struct udev *udev;
132         int refcount;
133         pid_t pid;
134         struct udev_monitor *monitor;
135         enum worker_state state;
136         struct event *event;
137 };
138
139 /* passed from worker to main process */
140 struct worker_message {
141         pid_t pid;
142         int exitcode;
143 };
144
145 static struct worker *node_to_worker(struct udev_list_node *node)
146 {
147         char *worker;
148
149         worker = (char *)node;
150         worker -= offsetof(struct worker, node);
151         return (struct worker *)worker;
152 }
153
154 static void event_queue_delete(struct event *event, bool export)
155 {
156         udev_list_node_remove(&event->node);
157
158         if (export) {
159                 /* mark as failed, if "add" event returns non-zero */
160                 if (event->exitcode != 0 && strcmp(udev_device_get_action(event->dev), "remove") != 0)
161                         udev_queue_export_device_failed(udev_queue_export, event->dev);
162                 else
163                         udev_queue_export_device_finished(udev_queue_export, event->dev);
164                 info(event->udev, "seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode);
165         }
166         udev_device_unref(event->dev);
167         free(event);
168 }
169
170 static struct worker *worker_ref(struct worker *worker)
171 {
172         worker->refcount++;
173         return worker;
174 }
175
176 static void worker_cleanup(struct worker *worker)
177 {
178         udev_list_node_remove(&worker->node);
179         udev_monitor_unref(worker->monitor);
180         children--;
181         free(worker);
182 }
183
184 static void worker_unref(struct worker *worker)
185 {
186         worker->refcount--;
187         if (worker->refcount > 0)
188                 return;
189         info(worker->udev, "worker [%u] cleaned up\n", worker->pid);
190         worker_cleanup(worker);
191 }
192
193 static void worker_list_cleanup(struct udev *udev)
194 {
195         struct udev_list_node *loop, *tmp;
196
197         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
198                 struct worker *worker = node_to_worker(loop);
199
200                 worker_cleanup(worker);
201         }
202 }
203
204 static void worker_new(struct event *event)
205 {
206         struct udev *udev = event->udev;
207         struct worker *worker;
208         struct udev_monitor *worker_monitor;
209         pid_t pid;
210
211         /* listen for new events */
212         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
213         if (worker_monitor == NULL)
214                 return;
215         /* allow the main daemon netlink address to send devices to the worker */
216         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
217         udev_monitor_enable_receiving(worker_monitor);
218
219         worker = calloc(1, sizeof(struct worker));
220         if (worker == NULL) {
221                 udev_monitor_unref(worker_monitor);
222                 return;
223         }
224         /* worker + event reference */
225         worker->refcount = 2;
226         worker->udev = udev;
227
228         pid = fork();
229         switch (pid) {
230         case 0: {
231                 struct udev_device *dev = NULL;
232                 int fd_monitor;
233                 struct epoll_event ep_signal, ep_monitor;
234                 sigset_t mask;
235                 int rc = EXIT_SUCCESS;
236
237                 /* move initial device from queue */
238                 dev = event->dev;
239                 event->dev = NULL;
240
241                 free(worker);
242                 worker_list_cleanup(udev);
243                 event_queue_cleanup(udev, EVENT_UNDEF);
244                 udev_queue_export_unref(udev_queue_export);
245                 udev_monitor_unref(monitor);
246                 udev_ctrl_unref(udev_ctrl);
247                 close(fd_signal);
248                 close(fd_ep);
249                 close(worker_watch[READ_END]);
250
251                 sigfillset(&mask);
252                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
253                 if (fd_signal < 0) {
254                         err(udev, "error creating signalfd %m\n");
255                         rc = 2;
256                         goto out;
257                 }
258
259                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
260                 if (fd_ep < 0) {
261                         err(udev, "error creating epoll fd: %m\n");
262                         rc = 3;
263                         goto out;
264                 }
265
266                 memset(&ep_signal, 0, sizeof(struct epoll_event));
267                 ep_signal.events = EPOLLIN;
268                 ep_signal.data.fd = fd_signal;
269
270                 fd_monitor = udev_monitor_get_fd(worker_monitor);
271                 memset(&ep_monitor, 0, sizeof(struct epoll_event));
272                 ep_monitor.events = EPOLLIN;
273                 ep_monitor.data.fd = fd_monitor;
274
275                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
276                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
277                         err(udev, "fail to add fds to epoll: %m\n");
278                         rc = 4;
279                         goto out;
280                 }
281
282                 /* request TERM signal if parent exits */
283                 prctl(PR_SET_PDEATHSIG, SIGTERM);
284
285                 for (;;) {
286                         struct udev_event *udev_event;
287                         struct worker_message msg;
288                         int failed = 0;
289                         int err;
290
291                         info(udev, "seq %llu running\n", udev_device_get_seqnum(dev));
292                         udev_event = udev_event_new(dev);
293                         if (udev_event == NULL) {
294                                 rc = 5;
295                                 goto out;
296                         }
297
298                         /* needed for SIGCHLD/SIGTERM in spawn() */
299                         udev_event->fd_signal = fd_signal;
300
301                         if (exec_delay > 0)
302                                 udev_event->exec_delay = exec_delay;
303
304                         /* apply rules, create node, symlinks */
305                         err = udev_event_execute_rules(udev_event, rules, &sigmask_orig);
306
307                         if (err == 0)
308                                 failed = udev_event_execute_run(udev_event, &sigmask_orig);
309
310                         /* apply/restore inotify watch */
311                         if (err == 0 && udev_event->inotify_watch) {
312                                 udev_watch_begin(udev, dev);
313                                 udev_device_update_db(dev);
314                         }
315
316                         /* send processed event back to libudev listeners */
317                         udev_monitor_send_device(worker_monitor, NULL, dev);
318
319                         /* send udevd the result of the event execution */
320                         memset(&msg, 0, sizeof(struct worker_message));
321                         if (err != 0)
322                                 msg.exitcode = err;
323                         else if (failed != 0)
324                                 msg.exitcode = failed;
325                         msg.pid = getpid();
326                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
327
328                         info(udev, "seq %llu processed with %i\n", udev_device_get_seqnum(dev), err);
329
330                         udev_device_unref(dev);
331                         dev = NULL;
332
333                         if (udev_event->sigterm) {
334                                 udev_event_unref(udev_event);
335                                 goto out;
336                         }
337
338                         udev_event_unref(udev_event);
339
340                         /* wait for more device messages from main udevd, or term signal */
341                         while (dev == NULL) {
342                                 struct epoll_event ev[4];
343                                 int fdcount;
344                                 int i;
345
346                                 fdcount = epoll_wait(fd_ep, ev, ARRAY_SIZE(ev), -1);
347                                 if (fdcount < 0) {
348                                         if (errno == EINTR)
349                                                 continue;
350                                         err = -errno;
351                                         err(udev, "failed to poll: %m\n");
352                                         goto out;
353                                 }
354
355                                 for (i = 0; i < fdcount; i++) {
356                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
357                                                 dev = udev_monitor_receive_device(worker_monitor);
358                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
359                                                 struct signalfd_siginfo fdsi;
360                                                 ssize_t size;
361
362                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
363                                                 if (size != sizeof(struct signalfd_siginfo))
364                                                         continue;
365                                                 switch (fdsi.ssi_signo) {
366                                                 case SIGTERM:
367                                                         goto out;
368                                                 case SIGALRM:
369                                                         rc = EXIT_FAILURE;
370                                                         goto out;
371                                                 }
372                                         }
373                                 }
374                         }
375                 }
376 out:
377                 udev_device_unref(dev);
378                 if (fd_signal >= 0)
379                         close(fd_signal);
380                 if (fd_ep >= 0)
381                         close(fd_ep);
382                 close(fd_inotify);
383                 close(worker_watch[WRITE_END]);
384                 udev_rules_unref(rules);
385                 udev_monitor_unref(worker_monitor);
386                 udev_unref(udev);
387                 udev_log_close();
388                 exit(rc);
389         }
390         case -1:
391                 udev_monitor_unref(worker_monitor);
392                 event->state = EVENT_QUEUED;
393                 free(worker);
394                 err(udev, "fork of child failed: %m\n");
395                 break;
396         default:
397                 /* close monitor, but keep address around */
398                 udev_monitor_disconnect(worker_monitor);
399                 worker->monitor = worker_monitor;
400                 worker->pid = pid;
401                 worker->state = WORKER_RUNNING;
402                 worker->event = event;
403                 event->state = EVENT_RUNNING;
404                 udev_list_node_append(&worker->node, &worker_list);
405                 children++;
406                 info(udev, "seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid);
407                 break;
408         }
409 }
410
411 static void event_run(struct event *event, bool force)
412 {
413         struct udev_list_node *loop;
414
415         udev_list_node_foreach(loop, &worker_list) {
416                 struct worker *worker = node_to_worker(loop);
417                 ssize_t count;
418
419                 if (worker->state != WORKER_IDLE)
420                         continue;
421
422                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
423                 if (count < 0) {
424                         err(event->udev, "worker [%u] did not accept message %zi (%m), kill it\n", worker->pid, count);
425                         kill(worker->pid, SIGKILL);
426                         worker->state = WORKER_KILLED;
427                         continue;
428                 }
429                 worker_ref(worker);
430                 worker->event = event;
431                 worker->state = WORKER_RUNNING;
432                 event->state = EVENT_RUNNING;
433                 return;
434         }
435
436         if (!force && children >= children_max) {
437                 if (children_max > 1)
438                         info(event->udev, "maximum number (%i) of children reached\n", children);
439                 return;
440         }
441
442         /* start new worker and pass initial device */
443         worker_new(event);
444 }
445
446 static int event_queue_insert(struct udev_device *dev)
447 {
448         struct event *event;
449
450         event = calloc(1, sizeof(struct event));
451         if (event == NULL)
452                 return -1;
453
454         event->udev = udev_device_get_udev(dev);
455         event->dev = dev;
456         event->seqnum = udev_device_get_seqnum(dev);
457         event->devpath = udev_device_get_devpath(dev);
458         event->devpath_len = strlen(event->devpath);
459         event->devpath_old = udev_device_get_devpath_old(dev);
460         event->devnum = udev_device_get_devnum(dev);
461         event->is_block = (strcmp("block", udev_device_get_subsystem(dev)) == 0);
462         event->ifindex = udev_device_get_ifindex(dev);
463
464         udev_queue_export_device_queued(udev_queue_export, dev);
465         info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev),
466              udev_device_get_action(dev), udev_device_get_subsystem(dev));
467
468         event->state = EVENT_QUEUED;
469         udev_list_node_append(&event->node, &event_list);
470
471         /* run all events with a timeout set immediately */
472         if (udev_device_get_timeout(dev) > 0) {
473                 event_run(event, true);
474                 return 0;
475         }
476
477         return 0;
478 }
479
480 static void worker_kill(struct udev *udev, int retain)
481 {
482         struct udev_list_node *loop;
483         int max;
484
485         if (children <= retain)
486                 return;
487
488         max = children - retain;
489
490         udev_list_node_foreach(loop, &worker_list) {
491                 struct worker *worker = node_to_worker(loop);
492
493                 if (max-- <= 0)
494                         break;
495
496                 if (worker->state == WORKER_KILLED)
497                         continue;
498
499                 worker->state = WORKER_KILLED;
500                 kill(worker->pid, SIGTERM);
501         }
502 }
503
504 /* lookup event for identical, parent, child device */
505 static bool is_devpath_busy(struct event *event)
506 {
507         struct udev_list_node *loop;
508         size_t common;
509
510         /* check if queue contains events we depend on */
511         udev_list_node_foreach(loop, &event_list) {
512                 struct event *loop_event = node_to_event(loop);
513
514                 /* we already found a later event, earlier can not block us, no need to check again */
515                 if (loop_event->seqnum < event->delaying_seqnum)
516                         continue;
517
518                 /* event we checked earlier still exists, no need to check again */
519                 if (loop_event->seqnum == event->delaying_seqnum)
520                         return true;
521
522                 /* found ourself, no later event can block us */
523                 if (loop_event->seqnum >= event->seqnum)
524                         break;
525
526                 /* check major/minor */
527                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
528                         return true;
529
530                 /* check network device ifindex */
531                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
532                         return true;
533
534                 /* check our old name */
535                 if (event->devpath_old != NULL && strcmp(loop_event->devpath, event->devpath_old) == 0) {
536                         event->delaying_seqnum = loop_event->seqnum;
537                         return true;
538                 }
539
540                 /* compare devpath */
541                 common = MIN(loop_event->devpath_len, event->devpath_len);
542
543                 /* one devpath is contained in the other? */
544                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
545                         continue;
546
547                 /* identical device event found */
548                 if (loop_event->devpath_len == event->devpath_len) {
549                         /* devices names might have changed/swapped in the meantime */
550                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
551                                 continue;
552                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
553                                 continue;
554                         event->delaying_seqnum = loop_event->seqnum;
555                         return true;
556                 }
557
558                 /* parent device event found */
559                 if (event->devpath[common] == '/') {
560                         event->delaying_seqnum = loop_event->seqnum;
561                         return true;
562                 }
563
564                 /* child device event found */
565                 if (loop_event->devpath[common] == '/') {
566                         event->delaying_seqnum = loop_event->seqnum;
567                         return true;
568                 }
569
570                 /* no matching device */
571                 continue;
572         }
573
574         return false;
575 }
576
577 static void event_queue_start(struct udev *udev)
578 {
579         struct udev_list_node *loop;
580
581         udev_list_node_foreach(loop, &event_list) {
582                 struct event *event = node_to_event(loop);
583
584                 if (event->state != EVENT_QUEUED)
585                         continue;
586
587                 /* do not start event if parent or child event is still running */
588                 if (is_devpath_busy(event)) {
589                         dbg(udev, "delay seq %llu (%s)\n", event->seqnum, event->devpath);
590                         continue;
591                 }
592
593                 event_run(event, false);
594         }
595 }
596
597 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
598 {
599         struct udev_list_node *loop, *tmp;
600
601         udev_list_node_foreach_safe(loop, tmp, &event_list) {
602                 struct event *event = node_to_event(loop);
603
604                 if (match_type != EVENT_UNDEF && match_type != event->state)
605                         continue;
606
607                 event_queue_delete(event, false);
608         }
609 }
610
611 static void worker_returned(int fd_worker)
612 {
613         for (;;) {
614                 struct worker_message msg;
615                 ssize_t size;
616                 struct udev_list_node *loop;
617
618                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
619                 if (size != sizeof(struct worker_message))
620                         break;
621
622                 /* lookup worker who sent the signal */
623                 udev_list_node_foreach(loop, &worker_list) {
624                         struct worker *worker = node_to_worker(loop);
625
626                         if (worker->pid != msg.pid)
627                                 continue;
628
629                         /* worker returned */
630                         worker->event->exitcode = msg.exitcode;
631                         event_queue_delete(worker->event, true);
632                         worker->event = NULL;
633                         if (worker->state != WORKER_KILLED)
634                                 worker->state = WORKER_IDLE;
635                         worker_unref(worker);
636                         break;
637                 }
638         }
639 }
640
641 /* receive the udevd message from userspace */
642 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
643 {
644         struct udev *udev = udev_ctrl_get_udev(uctrl);
645         struct udev_ctrl_connection *ctrl_conn;
646         struct udev_ctrl_msg *ctrl_msg = NULL;
647         const char *str;
648         int i;
649
650         ctrl_conn = udev_ctrl_get_connection(uctrl);
651         if (ctrl_conn == NULL)
652                 goto out;
653
654         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
655         if (ctrl_msg == NULL)
656                 goto out;
657
658         i = udev_ctrl_get_set_log_level(ctrl_msg);
659         if (i >= 0) {
660                 info(udev, "udevd message (SET_LOG_PRIORITY) received, log_priority=%i\n", i);
661                 udev_set_log_priority(udev, i);
662                 worker_kill(udev, 0);
663         }
664
665         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
666                 info(udev, "udevd message (STOP_EXEC_QUEUE) received\n");
667                 stop_exec_queue = true;
668         }
669
670         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
671                 info(udev, "udevd message (START_EXEC_QUEUE) received\n");
672                 stop_exec_queue = false;
673         }
674
675         if (udev_ctrl_get_reload_rules(ctrl_msg) > 0) {
676                 info(udev, "udevd message (RELOAD_RULES) received\n");
677                 reload_config = true;
678         }
679
680         str = udev_ctrl_get_set_env(ctrl_msg);
681         if (str != NULL) {
682                 char *key;
683
684                 key = strdup(str);
685                 if (key != NULL) {
686                         char *val;
687
688                         val = strchr(key, '=');
689                         if (val != NULL) {
690                                 val[0] = '\0';
691                                 val = &val[1];
692                                 if (val[0] == '\0') {
693                                         info(udev, "udevd message (ENV) received, unset '%s'\n", key);
694                                         udev_add_property(udev, key, NULL);
695                                 } else {
696                                         info(udev, "udevd message (ENV) received, set '%s=%s'\n", key, val);
697                                         udev_add_property(udev, key, val);
698                                 }
699                         } else {
700                                 err(udev, "wrong key format '%s'\n", key);
701                         }
702                         free(key);
703                 }
704                 worker_kill(udev, 0);
705         }
706
707         i = udev_ctrl_get_set_children_max(ctrl_msg);
708         if (i >= 0) {
709                 info(udev, "udevd message (SET_MAX_CHILDREN) received, children_max=%i\n", i);
710                 children_max = i;
711         }
712
713         if (udev_ctrl_get_ping(ctrl_msg) > 0)
714                 info(udev, "udevd message (SYNC) received\n");
715
716         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
717                 info(udev, "udevd message (EXIT) received\n");
718                 udev_exit = true;
719                 /* keep reference to block the client until we exit */
720                 udev_ctrl_connection_ref(ctrl_conn);
721         }
722 out:
723         udev_ctrl_msg_unref(ctrl_msg);
724         return udev_ctrl_connection_unref(ctrl_conn);
725 }
726
727 /* read inotify messages */
728 static int handle_inotify(struct udev *udev)
729 {
730         int nbytes, pos;
731         char *buf;
732         struct inotify_event *ev;
733
734         if ((ioctl(fd_inotify, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
735                 return 0;
736
737         buf = malloc(nbytes);
738         if (buf == NULL) {
739                 err(udev, "error getting buffer for inotify\n");
740                 return -1;
741         }
742
743         nbytes = read(fd_inotify, buf, nbytes);
744
745         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
746                 struct udev_device *dev;
747
748                 ev = (struct inotify_event *)(buf + pos);
749                 if (ev->len) {
750                         const char *s;
751
752                         info(udev, "inotify event: %x for %s\n", ev->mask, ev->name);
753                         s = strstr(ev->name, ".rules");
754                         if (s == NULL)
755                                 continue;
756                         if (strlen(s) != strlen(".rules"))
757                                 continue;
758                         reload_config = true;
759                         continue;
760                 }
761
762                 dev = udev_watch_lookup(udev, ev->wd);
763                 if (dev != NULL) {
764                         info(udev, "inotify event: %x for %s\n", ev->mask, udev_device_get_devnode(dev));
765                         if (ev->mask & IN_CLOSE_WRITE) {
766                                 char filename[UTIL_PATH_SIZE];
767                                 int fd;
768
769                                 info(udev, "device %s closed, synthesising 'change'\n", udev_device_get_devnode(dev));
770                                 util_strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
771                                 fd = open(filename, O_WRONLY);
772                                 if (fd >= 0) {
773                                         if (write(fd, "change", 6) < 0)
774                                                 info(udev, "error writing uevent: %m\n");
775                                         close(fd);
776                                 }
777                         }
778                         if (ev->mask & IN_IGNORED)
779                                 udev_watch_end(udev, dev);
780
781                         udev_device_unref(dev);
782                 }
783
784         }
785
786         free(buf);
787         return 0;
788 }
789
790 static void handle_signal(struct udev *udev, int signo)
791 {
792         switch (signo) {
793         case SIGINT:
794         case SIGTERM:
795                 udev_exit = true;
796                 break;
797         case SIGCHLD:
798                 for (;;) {
799                         pid_t pid;
800                         int status;
801                         struct udev_list_node *loop, *tmp;
802
803                         pid = waitpid(-1, &status, WNOHANG);
804                         if (pid <= 0)
805                                 break;
806
807                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
808                                 struct worker *worker = node_to_worker(loop);
809
810                                 if (worker->pid != pid)
811                                         continue;
812                                 info(udev, "worker [%u] exit\n", pid);
813
814                                 if (WIFEXITED(status)) {
815                                         if (WEXITSTATUS(status) != 0)
816                                                 err(udev, "worker [%u] exit with return code %i\n", pid, WEXITSTATUS(status));
817                                 } else if (WIFSIGNALED(status)) {
818                                         err(udev, "worker [%u] terminated by signal %i (%s)\n",
819                                             pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
820                                 } else if (WIFSTOPPED(status)) {
821                                         err(udev, "worker [%u] stopped\n", pid);
822                                 } else if (WIFCONTINUED(status)) {
823                                         err(udev, "worker [%u] continued\n", pid);
824                                 } else {
825                                         err(udev, "worker [%u] exit with status 0x%04x\n", pid, status);
826                                 }
827
828                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
829                                         if (worker->event != NULL) {
830                                                 err(udev, "worker [%u] failed while handling '%s'\n",
831                                                     pid, worker->event->devpath);
832                                                 worker->event->exitcode = -32;
833                                                 event_queue_delete(worker->event, true);
834                                                 /* drop reference taken for state 'running' */
835                                                 worker_unref(worker);
836                                         }
837                                 }
838                                 worker_unref(worker);
839                                 break;
840                         }
841                 }
842                 break;
843         case SIGHUP:
844                 reload_config = true;
845                 break;
846         }
847 }
848
849 static void static_dev_create_from_modules(struct udev *udev)
850 {
851         struct utsname kernel;
852         char modules[UTIL_PATH_SIZE];
853         char buf[4096];
854         FILE *f;
855
856         uname(&kernel);
857         util_strscpyl(modules, sizeof(modules), "/lib/modules/", kernel.release, "/modules.devname", NULL);
858         f = fopen(modules, "r");
859         if (f == NULL)
860                 return;
861
862         while (fgets(buf, sizeof(buf), f) != NULL) {
863                 char *s;
864                 const char *modname;
865                 const char *devname;
866                 const char *devno;
867                 int maj, min;
868                 char type;
869                 mode_t mode;
870                 char filename[UTIL_PATH_SIZE];
871
872                 if (buf[0] == '#')
873                         continue;
874
875                 modname = buf;
876                 s = strchr(modname, ' ');
877                 if (s == NULL)
878                         continue;
879                 s[0] = '\0';
880
881                 devname = &s[1];
882                 s = strchr(devname, ' ');
883                 if (s == NULL)
884                         continue;
885                 s[0] = '\0';
886
887                 devno = &s[1];
888                 s = strchr(devno, ' ');
889                 if (s == NULL)
890                         s = strchr(devno, '\n');
891                 if (s != NULL)
892                         s[0] = '\0';
893                 if (sscanf(devno, "%c%u:%u", &type, &maj, &min) != 3)
894                         continue;
895
896                 if (type == 'c')
897                         mode = 0600 | S_IFCHR;
898                 else if (type == 'b')
899                         mode = 0600 | S_IFBLK;
900                 else
901                         continue;
902
903                 util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/", devname, NULL);
904                 util_create_path_selinux(udev, filename);
905                 udev_selinux_setfscreatecon(udev, filename, mode);
906                 info(udev, "mknod '%s' %c%u:%u\n", filename, type, maj, min);
907                 if (mknod(filename, mode, makedev(maj, min)) < 0 && errno == EEXIST)
908                         utimensat(AT_FDCWD, filename, NULL, 0);
909                 udev_selinux_resetfscreatecon(udev);
910         }
911
912         fclose(f);
913 }
914
915 static int copy_dev_dir(struct udev *udev, DIR *dir_from, DIR *dir_to, int maxdepth)
916 {
917         struct dirent *dent;
918
919         for (dent = readdir(dir_from); dent != NULL; dent = readdir(dir_from)) {
920                 struct stat stats;
921
922                 if (dent->d_name[0] == '.')
923                         continue;
924                 if (fstatat(dirfd(dir_from), dent->d_name, &stats, AT_SYMLINK_NOFOLLOW) != 0)
925                         continue;
926
927                 if (S_ISBLK(stats.st_mode) || S_ISCHR(stats.st_mode)) {
928                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, stats.st_mode & 0777);
929                         if (mknodat(dirfd(dir_to), dent->d_name, stats.st_mode, stats.st_rdev) == 0) {
930                                 fchmodat(dirfd(dir_to), dent->d_name, stats.st_mode & 0777, 0);
931                                 fchownat(dirfd(dir_to), dent->d_name, stats.st_uid, stats.st_gid, 0);
932                         } else {
933                                 utimensat(dirfd(dir_to), dent->d_name, NULL, 0);
934                         }
935                         udev_selinux_resetfscreatecon(udev);
936                 } else if (S_ISLNK(stats.st_mode)) {
937                         char target[UTIL_PATH_SIZE];
938                         ssize_t len;
939
940                         len = readlinkat(dirfd(dir_from), dent->d_name, target, sizeof(target));
941                         if (len <= 0 || len == (ssize_t)sizeof(target))
942                                 continue;
943                         target[len] = '\0';
944                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFLNK);
945                         if (symlinkat(target, dirfd(dir_to), dent->d_name) < 0 && errno == EEXIST)
946                                 utimensat(dirfd(dir_to), dent->d_name, NULL, AT_SYMLINK_NOFOLLOW);
947                         udev_selinux_resetfscreatecon(udev);
948                 } else if (S_ISDIR(stats.st_mode)) {
949                         DIR *dir2_from, *dir2_to;
950
951                         if (maxdepth == 0)
952                                 continue;
953
954                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFDIR|0755);
955                         mkdirat(dirfd(dir_to), dent->d_name, 0755);
956                         udev_selinux_resetfscreatecon(udev);
957
958                         dir2_to = fdopendir(openat(dirfd(dir_to), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC));
959                         if (dir2_to == NULL)
960                                 continue;
961
962                         dir2_from = fdopendir(openat(dirfd(dir_from), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC));
963                         if (dir2_from == NULL) {
964                                 closedir(dir2_to);
965                                 continue;
966                         }
967
968                         copy_dev_dir(udev, dir2_from, dir2_to, maxdepth-1);
969
970                         closedir(dir2_to);
971                         closedir(dir2_from);
972                 }
973         }
974
975         return 0;
976 }
977
978 static void static_dev_create_links(struct udev *udev, DIR *dir)
979 {
980         struct stdlinks {
981                 const char *link;
982                 const char *target;
983         };
984         static const struct stdlinks stdlinks[] = {
985                 { "core", "/proc/kcore" },
986                 { "fd", "/proc/self/fd" },
987                 { "stdin", "/proc/self/fd/0" },
988                 { "stdout", "/proc/self/fd/1" },
989                 { "stderr", "/proc/self/fd/2" },
990         };
991         unsigned int i;
992
993         for (i = 0; i < ARRAY_SIZE(stdlinks); i++) {
994                 struct stat sb;
995
996                 if (stat(stdlinks[i].target, &sb) == 0) {
997                         udev_selinux_setfscreateconat(udev, dirfd(dir), stdlinks[i].link, S_IFLNK);
998                         if (symlinkat(stdlinks[i].target, dirfd(dir), stdlinks[i].link) < 0 && errno == EEXIST)
999                                 utimensat(dirfd(dir), stdlinks[i].link, NULL, AT_SYMLINK_NOFOLLOW);
1000                         udev_selinux_resetfscreatecon(udev);
1001                 }
1002         }
1003 }
1004
1005 static void static_dev_create_from_devices(struct udev *udev, DIR *dir)
1006 {
1007         DIR *dir_from;
1008
1009         dir_from = opendir(LIBEXECDIR "/devices");
1010         if (dir_from == NULL)
1011                 return;
1012         copy_dev_dir(udev, dir_from, dir, 8);
1013         closedir(dir_from);
1014 }
1015
1016 static void static_dev_create(struct udev *udev)
1017 {
1018         DIR *dir;
1019
1020         dir = opendir(udev_get_dev_path(udev));
1021         if (dir == NULL)
1022                 return;
1023
1024         static_dev_create_links(udev, dir);
1025         static_dev_create_from_devices(udev, dir);
1026
1027         closedir(dir);
1028 }
1029
1030 static int mem_size_mb(void)
1031 {
1032         FILE *f;
1033         char buf[4096];
1034         long int memsize = -1;
1035
1036         f = fopen("/proc/meminfo", "r");
1037         if (f == NULL)
1038                 return -1;
1039
1040         while (fgets(buf, sizeof(buf), f) != NULL) {
1041                 long int value;
1042
1043                 if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) {
1044                         memsize = value / 1024;
1045                         break;
1046                 }
1047         }
1048
1049         fclose(f);
1050         return memsize;
1051 }
1052
1053 static int convert_db(struct udev *udev)
1054 {
1055         char filename[UTIL_PATH_SIZE];
1056         FILE *f;
1057         struct udev_enumerate *udev_enumerate;
1058         struct udev_list_entry *list_entry;
1059
1060         /* current database */
1061         util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/data", NULL);
1062         if (access(filename, F_OK) >= 0)
1063                 return 0;
1064
1065         /* make sure we do not get here again */
1066         util_create_path(udev, filename);
1067         mkdir(filename, 0755);
1068
1069         /* old database */
1070         util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/db", NULL);
1071         if (access(filename, F_OK) < 0)
1072                 return 0;
1073
1074         f = fopen("/dev/kmsg", "w");
1075         if (f != NULL) {
1076                 fprintf(f, "<30>udevd[%u]: converting old udev database\n", getpid());
1077                 fclose(f);
1078         }
1079
1080         udev_enumerate = udev_enumerate_new(udev);
1081         if (udev_enumerate == NULL)
1082                 return -1;
1083         udev_enumerate_scan_devices(udev_enumerate);
1084         udev_list_entry_foreach(list_entry, udev_enumerate_get_list_entry(udev_enumerate)) {
1085                 struct udev_device *device;
1086
1087                 device = udev_device_new_from_syspath(udev, udev_list_entry_get_name(list_entry));
1088                 if (device == NULL)
1089                         continue;
1090
1091                 /* try to find the old database for devices without a current one */
1092                 if (udev_device_read_db(device, NULL) < 0) {
1093                         bool have_db;
1094                         const char *id;
1095                         struct stat stats;
1096                         char devpath[UTIL_PATH_SIZE];
1097                         char from[UTIL_PATH_SIZE];
1098
1099                         have_db = false;
1100
1101                         /* find database in old location */
1102                         id = udev_device_get_id_filename(device);
1103                         util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", id, NULL);
1104                         if (lstat(from, &stats) == 0) {
1105                                 if (!have_db) {
1106                                         udev_device_read_db(device, from);
1107                                         have_db = true;
1108                                 }
1109                                 unlink(from);
1110                         }
1111
1112                         /* find old database with $subsys:$sysname name */
1113                         util_strscpyl(from, sizeof(from), udev_get_dev_path(udev),
1114                                      "/.udev/db/", udev_device_get_subsystem(device), ":",
1115                                      udev_device_get_sysname(device), NULL);
1116                         if (lstat(from, &stats) == 0) {
1117                                 if (!have_db) {
1118                                         udev_device_read_db(device, from);
1119                                         have_db = true;
1120                                 }
1121                                 unlink(from);
1122                         }
1123
1124                         /* find old database with the encoded devpath name */
1125                         util_path_encode(udev_device_get_devpath(device), devpath, sizeof(devpath));
1126                         util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", devpath, NULL);
1127                         if (lstat(from, &stats) == 0) {
1128                                 if (!have_db) {
1129                                         udev_device_read_db(device, from);
1130                                         have_db = true;
1131                                 }
1132                                 unlink(from);
1133                         }
1134
1135                         /* write out new database */
1136                         if (have_db)
1137                                 udev_device_update_db(device);
1138                 }
1139                 udev_device_unref(device);
1140         }
1141         udev_enumerate_unref(udev_enumerate);
1142         return 0;
1143 }
1144
1145 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
1146 {
1147         int ctrl = -1, netlink = -1;
1148         int fd, n;
1149
1150         n = sd_listen_fds(true);
1151         if (n <= 0)
1152                 return -1;
1153
1154         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
1155                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
1156                         if (ctrl >= 0)
1157                                 return -1;
1158                         ctrl = fd;
1159                         continue;
1160                 }
1161
1162                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
1163                         if (netlink >= 0)
1164                                 return -1;
1165                         netlink = fd;
1166                         continue;
1167                 }
1168
1169                 return -1;
1170         }
1171
1172         if (ctrl < 0 || netlink < 0)
1173                 return -1;
1174
1175         info(udev, "ctrl=%i netlink=%i\n", ctrl, netlink);
1176         *rctrl = ctrl;
1177         *rnetlink = netlink;
1178         return 0;
1179 }
1180
1181 int main(int argc, char *argv[])
1182 {
1183         struct udev *udev;
1184         FILE *f;
1185         sigset_t mask;
1186         int daemonize = false;
1187         int resolve_names = 1;
1188         static const struct option options[] = {
1189                 { "daemon", no_argument, NULL, 'd' },
1190                 { "debug", no_argument, NULL, 'D' },
1191                 { "children-max", required_argument, NULL, 'c' },
1192                 { "exec-delay", required_argument, NULL, 'e' },
1193                 { "resolve-names", required_argument, NULL, 'N' },
1194                 { "help", no_argument, NULL, 'h' },
1195                 { "version", no_argument, NULL, 'V' },
1196                 {}
1197         };
1198         int fd_ctrl = -1;
1199         int fd_netlink = -1;
1200         int fd_worker = -1;
1201         struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
1202         struct udev_ctrl_connection *ctrl_conn = NULL;
1203         int rc = 1;
1204
1205         udev = udev_new();
1206         if (udev == NULL)
1207                 goto exit;
1208
1209         udev_log_init("udevd");
1210         udev_set_log_fn(udev, log_fn);
1211         info(udev, "version %s\n", VERSION);
1212         udev_selinux_init(udev);
1213
1214         /* make sure, that our runtime dir exists and is writable */
1215         if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) < 0) {
1216                 /* try to create our own subdirectory, do not create parent directories */
1217                 mkdir(udev_get_run_config_path(udev), 0755);
1218
1219                 if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) >= 0) {
1220                         /* directory seems writable now */
1221                         udev_set_run_path(udev, udev_get_run_config_path(udev));
1222                 } else {
1223                         /* fall back to /dev/.udev */
1224                         char filename[UTIL_PATH_SIZE];
1225
1226                         util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev", NULL);
1227                         if (udev_set_run_path(udev, filename) == NULL)
1228                                 goto exit;
1229                         mkdir(udev_get_run_path(udev), 0755);
1230                         err(udev, "error: runtime directory '%s' not writable, for now falling back to '%s'",
1231                             udev_get_run_config_path(udev), udev_get_run_path(udev));
1232                 }
1233         }
1234         /* relabel runtime dir only if it resides below /dev */
1235         if (strncmp(udev_get_run_path(udev), udev_get_dev_path(udev), strlen(udev_get_dev_path(udev))) == 0)
1236                 udev_selinux_lsetfilecon(udev, udev_get_run_path(udev), 0755);
1237         info(udev, "runtime dir '%s'\n", udev_get_run_path(udev));
1238
1239         for (;;) {
1240                 int option;
1241
1242                 option = getopt_long(argc, argv, "c:deDtN:hV", options, NULL);
1243                 if (option == -1)
1244                         break;
1245
1246                 switch (option) {
1247                 case 'd':
1248                         daemonize = true;
1249                         break;
1250                 case 'c':
1251                         children_max = strtoul(optarg, NULL, 0);
1252                         break;
1253                 case 'e':
1254                         exec_delay = strtoul(optarg, NULL, 0);
1255                         break;
1256                 case 'D':
1257                         debug = true;
1258                         if (udev_get_log_priority(udev) < LOG_INFO)
1259                                 udev_set_log_priority(udev, LOG_INFO);
1260                         break;
1261                 case 'N':
1262                         if (strcmp (optarg, "early") == 0) {
1263                                 resolve_names = 1;
1264                         } else if (strcmp (optarg, "late") == 0) {
1265                                 resolve_names = 0;
1266                         } else if (strcmp (optarg, "never") == 0) {
1267                                 resolve_names = -1;
1268                         } else {
1269                                 fprintf(stderr, "resolve-names must be early, late or never\n");
1270                                 err(udev, "resolve-names must be early, late or never\n");
1271                                 goto exit;
1272                         }
1273                         break;
1274                 case 'h':
1275                         printf("Usage: udevd OPTIONS\n"
1276                                "  --daemon\n"
1277                                "  --debug\n"
1278                                "  --children-max=<maximum number of workers>\n"
1279                                "  --exec-delay=<seconds to wait before executing RUN=>\n"
1280                                "  --resolve-names=early|late|never\n" 
1281                                "  --version\n"
1282                                "  --help\n"
1283                                "\n");
1284                         goto exit;
1285                 case 'V':
1286                         printf("%s\n", VERSION);
1287                         goto exit;
1288                 default:
1289                         goto exit;
1290                 }
1291         }
1292
1293         /*
1294          * read the kernel commandline, in case we need to get into debug mode
1295          *   udev.log-priority=<level>              syslog priority
1296          *   udev.children-max=<number of workers>  events are fully serialized if set to 1
1297          *
1298          */
1299         f = fopen("/proc/cmdline", "r");
1300         if (f != NULL) {
1301                 char cmdline[4096];
1302
1303                 if (fgets(cmdline, sizeof(cmdline), f) != NULL) {
1304                         char *pos;
1305
1306                         pos = strstr(cmdline, "udev.log-priority=");
1307                         if (pos != NULL) {
1308                                 pos += strlen("udev.log-priority=");
1309                                 udev_set_log_priority(udev, util_log_priority(pos));
1310                         }
1311
1312                         pos = strstr(cmdline, "udev.children-max=");
1313                         if (pos != NULL) {
1314                                 pos += strlen("udev.children-max=");
1315                                 children_max = strtoul(pos, NULL, 0);
1316                         }
1317
1318                         pos = strstr(cmdline, "udev.exec-delay=");
1319                         if (pos != NULL) {
1320                                 pos += strlen("udev.exec-delay=");
1321                                 exec_delay = strtoul(pos, NULL, 0);
1322                         }
1323                 }
1324                 fclose(f);
1325         }
1326
1327         if (getuid() != 0) {
1328                 fprintf(stderr, "root privileges required\n");
1329                 err(udev, "root privileges required\n");
1330                 goto exit;
1331         }
1332
1333         /* set umask before creating any file/directory */
1334         chdir("/");
1335         umask(022);
1336
1337         /* create standard links, copy static nodes, create nodes from modules */
1338         static_dev_create(udev);
1339         static_dev_create_from_modules(udev);
1340
1341         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1342         if (daemonize) {
1343                 int fd;
1344
1345                 fd = open("/dev/null", O_RDWR);
1346                 if (fd >= 0) {
1347                         if (write(STDOUT_FILENO, 0, 0) < 0)
1348                                 dup2(fd, STDOUT_FILENO);
1349                         if (write(STDERR_FILENO, 0, 0) < 0)
1350                                 dup2(fd, STDERR_FILENO);
1351                         if (fd > STDERR_FILENO)
1352                                 close(fd);
1353                 } else {
1354                         fprintf(stderr, "cannot open /dev/null\n");
1355                         err(udev, "cannot open /dev/null\n");
1356                 }
1357         }
1358
1359         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1360                 /* get control and netlink socket from from systemd */
1361                 udev_ctrl = udev_ctrl_new_from_socket_fd(udev, UDEV_CTRL_SOCK_PATH, fd_ctrl);
1362                 if (udev_ctrl == NULL) {
1363                         err(udev, "error taking over udev control socket");
1364                         rc = 1;
1365                         goto exit;
1366                 }
1367
1368                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1369                 if (monitor == NULL) {
1370                         err(udev, "error taking over netlink socket\n");
1371                         rc = 3;
1372                         goto exit;
1373                 }
1374         } else {
1375                 /* open control and netlink socket */
1376                 udev_ctrl = udev_ctrl_new_from_socket(udev, UDEV_CTRL_SOCK_PATH);
1377                 if (udev_ctrl == NULL) {
1378                         fprintf(stderr, "error initializing udev control socket");
1379                         err(udev, "error initializing udev control socket");
1380                         rc = 1;
1381                         goto exit;
1382                 }
1383                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1384
1385                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1386                 if (monitor == NULL) {
1387                         fprintf(stderr, "error initializing netlink socket\n");
1388                         err(udev, "error initializing netlink socket\n");
1389                         rc = 3;
1390                         goto exit;
1391                 }
1392                 fd_netlink = udev_monitor_get_fd(monitor);
1393         }
1394
1395         if (udev_monitor_enable_receiving(monitor) < 0) {
1396                 fprintf(stderr, "error binding netlink socket\n");
1397                 err(udev, "error binding netlink socket\n");
1398                 rc = 3;
1399                 goto exit;
1400         }
1401
1402         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1403                 fprintf(stderr, "error binding udev control socket\n");
1404                 err(udev, "error binding udev control socket\n");
1405                 rc = 1;
1406                 goto exit;
1407         }
1408
1409         udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
1410
1411         /* create queue file before signalling 'ready', to make sure we block 'settle' */
1412         udev_queue_export = udev_queue_export_new(udev);
1413         if (udev_queue_export == NULL) {
1414                 err(udev, "error creating queue file\n");
1415                 goto exit;
1416         }
1417
1418         if (daemonize) {
1419                 pid_t pid;
1420                 int fd;
1421
1422                 pid = fork();
1423                 switch (pid) {
1424                 case 0:
1425                         break;
1426                 case -1:
1427                         err(udev, "fork of daemon failed: %m\n");
1428                         rc = 4;
1429                         goto exit;
1430                 default:
1431                         rc = EXIT_SUCCESS;
1432                         goto exit_keep_queue;
1433                 }
1434
1435                 setsid();
1436
1437                 fd = open("/proc/self/oom_score_adj", O_RDWR);
1438                 if (fd < 0) {
1439                         /* Fallback to old interface */
1440                         fd = open("/proc/self/oom_adj", O_RDWR);
1441                         if (fd < 0) {
1442                                 err(udev, "error disabling OOM: %m\n");
1443                         } else {
1444                                 /* OOM_DISABLE == -17 */
1445                                 write(fd, "-17", 3);
1446                                 close(fd);
1447                         }
1448                 } else {
1449                         write(fd, "-1000", 5);
1450                         close(fd);
1451                 }
1452         } else {
1453                 sd_notify(1, "READY=1");
1454         }
1455
1456         f = fopen("/dev/kmsg", "w");
1457         if (f != NULL) {
1458                 fprintf(f, "<30>udevd[%u]: starting version " VERSION "\n", getpid());
1459                 fclose(f);
1460         }
1461
1462         if (!debug) {
1463                 int fd;
1464
1465                 fd = open("/dev/null", O_RDWR);
1466                 if (fd >= 0) {
1467                         dup2(fd, STDIN_FILENO);
1468                         dup2(fd, STDOUT_FILENO);
1469                         dup2(fd, STDERR_FILENO);
1470                         close(fd);
1471                 }
1472         }
1473
1474         fd_inotify = udev_watch_init(udev);
1475         if (fd_inotify < 0) {
1476                 fprintf(stderr, "error initializing inotify\n");
1477                 err(udev, "error initializing inotify\n");
1478                 rc = 4;
1479                 goto exit;
1480         }
1481
1482         if (udev_get_rules_path(udev) != NULL) {
1483                 inotify_add_watch(fd_inotify, udev_get_rules_path(udev),
1484                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1485         } else {
1486                 char filename[UTIL_PATH_SIZE];
1487                 struct stat statbuf;
1488
1489                 inotify_add_watch(fd_inotify, LIBEXECDIR "/rules.d",
1490                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1491                 inotify_add_watch(fd_inotify, SYSCONFDIR "/udev/rules.d",
1492                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1493
1494                 /* watch dynamic rules directory */
1495                 util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/rules.d", NULL);
1496                 if (stat(filename, &statbuf) != 0) {
1497                         util_create_path(udev, filename);
1498                         mkdir(filename, 0755);
1499                 }
1500                 inotify_add_watch(fd_inotify, filename,
1501                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1502         }
1503         udev_watch_restore(udev);
1504
1505         /* block and listen to all signals on signalfd */
1506         sigfillset(&mask);
1507         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1508         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1509         if (fd_signal < 0) {
1510                 fprintf(stderr, "error creating signalfd\n");
1511                 err(udev, "error creating signalfd\n");
1512                 rc = 5;
1513                 goto exit;
1514         }
1515
1516         /* unnamed socket from workers to the main daemon */
1517         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1518                 fprintf(stderr, "error creating socketpair\n");
1519                 err(udev, "error creating socketpair\n");
1520                 rc = 6;
1521                 goto exit;
1522         }
1523         fd_worker = worker_watch[READ_END];
1524
1525         rules = udev_rules_new(udev, resolve_names);
1526         if (rules == NULL) {
1527                 err(udev, "error reading rules\n");
1528                 goto exit;
1529         }
1530
1531         memset(&ep_ctrl, 0, sizeof(struct epoll_event));
1532         ep_ctrl.events = EPOLLIN;
1533         ep_ctrl.data.fd = fd_ctrl;
1534
1535         memset(&ep_inotify, 0, sizeof(struct epoll_event));
1536         ep_inotify.events = EPOLLIN;
1537         ep_inotify.data.fd = fd_inotify;
1538
1539         memset(&ep_signal, 0, sizeof(struct epoll_event));
1540         ep_signal.events = EPOLLIN;
1541         ep_signal.data.fd = fd_signal;
1542
1543         memset(&ep_netlink, 0, sizeof(struct epoll_event));
1544         ep_netlink.events = EPOLLIN;
1545         ep_netlink.data.fd = fd_netlink;
1546
1547         memset(&ep_worker, 0, sizeof(struct epoll_event));
1548         ep_worker.events = EPOLLIN;
1549         ep_worker.data.fd = fd_worker;
1550
1551         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1552         if (fd_ep < 0) {
1553                 err(udev, "error creating epoll fd: %m\n");
1554                 goto exit;
1555         }
1556         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1557             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1558             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1559             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1560             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1561                 err(udev, "fail to add fds to epoll: %m\n");
1562                 goto exit;
1563         }
1564
1565         /* if needed, convert old database from earlier udev version */
1566         convert_db(udev);
1567
1568         if (children_max <= 0) {
1569                 int memsize = mem_size_mb();
1570
1571                 /* set value depending on the amount of RAM */
1572                 if (memsize > 0)
1573                         children_max = 128 + (memsize / 8);
1574                 else
1575                         children_max = 128;
1576         }
1577         info(udev, "set children_max to %u\n", children_max);
1578
1579         udev_rules_apply_static_dev_perms(rules);
1580
1581         udev_list_init(&event_list);
1582         udev_list_init(&worker_list);
1583
1584         for (;;) {
1585                 struct epoll_event ev[8];
1586                 int fdcount;
1587                 int timeout;
1588                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1589                 int i;
1590
1591                 if (udev_exit) {
1592                         /* close sources of new events and discard buffered events */
1593                         if (fd_ctrl >= 0) {
1594                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1595                                 fd_ctrl = -1;
1596                         }
1597                         if (monitor != NULL) {
1598                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1599                                 udev_monitor_unref(monitor);
1600                                 monitor = NULL;
1601                         }
1602                         if (fd_inotify >= 0) {
1603                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1604                                 close(fd_inotify);
1605                                 fd_inotify = -1;
1606                         }
1607
1608                         /* discard queued events and kill workers */
1609                         event_queue_cleanup(udev, EVENT_QUEUED);
1610                         worker_kill(udev, 0);
1611
1612                         /* exit after all has cleaned up */
1613                         if (udev_list_is_empty(&event_list) && udev_list_is_empty(&worker_list))
1614                                 break;
1615
1616                         /* timeout at exit for workers to finish */
1617                         timeout = 60 * 1000;
1618                 } else if (udev_list_is_empty(&event_list) && children > 2) {
1619                         /* set timeout to kill idle workers */
1620                         timeout = 3 * 1000;
1621                 } else {
1622                         timeout = -1;
1623                 }
1624                 fdcount = epoll_wait(fd_ep, ev, ARRAY_SIZE(ev), timeout);
1625                 if (fdcount < 0)
1626                         continue;
1627
1628                 if (fdcount == 0) {
1629                         if (udev_exit) {
1630                                 info(udev, "timeout, giving up waiting for workers to finish\n");
1631                                 break;
1632                         }
1633
1634                         /* timeout - kill idle workers */
1635                         worker_kill(udev, 2);
1636                 }
1637
1638                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1639                 for (i = 0; i < fdcount; i++) {
1640                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1641                                 is_worker = true;
1642                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1643                                 is_netlink = true;
1644                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1645                                 is_signal = true;
1646                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1647                                 is_inotify = true;
1648                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1649                                 is_ctrl = true;
1650                 }
1651
1652                 /* event has finished */
1653                 if (is_worker)
1654                         worker_returned(fd_worker);
1655
1656                 if (is_netlink) {
1657                         struct udev_device *dev;
1658
1659                         dev = udev_monitor_receive_device(monitor);
1660                         if (dev != NULL)
1661                                 if (event_queue_insert(dev) < 0)
1662                                         udev_device_unref(dev);
1663                 }
1664
1665                 /* start new events */
1666                 if (!udev_list_is_empty(&event_list) && !udev_exit && !stop_exec_queue)
1667                         event_queue_start(udev);
1668
1669                 if (is_signal) {
1670                         struct signalfd_siginfo fdsi;
1671                         ssize_t size;
1672
1673                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1674                         if (size == sizeof(struct signalfd_siginfo))
1675                                 handle_signal(udev, fdsi.ssi_signo);
1676                 }
1677
1678                 /* we are shutting down, the events below are not handled anymore */
1679                 if (udev_exit)
1680                         continue;
1681
1682                 /* device node and rules directory inotify watch */
1683                 if (is_inotify)
1684                         handle_inotify(udev);
1685
1686                 /*
1687                  * This needs to be after the inotify handling, to make sure,
1688                  * that the ping is send back after the possibly generated
1689                  * "change" events by the inotify device node watch.
1690                  *
1691                  * A single time we may receive a client connection which we need to
1692                  * keep open to block the client. It will be closed right before we
1693                  * exit.
1694                  */
1695                 if (is_ctrl)
1696                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1697
1698                 /* rules changed, set by inotify or a HUP signal */
1699                 if (reload_config) {
1700                         struct udev_rules *rules_new;
1701
1702                         worker_kill(udev, 0);
1703                         rules_new = udev_rules_new(udev, resolve_names);
1704                         if (rules_new != NULL) {
1705                                 udev_rules_unref(rules);
1706                                 rules = rules_new;
1707                         }
1708                         reload_config = 0;
1709                 }
1710         }
1711
1712         rc = EXIT_SUCCESS;
1713 exit:
1714         udev_queue_export_cleanup(udev_queue_export);
1715 exit_keep_queue:
1716         if (fd_ep >= 0)
1717                 close(fd_ep);
1718         worker_list_cleanup(udev);
1719         event_queue_cleanup(udev, EVENT_UNDEF);
1720         udev_rules_unref(rules);
1721         if (fd_signal >= 0)
1722                 close(fd_signal);
1723         if (worker_watch[READ_END] >= 0)
1724                 close(worker_watch[READ_END]);
1725         if (worker_watch[WRITE_END] >= 0)
1726                 close(worker_watch[WRITE_END]);
1727         udev_monitor_unref(monitor);
1728         udev_queue_export_unref(udev_queue_export);
1729         udev_ctrl_connection_unref(ctrl_conn);
1730         udev_ctrl_unref(udev_ctrl);
1731         udev_selinux_exit(udev);
1732         udev_unref(udev);
1733         udev_log_close();
1734         return rc;
1735 }