2 This file is part of systemd.
4 Copyright 2008-2012 Kay Sievers <kay@vrfy.org>
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
29 #include <sys/socket.h>
31 #include <arpa/inet.h>
32 #include <linux/netlink.h>
33 #include <linux/filter.h>
36 #include "libudev-private.h"
37 #include "socket-util.h"
41 * SECTION:libudev-monitor
42 * @short_description: device event source
44 * Connects to a device event source.
50 * Opaque object handling an event source.
56 union sockaddr_union snl;
57 union sockaddr_union snl_trusted_sender;
58 union sockaddr_union snl_destination;
60 struct udev_list filter_subsystem_list;
61 struct udev_list filter_tag_list;
65 enum udev_monitor_netlink_group {
71 #define UDEV_MONITOR_MAGIC 0xfeedcafe
72 struct udev_monitor_netlink_header {
73 /* "libudev" prefix to distinguish libudev and kernel messages */
76 * magic to protect against daemon <-> library message format mismatch
77 * used in the kernel from socket filter rules; needs to be stored in network order
80 /* total length of header structure known to the sender */
81 unsigned int header_size;
82 /* properties string buffer */
83 unsigned int properties_off;
84 unsigned int properties_len;
86 * hashes of primary device properties strings, to let libudev subscribers
87 * use in-kernel socket filters; values need to be stored in network order
89 unsigned int filter_subsystem_hash;
90 unsigned int filter_devtype_hash;
91 unsigned int filter_tag_bloom_hi;
92 unsigned int filter_tag_bloom_lo;
95 static struct udev_monitor *udev_monitor_new(struct udev *udev)
97 struct udev_monitor *udev_monitor;
99 udev_monitor = new0(struct udev_monitor, 1);
100 if (udev_monitor == NULL)
102 udev_monitor->refcount = 1;
103 udev_monitor->udev = udev;
104 udev_list_init(udev, &udev_monitor->filter_subsystem_list, false);
105 udev_list_init(udev, &udev_monitor->filter_tag_list, true);
109 /* we consider udev running when /dev is on devtmpfs */
110 static bool udev_has_devtmpfs(struct udev *udev) {
112 union file_handle_union h = FILE_HANDLE_INIT;
113 _cleanup_fclose_ FILE *f = NULL;
114 char line[LINE_MAX], *e;
118 r = name_to_handle_at(AT_FDCWD, "/dev", &h.handle, &mount_id, 0);
120 if (errno != EOPNOTSUPP)
121 log_debug_errno(errno, "name_to_handle_at on /dev: %m");
125 f = fopen("/proc/self/mountinfo", "re");
129 FOREACH_LINE(line, f, return false) {
132 if (sscanf(line, "%i", &mid) != 1)
138 e = strstr(line, " - ");
142 /* accept any name that starts with the currently expected type */
143 if (startswith(e + 3, "devtmpfs"))
150 struct udev_monitor *udev_monitor_new_from_netlink_fd(struct udev *udev, const char *name, int fd)
152 struct udev_monitor *udev_monitor;
159 group = UDEV_MONITOR_NONE;
160 else if (streq(name, "udev")) {
162 * We do not support subscribing to uevents if no instance of
163 * udev is running. Uevents would otherwise broadcast the
164 * processing data of the host into containers, which is not
167 * Containers will currently not get any udev uevents, until
168 * a supporting infrastructure is available.
170 * We do not set a netlink multicast group here, so the socket
171 * will not receive any messages.
173 if (access("/run/udev/control", F_OK) < 0 && !udev_has_devtmpfs(udev)) {
174 log_debug("the udev service seems not to be active, disable the monitor");
175 group = UDEV_MONITOR_NONE;
177 group = UDEV_MONITOR_UDEV;
178 } else if (streq(name, "kernel"))
179 group = UDEV_MONITOR_KERNEL;
183 udev_monitor = udev_monitor_new(udev);
184 if (udev_monitor == NULL)
188 udev_monitor->sock = socket(PF_NETLINK, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, NETLINK_KOBJECT_UEVENT);
189 if (udev_monitor->sock == -1) {
190 log_debug_errno(errno, "error getting socket: %m");
195 udev_monitor->bound = true;
196 udev_monitor->sock = fd;
199 udev_monitor->snl.nl.nl_family = AF_NETLINK;
200 udev_monitor->snl.nl.nl_groups = group;
202 /* default destination for sending */
203 udev_monitor->snl_destination.nl.nl_family = AF_NETLINK;
204 udev_monitor->snl_destination.nl.nl_groups = UDEV_MONITOR_UDEV;
210 * udev_monitor_new_from_netlink:
211 * @udev: udev library context
212 * @name: name of event source
214 * Create new udev monitor and connect to a specified event
215 * source. Valid sources identifiers are "udev" and "kernel".
217 * Applications should usually not connect directly to the
218 * "kernel" events, because the devices might not be useable
219 * at that time, before udev has configured them, and created
220 * device nodes. Accessing devices at the same time as udev,
221 * might result in unpredictable behavior. The "udev" events
222 * are sent out after udev has finished its event processing,
223 * all rules have been processed, and needed device nodes are
226 * The initial refcount is 1, and needs to be decremented to
227 * release the resources of the udev monitor.
229 * Returns: a new udev monitor, or #NULL, in case of an error
231 _public_ struct udev_monitor *udev_monitor_new_from_netlink(struct udev *udev, const char *name)
233 return udev_monitor_new_from_netlink_fd(udev, name, -1);
236 static inline void bpf_stmt(struct sock_filter *inss, unsigned int *i,
237 unsigned short code, unsigned int data)
239 struct sock_filter *ins = &inss[*i];
246 static inline void bpf_jmp(struct sock_filter *inss, unsigned int *i,
247 unsigned short code, unsigned int data,
248 unsigned short jt, unsigned short jf)
250 struct sock_filter *ins = &inss[*i];
260 * udev_monitor_filter_update:
261 * @udev_monitor: monitor
263 * Update the installed socket filter. This is only needed,
264 * if the filter was removed or changed.
266 * Returns: 0 on success, otherwise a negative error value.
268 _public_ int udev_monitor_filter_update(struct udev_monitor *udev_monitor)
270 struct sock_filter ins[512];
271 struct sock_fprog filter;
273 struct udev_list_entry *list_entry;
276 if (udev_list_get_entry(&udev_monitor->filter_subsystem_list) == NULL &&
277 udev_list_get_entry(&udev_monitor->filter_tag_list) == NULL)
280 memzero(ins, sizeof(ins));
283 /* load magic in A */
284 bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, offsetof(struct udev_monitor_netlink_header, magic));
285 /* jump if magic matches */
286 bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, UDEV_MONITOR_MAGIC, 1, 0);
287 /* wrong magic, pass packet */
288 bpf_stmt(ins, &i, BPF_RET|BPF_K, 0xffffffff);
290 if (udev_list_get_entry(&udev_monitor->filter_tag_list) != NULL) {
293 /* count tag matches, to calculate end of tag match block */
295 udev_list_entry_foreach(list_entry, udev_list_get_entry(&udev_monitor->filter_tag_list))
298 /* add all tags matches */
299 udev_list_entry_foreach(list_entry, udev_list_get_entry(&udev_monitor->filter_tag_list)) {
300 uint64_t tag_bloom_bits = util_string_bloom64(udev_list_entry_get_name(list_entry));
301 uint32_t tag_bloom_hi = tag_bloom_bits >> 32;
302 uint32_t tag_bloom_lo = tag_bloom_bits & 0xffffffff;
304 /* load device bloom bits in A */
305 bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, offsetof(struct udev_monitor_netlink_header, filter_tag_bloom_hi));
306 /* clear bits (tag bits & bloom bits) */
307 bpf_stmt(ins, &i, BPF_ALU|BPF_AND|BPF_K, tag_bloom_hi);
308 /* jump to next tag if it does not match */
309 bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, tag_bloom_hi, 0, 3);
311 /* load device bloom bits in A */
312 bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, offsetof(struct udev_monitor_netlink_header, filter_tag_bloom_lo));
313 /* clear bits (tag bits & bloom bits) */
314 bpf_stmt(ins, &i, BPF_ALU|BPF_AND|BPF_K, tag_bloom_lo);
315 /* jump behind end of tag match block if tag matches */
317 bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, tag_bloom_lo, 1 + (tag_matches * 6), 0);
320 /* nothing matched, drop packet */
321 bpf_stmt(ins, &i, BPF_RET|BPF_K, 0);
324 /* add all subsystem matches */
325 if (udev_list_get_entry(&udev_monitor->filter_subsystem_list) != NULL) {
326 udev_list_entry_foreach(list_entry, udev_list_get_entry(&udev_monitor->filter_subsystem_list)) {
327 unsigned int hash = util_string_hash32(udev_list_entry_get_name(list_entry));
329 /* load device subsystem value in A */
330 bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, offsetof(struct udev_monitor_netlink_header, filter_subsystem_hash));
331 if (udev_list_entry_get_value(list_entry) == NULL) {
332 /* jump if subsystem does not match */
333 bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, hash, 0, 1);
335 /* jump if subsystem does not match */
336 bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, hash, 0, 3);
338 /* load device devtype value in A */
339 bpf_stmt(ins, &i, BPF_LD|BPF_W|BPF_ABS, offsetof(struct udev_monitor_netlink_header, filter_devtype_hash));
340 /* jump if value does not match */
341 hash = util_string_hash32(udev_list_entry_get_value(list_entry));
342 bpf_jmp(ins, &i, BPF_JMP|BPF_JEQ|BPF_K, hash, 0, 1);
345 /* matched, pass packet */
346 bpf_stmt(ins, &i, BPF_RET|BPF_K, 0xffffffff);
348 if (i+1 >= ELEMENTSOF(ins))
352 /* nothing matched, drop packet */
353 bpf_stmt(ins, &i, BPF_RET|BPF_K, 0);
356 /* matched, pass packet */
357 bpf_stmt(ins, &i, BPF_RET|BPF_K, 0xffffffff);
360 memzero(&filter, sizeof(filter));
363 err = setsockopt(udev_monitor->sock, SOL_SOCKET, SO_ATTACH_FILTER, &filter, sizeof(filter));
364 return err < 0 ? -errno : 0;
367 int udev_monitor_allow_unicast_sender(struct udev_monitor *udev_monitor, struct udev_monitor *sender)
369 udev_monitor->snl_trusted_sender.nl.nl_pid = sender->snl.nl.nl_pid;
373 * udev_monitor_enable_receiving:
374 * @udev_monitor: the monitor which should receive events
376 * Binds the @udev_monitor socket to the event source.
378 * Returns: 0 on success, otherwise a negative error value.
380 _public_ int udev_monitor_enable_receiving(struct udev_monitor *udev_monitor)
385 udev_monitor_filter_update(udev_monitor);
387 if (!udev_monitor->bound) {
388 err = bind(udev_monitor->sock,
389 &udev_monitor->snl.sa, sizeof(struct sockaddr_nl));
391 udev_monitor->bound = true;
395 union sockaddr_union snl;
399 * get the address the kernel has assigned us
400 * it is usually, but not necessarily the pid
402 addrlen = sizeof(struct sockaddr_nl);
403 err = getsockname(udev_monitor->sock, &snl.sa, &addrlen);
405 udev_monitor->snl.nl.nl_pid = snl.nl.nl_pid;
407 log_debug_errno(errno, "bind failed: %m");
411 /* enable receiving of sender credentials */
412 err = setsockopt(udev_monitor->sock, SOL_SOCKET, SO_PASSCRED, &on, sizeof(on));
414 log_debug_errno(errno, "setting SO_PASSCRED failed: %m");
420 * udev_monitor_set_receive_buffer_size:
421 * @udev_monitor: the monitor which should receive events
422 * @size: the size in bytes
424 * Set the size of the kernel socket buffer. This call needs the
425 * appropriate privileges to succeed.
427 * Returns: 0 on success, otherwise -1 on error.
429 _public_ int udev_monitor_set_receive_buffer_size(struct udev_monitor *udev_monitor, int size)
431 if (udev_monitor == NULL)
433 return setsockopt(udev_monitor->sock, SOL_SOCKET, SO_RCVBUFFORCE, &size, sizeof(size));
436 int udev_monitor_disconnect(struct udev_monitor *udev_monitor)
440 err = close(udev_monitor->sock);
441 udev_monitor->sock = -1;
442 return err < 0 ? -errno : 0;
447 * @udev_monitor: udev monitor
449 * Take a reference of a udev monitor.
451 * Returns: the passed udev monitor
453 _public_ struct udev_monitor *udev_monitor_ref(struct udev_monitor *udev_monitor)
455 if (udev_monitor == NULL)
457 udev_monitor->refcount++;
462 * udev_monitor_unref:
463 * @udev_monitor: udev monitor
465 * Drop a reference of a udev monitor. If the refcount reaches zero,
466 * the bound socket will be closed, and the resources of the monitor
471 _public_ struct udev_monitor *udev_monitor_unref(struct udev_monitor *udev_monitor)
473 if (udev_monitor == NULL)
475 udev_monitor->refcount--;
476 if (udev_monitor->refcount > 0)
478 if (udev_monitor->sock >= 0)
479 close(udev_monitor->sock);
480 udev_list_cleanup(&udev_monitor->filter_subsystem_list);
481 udev_list_cleanup(&udev_monitor->filter_tag_list);
487 * udev_monitor_get_udev:
488 * @udev_monitor: udev monitor
490 * Retrieve the udev library context the monitor was created with.
492 * Returns: the udev library context
494 _public_ struct udev *udev_monitor_get_udev(struct udev_monitor *udev_monitor)
496 if (udev_monitor == NULL)
498 return udev_monitor->udev;
502 * udev_monitor_get_fd:
503 * @udev_monitor: udev monitor
505 * Retrieve the socket file descriptor associated with the monitor.
507 * Returns: the socket file descriptor
509 _public_ int udev_monitor_get_fd(struct udev_monitor *udev_monitor)
511 if (udev_monitor == NULL)
513 return udev_monitor->sock;
516 static int passes_filter(struct udev_monitor *udev_monitor, struct udev_device *udev_device)
518 struct udev_list_entry *list_entry;
520 if (udev_list_get_entry(&udev_monitor->filter_subsystem_list) == NULL)
522 udev_list_entry_foreach(list_entry, udev_list_get_entry(&udev_monitor->filter_subsystem_list)) {
523 const char *subsys = udev_list_entry_get_name(list_entry);
524 const char *dsubsys = udev_device_get_subsystem(udev_device);
526 const char *ddevtype;
528 if (!streq(dsubsys, subsys))
531 devtype = udev_list_entry_get_value(list_entry);
534 ddevtype = udev_device_get_devtype(udev_device);
535 if (ddevtype == NULL)
537 if (streq(ddevtype, devtype))
543 if (udev_list_get_entry(&udev_monitor->filter_tag_list) == NULL)
545 udev_list_entry_foreach(list_entry, udev_list_get_entry(&udev_monitor->filter_tag_list)) {
546 const char *tag = udev_list_entry_get_name(list_entry);
548 if (udev_device_has_tag(udev_device, tag))
555 * udev_monitor_receive_device:
556 * @udev_monitor: udev monitor
558 * Receive data from the udev monitor socket, allocate a new udev
559 * device, fill in the received data, and return the device.
561 * Only socket connections with uid=0 are accepted.
563 * The monitor socket is by default set to NONBLOCK. A variant of poll() on
564 * the file descriptor returned by udev_monitor_get_fd() should to be used to
565 * wake up when new devices arrive, or alternatively the file descriptor
566 * switched into blocking mode.
568 * The initial refcount is 1, and needs to be decremented to
569 * release the resources of the udev device.
571 * Returns: a new udev device, or #NULL, in case of an error
573 _public_ struct udev_device *udev_monitor_receive_device(struct udev_monitor *udev_monitor)
575 struct udev_device *udev_device;
578 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
579 struct cmsghdr *cmsg;
580 union sockaddr_union snl;
583 struct udev_monitor_netlink_header nlh;
588 bool is_initialized = false;
591 if (udev_monitor == NULL)
594 iov.iov_len = sizeof(buf);
595 memzero(&smsg, sizeof(struct msghdr));
598 smsg.msg_control = cred_msg;
599 smsg.msg_controllen = sizeof(cred_msg);
600 smsg.msg_name = &snl;
601 smsg.msg_namelen = sizeof(snl);
603 buflen = recvmsg(udev_monitor->sock, &smsg, 0);
606 log_debug("unable to receive message");
610 if (buflen < 32 || (smsg.msg_flags & MSG_TRUNC)) {
611 log_debug("invalid message length");
615 if (snl.nl.nl_groups == 0) {
616 /* unicast message, check if we trust the sender */
617 if (udev_monitor->snl_trusted_sender.nl.nl_pid == 0 ||
618 snl.nl.nl_pid != udev_monitor->snl_trusted_sender.nl.nl_pid) {
619 log_debug("unicast netlink message ignored");
622 } else if (snl.nl.nl_groups == UDEV_MONITOR_KERNEL) {
623 if (snl.nl.nl_pid > 0) {
624 log_debug("multicast kernel netlink message from PID %"PRIu32" ignored",
630 cmsg = CMSG_FIRSTHDR(&smsg);
631 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
632 log_debug("no sender credentials received, message ignored");
636 cred = (struct ucred *)CMSG_DATA(cmsg);
637 if (cred->uid != 0) {
638 log_debug("sender uid="UID_FMT", message ignored", cred->uid);
642 if (memcmp(buf.raw, "libudev", 8) == 0) {
643 /* udev message needs proper version magic */
644 if (buf.nlh.magic != htonl(UDEV_MONITOR_MAGIC)) {
645 log_debug("unrecognized message signature (%x != %x)",
646 buf.nlh.magic, htonl(UDEV_MONITOR_MAGIC));
649 if (buf.nlh.properties_off+32 > (size_t)buflen) {
653 bufpos = buf.nlh.properties_off;
655 /* devices received from udev are always initialized */
656 is_initialized = true;
658 /* kernel message with header */
659 bufpos = strlen(buf.raw) + 1;
660 if ((size_t)bufpos < sizeof("a@/d") || bufpos >= buflen) {
661 log_debug("invalid message length");
665 /* check message header */
666 if (strstr(buf.raw, "@/") == NULL) {
667 log_debug("unrecognized message header");
672 udev_device = udev_device_new_from_nulstr(udev_monitor->udev, &buf.raw[bufpos], buflen - bufpos);
677 udev_device_set_is_initialized(udev_device);
679 /* skip device, if it does not pass the current filter */
680 if (!passes_filter(udev_monitor, udev_device)) {
681 struct pollfd pfd[1];
684 udev_device_unref(udev_device);
686 /* if something is queued, get next device */
687 pfd[0].fd = udev_monitor->sock;
688 pfd[0].events = POLLIN;
689 rc = poll(pfd, 1, 0);
698 int udev_monitor_send_device(struct udev_monitor *udev_monitor,
699 struct udev_monitor *destination, struct udev_device *udev_device)
707 struct udev_monitor_netlink_header nlh;
708 struct udev_list_entry *list_entry;
709 uint64_t tag_bloom_bits;
711 blen = udev_device_get_properties_monitor_buf(udev_device, &buf);
715 /* add versioned header */
716 memzero(&nlh, sizeof(struct udev_monitor_netlink_header));
717 memcpy(nlh.prefix, "libudev", 8);
718 nlh.magic = htonl(UDEV_MONITOR_MAGIC);
719 nlh.header_size = sizeof(struct udev_monitor_netlink_header);
720 val = udev_device_get_subsystem(udev_device);
721 nlh.filter_subsystem_hash = htonl(util_string_hash32(val));
722 val = udev_device_get_devtype(udev_device);
724 nlh.filter_devtype_hash = htonl(util_string_hash32(val));
725 iov[0].iov_base = &nlh;
726 iov[0].iov_len = sizeof(struct udev_monitor_netlink_header);
728 /* add tag bloom filter */
730 udev_list_entry_foreach(list_entry, udev_device_get_tags_list_entry(udev_device))
731 tag_bloom_bits |= util_string_bloom64(udev_list_entry_get_name(list_entry));
732 if (tag_bloom_bits > 0) {
733 nlh.filter_tag_bloom_hi = htonl(tag_bloom_bits >> 32);
734 nlh.filter_tag_bloom_lo = htonl(tag_bloom_bits & 0xffffffff);
737 /* add properties list */
738 nlh.properties_off = iov[0].iov_len;
739 nlh.properties_len = blen;
740 iov[1].iov_base = (char *)buf;
741 iov[1].iov_len = blen;
743 memzero(&smsg, sizeof(struct msghdr));
747 * Use custom address for target, or the default one.
749 * If we send to a multicast group, we will get
750 * ECONNREFUSED, which is expected.
752 if (destination != NULL)
753 smsg.msg_name = &destination->snl;
755 smsg.msg_name = &udev_monitor->snl_destination;
756 smsg.msg_namelen = sizeof(struct sockaddr_nl);
757 count = sendmsg(udev_monitor->sock, &smsg, 0);
758 log_debug("passed %zi bytes to netlink monitor %p", count, udev_monitor);
763 * udev_monitor_filter_add_match_subsystem_devtype:
764 * @udev_monitor: the monitor
765 * @subsystem: the subsystem value to match the incoming devices against
766 * @devtype: the devtype value to match the incoming devices against
768 * This filter is efficiently executed inside the kernel, and libudev subscribers
769 * will usually not be woken up for devices which do not match.
771 * The filter must be installed before the monitor is switched to listening mode.
773 * Returns: 0 on success, otherwise a negative error value.
775 _public_ int udev_monitor_filter_add_match_subsystem_devtype(struct udev_monitor *udev_monitor, const char *subsystem, const char *devtype)
777 if (udev_monitor == NULL)
779 if (subsystem == NULL)
781 if (udev_list_entry_add(&udev_monitor->filter_subsystem_list, subsystem, devtype) == NULL)
787 * udev_monitor_filter_add_match_tag:
788 * @udev_monitor: the monitor
789 * @tag: the name of a tag
791 * This filter is efficiently executed inside the kernel, and libudev subscribers
792 * will usually not be woken up for devices which do not match.
794 * The filter must be installed before the monitor is switched to listening mode.
796 * Returns: 0 on success, otherwise a negative error value.
798 _public_ int udev_monitor_filter_add_match_tag(struct udev_monitor *udev_monitor, const char *tag)
800 if (udev_monitor == NULL)
804 if (udev_list_entry_add(&udev_monitor->filter_tag_list, tag, NULL) == NULL)
810 * udev_monitor_filter_remove:
811 * @udev_monitor: monitor
813 * Remove all filters from monitor.
815 * Returns: 0 on success, otherwise a negative error value.
817 _public_ int udev_monitor_filter_remove(struct udev_monitor *udev_monitor)
819 static struct sock_fprog filter = { 0, NULL };
821 udev_list_cleanup(&udev_monitor->filter_subsystem_list);
822 return setsockopt(udev_monitor->sock, SOL_SOCKET, SO_ATTACH_FILTER, &filter, sizeof(filter));