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.
44 #include "netlink-protocol.h"
45 #include "dynamic-string.h"
49 #define THIS_MODULE VLM_netlink
51 /* Linux header file confusion causes this to be undefined. */
53 #define SOL_NETLINK 270
56 static void log_nlmsg(const char *function, int error,
57 const void *message, size_t size);
59 /* Netlink sockets. */
67 /* Next nlmsghdr sequence number.
69 * This implementation uses sequence numbers that are unique process-wide, to
70 * avoid a hypothetical race: send request, close socket, open new socket that
71 * reuses the old socket's PID value, send request on new socket, receive reply
72 * from kernel to old socket but with same PID and sequence number. (This race
73 * could be avoided other ways, e.g. by preventing PIDs from being quickly
75 static uint32_t next_seq;
77 static int alloc_pid(uint32_t *);
78 static void free_pid(uint32_t);
80 /* Creates a new netlink socket for the given netlink 'protocol'
81 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
82 * new socket if successful, otherwise returns a positive errno value.
84 * If 'multicast_group' is nonzero, the new socket subscribes to the specified
85 * netlink multicast group. (A netlink socket may listen to an arbitrary
86 * number of multicast groups, but so far we only need one at a time.)
88 * Nonzero 'so_sndbuf' or 'so_rcvbuf' override the kernel default send or
89 * receive buffer size, respectively.
92 nl_sock_create(int protocol, int multicast_group,
93 size_t so_sndbuf, size_t so_rcvbuf, struct nl_sock **sockp)
96 struct sockaddr_nl local, remote;
100 /* Pick initial sequence number. */
101 next_seq = getpid() ^ time(0);
105 sock = malloc(sizeof *sock);
110 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
112 VLOG_ERR("fcntl: %s", strerror(errno));
116 retval = alloc_pid(&sock->pid);
122 && setsockopt(sock->fd, SOL_SOCKET, SO_SNDBUF,
123 &so_sndbuf, sizeof so_sndbuf) < 0) {
124 VLOG_ERR("setsockopt(SO_SNDBUF,%zu): %s", so_sndbuf, strerror(errno));
129 && setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
130 &so_rcvbuf, sizeof so_rcvbuf) < 0) {
131 VLOG_ERR("setsockopt(SO_RCVBUF,%zu): %s", so_rcvbuf, strerror(errno));
135 /* Bind local address as our selected pid. */
136 memset(&local, 0, sizeof local);
137 local.nl_family = AF_NETLINK;
138 local.nl_pid = sock->pid;
139 if (multicast_group > 0 && multicast_group <= 32) {
140 /* This method of joining multicast groups is supported by old kernels,
141 * but it only allows 32 multicast groups per protocol. */
142 local.nl_groups |= 1ul << (multicast_group - 1);
144 if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) {
145 VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno));
149 /* Bind remote address as the kernel (pid 0). */
150 memset(&remote, 0, sizeof remote);
151 remote.nl_family = AF_NETLINK;
153 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
154 VLOG_ERR("connect(0): %s", strerror(errno));
158 /* Older kernel headers failed to define this macro. We want our programs
159 * to support the newer kernel features even if compiled with older
160 * headers, so define it ourselves in such a case. */
161 #ifndef NETLINK_ADD_MEMBERSHIP
162 #define NETLINK_ADD_MEMBERSHIP 1
165 /* This method of joining multicast groups is only supported by newish
166 * kernels, but it allows for an arbitrary number of multicast groups. */
167 if (multicast_group > 32
168 && setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
169 &multicast_group, sizeof multicast_group) < 0) {
170 VLOG_ERR("setsockopt(NETLINK_ADD_MEMBERSHIP,%d): %s",
171 multicast_group, strerror(errno));
194 /* Destroys netlink socket 'sock'. */
196 nl_sock_destroy(struct nl_sock *sock)
205 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
206 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size before the
209 * Returns 0 if successful, otherwise a positive errno value. If
210 * 'wait' is true, then the send will wait until buffer space is ready;
211 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
213 nl_sock_send(struct nl_sock *sock, const struct buffer *msg, bool wait)
217 nl_msg_nlmsghdr(msg)->nlmsg_len = msg->size;
220 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
221 error = retval < 0 ? errno : 0;
222 } while (error == EINTR);
223 log_nlmsg(__func__, error, msg->data, msg->size);
227 /* Tries to send the 'n_iov' chunks of data in 'iov' to the kernel on 'sock' as
228 * a single Netlink message. (The message must be fully formed and not require
229 * finalization of its nlmsg_len field.)
231 * Returns 0 if successful, otherwise a positive errno value. If 'wait' is
232 * true, then the send will wait until buffer space is ready; otherwise,
233 * returns EAGAIN if the 'sock' send buffer is full. */
235 nl_sock_sendv(struct nl_sock *sock, const struct iovec iov[], size_t n_iov,
241 memset(&msg, 0, sizeof msg);
242 msg.msg_iov = (struct iovec *) iov;
243 msg.msg_iovlen = n_iov;
246 retval = sendmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
247 error = retval < 0 ? errno : 0;
248 } while (error == EINTR);
249 if (error != EAGAIN) {
250 log_nlmsg(__func__, error, iov[0].iov_base, iov[0].iov_len);
255 /* Tries to receive a netlink message from the kernel on 'sock'. If
256 * successful, stores the received message into '*bufp' and returns 0. The
257 * caller is responsible for destroying the message with buffer_delete(). On
258 * failure, returns a positive errno value and stores a null pointer into
261 * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
262 * returns EAGAIN if the 'sock' receive buffer is empty. */
264 nl_sock_recv(struct nl_sock *sock, struct buffer **bufp, bool wait)
267 ssize_t bufsize = 2048;
268 ssize_t nbytes, nbytes2;
270 struct nlmsghdr *nlmsghdr;
272 struct msghdr msg = {
282 buf = buffer_new(bufsize);
286 /* Attempt to read the message. We don't know the size of the data
287 * yet, so we take a guess at 2048. If we're wrong, we keep trying
288 * and doubling the buffer size each time.
290 nlmsghdr = buffer_put_uninit(buf, bufsize);
291 iov.iov_base = nlmsghdr;
292 iov.iov_len = bufsize;
294 nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK);
295 } while (nbytes < 0 && errno == EINTR);
300 if (msg.msg_flags & MSG_TRUNC) {
302 buffer_reinit(buf, bufsize);
307 /* We successfully read the message, so recv again to clear the queue */
311 nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT);
312 } while (nbytes2 < 0 && errno == EINTR);
314 if (errno == ENOBUFS) {
315 /* The kernel is notifying us that a message it tried to send to us
316 * was dropped. We have to pass this along to the caller in case
317 * it wants to retry a request. So kill the buffer, which we can
318 * re-read next time. */
322 VLOG_ERR("failed to remove nlmsg from socket: %s\n",
326 if (nbytes < sizeof *nlmsghdr
327 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
328 || nlmsghdr->nlmsg_len > nbytes) {
329 VLOG_ERR("received invalid nlmsg (%zd bytes < %d)",
330 bufsize, NLMSG_HDRLEN);
335 log_nlmsg(__func__, 0, buf->data, buf->size);
339 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
340 * successful, stores the reply into '*replyp' and returns 0. The caller is
341 * responsible for destroying the reply with buffer_delete(). On failure,
342 * returns a positive errno value and stores a null pointer into '*replyp'.
344 * Bare Netlink is an unreliable transport protocol. This function layers
345 * reliable delivery and reply semantics on top of bare Netlink.
347 * In Netlink, sending a request to the kernel is reliable enough, because the
348 * kernel will tell us if the message cannot be queued (and we will in that
349 * case put it on the transmit queue and wait until it can be delivered).
351 * Receiving the reply is the real problem: if the socket buffer is full when
352 * the kernel tries to send the reply, the reply will be dropped. However, the
353 * kernel sets a flag that a reply has been dropped. The next call to recv
354 * then returns ENOBUFS. We can then re-send the request.
358 * 1. Netlink depends on sequence numbers to match up requests and
359 * replies. The sender of a request supplies a sequence number, and
360 * the reply echos back that sequence number.
362 * This is fine, but (1) some kernel netlink implementations are
363 * broken, in that they fail to echo sequence numbers and (2) this
364 * function will drop packets with non-matching sequence numbers, so
365 * that only a single request can be usefully transacted at a time.
367 * 2. Resending the request causes it to be re-executed, so the request
368 * needs to be idempotent.
371 nl_sock_transact(struct nl_sock *sock,
372 const struct buffer *request, struct buffer **replyp)
374 uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
375 struct nlmsghdr *nlmsghdr;
376 struct buffer *reply;
381 /* Ensure that we get a reply even if this message doesn't ordinarily call
383 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK;
386 retval = nl_sock_send(sock, request, true);
392 retval = nl_sock_recv(sock, &reply, true);
394 if (retval == ENOBUFS) {
395 VLOG_DBG("receive buffer overflow, resending request");
401 nlmsghdr = nl_msg_nlmsghdr(reply);
402 if (seq != nlmsghdr->nlmsg_seq) {
403 VLOG_DBG("ignoring seq %"PRIu32" != expected %"PRIu32,
404 nl_msg_nlmsghdr(reply)->nlmsg_seq, seq);
405 buffer_delete(reply);
408 if (nl_msg_nlmsgerr(reply, &retval)) {
410 VLOG_DBG("received NAK error=%d (%s)", retval, strerror(retval));
412 return retval != EAGAIN ? retval : EPROTO;
419 /* Returns 'sock''s underlying file descriptor. */
421 nl_sock_fd(const struct nl_sock *sock)
426 /* Netlink messages. */
428 /* Returns the nlmsghdr at the head of 'msg'.
430 * 'msg' must be at least as large as a nlmsghdr. */
432 nl_msg_nlmsghdr(const struct buffer *msg)
434 return buffer_at_assert(msg, 0, NLMSG_HDRLEN);
437 /* Returns the genlmsghdr just past 'msg''s nlmsghdr.
439 * Returns a null pointer if 'msg' is not large enough to contain an nlmsghdr
440 * and a genlmsghdr. */
442 nl_msg_genlmsghdr(const struct buffer *msg)
444 return buffer_at(msg, NLMSG_HDRLEN, GENL_HDRLEN);
447 /* If 'buffer' is a NLMSG_ERROR message, stores 0 in '*errorp' if it is an ACK
448 * message, otherwise a positive errno value, and returns true. If 'buffer' is
449 * not an NLMSG_ERROR message, returns false.
451 * 'msg' must be at least as large as a nlmsghdr. */
453 nl_msg_nlmsgerr(const struct buffer *msg, int *errorp)
455 if (nl_msg_nlmsghdr(msg)->nlmsg_type == NLMSG_ERROR) {
456 struct nlmsgerr *err = buffer_at(msg, NLMSG_HDRLEN, sizeof *err);
459 VLOG_ERR("received invalid nlmsgerr (%zd bytes < %zd)",
460 msg->size, NLMSG_HDRLEN + sizeof *err);
461 } else if (err->error <= 0 && err->error > INT_MIN) {
473 /* Ensures that 'b' has room for at least 'size' bytes plus netlink padding at
474 * its tail end, reallocating and copying its data if necessary. */
476 nl_msg_reserve(struct buffer *msg, size_t size)
478 buffer_prealloc_tailroom(msg, NLMSG_ALIGN(size));
481 /* Puts a nlmsghdr at the beginning of 'msg', which must be initially empty.
482 * Uses the given 'type' and 'flags'. 'sock' is used to obtain a PID and
483 * sequence number for proper routing of replies. 'expected_payload' should be
484 * an estimate of the number of payload bytes to be supplied; if the size of
485 * the payload is unknown a value of 0 is acceptable.
487 * 'type' is ordinarily an enumerated value specific to the Netlink protocol
488 * (e.g. RTM_NEWLINK, for NETLINK_ROUTE protocol). For Generic Netlink, 'type'
489 * is the family number obtained via nl_lookup_genl_family().
491 * 'flags' is a bit-mask that indicates what kind of request is being made. It
492 * is often NLM_F_REQUEST indicating that a request is being made, commonly
493 * or'd with NLM_F_ACK to request an acknowledgement.
495 * nl_msg_put_genlmsghdr is more convenient for composing a Generic Netlink
498 nl_msg_put_nlmsghdr(struct buffer *msg, struct nl_sock *sock,
499 size_t expected_payload, uint32_t type, uint32_t flags)
501 struct nlmsghdr *nlmsghdr;
503 assert(msg->size == 0);
505 nl_msg_reserve(msg, NLMSG_HDRLEN + expected_payload);
506 nlmsghdr = nl_msg_put_uninit(msg, NLMSG_HDRLEN);
507 nlmsghdr->nlmsg_len = 0;
508 nlmsghdr->nlmsg_type = type;
509 nlmsghdr->nlmsg_flags = flags;
510 nlmsghdr->nlmsg_seq = ++next_seq;
511 nlmsghdr->nlmsg_pid = sock->pid;
514 /* Puts a nlmsghdr and genlmsghdr at the beginning of 'msg', which must be
515 * initially empty. 'sock' is used to obtain a PID and sequence number for
516 * proper routing of replies. 'expected_payload' should be an estimate of the
517 * number of payload bytes to be supplied; if the size of the payload is
518 * unknown a value of 0 is acceptable.
520 * 'family' is the family number obtained via nl_lookup_genl_family().
522 * 'flags' is a bit-mask that indicates what kind of request is being made. It
523 * is often NLM_F_REQUEST indicating that a request is being made, commonly
524 * or'd with NLM_F_ACK to request an acknowledgement.
526 * 'cmd' is an enumerated value specific to the Generic Netlink family
527 * (e.g. CTRL_CMD_NEWFAMILY for the GENL_ID_CTRL family).
529 * 'version' is a version number specific to the family and command (often 1).
531 * nl_msg_put_nlmsghdr should be used to compose Netlink messages that are not
532 * Generic Netlink messages. */
534 nl_msg_put_genlmsghdr(struct buffer *msg, struct nl_sock *sock,
535 size_t expected_payload, int family, uint32_t flags,
536 uint8_t cmd, uint8_t version)
538 struct genlmsghdr *genlmsghdr;
540 nl_msg_put_nlmsghdr(msg, sock, GENL_HDRLEN + expected_payload,
542 assert(msg->size == NLMSG_HDRLEN);
543 genlmsghdr = nl_msg_put_uninit(msg, GENL_HDRLEN);
544 genlmsghdr->cmd = cmd;
545 genlmsghdr->version = version;
546 genlmsghdr->reserved = 0;
549 /* Appends the 'size' bytes of data in 'p', plus Netlink padding if needed, to
550 * the tail end of 'msg'. Data in 'msg' is reallocated and copied if
553 nl_msg_put(struct buffer *msg, const void *data, size_t size)
555 memcpy(nl_msg_put_uninit(msg, size), data, size);
558 /* Appends 'size' bytes of data, plus Netlink padding if needed, to the tail
559 * end of 'msg', reallocating and copying its data if necessary. Returns a
560 * pointer to the first byte of the new data, which is left uninitialized. */
562 nl_msg_put_uninit(struct buffer *msg, size_t size)
564 size_t pad = NLMSG_ALIGN(size) - size;
565 char *p = buffer_put_uninit(msg, size + pad);
567 memset(p + size, 0, pad);
572 /* Appends a Netlink attribute of the given 'type' and room for 'size' bytes of
573 * data as its payload, plus Netlink padding if needed, to the tail end of
574 * 'msg', reallocating and copying its data if necessary. Returns a pointer to
575 * the first byte of data in the attribute, which is left uninitialized. */
577 nl_msg_put_unspec_uninit(struct buffer *msg, uint16_t type, size_t size)
579 size_t total_size = NLA_HDRLEN + size;
580 struct nlattr* nla = nl_msg_put_uninit(msg, total_size);
581 assert(NLA_ALIGN(total_size) <= UINT16_MAX);
582 nla->nla_len = total_size;
583 nla->nla_type = type;
587 /* Appends a Netlink attribute of the given 'type' and the 'size' bytes of
588 * 'data' as its payload, to the tail end of 'msg', reallocating and copying
589 * its data if necessary. Returns a pointer to the first byte of data in the
590 * attribute, which is left uninitialized. */
592 nl_msg_put_unspec(struct buffer *msg, uint16_t type,
593 const void *data, size_t size)
595 memcpy(nl_msg_put_unspec_uninit(msg, type, size), data, size);
598 /* Appends a Netlink attribute of the given 'type' and no payload to 'msg'.
599 * (Some Netlink protocols use the presence or absence of an attribute as a
602 nl_msg_put_flag(struct buffer *msg, uint16_t type)
604 nl_msg_put_unspec(msg, type, NULL, 0);
607 /* Appends a Netlink attribute of the given 'type' and the given 8-bit 'value'
610 nl_msg_put_u8(struct buffer *msg, uint16_t type, uint8_t value)
612 nl_msg_put_unspec(msg, type, &value, sizeof value);
615 /* Appends a Netlink attribute of the given 'type' and the given 16-bit 'value'
618 nl_msg_put_u16(struct buffer *msg, uint16_t type, uint16_t value)
620 nl_msg_put_unspec(msg, type, &value, sizeof value);
623 /* Appends a Netlink attribute of the given 'type' and the given 32-bit 'value'
626 nl_msg_put_u32(struct buffer *msg, uint16_t type, uint32_t value)
628 nl_msg_put_unspec(msg, type, &value, sizeof value);
631 /* Appends a Netlink attribute of the given 'type' and the given 64-bit 'value'
634 nl_msg_put_u64(struct buffer *msg, uint16_t type, uint64_t value)
636 nl_msg_put_unspec(msg, type, &value, sizeof value);
639 /* Appends a Netlink attribute of the given 'type' and the given
640 * null-terminated string 'value' to 'msg'. */
642 nl_msg_put_string(struct buffer *msg, uint16_t type, const char *value)
644 nl_msg_put_unspec(msg, type, value, strlen(value) + 1);
647 /* Appends a Netlink attribute of the given 'type' and the given buffered
648 * netlink message in 'nested_msg' to 'msg'. The nlmsg_len field in
649 * 'nested_msg' is finalized to match 'nested_msg->size'. */
651 nl_msg_put_nested(struct buffer *msg,
652 uint16_t type, struct buffer *nested_msg)
654 nl_msg_nlmsghdr(nested_msg)->nlmsg_len = nested_msg->size;
655 nl_msg_put_unspec(msg, type, nested_msg->data, nested_msg->size);
658 /* Returns the first byte in the payload of attribute 'nla'. */
660 nl_attr_get(const struct nlattr *nla)
662 assert(nla->nla_len >= NLA_HDRLEN);
666 /* Returns the number of bytes in the payload of attribute 'nla'. */
668 nl_attr_get_size(const struct nlattr *nla)
670 assert(nla->nla_len >= NLA_HDRLEN);
671 return nla->nla_len - NLA_HDRLEN;
674 /* Asserts that 'nla''s payload is at least 'size' bytes long, and returns the
675 * first byte of the payload. */
677 nl_attr_get_unspec(const struct nlattr *nla, size_t size)
679 assert(nla->nla_len >= NLA_HDRLEN + size);
683 /* Returns true if 'nla' is nonnull. (Some Netlink protocols use the presence
684 * or absence of an attribute as a Boolean flag.) */
686 nl_attr_get_flag(const struct nlattr *nla)
691 #define NL_ATTR_GET_AS(NLA, TYPE) \
692 (*(TYPE*) nl_attr_get_unspec(nla, sizeof(TYPE)))
694 /* Returns the 8-bit value in 'nla''s payload.
696 * Asserts that 'nla''s payload is at least 1 byte long. */
698 nl_attr_get_u8(const struct nlattr *nla)
700 return NL_ATTR_GET_AS(nla, uint8_t);
703 /* Returns the 16-bit value in 'nla''s payload.
705 * Asserts that 'nla''s payload is at least 2 bytes long. */
707 nl_attr_get_u16(const struct nlattr *nla)
709 return NL_ATTR_GET_AS(nla, uint16_t);
712 /* Returns the 32-bit value in 'nla''s payload.
714 * Asserts that 'nla''s payload is at least 4 bytes long. */
716 nl_attr_get_u32(const struct nlattr *nla)
718 return NL_ATTR_GET_AS(nla, uint32_t);
721 /* Returns the 64-bit value in 'nla''s payload.
723 * Asserts that 'nla''s payload is at least 8 bytes long. */
725 nl_attr_get_u64(const struct nlattr *nla)
727 return NL_ATTR_GET_AS(nla, uint64_t);
730 /* Returns the null-terminated string value in 'nla''s payload.
732 * Asserts that 'nla''s payload contains a null-terminated string. */
734 nl_attr_get_string(const struct nlattr *nla)
736 assert(nla->nla_len > NLA_HDRLEN);
737 assert(memchr(nl_attr_get(nla), '\0', nla->nla_len - NLA_HDRLEN) != NULL);
738 return nl_attr_get(nla);
741 /* Default minimum and maximum payload sizes for each type of attribute. */
742 static const size_t attr_len_range[][2] = {
743 [0 ... N_NL_ATTR_TYPES - 1] = { 0, SIZE_MAX },
744 [NL_A_U8] = { 1, 1 },
745 [NL_A_U16] = { 2, 2 },
746 [NL_A_U32] = { 4, 4 },
747 [NL_A_U64] = { 8, 8 },
748 [NL_A_STRING] = { 1, SIZE_MAX },
749 [NL_A_FLAG] = { 0, SIZE_MAX },
750 [NL_A_NESTED] = { NLMSG_HDRLEN, SIZE_MAX },
753 /* Parses the Generic Netlink payload of 'msg' as a sequence of Netlink
754 * attributes. 'policy[i]', for 0 <= i < n_attrs, specifies how the attribute
755 * with nla_type == i is parsed; a pointer to attribute i is stored in
756 * attrs[i]. Returns true if successful, false on failure. */
758 nl_policy_parse(const struct buffer *msg, const struct nl_policy policy[],
759 struct nlattr *attrs[], size_t n_attrs)
766 for (i = 0; i < n_attrs; i++) {
769 assert(policy[i].type < N_NL_ATTR_TYPES);
770 if (policy[i].type != NL_A_NO_ATTR
771 && policy[i].type != NL_A_FLAG
772 && !policy[i].optional) {
777 p = buffer_at(msg, NLMSG_HDRLEN + GENL_HDRLEN, 0);
779 VLOG_DBG("missing headers in nl_policy_parse");
782 tail = buffer_tail(msg);
785 size_t offset = p - msg->data;
786 struct nlattr *nla = p;
787 size_t len, aligned_len;
790 /* Make sure its claimed length is plausible. */
791 if (nla->nla_len < NLA_HDRLEN) {
792 VLOG_DBG("%zu: attr shorter than NLA_HDRLEN (%"PRIu16")",
793 offset, nla->nla_len);
796 len = nla->nla_len - NLA_HDRLEN;
797 aligned_len = NLA_ALIGN(len);
798 if (aligned_len > tail - p) {
799 VLOG_DBG("%zu: attr %"PRIu16" aligned data len (%zu) "
800 "> bytes left (%tu)",
801 offset, nla->nla_type, aligned_len, tail - p);
805 type = nla->nla_type;
806 if (type < n_attrs && policy[type].type != NL_A_NO_ATTR) {
807 const struct nl_policy *p = &policy[type];
808 size_t min_len, max_len;
810 /* Validate length and content. */
811 min_len = p->min_len ? p->min_len : attr_len_range[p->type][0];
812 max_len = p->max_len ? p->max_len : attr_len_range[p->type][1];
813 if (len < min_len || len > max_len) {
814 VLOG_DBG("%zu: attr %"PRIu16" length %zu not in allowed range "
815 "%zu...%zu", offset, type, len, min_len, max_len);
818 if (p->type == NL_A_STRING) {
819 if (((char *) nla)[nla->nla_len - 1]) {
820 VLOG_DBG("%zu: attr %"PRIu16" lacks null terminator",
824 if (memchr(nla + 1, '\0', len - 1) != NULL) {
825 VLOG_DBG("%zu: attr %"PRIu16" lies about string length",
830 if (!p->optional && attrs[type] == NULL) {
831 assert(n_required > 0);
836 /* Skip attribute type that we don't care about. */
838 p += NLA_ALIGN(nla->nla_len);
841 VLOG_DBG("%zu required attrs missing", n_required);
849 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
850 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
853 static int do_lookup_genl_family(const char *name)
855 struct nl_sock *sock;
856 struct buffer request, *reply;
857 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
860 retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
865 buffer_init(&request, 0);
866 nl_msg_put_genlmsghdr(&request, sock, 0, GENL_ID_CTRL, NLM_F_REQUEST,
867 CTRL_CMD_GETFAMILY, 1);
868 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
869 retval = nl_sock_transact(sock, &request, &reply);
870 buffer_uninit(&request);
872 nl_sock_destroy(sock);
876 if (!nl_policy_parse(reply, family_policy, attrs,
877 ARRAY_SIZE(family_policy))) {
878 nl_sock_destroy(sock);
879 buffer_delete(reply);
883 retval = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
887 nl_sock_destroy(sock);
888 buffer_delete(reply);
892 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
893 * number and stores it in '*number'. If successful, returns 0 and the caller
894 * may use '*number' as the family number. On failure, returns a positive
895 * errno value and '*number' caches the errno value. */
897 nl_lookup_genl_family(const char *name, int *number)
900 *number = do_lookup_genl_family(name);
901 assert(*number != 0);
903 return *number > 0 ? 0 : -*number;
908 * Every Netlink socket must be bound to a unique 32-bit PID. By convention,
909 * programs that have a single Netlink socket use their Unix process ID as PID,
910 * and programs with multiple Netlink sockets add a unique per-socket
911 * identifier in the bits above the Unix process ID.
913 * The kernel has Netlink PID 0.
916 /* Parameters for how many bits in the PID should come from the Unix process ID
917 * and how many unique per-socket. */
918 #define SOCKET_BITS 10
919 #define MAX_SOCKETS (1u << SOCKET_BITS)
921 #define PROCESS_BITS (32 - SOCKET_BITS)
922 #define MAX_PROCESSES (1u << PROCESS_BITS)
923 #define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1))
925 /* Bit vector of unused socket identifiers. */
926 static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)];
928 /* Allocates and returns a new Netlink PID. */
930 alloc_pid(uint32_t *pid)
934 for (i = 0; i < MAX_SOCKETS; i++) {
935 if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) {
936 avail_sockets[i / 32] |= 1u << (i % 32);
937 *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS);
941 VLOG_ERR("netlink pid space exhausted");
945 /* Makes the specified 'pid' available for reuse. */
947 free_pid(uint32_t pid)
949 int sock = pid >> PROCESS_BITS;
950 assert(avail_sockets[sock / 32] & (1u << (sock % 32)));
951 avail_sockets[sock / 32] &= ~(1u << (sock % 32));
955 nlmsghdr_to_string(const struct nlmsghdr *h, struct ds *ds)
961 static const struct nlmsg_flag flags[] = {
962 { NLM_F_REQUEST, "REQUEST" },
963 { NLM_F_MULTI, "MULTI" },
964 { NLM_F_ACK, "ACK" },
965 { NLM_F_ECHO, "ECHO" },
966 { NLM_F_DUMP, "DUMP" },
967 { NLM_F_ROOT, "ROOT" },
968 { NLM_F_MATCH, "MATCH" },
969 { NLM_F_ATOMIC, "ATOMIC" },
971 const struct nlmsg_flag *flag;
974 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
975 h->nlmsg_len, h->nlmsg_type);
976 if (h->nlmsg_type == NLMSG_NOOP) {
977 ds_put_cstr(ds, "(no-op)");
978 } else if (h->nlmsg_type == NLMSG_ERROR) {
979 ds_put_cstr(ds, "(error)");
980 } else if (h->nlmsg_type == NLMSG_DONE) {
981 ds_put_cstr(ds, "(done)");
982 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
983 ds_put_cstr(ds, "(overrun)");
984 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
985 ds_put_cstr(ds, "(reserved)");
987 ds_put_cstr(ds, "(family-defined)");
989 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
990 flags_left = h->nlmsg_flags;
991 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
992 if ((flags_left & flag->bits) == flag->bits) {
993 ds_put_format(ds, "[%s]", flag->name);
994 flags_left &= ~flag->bits;
998 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1000 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32"(%d:%d))",
1001 h->nlmsg_seq, h->nlmsg_pid,
1002 (int) (h->nlmsg_pid & PROCESS_MASK),
1003 (int) (h->nlmsg_pid >> PROCESS_BITS));
1007 nlmsg_to_string(const struct buffer *buffer)
1009 struct ds ds = DS_EMPTY_INITIALIZER;
1010 const struct nlmsghdr *h = buffer_at(buffer, 0, NLMSG_HDRLEN);
1012 nlmsghdr_to_string(h, &ds);
1013 if (h->nlmsg_type == NLMSG_ERROR) {
1014 const struct nlmsgerr *e;
1015 e = buffer_at(buffer, NLMSG_HDRLEN,
1016 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1018 ds_put_format(&ds, " error(%d", e->error);
1020 ds_put_format(&ds, "(%s)", strerror(-e->error));
1022 ds_put_cstr(&ds, ", in-reply-to(");
1023 nlmsghdr_to_string(&e->msg, &ds);
1024 ds_put_cstr(&ds, "))");
1026 ds_put_cstr(&ds, " error(truncated)");
1028 } else if (h->nlmsg_type == NLMSG_DONE) {
1029 int *error = buffer_at(buffer, NLMSG_HDRLEN, sizeof *error);
1031 ds_put_format(&ds, " done(%d", *error);
1033 ds_put_format(&ds, "(%s)", strerror(-*error));
1035 ds_put_cstr(&ds, ")");
1037 ds_put_cstr(&ds, " done(truncated)");
1041 ds_put_cstr(&ds, "nl(truncated)");
1047 log_nlmsg(const char *function, int error,
1048 const void *message, size_t size)
1050 struct buffer buffer;
1053 if (!VLOG_IS_DBG_ENABLED()) {
1057 buffer.data = (void *) message;
1059 nlmsg = nlmsg_to_string(&buffer);
1060 VLOG_DBG("%s (%s): %s", function, strerror(error), nlmsg);