2 * uevent.c - trigger upon netlink uevents from the kernel
4 * Only kernels from version 2.6.10* on provide the uevent netlink socket.
5 * Until the libc-kernel-headers are updated, you need to compile with:
7 * gcc -I /lib/modules/`uname -r`/build/include -o uevent_listen uevent_listen.c
9 * Copyright (C) 2004 Kay Sievers <kay.sievers@vrfy.org>
11 * This program is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by the
13 * Free Software Foundation version 2 of the License.
15 * This program is distributed in the hope that it will be useful, but
16 * WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public License along
21 * with this program. If not, see <http://www.gnu.org/licenses/>.
33 #include <sys/socket.h>
37 #include <linux/types.h>
38 #include <linux/netlink.h>
52 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
56 pthread_mutex_t uevq_lock = PTHREAD_MUTEX_INITIALIZER;
57 pthread_mutex_t *uevq_lockp = &uevq_lock;
58 pthread_cond_t uev_cond = PTHREAD_COND_INITIALIZER;
59 pthread_cond_t *uev_condp = &uev_cond;
60 uev_trigger *my_uev_trigger;
61 void * my_trigger_data;
64 int is_uevent_busy(void)
68 pthread_mutex_lock(uevq_lockp);
69 empty = list_empty(&uevq);
70 pthread_mutex_unlock(uevq_lockp);
71 return (!empty || servicing_uev);
74 struct uevent * alloc_uevent (void)
76 struct uevent *uev = MALLOC(sizeof(struct uevent));
79 INIT_LIST_HEAD(&uev->node);
85 setup_thread_attr(pthread_attr_t *attr, size_t stacksize, int detached)
87 if (pthread_attr_init(attr)) {
88 fprintf(stderr, "can't initialize thread attr: %s\n",
92 if (stacksize < PTHREAD_STACK_MIN)
93 stacksize = PTHREAD_STACK_MIN;
95 if (pthread_attr_setstacksize(attr, stacksize)) {
96 fprintf(stderr, "can't set thread stack size to %lu: %s\n",
97 (unsigned long)stacksize, strerror(errno));
100 if (detached && pthread_attr_setdetachstate(attr,
101 PTHREAD_CREATE_DETACHED)) {
102 fprintf(stderr, "can't set thread to detached: %s\n",
109 * Called with uevq_lockp held
112 service_uevq(struct list_head *tmpq)
114 struct uevent *uev, *tmp;
116 list_for_each_entry_safe(uev, tmp, tmpq, node) {
117 list_del_init(&uev->node);
119 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
120 condlog(0, "uevent trigger error");
123 udev_device_unref(uev->udev);
128 static void uevq_stop(void *arg)
130 struct udev *udev = arg;
132 condlog(3, "Stopping uev queue");
133 pthread_mutex_lock(uevq_lockp);
134 my_uev_trigger = NULL;
135 pthread_cond_signal(uev_condp);
136 pthread_mutex_unlock(uevq_lockp);
141 uevq_cleanup(struct list_head *tmpq)
143 struct uevent *uev, *tmp;
145 list_for_each_entry_safe(uev, tmp, tmpq, node) {
146 list_del_init(&uev->node);
152 * Service the uevent queue.
154 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
157 my_uev_trigger = uev_trigger;
158 my_trigger_data = trigger_data;
160 mlockall(MCL_CURRENT | MCL_FUTURE);
165 pthread_mutex_lock(uevq_lockp);
168 * Condition signals are unreliable,
169 * so make sure we only wait if we have to.
171 if (list_empty(&uevq)) {
172 pthread_cond_wait(uev_condp, uevq_lockp);
175 list_splice_init(&uevq, &uevq_tmp);
176 pthread_mutex_unlock(uevq_lockp);
179 service_uevq(&uevq_tmp);
181 condlog(3, "Terminating uev service queue");
186 struct uevent *uevent_from_buffer(char *buf, ssize_t buflen)
194 uev = alloc_uevent();
196 condlog(1, "lost uevent, oom");
200 if ((size_t)buflen > sizeof(buf)-1)
201 buflen = sizeof(buf)-1;
204 * Copy the shared receive buffer contents to buffer private
205 * to this uevent so we can immediately reuse the shared buffer.
207 memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
208 buffer = uev->buffer;
209 buffer[buflen] = '\0';
211 /* save start of payload */
212 bufpos = strlen(buffer) + 1;
215 uev->action = buffer;
216 pos = strchr(buffer, '@');
218 condlog(3, "bad action string '%s'", buffer);
225 uev->devpath = &pos[1];
227 /* hotplug events have the environment attached - reconstruct envp[] */
228 for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
232 key = &buffer[bufpos];
233 keylen = strlen(key);
235 /* Filter out sequence number */
236 if (strncmp(key, "SEQNUM=", 7) == 0) {
239 uev->seqnum = strtoul(key + 7, &eptr, 10);
243 bufpos += keylen + 1;
247 condlog(3, "uevent %ld '%s' from '%s'", uev->seqnum,
248 uev->action, uev->devpath);
249 uev->kernel = strrchr(uev->devpath, '/');
253 /* print payload environment */
254 for (i = 0; uev->envp[i] != NULL; i++)
255 condlog(5, "%s", uev->envp[i]);
260 int failback_listen(void)
263 struct sockaddr_nl snl;
264 struct sockaddr_un sun;
267 int rcvbufsz = 128*1024;
269 int rcvszsz = sizeof(rcvsz);
270 unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
271 const int feature_on = 1;
273 * First check whether we have a udev socket
275 memset(&sun, 0x00, sizeof(struct sockaddr_un));
276 sun.sun_family = AF_LOCAL;
277 strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
278 addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
280 sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
283 condlog(3, "reading events from udev socket.");
285 /* the bind takes care of ensuring only one copy running */
286 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
288 condlog(0, "bind failed, exit");
292 /* enable receiving of the sender credentials */
293 retval = setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
294 &feature_on, sizeof(feature_on));
296 condlog(0, "failed to enable credential passing, exit");
301 /* Fallback to read kernel netlink events */
302 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
303 snl.nl_family = AF_NETLINK;
304 snl.nl_pid = getpid();
305 snl.nl_groups = 0x01;
307 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
309 condlog(0, "error getting socket, exit");
313 condlog(3, "reading events from kernel.");
316 * try to avoid dropping uevents, even so, this is not a guarantee,
317 * but it does help to change the netlink uevent socket's
318 * receive buffer threshold from the default value of 106,496 to
319 * the maximum value of 262,142.
321 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
325 condlog(0, "error setting receive buffer size for socket, exit");
328 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
330 condlog(0, "error setting receive buffer size for socket, exit");
333 condlog(3, "receive buffer size for socket is %u.", rcvsz);
335 /* enable receiving of the sender credentials */
336 if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
337 &feature_on, sizeof(feature_on)) < 0) {
338 condlog(0, "error on enabling credential passing for socket");
342 retval = bind(sock, (struct sockaddr *) &snl,
343 sizeof(struct sockaddr_nl));
345 condlog(0, "bind failed, exit");
356 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
357 struct cmsghdr *cmsg;
359 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
361 memset(buf, 0x00, sizeof(buf));
363 iov.iov_len = sizeof(buf);
364 memset (&smsg, 0x00, sizeof(struct msghdr));
367 smsg.msg_control = cred_msg;
368 smsg.msg_controllen = sizeof(cred_msg);
370 buflen = recvmsg(sock, &smsg, 0);
373 condlog(0, "error receiving message, errno %d", errno);
377 cmsg = CMSG_FIRSTHDR(&smsg);
378 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
379 condlog(3, "no sender credentials received, message ignored");
383 cred = (struct ucred *)CMSG_DATA(cmsg);
384 if (cred->uid != 0) {
385 condlog(3, "sender uid=%d, message ignored", cred->uid);
390 bufpos = strlen(buf) + 1;
391 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
392 condlog(3, "invalid message length");
396 /* check message header */
397 if (strstr(buf, "@/") == NULL) {
398 condlog(3, "unrecognized message header");
401 if ((size_t)buflen > sizeof(buf)-1) {
402 condlog(2, "buffer overflow for received uevent");
403 buflen = sizeof(buf)-1;
406 uev = uevent_from_buffer(buf, buflen);
410 * Queue uevent and poke service pthread.
412 pthread_mutex_lock(uevq_lockp);
413 list_add_tail(&uev->node, &uevq);
414 pthread_cond_signal(uev_condp);
415 pthread_mutex_unlock(uevq_lockp);
423 struct uevent *uevent_from_udev_device(struct udev_device *dev)
428 struct udev_list_entry *list_entry;
430 uev = alloc_uevent();
432 udev_device_unref(dev);
433 condlog(1, "lost uevent, oom");
437 end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
438 udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
439 const char *name, *value;
442 name = udev_list_entry_get_name(list_entry);
445 value = udev_list_entry_get_value(list_entry);
448 bytes = snprintf(pos, end - pos, "%s=%s", name, value);
449 if (pos + bytes >= end) {
450 condlog(2, "buffer overflow for uevent");
457 if (strcmp(name, "DEVPATH") == 0)
458 uev->devpath = uev->envp[i] + 8;
459 if (strcmp(name, "ACTION") == 0)
460 uev->action = uev->envp[i] + 7;
462 if (i == HOTPLUG_NUM_ENVP - 1)
468 condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
469 uev->kernel = strrchr(uev->devpath, '/');
473 /* print payload environment */
474 for (i = 0; uev->envp[i] != NULL; i++)
475 condlog(5, "%s", uev->envp[i]);
479 int uevent_listen(struct udev *udev)
482 struct udev_monitor *monitor = NULL;
483 int fd, socket_flags, events;
484 int need_failback = 1;
486 LIST_HEAD(uevlisten_tmp);
489 * Queue uevents for service by dedicated thread so that the uevent
490 * listening thread does not block on multipathd locks (vecs->lock)
491 * thereby not getting to empty the socket's receive buffer queue
495 condlog(1, "no udev context");
499 pthread_cleanup_push(uevq_stop, udev);
501 monitor = udev_monitor_new_from_netlink(udev, "udev");
503 condlog(2, "failed to create udev monitor");
506 #ifdef LIBUDEV_API_RECVBUF
507 if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
508 condlog(2, "failed to increase buffer size");
510 fd = udev_monitor_get_fd(monitor);
512 condlog(2, "failed to get monitor fd");
515 socket_flags = fcntl(fd, F_GETFL);
516 if (socket_flags < 0) {
517 condlog(2, "failed to get monitor socket flags : %s",
521 if (fcntl(fd, F_SETFL, socket_flags & ~O_NONBLOCK) < 0) {
522 condlog(2, "failed to set monitor socket flags : %s",
526 err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
529 condlog(2, "failed to create filter : %s", strerror(-err));
530 err = udev_monitor_enable_receiving(monitor);
532 condlog(2, "failed to enable receiving : %s", strerror(-err));
539 struct udev_device *dev;
540 struct pollfd ev_poll;
544 memset(&ev_poll, 0, sizeof(struct pollfd));
546 ev_poll.events = POLLIN;
547 poll_timeout = timeout * 1000;
549 fdcount = poll(&ev_poll, 1, poll_timeout);
550 if (fdcount && ev_poll.revents & POLLIN) {
552 dev = udev_monitor_receive_device(monitor);
554 condlog(0, "failed getting udev device");
557 uev = uevent_from_udev_device(dev);
560 list_add_tail(&uev->node, &uevlisten_tmp);
568 condlog(0, "error receiving "
569 "uevent message: %m");
573 if (!list_empty(&uevlisten_tmp)) {
575 * Queue uevents and poke service pthread.
577 condlog(3, "Forwarding %d uevents", events);
578 pthread_mutex_lock(uevq_lockp);
579 list_splice_tail_init(&uevlisten_tmp, &uevq);
580 pthread_cond_signal(uev_condp);
581 pthread_mutex_unlock(uevq_lockp);
589 udev_monitor_unref(monitor);
591 err = failback_listen();
592 pthread_cleanup_pop(1);
597 uevent_get_major(struct uevent *uev)
602 for (i = 0; uev->envp[i] != NULL; i++) {
603 if (!strncmp(uev->envp[i], "MAJOR", 5) && strlen(uev->envp[i]) > 6) {
604 p = uev->envp[i] + 6;
605 major = strtoul(p, &q, 10);
607 condlog(2, "invalid major '%s'", p);
617 uevent_get_minor(struct uevent *uev)
622 for (i = 0; uev->envp[i] != NULL; i++) {
623 if (!strncmp(uev->envp[i], "MINOR", 5) && strlen(uev->envp[i]) > 6) {
624 p = uev->envp[i] + 6;
625 minor = strtoul(p, &q, 10);
627 condlog(2, "invalid minor '%s'", p);
637 uevent_get_disk_ro(struct uevent *uev)
642 for (i = 0; uev->envp[i] != NULL; i++) {
643 if (!strncmp(uev->envp[i], "DISK_RO", 6) && strlen(uev->envp[i]) > 7) {
644 p = uev->envp[i] + 8;
645 ro = strtoul(p, &q, 10);
647 condlog(2, "invalid read_only setting '%s'", p);
657 uevent_get_dm_name(struct uevent *uev)
662 for (i = 0; uev->envp[i] != NULL; i++) {
663 if (!strncmp(uev->envp[i], "DM_NAME", 6) &&
664 strlen(uev->envp[i]) > 7) {
665 p = MALLOC(strlen(uev->envp[i] + 8) + 1);
666 strcpy(p, uev->envp[i] + 8);