]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/udev/udevd.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
tree-wide: make sure our control buffers are properly aligned
[thirdparty/systemd.git] / src / udev / udevd.c
1 /* SPDX-License-Identifier: GPL-2.0+ */
2 /*
3 * Copyright © 2004 Chris Friesen <chris_friesen@sympatico.ca>
4 * Copyright © 2009 Canonical Ltd.
5 * Copyright © 2009 Scott James Remnant <scott@netsplit.com>
6 */
7
8 #include <errno.h>
9 #include <fcntl.h>
10 #include <getopt.h>
11 #include <stdbool.h>
12 #include <stddef.h>
13 #include <stdio.h>
14 #include <stdlib.h>
15 #include <sys/epoll.h>
16 #include <sys/file.h>
17 #include <sys/inotify.h>
18 #include <sys/ioctl.h>
19 #include <sys/mount.h>
20 #include <sys/prctl.h>
21 #include <sys/signalfd.h>
22 #include <sys/stat.h>
23 #include <sys/time.h>
24 #include <sys/wait.h>
25 #include <unistd.h>
26
27 #include "sd-daemon.h"
28 #include "sd-event.h"
29
30 #include "alloc-util.h"
31 #include "build.h"
32 #include "cgroup-util.h"
33 #include "cpu-set-util.h"
34 #include "dev-setup.h"
35 #include "device-monitor-private.h"
36 #include "device-private.h"
37 #include "device-util.h"
38 #include "event-util.h"
39 #include "fd-util.h"
40 #include "fileio.h"
41 #include "format-util.h"
42 #include "fs-util.h"
43 #include "hashmap.h"
44 #include "io-util.h"
45 #include "libudev-device-internal.h"
46 #include "limits-util.h"
47 #include "list.h"
48 #include "main-func.h"
49 #include "mkdir.h"
50 #include "netlink-util.h"
51 #include "parse-util.h"
52 #include "pretty-print.h"
53 #include "proc-cmdline.h"
54 #include "process-util.h"
55 #include "selinux-util.h"
56 #include "signal-util.h"
57 #include "socket-util.h"
58 #include "string-util.h"
59 #include "strv.h"
60 #include "strxcpyx.h"
61 #include "syslog-util.h"
62 #include "udev-builtin.h"
63 #include "udev-ctrl.h"
64 #include "udev-event.h"
65 #include "udev-util.h"
66 #include "udev-watch.h"
67 #include "user-util.h"
68
69 #define WORKER_NUM_MAX 2048U
70
71 static bool arg_debug = false;
72 static int arg_daemonize = false;
73 static ResolveNameTiming arg_resolve_name_timing = RESOLVE_NAME_EARLY;
74 static unsigned arg_children_max = 0;
75 static usec_t arg_exec_delay_usec = 0;
76 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
77
78 typedef struct Manager {
79 sd_event *event;
80 Hashmap *workers;
81 LIST_HEAD(struct event, events);
82 const char *cgroup;
83 pid_t pid; /* the process that originally allocated the manager object */
84
85 UdevRules *rules;
86 Hashmap *properties;
87
88 sd_netlink *rtnl;
89
90 sd_device_monitor *monitor;
91 struct udev_ctrl *ctrl;
92 int fd_inotify;
93 int worker_watch[2];
94
95 sd_event_source *inotify_event;
96 sd_event_source *kill_workers_event;
97
98 usec_t last_usec;
99
100 bool stop_exec_queue:1;
101 bool exit:1;
102 } Manager;
103
104 enum event_state {
105 EVENT_UNDEF,
106 EVENT_QUEUED,
107 EVENT_RUNNING,
108 };
109
110 struct event {
111 Manager *manager;
112 struct worker *worker;
113 enum event_state state;
114
115 sd_device *dev;
116 sd_device *dev_kernel; /* clone of originally received device */
117
118 uint64_t seqnum;
119 uint64_t delaying_seqnum;
120
121 sd_event_source *timeout_warning_event;
122 sd_event_source *timeout_event;
123
124 LIST_FIELDS(struct event, event);
125 };
126
127 static void event_queue_cleanup(Manager *manager, enum event_state type);
128
129 enum worker_state {
130 WORKER_UNDEF,
131 WORKER_RUNNING,
132 WORKER_IDLE,
133 WORKER_KILLED,
134 };
135
136 struct worker {
137 Manager *manager;
138 pid_t pid;
139 sd_device_monitor *monitor;
140 enum worker_state state;
141 struct event *event;
142 };
143
144 /* passed from worker to main process */
145 struct worker_message {
146 };
147
148 static void event_free(struct event *event) {
149 if (!event)
150 return;
151
152 assert(event->manager);
153
154 LIST_REMOVE(event, event->manager->events, event);
155 sd_device_unref(event->dev);
156 sd_device_unref(event->dev_kernel);
157
158 sd_event_source_unref(event->timeout_warning_event);
159 sd_event_source_unref(event->timeout_event);
160
161 if (event->worker)
162 event->worker->event = NULL;
163
164 /* only clean up the queue from the process that created it */
165 if (LIST_IS_EMPTY(event->manager->events) &&
166 event->manager->pid == getpid_cached())
167 if (unlink("/run/udev/queue") < 0)
168 log_warning_errno(errno, "Failed to unlink /run/udev/queue: %m");
169
170 free(event);
171 }
172
173 static void worker_free(struct worker *worker) {
174 if (!worker)
175 return;
176
177 assert(worker->manager);
178
179 hashmap_remove(worker->manager->workers, PID_TO_PTR(worker->pid));
180 sd_device_monitor_unref(worker->monitor);
181 event_free(worker->event);
182
183 free(worker);
184 }
185
186 DEFINE_TRIVIAL_CLEANUP_FUNC(struct worker *, worker_free);
187 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(worker_hash_op, void, trivial_hash_func, trivial_compare_func, struct worker, worker_free);
188
189 static int worker_new(struct worker **ret, Manager *manager, sd_device_monitor *worker_monitor, pid_t pid) {
190 _cleanup_(worker_freep) struct worker *worker = NULL;
191 int r;
192
193 assert(ret);
194 assert(manager);
195 assert(worker_monitor);
196 assert(pid > 1);
197
198 /* close monitor, but keep address around */
199 device_monitor_disconnect(worker_monitor);
200
201 worker = new(struct worker, 1);
202 if (!worker)
203 return -ENOMEM;
204
205 *worker = (struct worker) {
206 .manager = manager,
207 .monitor = sd_device_monitor_ref(worker_monitor),
208 .pid = pid,
209 };
210
211 r = hashmap_ensure_allocated(&manager->workers, &worker_hash_op);
212 if (r < 0)
213 return r;
214
215 r = hashmap_put(manager->workers, PID_TO_PTR(pid), worker);
216 if (r < 0)
217 return r;
218
219 *ret = TAKE_PTR(worker);
220
221 return 0;
222 }
223
224 static int on_event_timeout(sd_event_source *s, uint64_t usec, void *userdata) {
225 struct event *event = userdata;
226
227 assert(event);
228 assert(event->worker);
229
230 kill_and_sigcont(event->worker->pid, SIGKILL);
231 event->worker->state = WORKER_KILLED;
232
233 log_device_error(event->dev, "Worker ["PID_FMT"] processing SEQNUM=%"PRIu64" killed", event->worker->pid, event->seqnum);
234
235 return 1;
236 }
237
238 static int on_event_timeout_warning(sd_event_source *s, uint64_t usec, void *userdata) {
239 struct event *event = userdata;
240
241 assert(event);
242 assert(event->worker);
243
244 log_device_warning(event->dev, "Worker ["PID_FMT"] processing SEQNUM=%"PRIu64" is taking a long time", event->worker->pid, event->seqnum);
245
246 return 1;
247 }
248
249 static void worker_attach_event(struct worker *worker, struct event *event) {
250 sd_event *e;
251 uint64_t usec;
252
253 assert(worker);
254 assert(worker->manager);
255 assert(event);
256 assert(!event->worker);
257 assert(!worker->event);
258
259 worker->state = WORKER_RUNNING;
260 worker->event = event;
261 event->state = EVENT_RUNNING;
262 event->worker = worker;
263
264 e = worker->manager->event;
265
266 assert_se(sd_event_now(e, CLOCK_MONOTONIC, &usec) >= 0);
267
268 (void) sd_event_add_time(e, &event->timeout_warning_event, CLOCK_MONOTONIC,
269 usec + udev_warn_timeout(arg_event_timeout_usec), USEC_PER_SEC, on_event_timeout_warning, event);
270
271 (void) sd_event_add_time(e, &event->timeout_event, CLOCK_MONOTONIC,
272 usec + arg_event_timeout_usec, USEC_PER_SEC, on_event_timeout, event);
273 }
274
275 static void manager_clear_for_worker(Manager *manager) {
276 assert(manager);
277
278 manager->inotify_event = sd_event_source_unref(manager->inotify_event);
279 manager->kill_workers_event = sd_event_source_unref(manager->kill_workers_event);
280
281 manager->event = sd_event_unref(manager->event);
282
283 manager->workers = hashmap_free(manager->workers);
284 event_queue_cleanup(manager, EVENT_UNDEF);
285
286 manager->monitor = sd_device_monitor_unref(manager->monitor);
287 manager->ctrl = udev_ctrl_unref(manager->ctrl);
288
289 manager->worker_watch[READ_END] = safe_close(manager->worker_watch[READ_END]);
290 }
291
292 static void manager_free(Manager *manager) {
293 if (!manager)
294 return;
295
296 udev_builtin_exit();
297
298 if (manager->pid == getpid_cached())
299 udev_ctrl_cleanup(manager->ctrl);
300
301 manager_clear_for_worker(manager);
302
303 sd_netlink_unref(manager->rtnl);
304
305 hashmap_free_free_free(manager->properties);
306 udev_rules_free(manager->rules);
307
308 safe_close(manager->fd_inotify);
309 safe_close_pair(manager->worker_watch);
310
311 free(manager);
312 }
313
314 DEFINE_TRIVIAL_CLEANUP_FUNC(Manager*, manager_free);
315
316 static int worker_send_message(int fd) {
317 struct worker_message message = {};
318
319 return loop_write(fd, &message, sizeof(message), false);
320 }
321
322 static int worker_lock_block_device(sd_device *dev, int *ret_fd) {
323 _cleanup_close_ int fd = -1;
324 const char *val;
325 int r;
326
327 assert(dev);
328 assert(ret_fd);
329
330 /*
331 * Take a shared lock on the device node; this establishes
332 * a concept of device "ownership" to serialize device
333 * access. External processes holding an exclusive lock will
334 * cause udev to skip the event handling; in the case udev
335 * acquired the lock, the external process can block until
336 * udev has finished its event handling.
337 */
338
339 if (device_for_action(dev, DEVICE_ACTION_REMOVE))
340 return 0;
341
342 r = sd_device_get_subsystem(dev, &val);
343 if (r < 0)
344 return log_device_debug_errno(dev, r, "Failed to get subsystem: %m");
345
346 if (!streq(val, "block"))
347 return 0;
348
349 r = sd_device_get_sysname(dev, &val);
350 if (r < 0)
351 return log_device_debug_errno(dev, r, "Failed to get sysname: %m");
352
353 if (STARTSWITH_SET(val, "dm-", "md", "drbd"))
354 return 0;
355
356 r = sd_device_get_devtype(dev, &val);
357 if (r < 0 && r != -ENOENT)
358 return log_device_debug_errno(dev, r, "Failed to get devtype: %m");
359 if (r >= 0 && streq(val, "partition")) {
360 r = sd_device_get_parent(dev, &dev);
361 if (r < 0)
362 return log_device_debug_errno(dev, r, "Failed to get parent device: %m");
363 }
364
365 r = sd_device_get_devname(dev, &val);
366 if (r == -ENOENT)
367 return 0;
368 if (r < 0)
369 return log_device_debug_errno(dev, r, "Failed to get devname: %m");
370
371 fd = open(val, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
372 if (fd < 0) {
373 log_device_debug_errno(dev, errno, "Failed to open '%s', ignoring: %m", val);
374 return 0;
375 }
376
377 if (flock(fd, LOCK_SH|LOCK_NB) < 0)
378 return log_device_debug_errno(dev, errno, "Failed to flock(%s): %m", val);
379
380 *ret_fd = TAKE_FD(fd);
381 return 1;
382 }
383
384 static int worker_process_device(Manager *manager, sd_device *dev) {
385 _cleanup_(udev_event_freep) UdevEvent *udev_event = NULL;
386 _cleanup_close_ int fd_lock = -1;
387 DeviceAction action;
388 uint64_t seqnum;
389 int r;
390
391 assert(manager);
392 assert(dev);
393
394 r = device_get_seqnum(dev, &seqnum);
395 if (r < 0)
396 return log_device_debug_errno(dev, r, "Failed to get SEQNUM: %m");
397
398 r = device_get_action(dev, &action);
399 if (r < 0)
400 return log_device_debug_errno(dev, r, "Failed to get ACTION: %m");
401
402 log_device_debug(dev, "Processing device (SEQNUM=%"PRIu64", ACTION=%s)",
403 seqnum, device_action_to_string(action));
404
405 udev_event = udev_event_new(dev, arg_exec_delay_usec, manager->rtnl);
406 if (!udev_event)
407 return -ENOMEM;
408
409 r = worker_lock_block_device(dev, &fd_lock);
410 if (r < 0)
411 return r;
412
413 /* apply rules, create node, symlinks */
414 r = udev_event_execute_rules(udev_event, arg_event_timeout_usec, manager->properties, manager->rules);
415 if (r < 0)
416 return r;
417
418 udev_event_execute_run(udev_event, arg_event_timeout_usec);
419
420 if (!manager->rtnl)
421 /* in case rtnl was initialized */
422 manager->rtnl = sd_netlink_ref(udev_event->rtnl);
423
424 /* apply/restore inotify watch */
425 if (udev_event->inotify_watch) {
426 (void) udev_watch_begin(dev);
427 r = device_update_db(dev);
428 if (r < 0)
429 return log_device_debug_errno(dev, r, "Failed to update database under /run/udev/data/: %m");
430 }
431
432 log_device_debug(dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) processed",
433 seqnum, device_action_to_string(action));
434
435 return 0;
436 }
437
438 static int worker_device_monitor_handler(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
439 Manager *manager = userdata;
440 int r;
441
442 assert(dev);
443 assert(manager);
444
445 r = worker_process_device(manager, dev);
446 if (r < 0)
447 log_device_warning_errno(dev, r, "Failed to process device, ignoring: %m");
448
449 /* send processed event back to libudev listeners */
450 r = device_monitor_send_device(monitor, NULL, dev);
451 if (r < 0)
452 log_device_warning_errno(dev, r, "Failed to send device, ignoring: %m");
453
454 /* send udevd the result of the event execution */
455 r = worker_send_message(manager->worker_watch[WRITE_END]);
456 if (r < 0)
457 log_device_warning_errno(dev, r, "Failed to send signal to main daemon, ignoring: %m");
458
459 return 1;
460 }
461
462 static int worker_main(Manager *_manager, sd_device_monitor *monitor, sd_device *first_device) {
463 _cleanup_(sd_device_unrefp) sd_device *dev = first_device;
464 _cleanup_(manager_freep) Manager *manager = _manager;
465 int r;
466
467 assert(manager);
468 assert(monitor);
469 assert(dev);
470
471 unsetenv("NOTIFY_SOCKET");
472
473 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, -1) >= 0);
474
475 /* Reset OOM score, we only protect the main daemon. */
476 r = set_oom_score_adjust(0);
477 if (r < 0)
478 log_debug_errno(r, "Failed to reset OOM score, ignoring: %m");
479
480 /* Clear unnecessary data in Manager object.*/
481 manager_clear_for_worker(manager);
482
483 r = sd_event_new(&manager->event);
484 if (r < 0)
485 return log_error_errno(r, "Failed to allocate event loop: %m");
486
487 r = sd_event_add_signal(manager->event, NULL, SIGTERM, NULL, NULL);
488 if (r < 0)
489 return log_error_errno(r, "Failed to set SIGTERM event: %m");
490
491 r = sd_device_monitor_attach_event(monitor, manager->event);
492 if (r < 0)
493 return log_error_errno(r, "Failed to attach event loop to device monitor: %m");
494
495 r = sd_device_monitor_start(monitor, worker_device_monitor_handler, manager);
496 if (r < 0)
497 return log_error_errno(r, "Failed to start device monitor: %m");
498
499 (void) sd_event_source_set_description(sd_device_monitor_get_event_source(monitor), "worker-device-monitor");
500
501 /* Process first device */
502 (void) worker_device_monitor_handler(monitor, dev, manager);
503
504 r = sd_event_loop(manager->event);
505 if (r < 0)
506 return log_error_errno(r, "Event loop failed: %m");
507
508 return 0;
509 }
510
511 static int worker_spawn(Manager *manager, struct event *event) {
512 _cleanup_(sd_device_monitor_unrefp) sd_device_monitor *worker_monitor = NULL;
513 struct worker *worker;
514 pid_t pid;
515 int r;
516
517 /* listen for new events */
518 r = device_monitor_new_full(&worker_monitor, MONITOR_GROUP_NONE, -1);
519 if (r < 0)
520 return r;
521
522 /* allow the main daemon netlink address to send devices to the worker */
523 r = device_monitor_allow_unicast_sender(worker_monitor, manager->monitor);
524 if (r < 0)
525 return log_error_errno(r, "Worker: Failed to set unicast sender: %m");
526
527 r = device_monitor_enable_receiving(worker_monitor);
528 if (r < 0)
529 return log_error_errno(r, "Worker: Failed to enable receiving of device: %m");
530
531 r = safe_fork(NULL, FORK_DEATHSIG, &pid);
532 if (r < 0) {
533 event->state = EVENT_QUEUED;
534 return log_error_errno(r, "Failed to fork() worker: %m");
535 }
536 if (r == 0) {
537 /* Worker process */
538 r = worker_main(manager, worker_monitor, sd_device_ref(event->dev));
539 log_close();
540 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
541 }
542
543 r = worker_new(&worker, manager, worker_monitor, pid);
544 if (r < 0)
545 return log_error_errno(r, "Failed to create worker object: %m");
546
547 worker_attach_event(worker, event);
548
549 log_device_debug(event->dev, "Worker ["PID_FMT"] is forked for processing SEQNUM=%"PRIu64".", pid, event->seqnum);
550 return 0;
551 }
552
553 static void event_run(Manager *manager, struct event *event) {
554 static bool log_children_max_reached = true;
555 struct worker *worker;
556 Iterator i;
557 int r;
558
559 assert(manager);
560 assert(event);
561
562 if (DEBUG_LOGGING) {
563 DeviceAction action;
564
565 r = device_get_action(event->dev, &action);
566 log_device_debug(event->dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) ready for processing",
567 event->seqnum, r >= 0 ? device_action_to_string(action) : "<unknown>");
568 }
569
570 HASHMAP_FOREACH(worker, manager->workers, i) {
571 if (worker->state != WORKER_IDLE)
572 continue;
573
574 r = device_monitor_send_device(manager->monitor, worker->monitor, event->dev);
575 if (r < 0) {
576 log_device_error_errno(event->dev, r, "Worker ["PID_FMT"] did not accept message, killing the worker: %m",
577 worker->pid);
578 (void) kill(worker->pid, SIGKILL);
579 worker->state = WORKER_KILLED;
580 continue;
581 }
582 worker_attach_event(worker, event);
583 return;
584 }
585
586 if (hashmap_size(manager->workers) >= arg_children_max) {
587
588 /* Avoid spamming the debug logs if the limit is already reached and
589 * many events still need to be processed */
590 if (log_children_max_reached && arg_children_max > 1) {
591 log_debug("Maximum number (%u) of children reached.", hashmap_size(manager->workers));
592 log_children_max_reached = false;
593 }
594 return;
595 }
596
597 /* Re-enable the debug message for the next batch of events */
598 log_children_max_reached = true;
599
600 /* start new worker and pass initial device */
601 worker_spawn(manager, event);
602 }
603
604 static int event_queue_insert(Manager *manager, sd_device *dev) {
605 _cleanup_(sd_device_unrefp) sd_device *clone = NULL;
606 struct event *event;
607 DeviceAction action;
608 uint64_t seqnum;
609 int r;
610
611 assert(manager);
612 assert(dev);
613
614 /* only one process can add events to the queue */
615 assert(manager->pid == getpid_cached());
616
617 /* We only accepts devices received by device monitor. */
618 r = device_get_seqnum(dev, &seqnum);
619 if (r < 0)
620 return r;
621
622 /* Refuse devices do not have ACTION property. */
623 r = device_get_action(dev, &action);
624 if (r < 0)
625 return r;
626
627 /* Save original device to restore the state on failures. */
628 r = device_shallow_clone(dev, &clone);
629 if (r < 0)
630 return r;
631
632 r = device_copy_properties(clone, dev);
633 if (r < 0)
634 return r;
635
636 event = new(struct event, 1);
637 if (!event)
638 return -ENOMEM;
639
640 *event = (struct event) {
641 .manager = manager,
642 .dev = sd_device_ref(dev),
643 .dev_kernel = TAKE_PTR(clone),
644 .seqnum = seqnum,
645 .state = EVENT_QUEUED,
646 };
647
648 if (LIST_IS_EMPTY(manager->events)) {
649 r = touch("/run/udev/queue");
650 if (r < 0)
651 log_warning_errno(r, "Failed to touch /run/udev/queue: %m");
652 }
653
654 LIST_APPEND(event, manager->events, event);
655
656 log_device_debug(dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) is queued",
657 seqnum, device_action_to_string(action));
658
659 return 0;
660 }
661
662 static void manager_kill_workers(Manager *manager) {
663 struct worker *worker;
664 Iterator i;
665
666 assert(manager);
667
668 HASHMAP_FOREACH(worker, manager->workers, i) {
669 if (worker->state == WORKER_KILLED)
670 continue;
671
672 worker->state = WORKER_KILLED;
673 (void) kill(worker->pid, SIGTERM);
674 }
675 }
676
677 /* lookup event for identical, parent, child device */
678 static int is_device_busy(Manager *manager, struct event *event) {
679 const char *subsystem, *devpath, *devpath_old = NULL;
680 dev_t devnum = makedev(0, 0);
681 struct event *loop_event;
682 size_t devpath_len;
683 int r, ifindex = 0;
684 bool is_block;
685
686 r = sd_device_get_subsystem(event->dev, &subsystem);
687 if (r < 0)
688 return r;
689
690 is_block = streq(subsystem, "block");
691
692 r = sd_device_get_devpath(event->dev, &devpath);
693 if (r < 0)
694 return r;
695
696 devpath_len = strlen(devpath);
697
698 r = sd_device_get_property_value(event->dev, "DEVPATH_OLD", &devpath_old);
699 if (r < 0 && r != -ENOENT)
700 return r;
701
702 r = sd_device_get_devnum(event->dev, &devnum);
703 if (r < 0 && r != -ENOENT)
704 return r;
705
706 r = sd_device_get_ifindex(event->dev, &ifindex);
707 if (r < 0 && r != -ENOENT)
708 return r;
709
710 /* check if queue contains events we depend on */
711 LIST_FOREACH(event, loop_event, manager->events) {
712 size_t loop_devpath_len, common;
713 const char *loop_devpath;
714
715 /* we already found a later event, earlier cannot block us, no need to check again */
716 if (loop_event->seqnum < event->delaying_seqnum)
717 continue;
718
719 /* event we checked earlier still exists, no need to check again */
720 if (loop_event->seqnum == event->delaying_seqnum)
721 return true;
722
723 /* found ourself, no later event can block us */
724 if (loop_event->seqnum >= event->seqnum)
725 break;
726
727 /* check major/minor */
728 if (major(devnum) != 0) {
729 const char *s;
730 dev_t d;
731
732 if (sd_device_get_subsystem(loop_event->dev, &s) < 0)
733 continue;
734
735 if (sd_device_get_devnum(loop_event->dev, &d) >= 0 &&
736 devnum == d && is_block == streq(s, "block"))
737 goto set_delaying_seqnum;
738 }
739
740 /* check network device ifindex */
741 if (ifindex > 0) {
742 int i;
743
744 if (sd_device_get_ifindex(loop_event->dev, &i) >= 0 &&
745 ifindex == i)
746 goto set_delaying_seqnum;
747 }
748
749 if (sd_device_get_devpath(loop_event->dev, &loop_devpath) < 0)
750 continue;
751
752 /* check our old name */
753 if (devpath_old && streq(devpath_old, loop_devpath))
754 goto set_delaying_seqnum;
755
756 loop_devpath_len = strlen(loop_devpath);
757
758 /* compare devpath */
759 common = MIN(devpath_len, loop_devpath_len);
760
761 /* one devpath is contained in the other? */
762 if (!strneq(devpath, loop_devpath, common))
763 continue;
764
765 /* identical device event found */
766 if (devpath_len == loop_devpath_len)
767 goto set_delaying_seqnum;
768
769 /* parent device event found */
770 if (devpath[common] == '/')
771 goto set_delaying_seqnum;
772
773 /* child device event found */
774 if (loop_devpath[common] == '/')
775 goto set_delaying_seqnum;
776 }
777
778 return false;
779
780 set_delaying_seqnum:
781 log_device_debug(event->dev, "SEQNUM=%" PRIu64 " blocked by SEQNUM=%" PRIu64,
782 event->seqnum, loop_event->seqnum);
783
784 event->delaying_seqnum = loop_event->seqnum;
785 return true;
786 }
787
788 static void manager_exit(Manager *manager) {
789 assert(manager);
790
791 manager->exit = true;
792
793 sd_notify(false,
794 "STOPPING=1\n"
795 "STATUS=Starting shutdown...");
796
797 /* close sources of new events and discard buffered events */
798 manager->ctrl = udev_ctrl_unref(manager->ctrl);
799
800 manager->inotify_event = sd_event_source_unref(manager->inotify_event);
801 manager->fd_inotify = safe_close(manager->fd_inotify);
802
803 manager->monitor = sd_device_monitor_unref(manager->monitor);
804
805 /* discard queued events and kill workers */
806 event_queue_cleanup(manager, EVENT_QUEUED);
807 manager_kill_workers(manager);
808 }
809
810 /* reload requested, HUP signal received, rules changed, builtin changed */
811 static void manager_reload(Manager *manager) {
812
813 assert(manager);
814
815 sd_notify(false,
816 "RELOADING=1\n"
817 "STATUS=Flushing configuration...");
818
819 manager_kill_workers(manager);
820 manager->rules = udev_rules_free(manager->rules);
821 udev_builtin_exit();
822
823 sd_notifyf(false,
824 "READY=1\n"
825 "STATUS=Processing with %u children at max", arg_children_max);
826 }
827
828 static int on_kill_workers_event(sd_event_source *s, uint64_t usec, void *userdata) {
829 Manager *manager = userdata;
830
831 assert(manager);
832
833 log_debug("Cleanup idle workers");
834 manager_kill_workers(manager);
835
836 return 1;
837 }
838
839 static void event_queue_start(Manager *manager) {
840 struct event *event;
841 usec_t usec;
842 int r;
843
844 assert(manager);
845
846 if (LIST_IS_EMPTY(manager->events) ||
847 manager->exit || manager->stop_exec_queue)
848 return;
849
850 assert_se(sd_event_now(manager->event, CLOCK_MONOTONIC, &usec) >= 0);
851 /* check for changed config, every 3 seconds at most */
852 if (manager->last_usec == 0 ||
853 usec - manager->last_usec > 3 * USEC_PER_SEC) {
854 if (udev_rules_check_timestamp(manager->rules) ||
855 udev_builtin_validate())
856 manager_reload(manager);
857
858 manager->last_usec = usec;
859 }
860
861 r = event_source_disable(manager->kill_workers_event);
862 if (r < 0)
863 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
864
865 udev_builtin_init();
866
867 if (!manager->rules) {
868 r = udev_rules_new(&manager->rules, arg_resolve_name_timing);
869 if (r < 0) {
870 log_warning_errno(r, "Failed to read udev rules: %m");
871 return;
872 }
873 }
874
875 LIST_FOREACH(event, event, manager->events) {
876 if (event->state != EVENT_QUEUED)
877 continue;
878
879 /* do not start event if parent or child event is still running */
880 if (is_device_busy(manager, event) != 0)
881 continue;
882
883 event_run(manager, event);
884 }
885 }
886
887 static void event_queue_cleanup(Manager *manager, enum event_state match_type) {
888 struct event *event, *tmp;
889
890 LIST_FOREACH_SAFE(event, event, tmp, manager->events) {
891 if (match_type != EVENT_UNDEF && match_type != event->state)
892 continue;
893
894 event_free(event);
895 }
896 }
897
898 static int on_worker(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
899 Manager *manager = userdata;
900
901 assert(manager);
902
903 for (;;) {
904 struct worker_message msg;
905 struct iovec iovec = {
906 .iov_base = &msg,
907 .iov_len = sizeof(msg),
908 };
909 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred))) control;
910 struct msghdr msghdr = {
911 .msg_iov = &iovec,
912 .msg_iovlen = 1,
913 .msg_control = &control,
914 .msg_controllen = sizeof(control),
915 };
916 ssize_t size;
917 struct ucred *ucred;
918 struct worker *worker;
919
920 size = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT);
921 if (size == -EINTR)
922 continue;
923 if (size == -EAGAIN)
924 /* nothing more to read */
925 break;
926 if (size < 0)
927 return log_error_errno(size, "Failed to receive message: %m");
928
929 cmsg_close_all(&msghdr);
930
931 if (size != sizeof(struct worker_message)) {
932 log_warning("Ignoring worker message with invalid size %zi bytes", size);
933 continue;
934 }
935
936 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
937 if (!ucred || ucred->pid <= 0) {
938 log_warning("Ignoring worker message without valid PID");
939 continue;
940 }
941
942 /* lookup worker who sent the signal */
943 worker = hashmap_get(manager->workers, PID_TO_PTR(ucred->pid));
944 if (!worker) {
945 log_debug("Worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid);
946 continue;
947 }
948
949 if (worker->state != WORKER_KILLED)
950 worker->state = WORKER_IDLE;
951
952 /* worker returned */
953 event_free(worker->event);
954 }
955
956 /* we have free workers, try to schedule events */
957 event_queue_start(manager);
958
959 return 1;
960 }
961
962 static int on_uevent(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
963 Manager *manager = userdata;
964 int r;
965
966 assert(manager);
967
968 device_ensure_usec_initialized(dev, NULL);
969
970 r = event_queue_insert(manager, dev);
971 if (r < 0) {
972 log_device_error_errno(dev, r, "Failed to insert device into event queue: %m");
973 return 1;
974 }
975
976 /* we have fresh events, try to schedule them */
977 event_queue_start(manager);
978
979 return 1;
980 }
981
982 /* receive the udevd message from userspace */
983 static int on_ctrl_msg(struct udev_ctrl *uctrl, enum udev_ctrl_msg_type type, const union udev_ctrl_msg_value *value, void *userdata) {
984 Manager *manager = userdata;
985 int r;
986
987 assert(value);
988 assert(manager);
989
990 switch (type) {
991 case UDEV_CTRL_SET_LOG_LEVEL:
992 log_debug("Received udev control message (SET_LOG_LEVEL), setting log_priority=%i", value->intval);
993 log_set_max_level_realm(LOG_REALM_UDEV, value->intval);
994 log_set_max_level_realm(LOG_REALM_SYSTEMD, value->intval);
995 manager_kill_workers(manager);
996 break;
997 case UDEV_CTRL_STOP_EXEC_QUEUE:
998 log_debug("Received udev control message (STOP_EXEC_QUEUE)");
999 manager->stop_exec_queue = true;
1000 break;
1001 case UDEV_CTRL_START_EXEC_QUEUE:
1002 log_debug("Received udev control message (START_EXEC_QUEUE)");
1003 manager->stop_exec_queue = false;
1004 event_queue_start(manager);
1005 break;
1006 case UDEV_CTRL_RELOAD:
1007 log_debug("Received udev control message (RELOAD)");
1008 manager_reload(manager);
1009 break;
1010 case UDEV_CTRL_SET_ENV: {
1011 _cleanup_free_ char *key = NULL, *val = NULL, *old_key = NULL, *old_val = NULL;
1012 const char *eq;
1013
1014 eq = strchr(value->buf, '=');
1015 if (!eq) {
1016 log_error("Invalid key format '%s'", value->buf);
1017 return 1;
1018 }
1019
1020 key = strndup(value->buf, eq - value->buf);
1021 if (!key) {
1022 log_oom();
1023 return 1;
1024 }
1025
1026 old_val = hashmap_remove2(manager->properties, key, (void **) &old_key);
1027
1028 r = hashmap_ensure_allocated(&manager->properties, &string_hash_ops);
1029 if (r < 0) {
1030 log_oom();
1031 return 1;
1032 }
1033
1034 eq++;
1035 if (isempty(eq)) {
1036 log_debug("Received udev control message (ENV), unsetting '%s'", key);
1037
1038 r = hashmap_put(manager->properties, key, NULL);
1039 if (r < 0) {
1040 log_oom();
1041 return 1;
1042 }
1043 } else {
1044 val = strdup(eq);
1045 if (!val) {
1046 log_oom();
1047 return 1;
1048 }
1049
1050 log_debug("Received udev control message (ENV), setting '%s=%s'", key, val);
1051
1052 r = hashmap_put(manager->properties, key, val);
1053 if (r < 0) {
1054 log_oom();
1055 return 1;
1056 }
1057 }
1058
1059 key = val = NULL;
1060 manager_kill_workers(manager);
1061 break;
1062 }
1063 case UDEV_CTRL_SET_CHILDREN_MAX:
1064 if (value->intval <= 0) {
1065 log_debug("Received invalid udev control message (SET_MAX_CHILDREN, %i), ignoring.", value->intval);
1066 return 0;
1067 }
1068
1069 log_debug("Received udev control message (SET_MAX_CHILDREN), setting children_max=%i", value->intval);
1070 arg_children_max = value->intval;
1071
1072 (void) sd_notifyf(false,
1073 "READY=1\n"
1074 "STATUS=Processing with %u children at max", arg_children_max);
1075 break;
1076 case UDEV_CTRL_PING:
1077 log_debug("Received udev control message (PING)");
1078 break;
1079 case UDEV_CTRL_EXIT:
1080 log_debug("Received udev control message (EXIT)");
1081 manager_exit(manager);
1082 break;
1083 default:
1084 log_debug("Received unknown udev control message, ignoring");
1085 }
1086
1087 return 1;
1088 }
1089
1090 static int synthesize_change_one(sd_device *dev, const char *syspath) {
1091 const char *filename;
1092 int r;
1093
1094 filename = strjoina(syspath, "/uevent");
1095 log_device_debug(dev, "device is closed, synthesising 'change' on %s", syspath);
1096 r = write_string_file(filename, "change", WRITE_STRING_FILE_DISABLE_BUFFER);
1097 if (r < 0)
1098 return log_device_debug_errno(dev, r, "Failed to write 'change' to %s: %m", filename);
1099 return 0;
1100 }
1101
1102 static int synthesize_change(sd_device *dev) {
1103 const char *subsystem, *sysname, *devname, *syspath, *devtype;
1104 int r;
1105
1106 r = sd_device_get_subsystem(dev, &subsystem);
1107 if (r < 0)
1108 return r;
1109
1110 r = sd_device_get_sysname(dev, &sysname);
1111 if (r < 0)
1112 return r;
1113
1114 r = sd_device_get_devname(dev, &devname);
1115 if (r < 0)
1116 return r;
1117
1118 r = sd_device_get_syspath(dev, &syspath);
1119 if (r < 0)
1120 return r;
1121
1122 r = sd_device_get_devtype(dev, &devtype);
1123 if (r < 0)
1124 return r;
1125
1126 if (streq_ptr("block", subsystem) &&
1127 streq_ptr("disk", devtype) &&
1128 !startswith(sysname, "dm-")) {
1129 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
1130 bool part_table_read = false, has_partitions = false;
1131 sd_device *d;
1132 int fd;
1133
1134 /*
1135 * Try to re-read the partition table. This only succeeds if
1136 * none of the devices is busy. The kernel returns 0 if no
1137 * partition table is found, and we will not get an event for
1138 * the disk.
1139 */
1140 fd = open(devname, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
1141 if (fd >= 0) {
1142 r = flock(fd, LOCK_EX|LOCK_NB);
1143 if (r >= 0)
1144 r = ioctl(fd, BLKRRPART, 0);
1145
1146 close(fd);
1147 if (r >= 0)
1148 part_table_read = true;
1149 }
1150
1151 /* search for partitions */
1152 r = sd_device_enumerator_new(&e);
1153 if (r < 0)
1154 return r;
1155
1156 r = sd_device_enumerator_allow_uninitialized(e);
1157 if (r < 0)
1158 return r;
1159
1160 r = sd_device_enumerator_add_match_parent(e, dev);
1161 if (r < 0)
1162 return r;
1163
1164 r = sd_device_enumerator_add_match_subsystem(e, "block", true);
1165 if (r < 0)
1166 return r;
1167
1168 FOREACH_DEVICE(e, d) {
1169 const char *t;
1170
1171 if (sd_device_get_devtype(d, &t) < 0 ||
1172 !streq("partition", t))
1173 continue;
1174
1175 has_partitions = true;
1176 break;
1177 }
1178
1179 /*
1180 * We have partitions and re-read the table, the kernel already sent
1181 * out a "change" event for the disk, and "remove/add" for all
1182 * partitions.
1183 */
1184 if (part_table_read && has_partitions)
1185 return 0;
1186
1187 /*
1188 * We have partitions but re-reading the partition table did not
1189 * work, synthesize "change" for the disk and all partitions.
1190 */
1191 (void) synthesize_change_one(dev, syspath);
1192
1193 FOREACH_DEVICE(e, d) {
1194 const char *t, *n, *s;
1195
1196 if (sd_device_get_devtype(d, &t) < 0 ||
1197 !streq("partition", t))
1198 continue;
1199
1200 if (sd_device_get_devname(d, &n) < 0 ||
1201 sd_device_get_syspath(d, &s) < 0)
1202 continue;
1203
1204 (void) synthesize_change_one(dev, s);
1205 }
1206
1207 } else
1208 (void) synthesize_change_one(dev, syspath);
1209
1210 return 0;
1211 }
1212
1213 static int on_inotify(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1214 Manager *manager = userdata;
1215 union inotify_event_buffer buffer;
1216 struct inotify_event *e;
1217 ssize_t l;
1218 int r;
1219
1220 assert(manager);
1221
1222 r = event_source_disable(manager->kill_workers_event);
1223 if (r < 0)
1224 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1225
1226 l = read(fd, &buffer, sizeof(buffer));
1227 if (l < 0) {
1228 if (IN_SET(errno, EAGAIN, EINTR))
1229 return 1;
1230
1231 return log_error_errno(errno, "Failed to read inotify fd: %m");
1232 }
1233
1234 FOREACH_INOTIFY_EVENT(e, buffer, l) {
1235 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
1236 const char *devnode;
1237
1238 if (udev_watch_lookup(e->wd, &dev) <= 0)
1239 continue;
1240
1241 if (sd_device_get_devname(dev, &devnode) < 0)
1242 continue;
1243
1244 log_device_debug(dev, "Inotify event: %x for %s", e->mask, devnode);
1245 if (e->mask & IN_CLOSE_WRITE)
1246 synthesize_change(dev);
1247 else if (e->mask & IN_IGNORED)
1248 udev_watch_end(dev);
1249 }
1250
1251 return 1;
1252 }
1253
1254 static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1255 Manager *manager = userdata;
1256
1257 assert(manager);
1258
1259 manager_exit(manager);
1260
1261 return 1;
1262 }
1263
1264 static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1265 Manager *manager = userdata;
1266
1267 assert(manager);
1268
1269 manager_reload(manager);
1270
1271 return 1;
1272 }
1273
1274 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1275 Manager *manager = userdata;
1276 int r;
1277
1278 assert(manager);
1279
1280 for (;;) {
1281 pid_t pid;
1282 int status;
1283 struct worker *worker;
1284
1285 pid = waitpid(-1, &status, WNOHANG);
1286 if (pid <= 0)
1287 break;
1288
1289 worker = hashmap_get(manager->workers, PID_TO_PTR(pid));
1290 if (!worker) {
1291 log_warning("Worker ["PID_FMT"] is unknown, ignoring", pid);
1292 continue;
1293 }
1294
1295 if (WIFEXITED(status)) {
1296 if (WEXITSTATUS(status) == 0)
1297 log_debug("Worker ["PID_FMT"] exited", pid);
1298 else
1299 log_warning("Worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status));
1300 } else if (WIFSIGNALED(status))
1301 log_warning("Worker ["PID_FMT"] terminated by signal %i (%s)", pid, WTERMSIG(status), signal_to_string(WTERMSIG(status)));
1302 else if (WIFSTOPPED(status)) {
1303 log_info("Worker ["PID_FMT"] stopped", pid);
1304 continue;
1305 } else if (WIFCONTINUED(status)) {
1306 log_info("Worker ["PID_FMT"] continued", pid);
1307 continue;
1308 } else
1309 log_warning("Worker ["PID_FMT"] exit with status 0x%04x", pid, status);
1310
1311 if ((!WIFEXITED(status) || WEXITSTATUS(status) != 0) && worker->event) {
1312 log_device_error(worker->event->dev, "Worker ["PID_FMT"] failed", pid);
1313
1314 /* delete state from disk */
1315 device_delete_db(worker->event->dev);
1316 device_tag_index(worker->event->dev, NULL, false);
1317
1318 if (manager->monitor) {
1319 /* forward kernel event without amending it */
1320 r = device_monitor_send_device(manager->monitor, NULL, worker->event->dev_kernel);
1321 if (r < 0)
1322 log_device_error_errno(worker->event->dev_kernel, r, "Failed to send back device to kernel: %m");
1323 }
1324 }
1325
1326 worker_free(worker);
1327 }
1328
1329 /* we can start new workers, try to schedule events */
1330 event_queue_start(manager);
1331
1332 /* Disable unnecessary cleanup event */
1333 if (hashmap_isempty(manager->workers)) {
1334 r = event_source_disable(manager->kill_workers_event);
1335 if (r < 0)
1336 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1337 }
1338
1339 return 1;
1340 }
1341
1342 static int on_post(sd_event_source *s, void *userdata) {
1343 Manager *manager = userdata;
1344
1345 assert(manager);
1346
1347 if (!LIST_IS_EMPTY(manager->events))
1348 return 1;
1349
1350 /* There are no pending events. Let's cleanup idle process. */
1351
1352 if (!hashmap_isempty(manager->workers)) {
1353 /* There are idle workers */
1354 (void) event_reset_time(manager->event, &manager->kill_workers_event, CLOCK_MONOTONIC,
1355 now(CLOCK_MONOTONIC) + 3 * USEC_PER_SEC, USEC_PER_SEC,
1356 on_kill_workers_event, manager, 0, "kill-workers-event", false);
1357 return 1;
1358 }
1359
1360 /* There are no idle workers. */
1361
1362 if (manager->exit)
1363 return sd_event_exit(manager->event, 0);
1364
1365 if (manager->cgroup)
1366 /* cleanup possible left-over processes in our cgroup */
1367 (void) cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, CGROUP_IGNORE_SELF, NULL, NULL, NULL);
1368
1369 return 1;
1370 }
1371
1372 static int listen_fds(int *ret_ctrl, int *ret_netlink) {
1373 int ctrl_fd = -1, netlink_fd = -1;
1374 int fd, n;
1375
1376 assert(ret_ctrl);
1377 assert(ret_netlink);
1378
1379 n = sd_listen_fds(true);
1380 if (n < 0)
1381 return n;
1382
1383 for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
1384 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1) > 0) {
1385 if (ctrl_fd >= 0)
1386 return -EINVAL;
1387 ctrl_fd = fd;
1388 continue;
1389 }
1390
1391 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1) > 0) {
1392 if (netlink_fd >= 0)
1393 return -EINVAL;
1394 netlink_fd = fd;
1395 continue;
1396 }
1397
1398 return -EINVAL;
1399 }
1400
1401 *ret_ctrl = ctrl_fd;
1402 *ret_netlink = netlink_fd;
1403
1404 return 0;
1405 }
1406
1407 /*
1408 * read the kernel command line, in case we need to get into debug mode
1409 * udev.log_priority=<level> syslog priority
1410 * udev.children_max=<number of workers> events are fully serialized if set to 1
1411 * udev.exec_delay=<number of seconds> delay execution of every executed program
1412 * udev.event_timeout=<number of seconds> seconds to wait before terminating an event
1413 */
1414 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
1415 int r = 0;
1416
1417 assert(key);
1418
1419 if (!value)
1420 return 0;
1421
1422 if (proc_cmdline_key_streq(key, "udev.log_priority")) {
1423
1424 if (proc_cmdline_value_missing(key, value))
1425 return 0;
1426
1427 r = log_level_from_string(value);
1428 if (r >= 0)
1429 log_set_max_level(r);
1430
1431 } else if (proc_cmdline_key_streq(key, "udev.event_timeout")) {
1432
1433 if (proc_cmdline_value_missing(key, value))
1434 return 0;
1435
1436 r = parse_sec(value, &arg_event_timeout_usec);
1437
1438 } else if (proc_cmdline_key_streq(key, "udev.children_max")) {
1439
1440 if (proc_cmdline_value_missing(key, value))
1441 return 0;
1442
1443 r = safe_atou(value, &arg_children_max);
1444
1445 } else if (proc_cmdline_key_streq(key, "udev.exec_delay")) {
1446
1447 if (proc_cmdline_value_missing(key, value))
1448 return 0;
1449
1450 r = parse_sec(value, &arg_exec_delay_usec);
1451
1452 } else if (startswith(key, "udev."))
1453 log_warning("Unknown udev kernel command line option \"%s\", ignoring", key);
1454
1455 if (r < 0)
1456 log_warning_errno(r, "Failed to parse \"%s=%s\", ignoring: %m", key, value);
1457
1458 return 0;
1459 }
1460
1461 static int help(void) {
1462 _cleanup_free_ char *link = NULL;
1463 int r;
1464
1465 r = terminal_urlify_man("systemd-udevd.service", "8", &link);
1466 if (r < 0)
1467 return log_oom();
1468
1469 printf("%s [OPTIONS...]\n\n"
1470 "Manages devices.\n\n"
1471 " -h --help Print this message\n"
1472 " -V --version Print version of the program\n"
1473 " -d --daemon Detach and run in the background\n"
1474 " -D --debug Enable debug output\n"
1475 " -c --children-max=INT Set maximum number of workers\n"
1476 " -e --exec-delay=SECONDS Seconds to wait before executing RUN=\n"
1477 " -t --event-timeout=SECONDS Seconds to wait before terminating an event\n"
1478 " -N --resolve-names=early|late|never\n"
1479 " When to resolve users and groups\n"
1480 "\nSee the %s for details.\n"
1481 , program_invocation_short_name
1482 , link
1483 );
1484
1485 return 0;
1486 }
1487
1488 static int parse_argv(int argc, char *argv[]) {
1489 static const struct option options[] = {
1490 { "daemon", no_argument, NULL, 'd' },
1491 { "debug", no_argument, NULL, 'D' },
1492 { "children-max", required_argument, NULL, 'c' },
1493 { "exec-delay", required_argument, NULL, 'e' },
1494 { "event-timeout", required_argument, NULL, 't' },
1495 { "resolve-names", required_argument, NULL, 'N' },
1496 { "help", no_argument, NULL, 'h' },
1497 { "version", no_argument, NULL, 'V' },
1498 {}
1499 };
1500
1501 int c, r;
1502
1503 assert(argc >= 0);
1504 assert(argv);
1505
1506 while ((c = getopt_long(argc, argv, "c:de:Dt:N:hV", options, NULL)) >= 0) {
1507 switch (c) {
1508
1509 case 'd':
1510 arg_daemonize = true;
1511 break;
1512 case 'c':
1513 r = safe_atou(optarg, &arg_children_max);
1514 if (r < 0)
1515 log_warning_errno(r, "Failed to parse --children-max= value '%s', ignoring: %m", optarg);
1516 break;
1517 case 'e':
1518 r = parse_sec(optarg, &arg_exec_delay_usec);
1519 if (r < 0)
1520 log_warning_errno(r, "Failed to parse --exec-delay= value '%s', ignoring: %m", optarg);
1521 break;
1522 case 't':
1523 r = parse_sec(optarg, &arg_event_timeout_usec);
1524 if (r < 0)
1525 log_warning_errno(r, "Failed to parse --event-timeout= value '%s', ignoring: %m", optarg);
1526 break;
1527 case 'D':
1528 arg_debug = true;
1529 break;
1530 case 'N': {
1531 ResolveNameTiming t;
1532
1533 t = resolve_name_timing_from_string(optarg);
1534 if (t < 0)
1535 log_warning("Invalid --resolve-names= value '%s', ignoring.", optarg);
1536 else
1537 arg_resolve_name_timing = t;
1538 break;
1539 }
1540 case 'h':
1541 return help();
1542 case 'V':
1543 printf("%s\n", GIT_VERSION);
1544 return 0;
1545 case '?':
1546 return -EINVAL;
1547 default:
1548 assert_not_reached("Unhandled option");
1549
1550 }
1551 }
1552
1553 return 1;
1554 }
1555
1556 static int manager_new(Manager **ret, int fd_ctrl, int fd_uevent, const char *cgroup) {
1557 _cleanup_(manager_freep) Manager *manager = NULL;
1558 int r;
1559
1560 assert(ret);
1561
1562 manager = new(Manager, 1);
1563 if (!manager)
1564 return log_oom();
1565
1566 *manager = (Manager) {
1567 .fd_inotify = -1,
1568 .worker_watch = { -1, -1 },
1569 .cgroup = cgroup,
1570 };
1571
1572 r = udev_ctrl_new_from_fd(&manager->ctrl, fd_ctrl);
1573 if (r < 0)
1574 return log_error_errno(r, "Failed to initialize udev control socket: %m");
1575
1576 r = udev_ctrl_enable_receiving(manager->ctrl);
1577 if (r < 0)
1578 return log_error_errno(r, "Failed to bind udev control socket: %m");
1579
1580 r = device_monitor_new_full(&manager->monitor, MONITOR_GROUP_KERNEL, fd_uevent);
1581 if (r < 0)
1582 return log_error_errno(r, "Failed to initialize device monitor: %m");
1583
1584 /* Bump receiver buffer, but only if we are not called via socket activation, as in that
1585 * case systemd sets the receive buffer size for us, and the value in the .socket unit
1586 * should take full effect. */
1587 if (fd_uevent < 0)
1588 (void) sd_device_monitor_set_receive_buffer_size(manager->monitor, 128 * 1024 * 1024);
1589
1590 r = device_monitor_enable_receiving(manager->monitor);
1591 if (r < 0)
1592 return log_error_errno(r, "Failed to bind netlink socket: %m");
1593
1594 *ret = TAKE_PTR(manager);
1595
1596 return 0;
1597 }
1598
1599 static int main_loop(Manager *manager) {
1600 int fd_worker, r;
1601
1602 manager->pid = getpid_cached();
1603
1604 /* unnamed socket from workers to the main daemon */
1605 r = socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, manager->worker_watch);
1606 if (r < 0)
1607 return log_error_errno(errno, "Failed to create socketpair for communicating with workers: %m");
1608
1609 fd_worker = manager->worker_watch[READ_END];
1610
1611 r = setsockopt_int(fd_worker, SOL_SOCKET, SO_PASSCRED, true);
1612 if (r < 0)
1613 return log_error_errno(r, "Failed to enable SO_PASSCRED: %m");
1614
1615 r = udev_watch_init();
1616 if (r < 0)
1617 return log_error_errno(r, "Failed to create inotify descriptor: %m");
1618 manager->fd_inotify = r;
1619
1620 udev_watch_restore();
1621
1622 /* block and listen to all signals on signalfd */
1623 assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, SIGINT, SIGHUP, SIGCHLD, -1) >= 0);
1624
1625 r = sd_event_default(&manager->event);
1626 if (r < 0)
1627 return log_error_errno(r, "Failed to allocate event loop: %m");
1628
1629 r = sd_event_add_signal(manager->event, NULL, SIGINT, on_sigterm, manager);
1630 if (r < 0)
1631 return log_error_errno(r, "Failed to create SIGINT event source: %m");
1632
1633 r = sd_event_add_signal(manager->event, NULL, SIGTERM, on_sigterm, manager);
1634 if (r < 0)
1635 return log_error_errno(r, "Failed to create SIGTERM event source: %m");
1636
1637 r = sd_event_add_signal(manager->event, NULL, SIGHUP, on_sighup, manager);
1638 if (r < 0)
1639 return log_error_errno(r, "Failed to create SIGHUP event source: %m");
1640
1641 r = sd_event_add_signal(manager->event, NULL, SIGCHLD, on_sigchld, manager);
1642 if (r < 0)
1643 return log_error_errno(r, "Failed to create SIGCHLD event source: %m");
1644
1645 r = sd_event_set_watchdog(manager->event, true);
1646 if (r < 0)
1647 return log_error_errno(r, "Failed to create watchdog event source: %m");
1648
1649 r = udev_ctrl_attach_event(manager->ctrl, manager->event);
1650 if (r < 0)
1651 return log_error_errno(r, "Failed to attach event to udev control: %m");
1652
1653 r = udev_ctrl_start(manager->ctrl, on_ctrl_msg, manager);
1654 if (r < 0)
1655 return log_error_errno(r, "Failed to start device monitor: %m");
1656
1657 /* This needs to be after the inotify and uevent handling, to make sure
1658 * that the ping is send back after fully processing the pending uevents
1659 * (including the synthetic ones we may create due to inotify events).
1660 */
1661 r = sd_event_source_set_priority(udev_ctrl_get_event_source(manager->ctrl), SD_EVENT_PRIORITY_IDLE);
1662 if (r < 0)
1663 return log_error_errno(r, "Failed to set IDLE event priority for udev control event source: %m");
1664
1665 r = sd_event_add_io(manager->event, &manager->inotify_event, manager->fd_inotify, EPOLLIN, on_inotify, manager);
1666 if (r < 0)
1667 return log_error_errno(r, "Failed to create inotify event source: %m");
1668
1669 r = sd_device_monitor_attach_event(manager->monitor, manager->event);
1670 if (r < 0)
1671 return log_error_errno(r, "Failed to attach event to device monitor: %m");
1672
1673 r = sd_device_monitor_start(manager->monitor, on_uevent, manager);
1674 if (r < 0)
1675 return log_error_errno(r, "Failed to start device monitor: %m");
1676
1677 (void) sd_event_source_set_description(sd_device_monitor_get_event_source(manager->monitor), "device-monitor");
1678
1679 r = sd_event_add_io(manager->event, NULL, fd_worker, EPOLLIN, on_worker, manager);
1680 if (r < 0)
1681 return log_error_errno(r, "Failed to create worker event source: %m");
1682
1683 r = sd_event_add_post(manager->event, NULL, on_post, manager);
1684 if (r < 0)
1685 return log_error_errno(r, "Failed to create post event source: %m");
1686
1687 udev_builtin_init();
1688
1689 r = udev_rules_new(&manager->rules, arg_resolve_name_timing);
1690 if (!manager->rules)
1691 return log_error_errno(r, "Failed to read udev rules: %m");
1692
1693 r = udev_rules_apply_static_dev_perms(manager->rules);
1694 if (r < 0)
1695 log_error_errno(r, "Failed to apply permissions on static device nodes: %m");
1696
1697 (void) sd_notifyf(false,
1698 "READY=1\n"
1699 "STATUS=Processing with %u children at max", arg_children_max);
1700
1701 r = sd_event_loop(manager->event);
1702 if (r < 0)
1703 log_error_errno(r, "Event loop failed: %m");
1704
1705 sd_notify(false,
1706 "STOPPING=1\n"
1707 "STATUS=Shutting down...");
1708 return r;
1709 }
1710
1711 static int run(int argc, char *argv[]) {
1712 _cleanup_free_ char *cgroup = NULL;
1713 _cleanup_(manager_freep) Manager *manager = NULL;
1714 int fd_ctrl = -1, fd_uevent = -1;
1715 int r;
1716
1717 log_set_target(LOG_TARGET_AUTO);
1718 log_open();
1719 udev_parse_config_full(&arg_children_max, &arg_exec_delay_usec, &arg_event_timeout_usec, &arg_resolve_name_timing);
1720 log_parse_environment();
1721 log_open(); /* Done again to update after reading configuration. */
1722
1723 r = parse_argv(argc, argv);
1724 if (r <= 0)
1725 return r;
1726
1727 r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, PROC_CMDLINE_STRIP_RD_PREFIX);
1728 if (r < 0)
1729 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
1730
1731 if (arg_debug) {
1732 log_set_target(LOG_TARGET_CONSOLE);
1733 log_set_max_level(LOG_DEBUG);
1734 }
1735
1736 log_set_max_level_realm(LOG_REALM_SYSTEMD, log_get_max_level());
1737
1738 r = must_be_root();
1739 if (r < 0)
1740 return r;
1741
1742 if (arg_children_max == 0) {
1743 unsigned long cpu_limit, mem_limit;
1744 unsigned long cpu_count = 1;
1745 cpu_set_t cpu_set;
1746
1747 if (sched_getaffinity(0, sizeof(cpu_set), &cpu_set) == 0)
1748 cpu_count = CPU_COUNT(&cpu_set);
1749
1750 cpu_limit = cpu_count * 2 + 16;
1751 mem_limit = MAX(physical_memory() / (128UL*1024*1024), 10U);
1752
1753 arg_children_max = MIN(cpu_limit, mem_limit);
1754 arg_children_max = MIN(WORKER_NUM_MAX, arg_children_max);
1755
1756 log_debug("Set children_max to %u", arg_children_max);
1757 }
1758
1759 /* set umask before creating any file/directory */
1760 r = chdir("/");
1761 if (r < 0)
1762 return log_error_errno(errno, "Failed to change dir to '/': %m");
1763
1764 umask(022);
1765
1766 r = mac_selinux_init();
1767 if (r < 0)
1768 return log_error_errno(r, "Could not initialize labelling: %m");
1769
1770 r = mkdir_errno_wrapper("/run/udev", 0755);
1771 if (r < 0 && r != -EEXIST)
1772 return log_error_errno(r, "Failed to create /run/udev: %m");
1773
1774 dev_setup(NULL, UID_INVALID, GID_INVALID);
1775
1776 if (getppid() == 1 && sd_booted() > 0) {
1777 /* Get our own cgroup, we regularly kill everything udev has left behind.
1778 * We only do this on systemd systems, and only if we are directly spawned
1779 * by PID1. Otherwise we are not guaranteed to have a dedicated cgroup. */
1780 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
1781 if (r < 0) {
1782 if (IN_SET(r, -ENOENT, -ENOMEDIUM))
1783 log_debug_errno(r, "Dedicated cgroup not found: %m");
1784 else
1785 log_warning_errno(r, "Failed to get cgroup: %m");
1786 }
1787 }
1788
1789 r = listen_fds(&fd_ctrl, &fd_uevent);
1790 if (r < 0)
1791 return log_error_errno(r, "Failed to listen on fds: %m");
1792
1793 r = manager_new(&manager, fd_ctrl, fd_uevent, cgroup);
1794 if (r < 0)
1795 return log_error_errno(r, "Failed to create manager: %m");
1796
1797 if (arg_daemonize) {
1798 pid_t pid;
1799
1800 log_info("Starting version " GIT_VERSION);
1801
1802 /* connect /dev/null to stdin, stdout, stderr */
1803 if (log_get_max_level() < LOG_DEBUG) {
1804 r = make_null_stdio();
1805 if (r < 0)
1806 log_warning_errno(r, "Failed to redirect standard streams to /dev/null: %m");
1807 }
1808
1809 pid = fork();
1810 if (pid < 0)
1811 return log_error_errno(errno, "Failed to fork daemon: %m");
1812 if (pid > 0)
1813 /* parent */
1814 return 0;
1815
1816 /* child */
1817 (void) setsid();
1818
1819 r = set_oom_score_adjust(-1000);
1820 if (r < 0)
1821 log_debug_errno(r, "Failed to adjust OOM score, ignoring: %m");
1822 }
1823
1824 return main_loop(manager);
1825 }
1826
1827 DEFINE_MAIN_FUNCTION(run);
Unnamed repository; edit this file 'description' to name the repository.