1 /* Copyright (c) 2008 The Board of Trustees of The Leland Stanford
4 * We are making the OpenFlow specification and associated documentation
5 * (Software) available for public use and benefit with the expectation
6 * that others will use, modify and enhance the Software and contribute
7 * those enhancements back to the community. However, since we would
8 * like to make the Software available for broadest use, with as few
9 * restrictions as possible permission is hereby granted, free of
10 * charge, to any person obtaining a copy of this Software to deal in
11 * the Software under the copyrights without restriction, including
12 * without limitation the rights to use, copy, modify, merge, publish,
13 * distribute, sublicense, and/or sell copies of the Software, and to
14 * permit persons to whom the Software is furnished to do so, subject to
15 * the following conditions:
17 * The above copyright notice and this permission notice shall be
18 * included in all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
23 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
24 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
25 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
26 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 * The name and trademarks of copyright holder(s) may NOT be used in
30 * advertising or publicity pertaining to the Software or any
31 * derivatives without specific, written prior permission.
47 #define THIS_MODULE VLM_netlink
49 /* Linux header file confusion causes this to be undefined. */
51 #define SOL_NETLINK 270
54 /* Netlink sockets. */
62 /* Next nlmsghdr sequence number.
64 * This implementation uses sequence numbers that are unique process-wide, to
65 * avoid a hypothetical race: send request, close socket, open new socket that
66 * reuses the old socket's PID value, send request on new socket, receive reply
67 * from kernel to old socket but with same PID and sequence number. (This race
68 * could be avoided other ways, e.g. by preventing PIDs from being quickly
70 static uint32_t next_seq;
72 static int alloc_pid(uint32_t *);
73 static void free_pid(uint32_t);
75 /* Creates a new netlink socket for the given netlink 'protocol'
76 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
77 * new socket if successful, otherwise returns a positive errno value.
79 * If 'multicast_group' is nonzero, the new socket subscribes to the specified
80 * netlink multicast group. (A netlink socket may listen to an arbitrary
81 * number of multicast groups, but so far we only need one at a time.)
83 * Nonzero 'so_sndbuf' or 'so_rcvbuf' override the kernel default send or
84 * receive buffer size, respectively.
87 nl_sock_create(int protocol, int multicast_group,
88 size_t so_sndbuf, size_t so_rcvbuf, struct nl_sock **sockp)
91 struct sockaddr_nl local, remote;
95 /* Pick initial sequence number. */
96 next_seq = getpid() ^ time(0);
100 sock = malloc(sizeof *sock);
105 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
107 VLOG_ERR("fcntl: %s", strerror(errno));
111 retval = alloc_pid(&sock->pid);
117 && setsockopt(sock->fd, SOL_SOCKET, SO_SNDBUF,
118 &so_sndbuf, sizeof so_sndbuf) < 0) {
119 VLOG_ERR("setsockopt(SO_SNDBUF,%zu): %s", so_sndbuf, strerror(errno));
124 && setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
125 &so_rcvbuf, sizeof so_rcvbuf) < 0) {
126 VLOG_ERR("setsockopt(SO_RCVBUF,%zu): %s", so_rcvbuf, strerror(errno));
130 /* Bind local address as our selected pid. */
131 memset(&local, 0, sizeof local);
132 local.nl_family = AF_NETLINK;
133 local.nl_pid = sock->pid;
134 if (multicast_group > 0 && multicast_group <= 32) {
135 /* This method of joining multicast groups is supported by old kernels,
136 * but it only allows 32 multicast groups per protocol. */
137 local.nl_groups |= 1ul << (multicast_group - 1);
139 if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) {
140 VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno));
144 /* Bind remote address as the kernel (pid 0). */
145 memset(&remote, 0, sizeof remote);
146 remote.nl_family = AF_NETLINK;
148 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
149 VLOG_ERR("connect(0): %s", strerror(errno));
153 #ifdef NETLINK_ADD_MEMBERSHIP
154 /* This method of joining multicast groups is only supported by newish
155 * kernels, but it allows for an arbitrary number of multicast groups. */
156 if (multicast_group > 32
157 && setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
158 &multicast_group, sizeof multicast_group) < 0) {
159 VLOG_ERR("setsockopt(NETLINK_ADD_MEMBERSHIP,%d): %s",
160 multicast_group, strerror(errno));
184 /* Destroys netlink socket 'sock'. */
186 nl_sock_destroy(struct nl_sock *sock)
195 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
196 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size before the
199 * Returns 0 if successful, otherwise a positive errno value. If
200 * 'wait' is true, then the send will wait until buffer space is ready;
201 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
203 nl_sock_send(struct nl_sock *sock, const struct buffer *msg, bool wait)
207 nl_msg_nlmsghdr(msg)->nlmsg_len = msg->size;
209 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
210 } while (retval < 0 && errno == EINTR);
211 return retval < 0 ? errno : 0;
214 /* Tries to send the 'n_iov' chunks of data in 'iov' to the kernel on 'sock' as
215 * a single Netlink message. (The message must be fully formed and not require
216 * finalization of its nlmsg_len field.)
218 * Returns 0 if successful, otherwise a positive errno value. If 'wait' is
219 * true, then the send will wait until buffer space is ready; otherwise,
220 * returns EAGAIN if the 'sock' send buffer is full. */
222 nl_sock_sendv(struct nl_sock *sock, const struct iovec iov[], size_t n_iov,
228 memset(&msg, 0, sizeof msg);
229 msg.msg_iov = (struct iovec *) iov;
230 msg.msg_iovlen = n_iov;
232 retval = sendmsg(sock->fd, &msg, MSG_DONTWAIT);
233 } while (retval < 0 && errno == EINTR);
234 return retval < 0 ? errno : 0;
237 /* Tries to receive a netlink message from the kernel on 'sock'. If
238 * successful, stores the received message into '*bufp' and returns 0. The
239 * caller is responsible for destroying the message with buffer_delete(). On
240 * failure, returns a positive errno value and stores a null pointer into
243 * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
244 * returns EAGAIN if the 'sock' receive buffer is empty. */
246 nl_sock_recv(struct nl_sock *sock, struct buffer **bufp, bool wait)
249 ssize_t bufsize = 2048;
250 ssize_t nbytes, nbytes2;
252 struct nlmsghdr *nlmsghdr;
254 struct msghdr msg = {
264 buf = buffer_new(bufsize);
268 /* Attempt to read the message. We don't know the size of the data
269 * yet, so we take a guess at 2048. If we're wrong, we keep trying
270 * and doubling the buffer size each time.
272 nlmsghdr = buffer_put_uninit(buf, bufsize);
273 iov.iov_base = nlmsghdr;
274 iov.iov_len = bufsize;
276 nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK);
277 } while (nbytes < 0 && errno == EINTR);
282 if (msg.msg_flags & MSG_TRUNC) {
284 buffer_reinit(buf, bufsize);
289 /* We successfully read the message, so recv again to clear the queue */
293 nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT);
295 VLOG_ERR("failed to remove nlmsg from socket: %d\n", errno);
297 } while (nbytes2 < 0 && errno == EINTR);
299 if (!NLMSG_OK(nlmsghdr, nbytes)) {
300 VLOG_ERR("received invalid nlmsg (%zd bytes < %d)",
301 bufsize, NLMSG_HDRLEN);
309 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
310 * successful, stores the reply into '*replyp' and returns 0. The caller is
311 * responsible for destroying the reply with buffer_delete(). On failure,
312 * returns a positive errno value and stores a null pointer into '*replyp'.
314 * Bare Netlink is an unreliable transport protocol. This function layers
315 * reliable delivery and reply semantics on top of bare Netlink.
317 * In Netlink, sending a request to the kernel is reliable enough, because the
318 * kernel will tell us if the message cannot be queued (and we will in that
319 * case put it on the transmit queue and wait until it can be delivered).
321 * Receiving the reply is the real problem: if the socket buffer is full when
322 * the kernel tries to send the reply, the reply will be dropped. However, the
323 * kernel sets a flag that a reply has been dropped. The next call to recv
324 * then returns ENOBUFS. We can then re-send the request.
328 * 1. Netlink depends on sequence numbers to match up requests and
329 * replies. The sender of a request supplies a sequence number, and
330 * the reply echos back that sequence number.
332 * This is fine, but (1) some kernel netlink implementations are
333 * broken, in that they fail to echo sequence numbers and (2) this
334 * function will drop packets with non-matching sequence numbers, so
335 * that only a single request can be usefully transacted at a time.
337 * 2. Resending the request causes it to be re-executed, so the request
338 * needs to be idempotent.
341 nl_sock_transact(struct nl_sock *sock,
342 const struct buffer *request, struct buffer **replyp)
344 uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
345 struct nlmsghdr *nlmsghdr;
346 struct buffer *reply;
351 /* Ensure that we get a reply even if this message doesn't ordinarily call
353 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK;
356 retval = nl_sock_send(sock, request, true);
362 retval = nl_sock_recv(sock, &reply, true);
364 if (retval == ENOBUFS) {
365 VLOG_DBG("receive buffer overflow, resending request");
371 nlmsghdr = nl_msg_nlmsghdr(reply);
372 if (seq != nlmsghdr->nlmsg_seq) {
373 VLOG_DBG("ignoring seq %"PRIu32" != expected %"PRIu32,
374 nl_msg_nlmsghdr(reply)->nlmsg_seq, seq);
375 buffer_delete(reply);
378 if (nl_msg_nlmsgerr(reply, &retval)) {
380 VLOG_DBG("received NAK error=%d (%s)", retval, strerror(retval));
382 return retval != EAGAIN ? retval : EPROTO;
389 /* Returns 'sock''s underlying file descriptor. */
391 nl_sock_fd(const struct nl_sock *sock)
396 /* Netlink messages. */
398 /* Returns the nlmsghdr at the head of 'msg'.
400 * 'msg' must be at least as large as a nlmsghdr. */
402 nl_msg_nlmsghdr(const struct buffer *msg)
404 return buffer_at_assert(msg, 0, NLMSG_HDRLEN);
407 /* Returns the genlmsghdr just past 'msg''s nlmsghdr.
409 * Returns a null pointer if 'msg' is not large enough to contain an nlmsghdr
410 * and a genlmsghdr. */
412 nl_msg_genlmsghdr(const struct buffer *msg)
414 return buffer_at(msg, NLMSG_HDRLEN, GENL_HDRLEN);
417 /* If 'buffer' is a NLMSG_ERROR message, stores 0 in '*errorp' if it is an ACK
418 * message, otherwise a positive errno value, and returns true. If 'buffer' is
419 * not an NLMSG_ERROR message, returns false.
421 * 'msg' must be at least as large as a nlmsghdr. */
423 nl_msg_nlmsgerr(const struct buffer *msg, int *errorp)
425 if (nl_msg_nlmsghdr(msg)->nlmsg_type == NLMSG_ERROR) {
426 struct nlmsgerr *err = buffer_at(msg, NLMSG_HDRLEN, sizeof *err);
429 VLOG_ERR("received invalid nlmsgerr (%zd bytes < %zd)",
430 msg->size, NLMSG_HDRLEN + sizeof *err);
431 } else if (err->error <= 0 && err->error > INT_MIN) {
443 /* Ensures that 'b' has room for at least 'size' bytes plus netlink pading at
444 * its tail end, reallocating and copying its data if necessary. */
446 nl_msg_reserve(struct buffer *msg, size_t size)
448 buffer_reserve_tailroom(msg, NLMSG_ALIGN(size));
451 /* Puts a nlmsghdr at the beginning of 'msg', which must be initially empty.
452 * Uses the given 'type' and 'flags'. 'sock' is used to obtain a PID and
453 * sequence number for proper routing of replies. 'expected_payload' should be
454 * an estimate of the number of payload bytes to be supplied; if the size of
455 * the payload is unknown a value of 0 is acceptable.
457 * 'type' is ordinarily an enumerated value specific to the Netlink protocol
458 * (e.g. RTM_NEWLINK, for NETLINK_ROUTE protocol). For Generic Netlink, 'type'
459 * is the family number obtained via nl_lookup_genl_family().
461 * 'flags' is a bit-mask that indicates what kind of request is being made. It
462 * is often NLM_F_REQUEST indicating that a request is being made, commonly
463 * or'd with NLM_F_ACK to request an acknowledgement.
465 * nl_msg_put_genlmsghdr is more convenient for composing a Generic Netlink
468 nl_msg_put_nlmsghdr(struct buffer *msg, struct nl_sock *sock,
469 size_t expected_payload, uint32_t type, uint32_t flags)
471 struct nlmsghdr *nlmsghdr;
473 assert(msg->size == 0);
475 nl_msg_reserve(msg, NLMSG_HDRLEN + expected_payload);
476 nlmsghdr = nl_msg_put_uninit(msg, NLMSG_HDRLEN);
477 nlmsghdr->nlmsg_len = 0;
478 nlmsghdr->nlmsg_type = type;
479 nlmsghdr->nlmsg_flags = flags;
480 nlmsghdr->nlmsg_seq = ++next_seq;
481 nlmsghdr->nlmsg_pid = sock->pid;
484 /* Puts a nlmsghdr and genlmsghdr at the beginning of 'msg', which must be
485 * initially empty. 'sock' is used to obtain a PID and sequence number for
486 * proper routing of replies. 'expected_payload' should be an estimate of the
487 * number of payload bytes to be supplied; if the size of the payload is
488 * unknown a value of 0 is acceptable.
490 * 'family' is the family number obtained via nl_lookup_genl_family().
492 * 'flags' is a bit-mask that indicates what kind of request is being made. It
493 * is often NLM_F_REQUEST indicating that a request is being made, commonly
494 * or'd with NLM_F_ACK to request an acknowledgement.
496 * 'cmd' is an enumerated value specific to the Generic Netlink family
497 * (e.g. CTRL_CMD_NEWFAMILY for the GENL_ID_CTRL family).
499 * 'version' is a version number specific to the family and command (often 1).
501 * nl_msg_put_nlmsghdr should be used to compose Netlink messages that are not
502 * Generic Netlink messages. */
504 nl_msg_put_genlmsghdr(struct buffer *msg, struct nl_sock *sock,
505 size_t expected_payload, int family, uint32_t flags,
506 uint8_t cmd, uint8_t version)
508 struct genlmsghdr *genlmsghdr;
510 nl_msg_put_nlmsghdr(msg, sock, GENL_HDRLEN + expected_payload,
512 assert(msg->size == NLMSG_HDRLEN);
513 genlmsghdr = nl_msg_put_uninit(msg, GENL_HDRLEN);
514 genlmsghdr->cmd = cmd;
515 genlmsghdr->version = version;
516 genlmsghdr->reserved = 0;
519 /* Appends the 'size' bytes of data in 'p', plus Netlink padding if needed, to
520 * the tail end of 'msg'. Data in 'msg' is reallocated and copied if
523 nl_msg_put(struct buffer *msg, const void *data, size_t size)
525 memcpy(nl_msg_put_uninit(msg, size), data, size);
528 /* Appends 'size' bytes of data, plus Netlink padding if needed, to the tail
529 * end of 'msg', reallocating and copying its data if necessary. Returns a
530 * pointer to the first byte of the new data, which is left uninitialized. */
532 nl_msg_put_uninit(struct buffer *msg, size_t size)
534 size_t pad = NLMSG_ALIGN(size) - size;
535 char *p = buffer_put_uninit(msg, size + pad);
537 memset(p + size, 0, pad);
542 /* Appends a Netlink attribute of the given 'type' and room for 'size' bytes of
543 * data as its payload, plus Netlink padding if needed, to the tail end of
544 * 'msg', reallocating and copying its data if necessary. Returns a pointer to
545 * the first byte of data in the attribute, which is left uninitialized. */
547 nl_msg_put_unspec_uninit(struct buffer *msg, uint16_t type, size_t size)
549 size_t total_size = NLA_HDRLEN + size;
550 struct nlattr* nla = nl_msg_put_uninit(msg, total_size);
551 assert(NLA_ALIGN(total_size) <= UINT16_MAX);
552 nla->nla_len = total_size;
553 nla->nla_type = type;
557 /* Appends a Netlink attribute of the given 'type' and the 'size' bytes of
558 * 'data' as its payload, to the tail end of 'msg', reallocating and copying
559 * its data if necessary. Returns a pointer to the first byte of data in the
560 * attribute, which is left uninitialized. */
562 nl_msg_put_unspec(struct buffer *msg, uint16_t type,
563 const void *data, size_t size)
565 memcpy(nl_msg_put_unspec_uninit(msg, type, size), data, size);
568 /* Appends a Netlink attribute of the given 'type' and no payload to 'msg'.
569 * (Some Netlink protocols use the presence or absence of an attribute as a
572 nl_msg_put_flag(struct buffer *msg, uint16_t type)
574 nl_msg_put_unspec(msg, type, NULL, 0);
577 /* Appends a Netlink attribute of the given 'type' and the given 8-bit 'value'
580 nl_msg_put_u8(struct buffer *msg, uint16_t type, uint8_t value)
582 nl_msg_put_unspec(msg, type, &value, sizeof value);
585 /* Appends a Netlink attribute of the given 'type' and the given 16-bit 'value'
588 nl_msg_put_u16(struct buffer *msg, uint16_t type, uint16_t value)
590 nl_msg_put_unspec(msg, type, &value, sizeof value);
593 /* Appends a Netlink attribute of the given 'type' and the given 32-bit 'value'
596 nl_msg_put_u32(struct buffer *msg, uint16_t type, uint32_t value)
598 nl_msg_put_unspec(msg, type, &value, sizeof value);
601 /* Appends a Netlink attribute of the given 'type' and the given 64-bit 'value'
604 nl_msg_put_u64(struct buffer *msg, uint16_t type, uint64_t value)
606 nl_msg_put_unspec(msg, type, &value, sizeof value);
609 /* Appends a Netlink attribute of the given 'type' and the given
610 * null-terminated string 'value' to 'msg'. */
612 nl_msg_put_string(struct buffer *msg, uint16_t type, const char *value)
614 nl_msg_put_unspec(msg, type, value, strlen(value) + 1);
617 /* Appends a Netlink attribute of the given 'type' and the given buffered
618 * netlink message in 'nested_msg' to 'msg'. The nlmsg_len field in
619 * 'nested_msg' is finalized to match 'nested_msg->size'. */
621 nl_msg_put_nested(struct buffer *msg,
622 uint16_t type, struct buffer *nested_msg)
624 nl_msg_nlmsghdr(nested_msg)->nlmsg_len = nested_msg->size;
625 nl_msg_put_unspec(msg, type, nested_msg->data, nested_msg->size);
628 /* Returns the first byte in the payload of attribute 'nla'. */
630 nl_attr_get(const struct nlattr *nla)
632 assert(nla->nla_len >= NLA_HDRLEN);
636 /* Returns the number of bytes in the payload of attribute 'nla'. */
638 nl_attr_get_size(const struct nlattr *nla)
640 assert(nla->nla_len >= NLA_HDRLEN);
641 return nla->nla_len - NLA_HDRLEN;
644 /* Asserts that 'nla''s payload is at least 'size' bytes long, and returns the
645 * first byte of the payload. */
647 nl_attr_get_unspec(const struct nlattr *nla, size_t size)
649 assert(nla->nla_len >= NLA_HDRLEN + size);
653 /* Returns true if 'nla' is nonnull. (Some Netlink protocols use the presence
654 * or absence of an attribute as a Boolean flag.) */
656 nl_attr_get_flag(const struct nlattr *nla)
661 #define NL_ATTR_GET_AS(NLA, TYPE) \
662 (*(TYPE*) nl_attr_get_unspec(nla, sizeof(TYPE)))
664 /* Returns the 8-bit value in 'nla''s payload.
666 * Asserts that 'nla''s payload is at least 1 byte long. */
668 nl_attr_get_u8(const struct nlattr *nla)
670 return NL_ATTR_GET_AS(nla, uint8_t);
673 /* Returns the 16-bit value in 'nla''s payload.
675 * Asserts that 'nla''s payload is at least 2 bytes long. */
677 nl_attr_get_u16(const struct nlattr *nla)
679 return NL_ATTR_GET_AS(nla, uint16_t);
682 /* Returns the 32-bit value in 'nla''s payload.
684 * Asserts that 'nla''s payload is at least 4 bytes long. */
686 nl_attr_get_u32(const struct nlattr *nla)
688 return NL_ATTR_GET_AS(nla, uint32_t);
691 /* Returns the 64-bit value in 'nla''s payload.
693 * Asserts that 'nla''s payload is at least 8 bytes long. */
695 nl_attr_get_u64(const struct nlattr *nla)
697 return NL_ATTR_GET_AS(nla, uint64_t);
700 /* Returns the null-terminated string value in 'nla''s payload.
702 * Asserts that 'nla''s payload contains a null-terminated string. */
704 nl_attr_get_string(const struct nlattr *nla)
706 assert(nla->nla_len > NLA_HDRLEN);
707 assert(memchr(nl_attr_get(nla), '\0', nla->nla_len - NLA_HDRLEN) != NULL);
708 return nl_attr_get(nla);
711 /* Default minimum and maximum payload sizes for each type of attribute. */
712 static const size_t attr_len_range[][2] = {
713 [0 ... N_NL_ATTR_TYPES - 1] = { 0, SIZE_MAX },
714 [NL_A_U8] = { 1, 1 },
715 [NL_A_U16] = { 2, 2 },
716 [NL_A_U32] = { 4, 4 },
717 [NL_A_U64] = { 8, 8 },
718 [NL_A_STRING] = { 1, SIZE_MAX },
719 [NL_A_FLAG] = { 0, SIZE_MAX },
720 [NL_A_NESTED] = { NLMSG_HDRLEN, SIZE_MAX },
723 /* Parses the Generic Netlink payload of 'msg' as a sequence of Netlink
724 * attributes. 'policy[i]', for 0 <= i < n_attrs, specifies how the attribute
725 * with nla_type == i is parsed; a pointer to attribute i is stored in
726 * attrs[i]. Returns true if successful, false on failure. */
728 nl_policy_parse(const struct buffer *msg, const struct nl_policy policy[],
729 struct nlattr *attrs[], size_t n_attrs)
736 for (i = 0; i < n_attrs; i++) {
739 assert(policy[i].type < N_NL_ATTR_TYPES);
740 if (policy[i].type != NL_A_NO_ATTR
741 && policy[i].type != NL_A_FLAG
742 && !policy[i].optional) {
747 p = buffer_at(msg, NLMSG_HDRLEN + GENL_HDRLEN, 0);
749 VLOG_DBG("missing headers in nl_policy_parse");
752 tail = buffer_tail(msg);
755 size_t offset = p - msg->data;
756 struct nlattr *nla = p;
757 size_t len, aligned_len;
760 /* Make sure its claimed length is plausible. */
761 if (nla->nla_len < NLA_HDRLEN) {
762 VLOG_DBG("%zu: attr shorter than NLA_HDRLEN (%"PRIu16")",
763 offset, nla->nla_len);
766 len = nla->nla_len - NLA_HDRLEN;
767 aligned_len = NLA_ALIGN(len);
768 if (aligned_len > tail - p) {
769 VLOG_DBG("%zu: attr %"PRIu16" aligned data len (%zu) "
770 "> bytes left (%tu)",
771 offset, nla->nla_type, aligned_len, tail - p);
775 type = nla->nla_type;
776 if (type < n_attrs && policy[type].type != NL_A_NO_ATTR) {
777 const struct nl_policy *p = &policy[type];
778 size_t min_len, max_len;
780 /* Validate length and content. */
781 min_len = p->min_len ? p->min_len : attr_len_range[p->type][0];
782 max_len = p->max_len ? p->max_len : attr_len_range[p->type][1];
783 if (len < min_len || len > max_len) {
784 VLOG_DBG("%zu: attr %"PRIu16" length %zu not in allowed range "
785 "%zu...%zu", offset, type, len, min_len, max_len);
788 if (p->type == NL_A_STRING) {
789 if (((char *) nla)[nla->nla_len - 1]) {
790 VLOG_DBG("%zu: attr %"PRIu16" lacks null terminator",
794 if (memchr(nla + 1, '\0', len - 1) != NULL) {
795 VLOG_DBG("%zu: attr %"PRIu16" lies about string length",
800 if (!p->optional && attrs[type] == NULL) {
801 assert(n_required > 0);
806 /* Skip attribute type that we don't care about. */
808 p += NLA_ALIGN(nla->nla_len);
811 VLOG_DBG("%zu required attrs missing", n_required);
819 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
820 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
823 static int do_lookup_genl_family(const char *name)
825 struct nl_sock *sock;
826 struct buffer request, *reply;
827 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
830 retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
835 buffer_init(&request, 0);
836 nl_msg_put_genlmsghdr(&request, sock, 0, GENL_ID_CTRL, NLM_F_REQUEST,
837 CTRL_CMD_GETFAMILY, 1);
838 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
839 retval = nl_sock_transact(sock, &request, &reply);
840 buffer_uninit(&request);
842 nl_sock_destroy(sock);
846 if (!nl_policy_parse(reply, family_policy, attrs,
847 ARRAY_SIZE(family_policy))) {
848 nl_sock_destroy(sock);
849 buffer_delete(reply);
853 retval = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
857 nl_sock_destroy(sock);
858 buffer_delete(reply);
862 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
863 * number and stores it in '*number'. If successful, returns 0 and the caller
864 * may use '*number' as the family number. On failure, returns a positive
865 * errno value and '*number' caches the errno value. */
867 nl_lookup_genl_family(const char *name, int *number)
870 *number = do_lookup_genl_family(name);
871 assert(*number != 0);
873 return *number > 0 ? 0 : -*number;
878 * Every Netlink socket must be bound to a unique 32-bit PID. By convention,
879 * programs that have a single Netlink socket use their Unix process ID as PID,
880 * and programs with multiple Netlink sockets add a unique per-socket
881 * identifier in the bits above the Unix process ID.
883 * The kernel has Netlink PID 0.
886 /* Parameters for how many bits in the PID should come from the Unix process ID
887 * and how many unique per-socket. */
888 #define SOCKET_BITS 10
889 #define MAX_SOCKETS (1u << SOCKET_BITS)
891 #define PROCESS_BITS (32 - SOCKET_BITS)
892 #define MAX_PROCESSES (1u << PROCESS_BITS)
893 #define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1))
895 /* Bit vector of unused socket identifiers. */
896 static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)];
898 /* Allocates and returns a new Netlink PID. */
900 alloc_pid(uint32_t *pid)
904 for (i = 0; i < MAX_SOCKETS; i++) {
905 if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) {
906 avail_sockets[i / 32] |= 1u << (i % 32);
907 *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS);
911 VLOG_ERR("netlink pid space exhausted");
915 /* Makes the specified 'pid' available for reuse. */
917 free_pid(uint32_t pid)
919 int sock = pid >> PROCESS_BITS;
920 assert(avail_sockets[sock / 32] & (1u << (sock % 32)));
921 avail_sockets[sock / 32] &= ~(1u << (sock % 32));