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, write to the Free Software Foundation, Inc.,
22 * 675 Mass Ave, Cambridge, MA 02139, USA.
34 #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, *uevq_lockp = &uevq_lock;
57 pthread_cond_t uev_cond, *uev_condp = &uev_cond;
58 uev_trigger *my_uev_trigger;
59 void * my_trigger_data;
62 int is_uevent_busy(void)
66 pthread_mutex_lock(uevq_lockp);
67 empty = list_empty(&uevq);
68 pthread_mutex_unlock(uevq_lockp);
69 return (!empty || servicing_uev);
72 struct uevent * alloc_uevent (void)
74 struct uevent *uev = MALLOC(sizeof(struct uevent));
77 INIT_LIST_HEAD(&uev->node);
83 setup_thread_attr(pthread_attr_t *attr, size_t stacksize, int detached)
85 if (pthread_attr_init(attr)) {
86 fprintf(stderr, "can't initialize thread attr: %s\n",
90 if (stacksize < PTHREAD_STACK_MIN)
91 stacksize = PTHREAD_STACK_MIN;
93 if (pthread_attr_setstacksize(attr, stacksize)) {
94 fprintf(stderr, "can't set thread stack size to %lu: %s\n",
95 (unsigned long)stacksize, strerror(errno));
98 if (detached && pthread_attr_setdetachstate(attr,
99 PTHREAD_CREATE_DETACHED)) {
100 fprintf(stderr, "can't set thread to detached: %s\n",
107 * Called with uevq_lockp held
110 service_uevq(struct list_head *tmpq)
112 struct uevent *uev, *tmp;
114 list_for_each_entry_safe(uev, tmp, tmpq, node) {
115 list_del_init(&uev->node);
117 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
118 condlog(0, "uevent trigger error");
121 udev_device_unref(uev->udev);
126 static void uevq_stop(void *arg)
128 condlog(3, "Stopping uev queue");
129 pthread_mutex_lock(uevq_lockp);
130 my_uev_trigger = NULL;
131 pthread_cond_signal(uev_condp);
132 pthread_mutex_unlock(uevq_lockp);
136 * Service the uevent queue.
138 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
141 my_uev_trigger = uev_trigger;
142 my_trigger_data = trigger_data;
144 mlockall(MCL_CURRENT | MCL_FUTURE);
149 pthread_mutex_lock(uevq_lockp);
152 * Condition signals are unreliable,
153 * so make sure we only wait if we have to.
155 if (list_empty(&uevq)) {
156 pthread_cond_wait(uev_condp, uevq_lockp);
159 list_splice_init(&uevq, &uevq_tmp);
160 pthread_mutex_unlock(uevq_lockp);
163 service_uevq(&uevq_tmp);
165 condlog(3, "Terminating uev service queue");
169 int failback_listen(void)
172 struct sockaddr_nl snl;
173 struct sockaddr_un sun;
176 int rcvbufsz = 128*1024;
178 int rcvszsz = sizeof(rcvsz);
179 unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
180 const int feature_on = 1;
182 * First check whether we have a udev socket
184 memset(&sun, 0x00, sizeof(struct sockaddr_un));
185 sun.sun_family = AF_LOCAL;
186 strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
187 addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
189 sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
192 condlog(3, "reading events from udev socket.");
194 /* the bind takes care of ensuring only one copy running */
195 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
197 condlog(0, "bind failed, exit");
201 /* enable receiving of the sender credentials */
202 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
203 &feature_on, sizeof(feature_on));
206 /* Fallback to read kernel netlink events */
207 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
208 snl.nl_family = AF_NETLINK;
209 snl.nl_pid = getpid();
210 snl.nl_groups = 0x01;
212 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
214 condlog(0, "error getting socket, exit");
218 condlog(3, "reading events from kernel.");
221 * try to avoid dropping uevents, even so, this is not a guarantee,
222 * but it does help to change the netlink uevent socket's
223 * receive buffer threshold from the default value of 106,496 to
224 * the maximum value of 262,142.
226 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
230 condlog(0, "error setting receive buffer size for socket, exit");
233 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
235 condlog(0, "error setting receive buffer size for socket, exit");
238 condlog(3, "receive buffer size for socket is %u.", rcvsz);
240 /* enable receiving of the sender credentials */
241 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
242 &feature_on, sizeof(feature_on));
244 retval = bind(sock, (struct sockaddr *) &snl,
245 sizeof(struct sockaddr_nl));
247 condlog(0, "bind failed, exit");
261 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
262 struct cmsghdr *cmsg;
264 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
266 memset(buf, 0x00, sizeof(buf));
268 iov.iov_len = sizeof(buf);
269 memset (&smsg, 0x00, sizeof(struct msghdr));
272 smsg.msg_control = cred_msg;
273 smsg.msg_controllen = sizeof(cred_msg);
275 buflen = recvmsg(sock, &smsg, 0);
278 condlog(0, "error receiving message, errno %d", errno);
282 cmsg = CMSG_FIRSTHDR(&smsg);
283 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
284 condlog(3, "no sender credentials received, message ignored");
288 cred = (struct ucred *)CMSG_DATA(cmsg);
289 if (cred->uid != 0) {
290 condlog(3, "sender uid=%d, message ignored", cred->uid);
295 bufpos = strlen(buf) + 1;
296 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
297 condlog(3, "invalid message length");
301 /* check message header */
302 if (strstr(buf, "@/") == NULL) {
303 condlog(3, "unrecognized message header");
306 if ((size_t)buflen > sizeof(buf)-1) {
307 condlog(2, "buffer overflow for received uevent");
308 buflen = sizeof(buf)-1;
311 uev = alloc_uevent();
314 condlog(1, "lost uevent, oom");
318 if ((size_t)buflen > sizeof(buf)-1)
319 buflen = sizeof(buf)-1;
322 * Copy the shared receive buffer contents to buffer private
323 * to this uevent so we can immediately reuse the shared buffer.
325 memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
326 buffer = uev->buffer;
327 buffer[buflen] = '\0';
329 /* save start of payload */
330 bufpos = strlen(buffer) + 1;
333 uev->action = buffer;
334 pos = strchr(buffer, '@');
336 condlog(3, "bad action string '%s'", buffer);
342 uev->devpath = &pos[1];
344 /* hotplug events have the environment attached - reconstruct envp[] */
345 for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
349 key = &buffer[bufpos];
350 keylen = strlen(key);
352 bufpos += keylen + 1;
356 condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
357 uev->kernel = strrchr(uev->devpath, '/');
361 /* print payload environment */
362 for (i = 0; uev->envp[i] != NULL; i++)
363 condlog(5, "%s", uev->envp[i]);
366 * Queue uevent and poke service pthread.
368 pthread_mutex_lock(uevq_lockp);
369 list_add_tail(&uev->node, &uevq);
370 pthread_cond_signal(uev_condp);
371 pthread_mutex_unlock(uevq_lockp);
379 int uevent_listen(void)
382 struct udev_monitor *monitor = NULL;
383 int fd, socket_flags;
384 int need_failback = 1;
386 * Queue uevents for service by dedicated thread so that the uevent
387 * listening thread does not block on multipathd locks (vecs->lock)
388 * thereby not getting to empty the socket's receive buffer queue
391 INIT_LIST_HEAD(&uevq);
393 pthread_mutex_init(uevq_lockp, NULL);
394 pthread_cond_init(uev_condp, NULL);
395 pthread_cleanup_push(uevq_stop, NULL);
397 monitor = udev_monitor_new_from_netlink(conf->udev, "udev");
399 condlog(2, "failed to create udev monitor");
402 if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
403 condlog(2, "failed to increase buffer size");
404 fd = udev_monitor_get_fd(monitor);
406 condlog(2, "failed to get monitor fd");
409 socket_flags = fcntl(fd, F_GETFL);
410 if (socket_flags < 0) {
411 condlog(2, "failed to get monitor socket flags : %s",
415 if (fcntl(fd, F_SETFL, socket_flags & ~O_NONBLOCK) < 0) {
416 condlog(2, "failed to set monitor socket flags : %s",
420 err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
423 condlog(2, "failed to create filter : %s\n", strerror(-err));
424 err = udev_monitor_enable_receiving(monitor);
426 condlog(2, "failed to enable receiving : %s\n", strerror(-err));
433 struct udev_device *dev;
434 struct udev_list_entry *list_entry;
436 dev = udev_monitor_receive_device(monitor);
438 condlog(0, "failed getting udev device");
442 uev = alloc_uevent();
444 condlog(1, "lost uevent, oom");
448 end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
449 udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
450 const char *name, *value;
453 name = udev_list_entry_get_name(list_entry);
456 value = udev_list_entry_get_value(list_entry);
459 bytes = snprintf(pos, end - pos, "%s=%s", name,
461 if (pos + bytes >= end) {
462 condlog(2, "buffer overflow for uevent");
469 if (strcmp(name, "DEVPATH") == 0)
470 uev->devpath = uev->envp[i] + 8;
471 if (strcmp(name, "ACTION") == 0)
472 uev->action = uev->envp[i] + 7;
474 if (i == HOTPLUG_NUM_ENVP - 1)
480 condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
481 uev->kernel = strrchr(uev->devpath, '/');
485 /* print payload environment */
486 for (i = 0; uev->envp[i] != NULL; i++)
487 condlog(5, "%s", uev->envp[i]);
490 * Queue uevent and poke service pthread.
492 pthread_mutex_lock(uevq_lockp);
493 list_add_tail(&uev->node, &uevq);
494 pthread_cond_signal(uev_condp);
495 pthread_mutex_unlock(uevq_lockp);
500 udev_monitor_unref(monitor);
502 err = failback_listen();
503 pthread_cleanup_pop(1);
504 pthread_mutex_destroy(uevq_lockp);
505 pthread_cond_destroy(uev_condp);
510 uevent_get_major(struct uevent *uev)
515 for (i = 0; uev->envp[i] != NULL; i++) {
516 if (!strncmp(uev->envp[i], "MAJOR", 5) && strlen(uev->envp[i]) > 6) {
517 p = uev->envp[i] + 6;
518 major = strtoul(p, &q, 10);
520 condlog(2, "invalid major '%s'", p);
530 uevent_get_minor(struct uevent *uev)
535 for (i = 0; uev->envp[i] != NULL; i++) {
536 if (!strncmp(uev->envp[i], "MINOR", 5) && strlen(uev->envp[i]) > 6) {
537 p = uev->envp[i] + 6;
538 minor = strtoul(p, &q, 10);
540 condlog(2, "invalid minor '%s'", p);
550 uevent_get_disk_ro(struct uevent *uev)
555 for (i = 0; uev->envp[i] != NULL; i++) {
556 if (!strncmp(uev->envp[i], "DISK_RO", 6) && strlen(uev->envp[i]) > 7) {
557 p = uev->envp[i] + 8;
558 ro = strtoul(p, &q, 10);
560 condlog(2, "invalid read_only setting '%s'", p);
570 uevent_get_dm_name(struct uevent *uev)
575 for (i = 0; uev->envp[i] != NULL; i++) {
576 if (!strncmp(uev->envp[i], "DM_NAME", 6) &&
577 strlen(uev->envp[i]) > 7) {
578 p = MALLOC(strlen(uev->envp[i] + 8) + 1);
579 strcpy(p, uev->envp[i] + 8);