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.
45 #include "netlink-protocol.h"
46 #include "dynamic-string.h"
50 #define THIS_MODULE VLM_netlink
52 /* Linux header file confusion causes this to be undefined. */
54 #define SOL_NETLINK 270
57 static void log_nlmsg(const char *function, int error,
58 const void *message, size_t size);
60 /* Netlink sockets. */
68 /* Next nlmsghdr sequence number.
70 * This implementation uses sequence numbers that are unique process-wide, to
71 * avoid a hypothetical race: send request, close socket, open new socket that
72 * reuses the old socket's PID value, send request on new socket, receive reply
73 * from kernel to old socket but with same PID and sequence number. (This race
74 * could be avoided other ways, e.g. by preventing PIDs from being quickly
76 static uint32_t next_seq;
78 static int alloc_pid(uint32_t *);
79 static void free_pid(uint32_t);
81 /* Creates a new netlink socket for the given netlink 'protocol'
82 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
83 * new socket if successful, otherwise returns a positive errno value.
85 * If 'multicast_group' is nonzero, the new socket subscribes to the specified
86 * netlink multicast group. (A netlink socket may listen to an arbitrary
87 * number of multicast groups, but so far we only need one at a time.)
89 * Nonzero 'so_sndbuf' or 'so_rcvbuf' override the kernel default send or
90 * receive buffer size, respectively.
93 nl_sock_create(int protocol, int multicast_group,
94 size_t so_sndbuf, size_t so_rcvbuf, struct nl_sock **sockp)
97 struct sockaddr_nl local, remote;
101 /* Pick initial sequence number. */
102 next_seq = getpid() ^ time(0);
106 sock = malloc(sizeof *sock);
111 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
113 VLOG_ERR("fcntl: %s", strerror(errno));
117 retval = alloc_pid(&sock->pid);
123 && setsockopt(sock->fd, SOL_SOCKET, SO_SNDBUF,
124 &so_sndbuf, sizeof so_sndbuf) < 0) {
125 VLOG_ERR("setsockopt(SO_SNDBUF,%zu): %s", so_sndbuf, strerror(errno));
130 && setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
131 &so_rcvbuf, sizeof so_rcvbuf) < 0) {
132 VLOG_ERR("setsockopt(SO_RCVBUF,%zu): %s", so_rcvbuf, strerror(errno));
136 /* Bind local address as our selected pid. */
137 memset(&local, 0, sizeof local);
138 local.nl_family = AF_NETLINK;
139 local.nl_pid = sock->pid;
140 if (multicast_group > 0 && multicast_group <= 32) {
141 /* This method of joining multicast groups is supported by old kernels,
142 * but it only allows 32 multicast groups per protocol. */
143 local.nl_groups |= 1ul << (multicast_group - 1);
145 if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) {
146 VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno));
150 /* Bind remote address as the kernel (pid 0). */
151 memset(&remote, 0, sizeof remote);
152 remote.nl_family = AF_NETLINK;
154 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
155 VLOG_ERR("connect(0): %s", strerror(errno));
159 /* Older kernel headers failed to define this macro. We want our programs
160 * to support the newer kernel features even if compiled with older
161 * headers, so define it ourselves in such a case. */
162 #ifndef NETLINK_ADD_MEMBERSHIP
163 #define NETLINK_ADD_MEMBERSHIP 1
166 /* This method of joining multicast groups is only supported by newish
167 * kernels, but it allows for an arbitrary number of multicast groups. */
168 if (multicast_group > 32
169 && setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
170 &multicast_group, sizeof multicast_group) < 0) {
171 VLOG_ERR("setsockopt(NETLINK_ADD_MEMBERSHIP,%d): %s",
172 multicast_group, strerror(errno));
195 /* Destroys netlink socket 'sock'. */
197 nl_sock_destroy(struct nl_sock *sock)
206 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
207 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size before the
210 * Returns 0 if successful, otherwise a positive errno value. If
211 * 'wait' is true, then the send will wait until buffer space is ready;
212 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
214 nl_sock_send(struct nl_sock *sock, const struct buffer *msg, bool wait)
218 nl_msg_nlmsghdr(msg)->nlmsg_len = msg->size;
221 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
222 error = retval < 0 ? errno : 0;
223 } while (error == EINTR);
224 log_nlmsg(__func__, error, msg->data, msg->size);
228 /* Tries to send the 'n_iov' chunks of data in 'iov' to the kernel on 'sock' as
229 * a single Netlink message. (The message must be fully formed and not require
230 * finalization of its nlmsg_len field.)
232 * Returns 0 if successful, otherwise a positive errno value. If 'wait' is
233 * true, then the send will wait until buffer space is ready; otherwise,
234 * returns EAGAIN if the 'sock' send buffer is full. */
236 nl_sock_sendv(struct nl_sock *sock, const struct iovec iov[], size_t n_iov,
242 memset(&msg, 0, sizeof msg);
243 msg.msg_iov = (struct iovec *) iov;
244 msg.msg_iovlen = n_iov;
247 retval = sendmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
248 error = retval < 0 ? errno : 0;
249 } while (error == EINTR);
250 if (error != EAGAIN) {
251 log_nlmsg(__func__, error, iov[0].iov_base, iov[0].iov_len);
256 /* Tries to receive a netlink message from the kernel on 'sock'. If
257 * successful, stores the received message into '*bufp' and returns 0. The
258 * caller is responsible for destroying the message with buffer_delete(). On
259 * failure, returns a positive errno value and stores a null pointer into
262 * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
263 * returns EAGAIN if the 'sock' receive buffer is empty. */
265 nl_sock_recv(struct nl_sock *sock, struct buffer **bufp, bool wait)
268 ssize_t bufsize = 2048;
269 ssize_t nbytes, nbytes2;
271 struct nlmsghdr *nlmsghdr;
273 struct msghdr msg = {
283 buf = buffer_new(bufsize);
287 /* Attempt to read the message. We don't know the size of the data
288 * yet, so we take a guess at 2048. If we're wrong, we keep trying
289 * and doubling the buffer size each time.
291 nlmsghdr = buffer_put_uninit(buf, bufsize);
292 iov.iov_base = nlmsghdr;
293 iov.iov_len = bufsize;
295 nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK);
296 } while (nbytes < 0 && errno == EINTR);
301 if (msg.msg_flags & MSG_TRUNC) {
303 buffer_reinit(buf, bufsize);
308 /* We successfully read the message, so recv again to clear the queue */
312 nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT);
313 } while (nbytes2 < 0 && errno == EINTR);
315 if (errno == ENOBUFS) {
316 /* The kernel is notifying us that a message it tried to send to us
317 * was dropped. We have to pass this along to the caller in case
318 * it wants to retry a request. So kill the buffer, which we can
319 * re-read next time. */
323 VLOG_ERR("failed to remove nlmsg from socket: %s\n",
327 if (nbytes < sizeof *nlmsghdr
328 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
329 || nlmsghdr->nlmsg_len > nbytes) {
330 VLOG_ERR("received invalid nlmsg (%zd bytes < %d)",
331 bufsize, NLMSG_HDRLEN);
336 log_nlmsg(__func__, 0, buf->data, buf->size);
340 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
341 * successful, stores the reply into '*replyp' and returns 0. The caller is
342 * responsible for destroying the reply with buffer_delete(). On failure,
343 * returns a positive errno value and stores a null pointer into '*replyp'.
345 * Bare Netlink is an unreliable transport protocol. This function layers
346 * reliable delivery and reply semantics on top of bare Netlink.
348 * In Netlink, sending a request to the kernel is reliable enough, because the
349 * kernel will tell us if the message cannot be queued (and we will in that
350 * case put it on the transmit queue and wait until it can be delivered).
352 * Receiving the reply is the real problem: if the socket buffer is full when
353 * the kernel tries to send the reply, the reply will be dropped. However, the
354 * kernel sets a flag that a reply has been dropped. The next call to recv
355 * then returns ENOBUFS. We can then re-send the request.
359 * 1. Netlink depends on sequence numbers to match up requests and
360 * replies. The sender of a request supplies a sequence number, and
361 * the reply echos back that sequence number.
363 * This is fine, but (1) some kernel netlink implementations are
364 * broken, in that they fail to echo sequence numbers and (2) this
365 * function will drop packets with non-matching sequence numbers, so
366 * that only a single request can be usefully transacted at a time.
368 * 2. Resending the request causes it to be re-executed, so the request
369 * needs to be idempotent.
372 nl_sock_transact(struct nl_sock *sock,
373 const struct buffer *request, struct buffer **replyp)
375 uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
376 struct nlmsghdr *nlmsghdr;
377 struct buffer *reply;
382 /* Ensure that we get a reply even if this message doesn't ordinarily call
384 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK;
387 retval = nl_sock_send(sock, request, true);
393 retval = nl_sock_recv(sock, &reply, true);
395 if (retval == ENOBUFS) {
396 VLOG_DBG("receive buffer overflow, resending request");
402 nlmsghdr = nl_msg_nlmsghdr(reply);
403 if (seq != nlmsghdr->nlmsg_seq) {
404 VLOG_DBG("ignoring seq %"PRIu32" != expected %"PRIu32,
405 nl_msg_nlmsghdr(reply)->nlmsg_seq, seq);
406 buffer_delete(reply);
409 if (nl_msg_nlmsgerr(reply, &retval)) {
411 VLOG_DBG("received NAK error=%d (%s)", retval, strerror(retval));
413 return retval != EAGAIN ? retval : EPROTO;
420 /* Returns 'sock''s underlying file descriptor. */
422 nl_sock_fd(const struct nl_sock *sock)
427 /* Netlink messages. */
429 /* Returns the nlmsghdr at the head of 'msg'.
431 * 'msg' must be at least as large as a nlmsghdr. */
433 nl_msg_nlmsghdr(const struct buffer *msg)
435 return buffer_at_assert(msg, 0, NLMSG_HDRLEN);
438 /* Returns the genlmsghdr just past 'msg''s nlmsghdr.
440 * Returns a null pointer if 'msg' is not large enough to contain an nlmsghdr
441 * and a genlmsghdr. */
443 nl_msg_genlmsghdr(const struct buffer *msg)
445 return buffer_at(msg, NLMSG_HDRLEN, GENL_HDRLEN);
448 /* If 'buffer' is a NLMSG_ERROR message, stores 0 in '*errorp' if it is an ACK
449 * message, otherwise a positive errno value, and returns true. If 'buffer' is
450 * not an NLMSG_ERROR message, returns false.
452 * 'msg' must be at least as large as a nlmsghdr. */
454 nl_msg_nlmsgerr(const struct buffer *msg, int *errorp)
456 if (nl_msg_nlmsghdr(msg)->nlmsg_type == NLMSG_ERROR) {
457 struct nlmsgerr *err = buffer_at(msg, NLMSG_HDRLEN, sizeof *err);
460 VLOG_ERR("received invalid nlmsgerr (%zd bytes < %zd)",
461 msg->size, NLMSG_HDRLEN + sizeof *err);
462 } else if (err->error <= 0 && err->error > INT_MIN) {
474 /* Ensures that 'b' has room for at least 'size' bytes plus netlink padding at
475 * its tail end, reallocating and copying its data if necessary. */
477 nl_msg_reserve(struct buffer *msg, size_t size)
479 buffer_prealloc_tailroom(msg, NLMSG_ALIGN(size));
482 /* Puts a nlmsghdr at the beginning of 'msg', which must be initially empty.
483 * Uses the given 'type' and 'flags'. 'sock' is used to obtain a PID and
484 * sequence number for proper routing of replies. 'expected_payload' should be
485 * an estimate of the number of payload bytes to be supplied; if the size of
486 * the payload is unknown a value of 0 is acceptable.
488 * 'type' is ordinarily an enumerated value specific to the Netlink protocol
489 * (e.g. RTM_NEWLINK, for NETLINK_ROUTE protocol). For Generic Netlink, 'type'
490 * 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 * nl_msg_put_genlmsghdr is more convenient for composing a Generic Netlink
499 nl_msg_put_nlmsghdr(struct buffer *msg, struct nl_sock *sock,
500 size_t expected_payload, uint32_t type, uint32_t flags)
502 struct nlmsghdr *nlmsghdr;
504 assert(msg->size == 0);
506 nl_msg_reserve(msg, NLMSG_HDRLEN + expected_payload);
507 nlmsghdr = nl_msg_put_uninit(msg, NLMSG_HDRLEN);
508 nlmsghdr->nlmsg_len = 0;
509 nlmsghdr->nlmsg_type = type;
510 nlmsghdr->nlmsg_flags = flags;
511 nlmsghdr->nlmsg_seq = ++next_seq;
512 nlmsghdr->nlmsg_pid = sock->pid;
515 /* Puts a nlmsghdr and genlmsghdr at the beginning of 'msg', which must be
516 * initially empty. 'sock' is used to obtain a PID and sequence number for
517 * proper routing of replies. 'expected_payload' should be an estimate of the
518 * number of payload bytes to be supplied; if the size of the payload is
519 * unknown a value of 0 is acceptable.
521 * 'family' is the family number obtained via nl_lookup_genl_family().
523 * 'flags' is a bit-mask that indicates what kind of request is being made. It
524 * is often NLM_F_REQUEST indicating that a request is being made, commonly
525 * or'd with NLM_F_ACK to request an acknowledgement.
527 * 'cmd' is an enumerated value specific to the Generic Netlink family
528 * (e.g. CTRL_CMD_NEWFAMILY for the GENL_ID_CTRL family).
530 * 'version' is a version number specific to the family and command (often 1).
532 * nl_msg_put_nlmsghdr should be used to compose Netlink messages that are not
533 * Generic Netlink messages. */
535 nl_msg_put_genlmsghdr(struct buffer *msg, struct nl_sock *sock,
536 size_t expected_payload, int family, uint32_t flags,
537 uint8_t cmd, uint8_t version)
539 struct genlmsghdr *genlmsghdr;
541 nl_msg_put_nlmsghdr(msg, sock, GENL_HDRLEN + expected_payload,
543 assert(msg->size == NLMSG_HDRLEN);
544 genlmsghdr = nl_msg_put_uninit(msg, GENL_HDRLEN);
545 genlmsghdr->cmd = cmd;
546 genlmsghdr->version = version;
547 genlmsghdr->reserved = 0;
550 /* Appends the 'size' bytes of data in 'p', plus Netlink padding if needed, to
551 * the tail end of 'msg'. Data in 'msg' is reallocated and copied if
554 nl_msg_put(struct buffer *msg, const void *data, size_t size)
556 memcpy(nl_msg_put_uninit(msg, size), data, size);
559 /* Appends 'size' bytes of data, plus Netlink padding if needed, to the tail
560 * end of 'msg', reallocating and copying its data if necessary. Returns a
561 * pointer to the first byte of the new data, which is left uninitialized. */
563 nl_msg_put_uninit(struct buffer *msg, size_t size)
565 size_t pad = NLMSG_ALIGN(size) - size;
566 char *p = buffer_put_uninit(msg, size + pad);
568 memset(p + size, 0, pad);
573 /* Appends a Netlink attribute of the given 'type' and room for 'size' bytes of
574 * data as its payload, plus Netlink padding if needed, to the tail end of
575 * 'msg', reallocating and copying its data if necessary. Returns a pointer to
576 * the first byte of data in the attribute, which is left uninitialized. */
578 nl_msg_put_unspec_uninit(struct buffer *msg, uint16_t type, size_t size)
580 size_t total_size = NLA_HDRLEN + size;
581 struct nlattr* nla = nl_msg_put_uninit(msg, total_size);
582 assert(NLA_ALIGN(total_size) <= UINT16_MAX);
583 nla->nla_len = total_size;
584 nla->nla_type = type;
588 /* Appends a Netlink attribute of the given 'type' and the 'size' bytes of
589 * 'data' as its payload, to the tail end of 'msg', reallocating and copying
590 * its data if necessary. Returns a pointer to the first byte of data in the
591 * attribute, which is left uninitialized. */
593 nl_msg_put_unspec(struct buffer *msg, uint16_t type,
594 const void *data, size_t size)
596 memcpy(nl_msg_put_unspec_uninit(msg, type, size), data, size);
599 /* Appends a Netlink attribute of the given 'type' and no payload to 'msg'.
600 * (Some Netlink protocols use the presence or absence of an attribute as a
603 nl_msg_put_flag(struct buffer *msg, uint16_t type)
605 nl_msg_put_unspec(msg, type, NULL, 0);
608 /* Appends a Netlink attribute of the given 'type' and the given 8-bit 'value'
611 nl_msg_put_u8(struct buffer *msg, uint16_t type, uint8_t value)
613 nl_msg_put_unspec(msg, type, &value, sizeof value);
616 /* Appends a Netlink attribute of the given 'type' and the given 16-bit 'value'
619 nl_msg_put_u16(struct buffer *msg, uint16_t type, uint16_t value)
621 nl_msg_put_unspec(msg, type, &value, sizeof value);
624 /* Appends a Netlink attribute of the given 'type' and the given 32-bit 'value'
627 nl_msg_put_u32(struct buffer *msg, uint16_t type, uint32_t value)
629 nl_msg_put_unspec(msg, type, &value, sizeof value);
632 /* Appends a Netlink attribute of the given 'type' and the given 64-bit 'value'
635 nl_msg_put_u64(struct buffer *msg, uint16_t type, uint64_t value)
637 nl_msg_put_unspec(msg, type, &value, sizeof value);
640 /* Appends a Netlink attribute of the given 'type' and the given
641 * null-terminated string 'value' to 'msg'. */
643 nl_msg_put_string(struct buffer *msg, uint16_t type, const char *value)
645 nl_msg_put_unspec(msg, type, value, strlen(value) + 1);
648 /* Appends a Netlink attribute of the given 'type' and the given buffered
649 * netlink message in 'nested_msg' to 'msg'. The nlmsg_len field in
650 * 'nested_msg' is finalized to match 'nested_msg->size'. */
652 nl_msg_put_nested(struct buffer *msg,
653 uint16_t type, struct buffer *nested_msg)
655 nl_msg_nlmsghdr(nested_msg)->nlmsg_len = nested_msg->size;
656 nl_msg_put_unspec(msg, type, nested_msg->data, nested_msg->size);
659 /* Returns the first byte in the payload of attribute 'nla'. */
661 nl_attr_get(const struct nlattr *nla)
663 assert(nla->nla_len >= NLA_HDRLEN);
667 /* Returns the number of bytes in the payload of attribute 'nla'. */
669 nl_attr_get_size(const struct nlattr *nla)
671 assert(nla->nla_len >= NLA_HDRLEN);
672 return nla->nla_len - NLA_HDRLEN;
675 /* Asserts that 'nla''s payload is at least 'size' bytes long, and returns the
676 * first byte of the payload. */
678 nl_attr_get_unspec(const struct nlattr *nla, size_t size)
680 assert(nla->nla_len >= NLA_HDRLEN + size);
684 /* Returns true if 'nla' is nonnull. (Some Netlink protocols use the presence
685 * or absence of an attribute as a Boolean flag.) */
687 nl_attr_get_flag(const struct nlattr *nla)
692 #define NL_ATTR_GET_AS(NLA, TYPE) \
693 (*(TYPE*) nl_attr_get_unspec(nla, sizeof(TYPE)))
695 /* Returns the 8-bit value in 'nla''s payload.
697 * Asserts that 'nla''s payload is at least 1 byte long. */
699 nl_attr_get_u8(const struct nlattr *nla)
701 return NL_ATTR_GET_AS(nla, uint8_t);
704 /* Returns the 16-bit value in 'nla''s payload.
706 * Asserts that 'nla''s payload is at least 2 bytes long. */
708 nl_attr_get_u16(const struct nlattr *nla)
710 return NL_ATTR_GET_AS(nla, uint16_t);
713 /* Returns the 32-bit value in 'nla''s payload.
715 * Asserts that 'nla''s payload is at least 4 bytes long. */
717 nl_attr_get_u32(const struct nlattr *nla)
719 return NL_ATTR_GET_AS(nla, uint32_t);
722 /* Returns the 64-bit value in 'nla''s payload.
724 * Asserts that 'nla''s payload is at least 8 bytes long. */
726 nl_attr_get_u64(const struct nlattr *nla)
728 return NL_ATTR_GET_AS(nla, uint64_t);
731 /* Returns the null-terminated string value in 'nla''s payload.
733 * Asserts that 'nla''s payload contains a null-terminated string. */
735 nl_attr_get_string(const struct nlattr *nla)
737 assert(nla->nla_len > NLA_HDRLEN);
738 assert(memchr(nl_attr_get(nla), '\0', nla->nla_len - NLA_HDRLEN) != NULL);
739 return nl_attr_get(nla);
742 /* Default minimum and maximum payload sizes for each type of attribute. */
743 static const size_t attr_len_range[][2] = {
744 [0 ... N_NL_ATTR_TYPES - 1] = { 0, SIZE_MAX },
745 [NL_A_U8] = { 1, 1 },
746 [NL_A_U16] = { 2, 2 },
747 [NL_A_U32] = { 4, 4 },
748 [NL_A_U64] = { 8, 8 },
749 [NL_A_STRING] = { 1, SIZE_MAX },
750 [NL_A_FLAG] = { 0, SIZE_MAX },
751 [NL_A_NESTED] = { NLMSG_HDRLEN, SIZE_MAX },
754 /* Parses the Generic Netlink payload of 'msg' as a sequence of Netlink
755 * attributes. 'policy[i]', for 0 <= i < n_attrs, specifies how the attribute
756 * with nla_type == i is parsed; a pointer to attribute i is stored in
757 * attrs[i]. Returns true if successful, false on failure. */
759 nl_policy_parse(const struct buffer *msg, const struct nl_policy policy[],
760 struct nlattr *attrs[], size_t n_attrs)
767 for (i = 0; i < n_attrs; i++) {
770 assert(policy[i].type < N_NL_ATTR_TYPES);
771 if (policy[i].type != NL_A_NO_ATTR
772 && policy[i].type != NL_A_FLAG
773 && !policy[i].optional) {
778 p = buffer_at(msg, NLMSG_HDRLEN + GENL_HDRLEN, 0);
780 VLOG_DBG("missing headers in nl_policy_parse");
783 tail = buffer_tail(msg);
786 size_t offset = p - msg->data;
787 struct nlattr *nla = p;
788 size_t len, aligned_len;
791 /* Make sure its claimed length is plausible. */
792 if (nla->nla_len < NLA_HDRLEN) {
793 VLOG_DBG("%zu: attr shorter than NLA_HDRLEN (%"PRIu16")",
794 offset, nla->nla_len);
797 len = nla->nla_len - NLA_HDRLEN;
798 aligned_len = NLA_ALIGN(len);
799 if (aligned_len > tail - p) {
800 VLOG_DBG("%zu: attr %"PRIu16" aligned data len (%zu) "
801 "> bytes left (%tu)",
802 offset, nla->nla_type, aligned_len, tail - p);
806 type = nla->nla_type;
807 if (type < n_attrs && policy[type].type != NL_A_NO_ATTR) {
808 const struct nl_policy *p = &policy[type];
809 size_t min_len, max_len;
811 /* Validate length and content. */
812 min_len = p->min_len ? p->min_len : attr_len_range[p->type][0];
813 max_len = p->max_len ? p->max_len : attr_len_range[p->type][1];
814 if (len < min_len || len > max_len) {
815 VLOG_DBG("%zu: attr %"PRIu16" length %zu not in allowed range "
816 "%zu...%zu", offset, type, len, min_len, max_len);
819 if (p->type == NL_A_STRING) {
820 if (((char *) nla)[nla->nla_len - 1]) {
821 VLOG_DBG("%zu: attr %"PRIu16" lacks null terminator",
825 if (memchr(nla + 1, '\0', len - 1) != NULL) {
826 VLOG_DBG("%zu: attr %"PRIu16" lies about string length",
831 if (!p->optional && attrs[type] == NULL) {
832 assert(n_required > 0);
837 /* Skip attribute type that we don't care about. */
839 p += NLA_ALIGN(nla->nla_len);
842 VLOG_DBG("%zu required attrs missing", n_required);
850 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
851 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
854 static int do_lookup_genl_family(const char *name)
856 struct nl_sock *sock;
857 struct buffer request, *reply;
858 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
861 retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock);
866 buffer_init(&request, 0);
867 nl_msg_put_genlmsghdr(&request, sock, 0, GENL_ID_CTRL, NLM_F_REQUEST,
868 CTRL_CMD_GETFAMILY, 1);
869 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
870 retval = nl_sock_transact(sock, &request, &reply);
871 buffer_uninit(&request);
873 nl_sock_destroy(sock);
877 if (!nl_policy_parse(reply, family_policy, attrs,
878 ARRAY_SIZE(family_policy))) {
879 nl_sock_destroy(sock);
880 buffer_delete(reply);
884 retval = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
888 nl_sock_destroy(sock);
889 buffer_delete(reply);
893 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
894 * number and stores it in '*number'. If successful, returns 0 and the caller
895 * may use '*number' as the family number. On failure, returns a positive
896 * errno value and '*number' caches the errno value. */
898 nl_lookup_genl_family(const char *name, int *number)
901 *number = do_lookup_genl_family(name);
902 assert(*number != 0);
904 return *number > 0 ? 0 : -*number;
909 * Every Netlink socket must be bound to a unique 32-bit PID. By convention,
910 * programs that have a single Netlink socket use their Unix process ID as PID,
911 * and programs with multiple Netlink sockets add a unique per-socket
912 * identifier in the bits above the Unix process ID.
914 * The kernel has Netlink PID 0.
917 /* Parameters for how many bits in the PID should come from the Unix process ID
918 * and how many unique per-socket. */
919 #define SOCKET_BITS 10
920 #define MAX_SOCKETS (1u << SOCKET_BITS)
922 #define PROCESS_BITS (32 - SOCKET_BITS)
923 #define MAX_PROCESSES (1u << PROCESS_BITS)
924 #define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1))
926 /* Bit vector of unused socket identifiers. */
927 static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)];
929 /* Allocates and returns a new Netlink PID. */
931 alloc_pid(uint32_t *pid)
935 for (i = 0; i < MAX_SOCKETS; i++) {
936 if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) {
937 avail_sockets[i / 32] |= 1u << (i % 32);
938 *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS);
942 VLOG_ERR("netlink pid space exhausted");
946 /* Makes the specified 'pid' available for reuse. */
948 free_pid(uint32_t pid)
950 int sock = pid >> PROCESS_BITS;
951 assert(avail_sockets[sock / 32] & (1u << (sock % 32)));
952 avail_sockets[sock / 32] &= ~(1u << (sock % 32));
956 nlmsghdr_to_string(const struct nlmsghdr *h, struct ds *ds)
962 static const struct nlmsg_flag flags[] = {
963 { NLM_F_REQUEST, "REQUEST" },
964 { NLM_F_MULTI, "MULTI" },
965 { NLM_F_ACK, "ACK" },
966 { NLM_F_ECHO, "ECHO" },
967 { NLM_F_DUMP, "DUMP" },
968 { NLM_F_ROOT, "ROOT" },
969 { NLM_F_MATCH, "MATCH" },
970 { NLM_F_ATOMIC, "ATOMIC" },
972 const struct nlmsg_flag *flag;
975 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
976 h->nlmsg_len, h->nlmsg_type);
977 if (h->nlmsg_type == NLMSG_NOOP) {
978 ds_put_cstr(ds, "(no-op)");
979 } else if (h->nlmsg_type == NLMSG_ERROR) {
980 ds_put_cstr(ds, "(error)");
981 } else if (h->nlmsg_type == NLMSG_DONE) {
982 ds_put_cstr(ds, "(done)");
983 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
984 ds_put_cstr(ds, "(overrun)");
985 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
986 ds_put_cstr(ds, "(reserved)");
988 ds_put_cstr(ds, "(family-defined)");
990 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
991 flags_left = h->nlmsg_flags;
992 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
993 if ((flags_left & flag->bits) == flag->bits) {
994 ds_put_format(ds, "[%s]", flag->name);
995 flags_left &= ~flag->bits;
999 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1001 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32"(%d:%d))",
1002 h->nlmsg_seq, h->nlmsg_pid,
1003 (int) (h->nlmsg_pid & PROCESS_MASK),
1004 (int) (h->nlmsg_pid >> PROCESS_BITS));
1008 nlmsg_to_string(const struct buffer *buffer)
1010 struct ds ds = DS_EMPTY_INITIALIZER;
1011 const struct nlmsghdr *h = buffer_at(buffer, 0, NLMSG_HDRLEN);
1013 nlmsghdr_to_string(h, &ds);
1014 if (h->nlmsg_type == NLMSG_ERROR) {
1015 const struct nlmsgerr *e;
1016 e = buffer_at(buffer, NLMSG_HDRLEN,
1017 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1019 ds_put_format(&ds, " error(%d", e->error);
1021 ds_put_format(&ds, "(%s)", strerror(-e->error));
1023 ds_put_cstr(&ds, ", in-reply-to(");
1024 nlmsghdr_to_string(&e->msg, &ds);
1025 ds_put_cstr(&ds, "))");
1027 ds_put_cstr(&ds, " error(truncated)");
1029 } else if (h->nlmsg_type == NLMSG_DONE) {
1030 int *error = buffer_at(buffer, NLMSG_HDRLEN, sizeof *error);
1032 ds_put_format(&ds, " done(%d", *error);
1034 ds_put_format(&ds, "(%s)", strerror(-*error));
1036 ds_put_cstr(&ds, ")");
1038 ds_put_cstr(&ds, " done(truncated)");
1042 ds_put_cstr(&ds, "nl(truncated)");
1048 log_nlmsg(const char *function, int error,
1049 const void *message, size_t size)
1051 struct buffer buffer;
1054 if (!VLOG_IS_DBG_ENABLED()) {
1058 buffer.data = (void *) message;
1060 nlmsg = nlmsg_to_string(&buffer);
1061 VLOG_DBG("%s (%s): %s", function, strerror(error), nlmsg);