• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org>
3  * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
4  * Copyright (C) 2009 Canonical Ltd.
5  * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
6  *
7  * This program is free software: you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation, either version 2 of the License, or
10  * (at your option) any later version.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #include <stddef.h>
22 #include <signal.h>
23 #include <unistd.h>
24 #include <errno.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <stdbool.h>
28 #include <string.h>
29 #include <ctype.h>
30 #include <fcntl.h>
31 #include <time.h>
32 #include <getopt.h>
33 #include <dirent.h>
34 #include <sys/file.h>
35 #include <sys/time.h>
36 #include <sys/prctl.h>
37 #include <sys/socket.h>
38 #include <sys/un.h>
39 #include <sys/signalfd.h>
40 #include <sys/epoll.h>
41 #include <sys/mount.h>
42 #include <poll.h>
43 #include <sys/wait.h>
44 #include <sys/stat.h>
45 #include <sys/ioctl.h>
46 #include <sys/inotify.h>
47 #include <sys/utsname.h>
48 #include <sys/sysmacros.h>
49 
50 #include "udev.h"
51 #include "udev-util.h"
52 #include "def.h"
53 #include "dev-setup.h"
54 #include "fileio.h"
55 #include "hashmap.h"
56 
57 static struct udev_rules *rules;
58 static struct udev_ctrl *udev_ctrl;
59 static struct udev_ctrl_connection *udev_ctrl_conn;
60 static struct udev_monitor *monitor;
61 static int worker_watch[2] = { -1, -1 };
62 static int fd_signal = -1;
63 static int fd_ep = -1;
64 static int fd_inotify = -1;
65 static bool stop_exec_queue;
66 static bool reload;
67 static bool arg_debug = false;
68 static int arg_daemonize = false;
69 static int arg_resolve_names = 1;
70 static unsigned arg_children_max;
71 static int arg_exec_delay;
72 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
73 static usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
74 static sigset_t sigmask_orig;
75 static UDEV_LIST(event_list);
76 Hashmap *workers;
77 static struct udev_list properties_list;
78 static bool udev_exit;
79 
80 enum event_state {
81         EVENT_UNDEF,
82         EVENT_QUEUED,
83         EVENT_RUNNING,
84 };
85 
86 struct event {
87         struct udev_list_node node;
88         struct udev *udev;
89         struct udev_device *dev;
90         struct udev_device *dev_kernel;
91         struct worker *worker;
92         enum event_state state;
93         unsigned long long int delaying_seqnum;
94         unsigned long long int seqnum;
95         const char *devpath;
96         size_t devpath_len;
97         const char *devpath_old;
98         dev_t devnum;
99         int ifindex;
100         bool is_block;
101         usec_t start_usec;
102         bool warned;
103 };
104 
node_to_event(struct udev_list_node * node)105 static inline struct event *node_to_event(struct udev_list_node *node) {
106         return container_of(node, struct event, node);
107 }
108 
109 static void event_queue_cleanup(struct udev *udev, enum event_state type);
110 
111 enum worker_state {
112         WORKER_UNDEF,
113         WORKER_RUNNING,
114         WORKER_IDLE,
115         WORKER_KILLED,
116 };
117 
118 struct worker {
119         struct udev_list_node node;
120         struct udev *udev;
121         int refcount;
122         pid_t pid;
123         struct udev_monitor *monitor;
124         enum worker_state state;
125         struct event *event;
126 };
127 
128 /* passed from worker to main process */
129 struct worker_message {
130 };
131 
event_free(struct event * event)132 static void event_free(struct event *event) {
133         if (!event)
134                 return;
135 
136         udev_list_node_remove(&event->node);
137         udev_device_unref(event->dev);
138         udev_device_unref(event->dev_kernel);
139 
140         if (event->worker)
141                 event->worker->event = NULL;
142 
143         free(event);
144 }
145 
worker_free(struct worker * worker)146 static void worker_free(struct worker *worker) {
147         if (!worker)
148                 return;
149 
150         hashmap_remove(workers, UINT_TO_PTR(worker->pid));
151         udev_monitor_unref(worker->monitor);
152         udev_unref(worker->udev);
153         event_free(worker->event);
154 
155         free(worker);
156 }
157 
workers_free(void)158 static void workers_free(void) {
159         struct worker *worker;
160         Iterator i;
161 
162         HASHMAP_FOREACH(worker, workers, i)
163                 worker_free(worker);
164 
165         hashmap_free(workers);
166         workers = NULL;
167 }
168 
worker_new(struct worker ** ret,struct udev * udev,struct udev_monitor * worker_monitor,pid_t pid)169 static int worker_new(struct worker **ret, struct udev *udev, struct udev_monitor *worker_monitor, pid_t pid) {
170         _cleanup_free_ struct worker *worker = NULL;
171         int r;
172 
173         assert(ret);
174         assert(udev);
175         assert(worker_monitor);
176         assert(pid > 1);
177 
178         worker = new0(struct worker, 1);
179         if (!worker)
180                 return -ENOMEM;
181 
182         worker->refcount = 1;
183         worker->udev = udev_ref(udev);
184         /* close monitor, but keep address around */
185         udev_monitor_disconnect(worker_monitor);
186         worker->monitor = udev_monitor_ref(worker_monitor);
187         worker->pid = pid;
188 
189         r = hashmap_ensure_allocated(&workers, NULL);
190         if (r < 0)
191                 return r;
192 
193         r = hashmap_put(workers, UINT_TO_PTR(pid), worker);
194         if (r < 0)
195                 return r;
196 
197         *ret = worker;
198         worker = NULL;
199 
200         return 0;
201 }
202 
worker_attach_event(struct worker * worker,struct event * event)203 static void worker_attach_event(struct worker *worker, struct event *event) {
204         assert(worker);
205         assert(event);
206         assert(!event->worker);
207         assert(!worker->event);
208 
209         worker->state = WORKER_RUNNING;
210         worker->event = event;
211         event->state = EVENT_RUNNING;
212         event->start_usec = now(CLOCK_MONOTONIC);
213         event->warned = false;
214         event->worker = worker;
215 }
216 
worker_send_message(int fd)217 static int worker_send_message(int fd) {
218         struct worker_message message = {};
219 
220         return loop_write(fd, &message, sizeof(message), false);
221 }
222 
worker_spawn(struct event * event)223 static void worker_spawn(struct event *event) {
224         struct udev *udev = event->udev;
225         _cleanup_udev_monitor_unref_ struct udev_monitor *worker_monitor = NULL;
226         pid_t pid;
227 
228         /* listen for new events */
229         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
230         if (worker_monitor == NULL)
231                 return;
232         /* allow the main daemon netlink address to send devices to the worker */
233         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
234         udev_monitor_enable_receiving(worker_monitor);
235 
236         pid = fork();
237         switch (pid) {
238         case 0: {
239                 struct udev_device *dev = NULL;
240                 int fd_monitor;
241                 struct epoll_event ep_signal, ep_monitor;
242                 sigset_t mask;
243                 int r = 0;
244 
245                 /* take initial device from queue */
246                 dev = event->dev;
247                 event->dev = NULL;
248 
249                 workers_free();
250                 event_queue_cleanup(udev, EVENT_UNDEF);
251                 udev_monitor_unref(monitor);
252                 udev_ctrl_unref(udev_ctrl);
253                 close(fd_signal);
254                 close(fd_ep);
255                 close(worker_watch[READ_END]);
256 
257                 sigfillset(&mask);
258                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
259                 if (fd_signal < 0) {
260                         r = log_error_errno(errno, "error creating signalfd %m");
261                         goto out;
262                 }
263 
264                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
265                 if (fd_ep < 0) {
266                         r = log_error_errno(errno, "error creating epoll fd: %m");
267                         goto out;
268                 }
269 
270                 memzero(&ep_signal, sizeof(struct epoll_event));
271                 ep_signal.events = EPOLLIN;
272                 ep_signal.data.fd = fd_signal;
273 
274                 fd_monitor = udev_monitor_get_fd(worker_monitor);
275                 memzero(&ep_monitor, sizeof(struct epoll_event));
276                 ep_monitor.events = EPOLLIN;
277                 ep_monitor.data.fd = fd_monitor;
278 
279                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
280                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
281                         r = log_error_errno(errno, "fail to add fds to epoll: %m");
282                         goto out;
283                 }
284 
285                 /* request TERM signal if parent exits */
286                 prctl(PR_SET_PDEATHSIG, SIGTERM);
287 
288                 /* reset OOM score, we only protect the main daemon */
289                 write_string_file("/proc/self/oom_score_adj", "0");
290 
291                 for (;;) {
292                         struct udev_event *udev_event;
293                         int fd_lock = -1;
294 
295                         log_debug("seq %llu running", udev_device_get_seqnum(dev));
296                         udev_event = udev_event_new(dev);
297                         if (udev_event == NULL) {
298                                 r = -ENOMEM;
299                                 goto out;
300                         }
301 
302                         /* needed for SIGCHLD/SIGTERM in spawn() */
303                         udev_event->fd_signal = fd_signal;
304 
305                         if (arg_exec_delay > 0)
306                                 udev_event->exec_delay = arg_exec_delay;
307 
308                         /*
309                          * Take a shared lock on the device node; this establishes
310                          * a concept of device "ownership" to serialize device
311                          * access. External processes holding an exclusive lock will
312                          * cause udev to skip the event handling; in the case udev
313                          * acquired the lock, the external process can block until
314                          * udev has finished its event handling.
315                          */
316                         if (!streq_ptr(udev_device_get_action(dev), "remove") &&
317                             streq_ptr("block", udev_device_get_subsystem(dev)) &&
318                             !startswith(udev_device_get_sysname(dev), "dm-") &&
319                             !startswith(udev_device_get_sysname(dev), "md")) {
320                                 struct udev_device *d = dev;
321 
322                                 if (streq_ptr("partition", udev_device_get_devtype(d)))
323                                         d = udev_device_get_parent(d);
324 
325                                 if (d) {
326                                         fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
327                                         if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
328                                                 log_debug_errno(errno, "Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
329                                                 fd_lock = safe_close(fd_lock);
330                                                 r = -EAGAIN;
331                                                 goto skip;
332                                         }
333                                 }
334                         }
335 
336                         /* apply rules, create node, symlinks */
337                         udev_event_execute_rules(udev_event,
338                                                  arg_event_timeout_usec, arg_event_timeout_warn_usec,
339                                                  &properties_list,
340                                                  rules,
341                                                  &sigmask_orig);
342 
343                         udev_event_execute_run(udev_event,
344                                                arg_event_timeout_usec, arg_event_timeout_warn_usec,
345                                                &sigmask_orig);
346 
347                         /* apply/restore inotify watch */
348                         if (udev_event->inotify_watch) {
349                                 udev_watch_begin(udev, dev);
350                                 udev_device_update_db(dev);
351                         }
352 
353                         safe_close(fd_lock);
354 
355                         /* send processed event back to libudev listeners */
356                         udev_monitor_send_device(worker_monitor, NULL, dev);
357 
358 skip:
359                         log_debug("seq %llu processed", udev_device_get_seqnum(dev));
360 
361                         /* send udevd the result of the event execution */
362                         r = worker_send_message(worker_watch[WRITE_END]);
363                         if (r < 0)
364                                 log_error_errno(r, "failed to send result of seq %llu to main daemon: %m",
365                                                 udev_device_get_seqnum(dev));
366 
367                         udev_device_unref(dev);
368                         dev = NULL;
369 
370                         if (udev_event->sigterm) {
371                                 udev_event_unref(udev_event);
372                                 goto out;
373                         }
374 
375                         udev_event_unref(udev_event);
376 
377                         /* wait for more device messages from main udevd, or term signal */
378                         while (dev == NULL) {
379                                 struct epoll_event ev[4];
380                                 int fdcount;
381                                 int i;
382 
383                                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
384                                 if (fdcount < 0) {
385                                         if (errno == EINTR)
386                                                 continue;
387                                         r = log_error_errno(errno, "failed to poll: %m");
388                                         goto out;
389                                 }
390 
391                                 for (i = 0; i < fdcount; i++) {
392                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
393                                                 dev = udev_monitor_receive_device(worker_monitor);
394                                                 break;
395                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
396                                                 struct signalfd_siginfo fdsi;
397                                                 ssize_t size;
398 
399                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
400                                                 if (size != sizeof(struct signalfd_siginfo))
401                                                         continue;
402                                                 switch (fdsi.ssi_signo) {
403                                                 case SIGTERM:
404                                                         goto out;
405                                                 }
406                                         }
407                                 }
408                         }
409                 }
410 out:
411                 udev_device_unref(dev);
412                 safe_close(fd_signal);
413                 safe_close(fd_ep);
414                 close(fd_inotify);
415                 close(worker_watch[WRITE_END]);
416                 udev_rules_unref(rules);
417                 udev_builtin_exit(udev);
418                 udev_unref(udev);
419                 log_close();
420                 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
421         }
422         case -1:
423                 event->state = EVENT_QUEUED;
424                 log_error_errno(errno, "fork of child failed: %m");
425                 break;
426         default:
427         {
428                 struct worker *worker;
429                 int r;
430 
431                 r = worker_new(&worker, udev, worker_monitor, pid);
432                 if (r < 0)
433                         return;
434 
435                 worker_attach_event(worker, event);
436 
437                 log_debug("seq %llu forked new worker ["PID_FMT"]", udev_device_get_seqnum(event->dev), pid);
438                 break;
439         }
440         }
441 }
442 
event_run(struct event * event)443 static void event_run(struct event *event) {
444         struct worker *worker;
445         Iterator i;
446 
447         HASHMAP_FOREACH(worker, workers, i) {
448                 ssize_t count;
449 
450                 if (worker->state != WORKER_IDLE)
451                         continue;
452 
453                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
454                 if (count < 0) {
455                         log_error_errno(errno, "worker ["PID_FMT"] did not accept message %zi (%m), kill it",
456                                         worker->pid, count);
457                         kill(worker->pid, SIGKILL);
458                         worker->state = WORKER_KILLED;
459                         continue;
460                 }
461                 worker_attach_event(worker, event);
462                 return;
463         }
464 
465         if (hashmap_size(workers) >= arg_children_max) {
466                 if (arg_children_max > 1)
467                         log_debug("maximum number (%i) of children reached", hashmap_size(workers));
468                 return;
469         }
470 
471         /* start new worker and pass initial device */
472         worker_spawn(event);
473 }
474 
event_queue_insert(struct udev_device * dev)475 static int event_queue_insert(struct udev_device *dev) {
476         struct event *event;
477 
478         event = new0(struct event, 1);
479         if (event == NULL)
480                 return -1;
481 
482         event->udev = udev_device_get_udev(dev);
483         event->dev = dev;
484         event->dev_kernel = udev_device_shallow_clone(dev);
485         udev_device_copy_properties(event->dev_kernel, dev);
486         event->seqnum = udev_device_get_seqnum(dev);
487         event->devpath = udev_device_get_devpath(dev);
488         event->devpath_len = strlen(event->devpath);
489         event->devpath_old = udev_device_get_devpath_old(dev);
490         event->devnum = udev_device_get_devnum(dev);
491         event->is_block = streq("block", udev_device_get_subsystem(dev));
492         event->ifindex = udev_device_get_ifindex(dev);
493 
494         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
495              udev_device_get_action(dev), udev_device_get_subsystem(dev));
496 
497         event->state = EVENT_QUEUED;
498         udev_list_node_append(&event->node, &event_list);
499         return 0;
500 }
501 
worker_kill(void)502 static void worker_kill(void) {
503         struct worker *worker;
504         Iterator i;
505 
506         HASHMAP_FOREACH(worker, workers, i) {
507                 if (worker->state == WORKER_KILLED)
508                         continue;
509 
510                 worker->state = WORKER_KILLED;
511                 kill(worker->pid, SIGTERM);
512         }
513 }
514 
515 /* lookup event for identical, parent, child device */
is_devpath_busy(struct event * event)516 static bool is_devpath_busy(struct event *event) {
517         struct udev_list_node *loop;
518         size_t common;
519 
520         /* check if queue contains events we depend on */
521         udev_list_node_foreach(loop, &event_list) {
522                 struct event *loop_event = node_to_event(loop);
523 
524                 /* we already found a later event, earlier can not block us, no need to check again */
525                 if (loop_event->seqnum < event->delaying_seqnum)
526                         continue;
527 
528                 /* event we checked earlier still exists, no need to check again */
529                 if (loop_event->seqnum == event->delaying_seqnum)
530                         return true;
531 
532                 /* found ourself, no later event can block us */
533                 if (loop_event->seqnum >= event->seqnum)
534                         break;
535 
536                 /* check major/minor */
537                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
538                         return true;
539 
540                 /* check network device ifindex */
541                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
542                         return true;
543 
544                 /* check our old name */
545                 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
546                         event->delaying_seqnum = loop_event->seqnum;
547                         return true;
548                 }
549 
550                 /* compare devpath */
551                 common = MIN(loop_event->devpath_len, event->devpath_len);
552 
553                 /* one devpath is contained in the other? */
554                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
555                         continue;
556 
557                 /* identical device event found */
558                 if (loop_event->devpath_len == event->devpath_len) {
559                         /* devices names might have changed/swapped in the meantime */
560                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
561                                 continue;
562                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
563                                 continue;
564                         event->delaying_seqnum = loop_event->seqnum;
565                         return true;
566                 }
567 
568                 /* parent device event found */
569                 if (event->devpath[common] == '/') {
570                         event->delaying_seqnum = loop_event->seqnum;
571                         return true;
572                 }
573 
574                 /* child device event found */
575                 if (loop_event->devpath[common] == '/') {
576                         event->delaying_seqnum = loop_event->seqnum;
577                         return true;
578                 }
579 
580                 /* no matching device */
581                 continue;
582         }
583 
584         return false;
585 }
586 
event_queue_start(struct udev * udev)587 static void event_queue_start(struct udev *udev) {
588         struct udev_list_node *loop;
589 
590         udev_list_node_foreach(loop, &event_list) {
591                 struct event *event = node_to_event(loop);
592 
593                 if (event->state != EVENT_QUEUED)
594                         continue;
595 
596                 /* do not start event if parent or child event is still running */
597                 if (is_devpath_busy(event))
598                         continue;
599 
600                 event_run(event);
601         }
602 }
603 
event_queue_cleanup(struct udev * udev,enum event_state match_type)604 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
605         struct udev_list_node *loop, *tmp;
606 
607         udev_list_node_foreach_safe(loop, tmp, &event_list) {
608                 struct event *event = node_to_event(loop);
609 
610                 if (match_type != EVENT_UNDEF && match_type != event->state)
611                         continue;
612 
613                 event_free(event);
614         }
615 }
616 
worker_returned(int fd_worker)617 static void worker_returned(int fd_worker) {
618         for (;;) {
619                 struct worker_message msg;
620                 struct iovec iovec;
621                 union {
622                         struct cmsghdr cmsghdr;
623                         uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
624                 } control = {};
625                 struct msghdr msghdr;
626                 struct cmsghdr *cmsg;
627                 ssize_t size;
628                 struct ucred *ucred = NULL;
629                 struct worker *worker;
630 
631                 memzero(&iovec, sizeof(struct iovec));
632                 iovec.iov_base = &msg;
633                 iovec.iov_len = sizeof(msg);
634 
635                 memzero(&msghdr, sizeof(struct msghdr));
636                 msghdr.msg_iov = &iovec;
637                 msghdr.msg_iovlen = 1;
638                 msghdr.msg_control = &control;
639                 msghdr.msg_controllen = sizeof(control);
640 
641                 size = recvmsg(fd_worker, &msghdr, MSG_DONTWAIT);
642                 if (size < 0) {
643                         if (errno == EINTR)
644                                 continue;
645                         else if (errno == EAGAIN)
646                                 /* nothing more to read */
647                                 break;
648 
649                         log_error_errno(errno, "failed to receive message: %m");
650                         return;
651                 } else if (size != sizeof(struct worker_message)) {
652                         log_warning_errno(EIO, "ignoring worker message with invalid size %zi bytes", size);
653                         return;
654                 }
655 
656                 for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
657                         if (cmsg->cmsg_level == SOL_SOCKET &&
658                             cmsg->cmsg_type == SCM_CREDENTIALS &&
659                             cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
660                                 ucred = (struct ucred*) CMSG_DATA(cmsg);
661                 }
662 
663                 if (!ucred || ucred->pid <= 0) {
664                         log_warning_errno(EIO, "ignoring worker message without valid PID");
665                         continue;
666                 }
667 
668                 /* lookup worker who sent the signal */
669                 worker = hashmap_get(workers, UINT_TO_PTR(ucred->pid));
670                 if (!worker) {
671                         log_debug("worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid);
672                         continue;
673                 }
674 
675                 if (worker->state != WORKER_KILLED)
676                         worker->state = WORKER_IDLE;
677 
678                 /* worker returned */
679                 event_free(worker->event);
680         }
681 }
682 
event_queue_update(void)683 static void event_queue_update(void) {
684         int r;
685 
686         if (!udev_list_node_is_empty(&event_list)) {
687                 r = touch("/data/udev/queue");
688                 if (r < 0)
689                         log_warning_errno(r, "could not touch /run/udev/queue: %m");
690         } else {
691                 r = unlink("/data/udev/queue");
692                 if (r < 0 && errno != ENOENT)
693                         log_warning("could not unlink /run/udev/queue: %m");
694         }
695 }
696 
697 /* receive the udevd message from userspace */
handle_ctrl_msg(struct udev_ctrl * uctrl)698 static void handle_ctrl_msg(struct udev_ctrl *uctrl) {
699         _cleanup_udev_ctrl_connection_unref_ struct udev_ctrl_connection *ctrl_conn = NULL;
700         _cleanup_udev_ctrl_msg_unref_ struct udev_ctrl_msg *ctrl_msg = NULL;
701         const char *str;
702         int i;
703 
704         assert(uctrl);
705 
706         ctrl_conn = udev_ctrl_get_connection(uctrl);
707         if (!ctrl_conn)
708                 return;
709 
710         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
711         if (!ctrl_msg)
712                 return;
713 
714         i = udev_ctrl_get_set_log_level(ctrl_msg);
715         if (i >= 0) {
716                 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
717                 log_set_max_level(i);
718                 worker_kill();
719         }
720 
721         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
722                 log_debug("udevd message (STOP_EXEC_QUEUE) received");
723                 stop_exec_queue = true;
724         }
725 
726         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
727                 log_debug("udevd message (START_EXEC_QUEUE) received");
728                 stop_exec_queue = false;
729         }
730 
731         if (udev_ctrl_get_reload(ctrl_msg) > 0) {
732                 log_debug("udevd message (RELOAD) received");
733                 reload = true;
734         }
735 
736         str = udev_ctrl_get_set_env(ctrl_msg);
737         if (str != NULL) {
738                 char *key;
739 
740                 key = strdup(str);
741                 if (key != NULL) {
742                         char *val;
743 
744                         val = strchr(key, '=');
745                         if (val != NULL) {
746                                 val[0] = '\0';
747                                 val = &val[1];
748                                 if (val[0] == '\0') {
749                                         log_debug("udevd message (ENV) received, unset '%s'", key);
750                                         udev_list_entry_add(&properties_list, key, NULL);
751                                 } else {
752                                         log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
753                                         udev_list_entry_add(&properties_list, key, val);
754                                 }
755                         } else {
756                                 log_error("wrong key format '%s'", key);
757                         }
758                         free(key);
759                 }
760                 worker_kill();
761         }
762 
763         i = udev_ctrl_get_set_children_max(ctrl_msg);
764         if (i >= 0) {
765                 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
766                 arg_children_max = i;
767         }
768 
769         if (udev_ctrl_get_ping(ctrl_msg) > 0) {
770                 log_debug("udevd message (SYNC) received");
771                 /* tell settle that we are busy or idle, this needs to be before the
772                  * PING handling
773                  */
774                 event_queue_update();
775         }
776 
777         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
778                 log_debug("udevd message (EXIT) received");
779                 udev_exit = true;
780                 /* keep reference to block the client until we exit */
781                 udev_ctrl_conn = udev_ctrl_connection_ref(ctrl_conn);
782         }
783 
784         return;
785 }
786 
synthesize_change(struct udev_device * dev)787 static int synthesize_change(struct udev_device *dev) {
788         char filename[UTIL_PATH_SIZE];
789         int r;
790 
791         if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
792             streq_ptr("disk", udev_device_get_devtype(dev)) &&
793             !startswith(udev_device_get_sysname(dev), "dm-")) {
794                 bool part_table_read = false;
795                 bool has_partitions = false;
796                 int fd;
797                 struct udev *udev = udev_device_get_udev(dev);
798                 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
799                 struct udev_list_entry *item;
800 
801                 /*
802                  * Try to re-read the partition table. This only succeeds if
803                  * none of the devices is busy. The kernel returns 0 if no
804                  * partition table is found, and we will not get an event for
805                  * the disk.
806                  */
807                 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
808                 if (fd >= 0) {
809                         r = flock(fd, LOCK_EX|LOCK_NB);
810                         if (r >= 0)
811                                 r = ioctl(fd, BLKRRPART, 0);
812 
813                         close(fd);
814                         if (r >= 0)
815                                 part_table_read = true;
816                 }
817 
818                 /* search for partitions */
819                 e = udev_enumerate_new(udev);
820                 if (!e)
821                         return -ENOMEM;
822 
823                 r = udev_enumerate_add_match_parent(e, dev);
824                 if (r < 0)
825                         return r;
826 
827                 r = udev_enumerate_add_match_subsystem(e, "block");
828                 if (r < 0)
829                         return r;
830 
831                 r = udev_enumerate_scan_devices(e);
832                 if (r < 0)
833                         return r;
834 
835                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
836                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
837 
838                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
839                         if (!d)
840                                 continue;
841 
842                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
843                                 continue;
844 
845                         has_partitions = true;
846                         break;
847                 }
848 
849                 /*
850                  * We have partitions and re-read the table, the kernel already sent
851                  * out a "change" event for the disk, and "remove/add" for all
852                  * partitions.
853                  */
854                 if (part_table_read && has_partitions)
855                         return 0;
856 
857                 /*
858                  * We have partitions but re-reading the partition table did not
859                  * work, synthesize "change" for the disk and all partitions.
860                  */
861                 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
862                 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
863                 write_string_file(filename, "change");
864 
865                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
866                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
867 
868                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
869                         if (!d)
870                                 continue;
871 
872                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
873                                 continue;
874 
875                         log_debug("device %s closed, synthesising partition '%s' 'change'",
876                                   udev_device_get_devnode(dev), udev_device_get_devnode(d));
877                         strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
878                         write_string_file(filename, "change");
879                 }
880 
881                 return 0;
882         }
883 
884         log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
885         strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
886         write_string_file(filename, "change");
887 
888         return 0;
889 }
890 
handle_inotify(struct udev * udev)891 static int handle_inotify(struct udev *udev) {
892         union inotify_event_buffer buffer;
893         struct inotify_event *e;
894         ssize_t l;
895 
896         l = read(fd_inotify, &buffer, sizeof(buffer));
897         if (l < 0) {
898                 if (errno == EAGAIN || errno == EINTR)
899                         return 0;
900 
901                 return log_error_errno(errno, "Failed to read inotify fd: %m");
902         }
903 
904         FOREACH_INOTIFY_EVENT(e, buffer, l) {
905                 struct udev_device *dev;
906 
907                 dev = udev_watch_lookup(udev, e->wd);
908                 if (!dev)
909                         continue;
910 
911                 log_debug("inotify event: %x for %s", e->mask, udev_device_get_devnode(dev));
912                 if (e->mask & IN_CLOSE_WRITE)
913                         synthesize_change(dev);
914                 else if (e->mask & IN_IGNORED)
915                         udev_watch_end(udev, dev);
916 
917                 udev_device_unref(dev);
918         }
919 
920         return 0;
921 }
922 
handle_signal(struct udev * udev,int signo)923 static void handle_signal(struct udev *udev, int signo) {
924         switch (signo) {
925         case SIGINT:
926         case SIGTERM:
927                 udev_exit = true;
928                 break;
929         case SIGCHLD:
930                 for (;;) {
931                         pid_t pid;
932                         int status;
933                         struct worker *worker;
934 
935                         pid = waitpid(-1, &status, WNOHANG);
936                         if (pid <= 0)
937                                 break;
938 
939                         worker = hashmap_get(workers, UINT_TO_PTR(pid));
940                         if (!worker) {
941                                 log_warning("worker ["PID_FMT"] is unknown, ignoring", pid);
942                                 continue;
943                         }
944 
945                         if (WIFEXITED(status)) {
946                                 if (WEXITSTATUS(status) == 0)
947                                         log_debug("worker ["PID_FMT"] exited", pid);
948                                 else
949                                         log_warning("worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status));
950                         } else if (WIFSIGNALED(status)) {
951                                 log_warning("worker ["PID_FMT"] terminated by signal %i (%s)",
952                                             pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
953                         } else if (WIFSTOPPED(status)) {
954                                 log_info("worker ["PID_FMT"] stopped", pid);
955                                 break;
956                         } else if (WIFCONTINUED(status)) {
957                                 log_info("worker ["PID_FMT"] continued", pid);
958                                 break;
959                         } else {
960                                 log_warning("worker ["PID_FMT"] exit with status 0x%04x", pid, status);
961                         }
962 
963                         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
964                                 if (worker->event) {
965                                         log_error("worker ["PID_FMT"] failed while handling '%s'", pid, worker->event->devpath);
966                                         /* delete state from disk */
967                                         udev_device_delete_db(worker->event->dev);
968                                         udev_device_tag_index(worker->event->dev, NULL, false);
969                                         /* forward kernel event without amending it */
970                                         udev_monitor_send_device(monitor, NULL, worker->event->dev_kernel);
971                                 }
972                         }
973 
974                         worker_free(worker);
975                 }
976                 break;
977         case SIGHUP:
978                 reload = true;
979                 break;
980         }
981 }
982 
983 /*
984  * read the kernel command line, in case we need to get into debug mode
985  *   udev.log-priority=<level>                 syslog priority
986  *   udev.children-max=<number of workers>     events are fully serialized if set to 1
987  *   udev.exec-delay=<number of seconds>       delay execution of every executed program
988  *   udev.event-timeout=<number of seconds>    seconds to wait before terminating an event
989  */
parse_proc_cmdline_item(const char * key,const char * value)990 static int parse_proc_cmdline_item(const char *key, const char *value) {
991         int r;
992 
993         assert(key);
994 
995         if (!value)
996                 return 0;
997 
998         if (startswith(key, "rd."))
999                 key += strlen("rd.");
1000 
1001         if (startswith(key, "udev."))
1002                 key += strlen("udev.");
1003         else
1004                 return 0;
1005 
1006         if (streq(key, "log-priority")) {
1007                 int prio;
1008 
1009                 prio = util_log_priority(value);
1010                 log_set_max_level(prio);
1011         } else if (streq(key, "children-max")) {
1012                 r = safe_atou(value, &arg_children_max);
1013                 if (r < 0)
1014                         log_warning("invalid udev.children-max ignored: %s", value);
1015         } else if (streq(key, "exec-delay")) {
1016                 r = safe_atoi(value, &arg_exec_delay);
1017                 if (r < 0)
1018                         log_warning("invalid udev.exec-delay ignored: %s", value);
1019         } else if (streq(key, "event-timeout")) {
1020                 r = safe_atou64(value, &arg_event_timeout_usec);
1021                 if (r < 0)
1022                         log_warning("invalid udev.event-timeout ignored: %s", value);
1023                 else {
1024                         arg_event_timeout_usec *= USEC_PER_SEC;
1025                         arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1026                 }
1027         }
1028 
1029         return 0;
1030 }
1031 
help(void)1032 static void help(void) {
1033         printf("%s [OPTIONS...]\n\n"
1034                "Manages devices.\n\n"
1035                "  -h --help                   Print this message\n"
1036                "  -V --version                Print version of the program\n"
1037                "  -d --daemon                 Detach and run in the background\n"
1038                "  -D --debug                  Enable debug output\n"
1039                "  -c --children-max=INT       Set maximum number of workers\n"
1040                "  -e --exec-delay=SECONDS     Seconds to wait before executing RUN=\n"
1041                "  -t --event-timeout=SECONDS  Seconds to wait before terminating an event\n"
1042                "  -N --resolve-names=early|late|never\n"
1043                "                              When to resolve users and groups\n"
1044                , program_invocation_short_name);
1045 }
1046 
parse_argv(int argc,char * argv[])1047 static int parse_argv(int argc, char *argv[]) {
1048         static const struct option options[] = {
1049                 { "daemon",             no_argument,            NULL, 'd' },
1050                 { "debug",              no_argument,            NULL, 'D' },
1051                 { "children-max",       required_argument,      NULL, 'c' },
1052                 { "exec-delay",         required_argument,      NULL, 'e' },
1053                 { "event-timeout",      required_argument,      NULL, 't' },
1054                 { "resolve-names",      required_argument,      NULL, 'N' },
1055                 { "help",               no_argument,            NULL, 'h' },
1056                 { "version",            no_argument,            NULL, 'V' },
1057                 {}
1058         };
1059 
1060         int c;
1061 
1062         assert(argc >= 0);
1063         assert(argv);
1064 
1065         while ((c = getopt_long(argc, argv, "c:de:Dt:N:hV", options, NULL)) >= 0) {
1066                 int r;
1067 
1068                 switch (c) {
1069 
1070                 case 'd':
1071                         arg_daemonize = true;
1072                         break;
1073                 case 'c':
1074                         r = safe_atou(optarg, &arg_children_max);
1075                         if (r < 0)
1076                                 log_warning("Invalid --children-max ignored: %s", optarg);
1077                         break;
1078                 case 'e':
1079                         r = safe_atoi(optarg, &arg_exec_delay);
1080                         if (r < 0)
1081                                 log_warning("Invalid --exec-delay ignored: %s", optarg);
1082                         break;
1083                 case 't':
1084                         r = safe_atou64(optarg, &arg_event_timeout_usec);
1085                         if (r < 0)
1086                                 log_warning("Invalid --event-timeout ignored: %s", optarg);
1087                         else {
1088                                 arg_event_timeout_usec *= USEC_PER_SEC;
1089                                 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1090                         }
1091                         break;
1092                 case 'D':
1093                         arg_debug = true;
1094                         break;
1095                 case 'N':
1096                         if (streq(optarg, "early")) {
1097                                 arg_resolve_names = 1;
1098                         } else if (streq(optarg, "late")) {
1099                                 arg_resolve_names = 0;
1100                         } else if (streq(optarg, "never")) {
1101                                 arg_resolve_names = -1;
1102                         } else {
1103                                 log_error("resolve-names must be early, late or never");
1104                                 return 0;
1105                         }
1106                         break;
1107                 case 'h':
1108                         help();
1109                         return 0;
1110                 case 'V':
1111                         printf("%s\n", UDEV_VERSION);
1112                         return 0;
1113                 case '?':
1114                         return -EINVAL;
1115                 default:
1116                         assert_not_reached("Unhandled option");
1117 
1118                 }
1119         }
1120 
1121         return 1;
1122 }
1123 
main(int argc,char * argv[])1124 int main(int argc, char *argv[]) {
1125         struct udev *udev;
1126         sigset_t mask;
1127         FILE *f;
1128         int fd_ctrl = -1;
1129         int fd_netlink = -1;
1130         int fd_worker = -1;
1131         struct epoll_event ep_ctrl = { .events = EPOLLIN };
1132         struct epoll_event ep_inotify = { .events = EPOLLIN };
1133         struct epoll_event ep_signal = { .events = EPOLLIN };
1134         struct epoll_event ep_netlink = { .events = EPOLLIN };
1135         struct epoll_event ep_worker = { .events = EPOLLIN };
1136         int r = 0, one = 1;
1137 
1138         udev = udev_new();
1139         if (!udev) {
1140                 r = log_error_errno(errno, "could not allocate udev context: %m");
1141                 goto exit;
1142         }
1143 
1144         log_set_target(LOG_TARGET_AUTO);
1145         log_open();
1146 
1147         r = parse_argv(argc, argv);
1148         if (r <= 0)
1149                 goto exit;
1150 
1151         r = parse_proc_cmdline(parse_proc_cmdline_item);
1152         if (r < 0)
1153                 log_warning_errno(r, "failed to parse kernel command line, ignoring: %m");
1154 
1155         if (arg_debug) {
1156 		log_set_target(LOG_TARGET_CONSOLE);
1157                 log_set_max_level(LOG_DEBUG);
1158 	}
1159 
1160         if (getuid() != 0) {
1161                 r = log_error_errno(EPERM, "root privileges required");
1162                 goto exit;
1163         }
1164 
1165         r = mac_selinux_init("/dev");
1166         if (r < 0) {
1167                 log_error_errno(r, "could not initialize labelling: %m");
1168                 goto exit;
1169         }
1170 
1171         /* set umask before creating any file/directory */
1172         r = chdir("/");
1173         if (r < 0) {
1174                 r = log_error_errno(errno, "could not change dir to /: %m");
1175                 goto exit;
1176         }
1177 
1178         umask(022);
1179 
1180         udev_list_init(udev, &properties_list, true);
1181 
1182         r = mkdir(UDEV_ROOT_RUN "/udev", 0755);
1183         if (r < 0 && errno != EEXIST) {
1184                 r = log_error_errno(errno, "could not create " UDEV_ROOT_RUN "/udev: %m");
1185                 goto exit;
1186         }
1187 
1188         dev_setup(NULL, UID_INVALID, GID_INVALID);
1189 
1190         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1191         if (arg_daemonize) {
1192                 int fd;
1193 
1194                 fd = open("/dev/null", O_RDWR);
1195                 if (fd >= 0) {
1196                         if (write(STDOUT_FILENO, 0, 0) < 0)
1197                                 dup2(fd, STDOUT_FILENO);
1198                         if (write(STDERR_FILENO, 0, 0) < 0)
1199                                 dup2(fd, STDERR_FILENO);
1200                         if (fd > STDERR_FILENO)
1201                                 close(fd);
1202                 } else {
1203                         log_error("cannot open /dev/null");
1204                 }
1205         }
1206 
1207         /* open control and netlink socket */
1208         udev_ctrl = udev_ctrl_new(udev);
1209         if (!udev_ctrl) {
1210                 r = log_error_errno(EINVAL, "error initializing udev control socket");
1211                 goto exit;
1212         }
1213         fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1214 
1215         monitor = udev_monitor_new_from_netlink(udev, "kernel");
1216         if (!monitor) {
1217                 r = log_error_errno(EINVAL, "error initializing netlink socket");
1218                 goto exit;
1219         }
1220         fd_netlink = udev_monitor_get_fd(monitor);
1221 
1222         if (udev_monitor_enable_receiving(monitor) < 0) {
1223                 r = log_error_errno(EINVAL, "error binding netlink socket");
1224                 goto exit;
1225         }
1226 
1227         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1228                 r = log_error_errno(EINVAL, "error binding udev control socket");
1229                 goto exit;
1230         }
1231 
1232         udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1233 
1234         log_info("starting version " VERSION);
1235 
1236         udev_builtin_init(udev);
1237 
1238         rules = udev_rules_new(udev, arg_resolve_names);
1239         if (!rules) {
1240                 r = log_error_errno(ENOMEM, "error reading rules");
1241                 goto exit;
1242         }
1243 
1244         r = udev_rules_apply_static_dev_perms(rules);
1245         if (r < 0)
1246                 log_error_errno(r, "failed to apply permissions on static device nodes: %m");
1247 
1248         if (arg_daemonize) {
1249                 pid_t pid;
1250 
1251                 pid = fork();
1252                 switch (pid) {
1253                 case 0:
1254                         break;
1255                 case -1:
1256                         r = log_error_errno(errno, "fork of daemon failed: %m");
1257                         goto exit;
1258                 default:
1259                         mac_selinux_finish();
1260                         log_close();
1261                         _exit(EXIT_SUCCESS);
1262                 }
1263 
1264                 setsid();
1265 
1266                 write_string_file("/proc/self/oom_score_adj", "-1000");
1267         }
1268 
1269         if (arg_children_max == 0) {
1270                 cpu_set_t cpu_set;
1271 
1272                 arg_children_max = 8;
1273 
1274                 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1275                         arg_children_max +=  CPU_COUNT(&cpu_set) * 2;
1276                 }
1277         }
1278         log_debug("set children_max to %u", arg_children_max);
1279 
1280         udev_list_node_init(&event_list);
1281 
1282         f = fopen("/dev/kmsg", "w");
1283         if (f != NULL) {
1284                 fprintf(f, "<30>udevd[%u]: starting eudev-" VERSION "\n", getpid());
1285                 fclose(f);
1286         }
1287 
1288         if (!arg_debug) {
1289                 int fd;
1290 
1291                 fd = open("/dev/null", O_RDWR);
1292                 if (fd >= 0) {
1293                         dup2(fd, STDIN_FILENO);
1294                         dup2(fd, STDOUT_FILENO);
1295                         dup2(fd, STDERR_FILENO);
1296                         close(fd);
1297                 }
1298         }
1299 
1300         fd_inotify = udev_watch_init(udev);
1301         if (fd_inotify < 0) {
1302                 r = log_error_errno(ENOMEM, "error initializing inotify");
1303                 goto exit;
1304         }
1305         /* watch rules.d paths for changes */
1306         inotify_add_watch(fd_inotify, UDEV_RULES_DIR,
1307                           IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1308         inotify_add_watch(fd_inotify, UDEV_CONF_DIR "/rules.d",
1309                           IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1310 
1311         if (access(UDEV_ROOT_RUN "/udev/rules.d", F_OK) < 0) {
1312                 udev_mkdir_p(UDEV_ROOT_RUN "/udev/rules.d", 0755);
1313         }
1314         inotify_add_watch(fd_inotify, UDEV_ROOT_RUN "/udev/rules.d",
1315                           IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1316 
1317         udev_watch_restore(udev);
1318 
1319         /* block and listen to all signals on signalfd */
1320         sigfillset(&mask);
1321         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1322         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1323         if (fd_signal < 0) {
1324                 r = log_error_errno(errno, "error creating signalfd");
1325                 goto exit;
1326         }
1327 
1328         /* unnamed socket from workers to the main daemon */
1329         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1330                 r = log_error_errno(errno, "error creating socketpair");
1331                 goto exit;
1332         }
1333         fd_worker = worker_watch[READ_END];
1334 
1335         r = setsockopt(fd_worker, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
1336         if (r < 0)
1337                 return log_error_errno(errno, "could not enable SO_PASSCRED: %m");
1338 
1339         ep_ctrl.data.fd = fd_ctrl;
1340         ep_inotify.data.fd = fd_inotify;
1341         ep_signal.data.fd = fd_signal;
1342         ep_netlink.data.fd = fd_netlink;
1343         ep_worker.data.fd = fd_worker;
1344 
1345         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1346         if (fd_ep < 0) {
1347                 log_error_errno(errno, "error creating epoll fd: %m");
1348                 goto exit;
1349         }
1350         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1351             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1352             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1353             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1354             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1355                 log_error_errno(errno, "fail to add fds to epoll: %m");
1356                 goto exit;
1357         }
1358 
1359         for (;;) {
1360                 static usec_t last_usec;
1361                 struct epoll_event ev[8];
1362                 int fdcount;
1363                 int timeout;
1364                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1365                 int i;
1366 
1367                 if (udev_exit) {
1368                         /* close sources of new events and discard buffered events */
1369                         if (fd_ctrl >= 0) {
1370                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1371                                 fd_ctrl = -1;
1372                         }
1373                         if (monitor != NULL) {
1374                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1375                                 udev_monitor_unref(monitor);
1376                                 monitor = NULL;
1377                         }
1378                         if (fd_inotify >= 0) {
1379                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1380                                 close(fd_inotify);
1381                                 fd_inotify = -1;
1382                         }
1383 
1384                         /* discard queued events and kill workers */
1385                         event_queue_cleanup(udev, EVENT_QUEUED);
1386                         worker_kill();
1387 
1388                         /* exit after all has cleaned up */
1389                         if (udev_list_node_is_empty(&event_list) && hashmap_isempty(workers))
1390                                 break;
1391 
1392                         /* timeout at exit for workers to finish */
1393                         timeout = 30 * MSEC_PER_SEC;
1394                 } else if (udev_list_node_is_empty(&event_list) && hashmap_isempty(workers)) {
1395                         /* we are idle */
1396                         timeout = -1;
1397                 } else {
1398                         /* kill idle or hanging workers */
1399                         timeout = 3 * MSEC_PER_SEC;
1400                 }
1401 
1402                 /* tell settle that we are busy or idle */
1403                 event_queue_update();
1404 
1405                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1406                 if (fdcount < 0)
1407                         continue;
1408 
1409                 if (fdcount == 0) {
1410                         struct worker *worker;
1411                         Iterator j;
1412 
1413                         /* timeout */
1414                         if (udev_exit) {
1415                                 log_error("timeout, giving up waiting for workers to finish");
1416                                 break;
1417                         }
1418 
1419                         /* kill idle workers */
1420                         if (udev_list_node_is_empty(&event_list)) {
1421                                 log_debug("cleanup idle workers");
1422                                 worker_kill();
1423                         }
1424 
1425                         /* check for hanging events */
1426                         HASHMAP_FOREACH(worker, workers, j) {
1427                                 struct event *event = worker->event;
1428                                 usec_t ts;
1429 
1430                                 if (worker->state != WORKER_RUNNING)
1431                                         continue;
1432 
1433                                 assert(event);
1434 
1435                                 ts = now(CLOCK_MONOTONIC);
1436 
1437                                 if ((ts - event->start_usec) > arg_event_timeout_warn_usec) {
1438                                         if ((ts - event->start_usec) > arg_event_timeout_usec) {
1439                                                 log_error("worker ["PID_FMT"] %s timeout; kill it", worker->pid, event->devpath);
1440                                                 kill(worker->pid, SIGKILL);
1441                                                 worker->state = WORKER_KILLED;
1442 
1443                                                 log_error("seq %llu '%s' killed", udev_device_get_seqnum(event->dev), event->devpath);
1444                                         } else if (!event->warned) {
1445                                                 log_warning("worker ["PID_FMT"] %s is taking a long time", worker->pid, event->devpath);
1446                                                 event->warned = true;
1447                                         }
1448                                 }
1449                         }
1450 
1451                 }
1452 
1453                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1454                 for (i = 0; i < fdcount; i++) {
1455                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1456                                 is_worker = true;
1457                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1458                                 is_netlink = true;
1459                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1460                                 is_signal = true;
1461                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1462                                 is_inotify = true;
1463                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1464                                 is_ctrl = true;
1465                 }
1466 
1467                 /* check for changed config, every 3 seconds at most */
1468                 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1469                         if (udev_rules_check_timestamp(rules))
1470                                 reload = true;
1471                         if (udev_builtin_validate(udev))
1472                                 reload = true;
1473 
1474                         last_usec = now(CLOCK_MONOTONIC);
1475                 }
1476 
1477                 /* reload requested, HUP signal received, rules changed, builtin changed */
1478                 if (reload) {
1479                         worker_kill();
1480                         rules = udev_rules_unref(rules);
1481                         udev_builtin_exit(udev);
1482                         reload = false;
1483                 }
1484 
1485                 /* event has finished */
1486                 if (is_worker)
1487                         worker_returned(fd_worker);
1488 
1489                 if (is_netlink) {
1490                         struct udev_device *dev;
1491 
1492                         dev = udev_monitor_receive_device(monitor);
1493                         if (dev) {
1494                                 udev_device_ensure_usec_initialized(dev, NULL);
1495                                 if (event_queue_insert(dev) < 0)
1496                                         udev_device_unref(dev);
1497                         }
1498                 }
1499 
1500                 /* start new events */
1501                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1502                         udev_builtin_init(udev);
1503                         if (rules == NULL)
1504                                 rules = udev_rules_new(udev, arg_resolve_names);
1505                         if (rules != NULL)
1506                                 event_queue_start(udev);
1507                 }
1508 
1509                 if (is_signal) {
1510                         struct signalfd_siginfo fdsi;
1511                         ssize_t size;
1512 
1513                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1514                         if (size == sizeof(struct signalfd_siginfo))
1515                                 handle_signal(udev, fdsi.ssi_signo);
1516                 }
1517 
1518                 /* we are shutting down, the events below are not handled anymore */
1519                 if (udev_exit)
1520                         continue;
1521 
1522                 /* device node watch */
1523                 if (is_inotify) {
1524                         handle_inotify(udev);
1525 
1526                         /*
1527                          * settle might be waiting on us to determine the queue
1528                          * state. If we just handled an inotify event, we might have
1529                          * generated a "change" event, but we won't have queued up
1530                          * the resultant uevent yet.
1531                          *
1532                          * Before we go ahead and potentially tell settle that the
1533                          * queue is empty, lets loop one more time to update the
1534                          * queue state again before deciding.
1535                          */
1536                         continue;
1537                 }
1538 
1539                 /*
1540                  * This needs to be after the inotify handling, to make sure,
1541                  * that the ping is send back after the possibly generated
1542                  * "change" events by the inotify device node watch.
1543                  */
1544                 if (is_ctrl)
1545                         handle_ctrl_msg(udev_ctrl);
1546         }
1547 
1548 exit:
1549         udev_ctrl_cleanup(udev_ctrl);
1550         unlink(UDEV_ROOT_RUN "/udev/queue");
1551 
1552         if (fd_ep >= 0)
1553                 close(fd_ep);
1554         workers_free();
1555         event_queue_cleanup(udev, EVENT_UNDEF);
1556         udev_rules_unref(rules);
1557         udev_builtin_exit(udev);
1558         if (fd_signal >= 0)
1559                 close(fd_signal);
1560         if (worker_watch[READ_END] >= 0)
1561                 close(worker_watch[READ_END]);
1562         if (worker_watch[WRITE_END] >= 0)
1563                 close(worker_watch[WRITE_END]);
1564         udev_monitor_unref(monitor);
1565         udev_ctrl_connection_unref(udev_ctrl_conn);
1566         udev_ctrl_unref(udev_ctrl);
1567         mac_selinux_finish();
1568         udev_unref(udev);
1569         log_close();
1570         return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;
1571 }
1572