2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "netlink-socket.h"
23 #include <sys/types.h>
27 #include "dynamic-string.h"
31 #include "netlink-protocol.h"
33 #include "poll-loop.h"
34 #include "socket-util.h"
39 VLOG_DEFINE_THIS_MODULE(netlink_socket);
41 COVERAGE_DEFINE(netlink_overflow);
42 COVERAGE_DEFINE(netlink_received);
43 COVERAGE_DEFINE(netlink_recv_jumbo);
44 COVERAGE_DEFINE(netlink_send);
45 COVERAGE_DEFINE(netlink_sent);
47 /* Linux header file confusion causes this to be undefined. */
49 #define SOL_NETLINK 270
52 /* A single (bad) Netlink message can in theory dump out many, many log
53 * messages, so the burst size is set quite high here to avoid missing useful
54 * information. Also, at high logging levels we log *all* Netlink messages. */
55 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
57 static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
58 static void log_nlmsg(const char *function, int error,
59 const void *message, size_t size, int protocol);
61 /* Netlink sockets. */
70 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
73 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
74 * of iovecs on the stack. */
77 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
78 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
80 * Initialized by nl_sock_create(). */
83 static int nl_sock_cow__(struct nl_sock *);
85 /* Creates a new netlink socket for the given netlink 'protocol'
86 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
87 * new socket if successful, otherwise returns a positive errno value. */
89 nl_sock_create(int protocol, struct nl_sock **sockp)
92 struct sockaddr_nl local, remote;
98 int save_errno = errno;
101 max_iovs = sysconf(_SC_UIO_MAXIOV);
102 if (max_iovs < _XOPEN_IOV_MAX) {
103 if (max_iovs == -1 && errno) {
104 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", strerror(errno));
106 max_iovs = _XOPEN_IOV_MAX;
107 } else if (max_iovs > MAX_IOVS) {
115 sock = malloc(sizeof *sock);
120 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
122 VLOG_ERR("fcntl: %s", strerror(errno));
125 sock->protocol = protocol;
129 rcvbuf = 1024 * 1024;
130 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
131 &rcvbuf, sizeof rcvbuf)) {
132 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed (%s)",
133 rcvbuf, strerror(errno));
136 retval = get_socket_rcvbuf(sock->fd);
141 sock->rcvbuf = retval;
143 /* Connect to kernel (pid 0) as remote address. */
144 memset(&remote, 0, sizeof remote);
145 remote.nl_family = AF_NETLINK;
147 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
148 VLOG_ERR("connect(0): %s", strerror(errno));
152 /* Obtain pid assigned by kernel. */
153 local_size = sizeof local;
154 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
155 VLOG_ERR("getsockname: %s", strerror(errno));
158 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
159 VLOG_ERR("getsockname returned bad Netlink name");
163 sock->pid = local.nl_pid;
182 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
183 * sets '*sockp' to the new socket if successful, otherwise returns a positive
186 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
188 return nl_sock_create(src->protocol, sockp);
191 /* Destroys netlink socket 'sock'. */
193 nl_sock_destroy(struct nl_sock *sock)
205 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
206 * successful, otherwise a positive errno value.
208 * A socket that is subscribed to a multicast group that receives asynchronous
209 * notifications must not be used for Netlink transactions or dumps, because
210 * transactions and dumps can cause notifications to be lost.
212 * Multicast group numbers are always positive.
214 * It is not an error to attempt to join a multicast group to which a socket
215 * already belongs. */
217 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
219 int error = nl_sock_cow__(sock);
223 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
224 &multicast_group, sizeof multicast_group) < 0) {
225 VLOG_WARN("could not join multicast group %u (%s)",
226 multicast_group, strerror(errno));
232 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
233 * successful, otherwise a positive errno value.
235 * Multicast group numbers are always positive.
237 * It is not an error to attempt to leave a multicast group to which a socket
240 * On success, reading from 'sock' will still return any messages that were
241 * received on 'multicast_group' before the group was left. */
243 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
246 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
247 &multicast_group, sizeof multicast_group) < 0) {
248 VLOG_WARN("could not leave multicast group %u (%s)",
249 multicast_group, strerror(errno));
256 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
258 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
261 nlmsg->nlmsg_len = msg->size;
262 nlmsg->nlmsg_seq = nl_sock_allocate_seq(sock, 1);
263 nlmsg->nlmsg_pid = sock->pid;
266 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
267 error = retval < 0 ? errno : 0;
268 } while (error == EINTR);
269 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
271 COVERAGE_INC(netlink_sent);
276 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
277 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, and
278 * nlmsg_pid will be set to 'sock''s pid, before the message is sent.
280 * Returns 0 if successful, otherwise a positive errno value. If
281 * 'wait' is true, then the send will wait until buffer space is ready;
282 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
284 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
286 int error = nl_sock_cow__(sock);
290 return nl_sock_send__(sock, msg, wait);
293 /* This stress option is useful for testing that OVS properly tolerates
294 * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
295 * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
296 * reply to a request. They can also occur if messages arrive on a multicast
297 * channel faster than OVS can process them. */
299 netlink_overflow, "simulate netlink socket receive buffer overflow",
303 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
305 /* We can't accurately predict the size of the data to be received. Most
306 * received data will fit in a 2 kB buffer, so we allocate that much space.
307 * In case the data is actually bigger than that, we make available enough
308 * additional space to allow Netlink messages to be up to 64 kB long (a
309 * reasonable figure since that's the maximum length of a Netlink
311 enum { MAX_SIZE = 65536 };
312 enum { HEAD_SIZE = 2048 };
313 enum { TAIL_SIZE = MAX_SIZE - HEAD_SIZE };
315 struct nlmsghdr *nlmsghdr;
316 uint8_t tail[TAIL_SIZE];
324 buf = ofpbuf_new(HEAD_SIZE);
325 iov[0].iov_base = buf->data;
326 iov[0].iov_len = HEAD_SIZE;
327 iov[1].iov_base = tail;
328 iov[1].iov_len = TAIL_SIZE;
330 memset(&msg, 0, sizeof msg);
335 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
336 } while (retval < 0 && errno == EINTR);
340 if (error == ENOBUFS) {
341 /* Socket receive buffer overflow dropped one or more messages that
342 * the kernel tried to send to us. */
343 COVERAGE_INC(netlink_overflow);
349 if (msg.msg_flags & MSG_TRUNC) {
350 VLOG_ERR_RL(&rl, "truncated message (longer than %d bytes)", MAX_SIZE);
355 ofpbuf_put_uninit(buf, MIN(retval, HEAD_SIZE));
356 if (retval > HEAD_SIZE) {
357 COVERAGE_INC(netlink_recv_jumbo);
358 ofpbuf_put(buf, tail, retval - HEAD_SIZE);
361 nlmsghdr = buf->data;
362 if (retval < sizeof *nlmsghdr
363 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
364 || nlmsghdr->nlmsg_len > retval) {
365 VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)",
366 retval, NLMSG_HDRLEN);
371 if (STRESS(netlink_overflow)) {
377 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
378 COVERAGE_INC(netlink_received);
383 /* Tries to receive a netlink message from the kernel on 'sock'. If
384 * successful, stores the received message into '*bufp' and returns 0. The
385 * caller is responsible for destroying the message with ofpbuf_delete(). On
386 * failure, returns a positive errno value and stores a null pointer into
389 * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
390 * returns EAGAIN if the 'sock' receive buffer is empty. */
392 nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
394 int error = nl_sock_cow__(sock);
398 return nl_sock_recv__(sock, bufp, wait);
402 find_nl_transaction_by_seq(struct nl_transaction **transactions, size_t n,
407 for (i = 0; i < n; i++) {
408 struct nl_transaction *t = transactions[i];
410 if (seq == nl_msg_nlmsghdr(t->request)->nlmsg_seq) {
419 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
424 for (i = 0; i < n; i++) {
425 transactions[i]->error = error;
426 transactions[i]->reply = NULL;
431 nl_sock_transact_multiple__(struct nl_sock *sock,
432 struct nl_transaction **transactions, size_t n,
435 struct iovec iovs[MAX_IOVS];
441 for (i = 0; i < n; i++) {
442 struct ofpbuf *request = transactions[i]->request;
443 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(request);
445 nlmsg->nlmsg_len = request->size;
446 nlmsg->nlmsg_pid = sock->pid;
448 iovs[i].iov_base = request->data;
449 iovs[i].iov_len = request->size;
452 memset(&msg, 0, sizeof msg);
456 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
457 } while (error == EINTR);
459 for (i = 0; i < n; i++) {
460 struct ofpbuf *request = transactions[i]->request;
462 log_nlmsg(__func__, error, request->data, request->size,
466 COVERAGE_ADD(netlink_sent, n);
474 struct ofpbuf *reply;
476 error = nl_sock_recv__(sock, &reply, false);
477 if (error == EAGAIN) {
478 nl_sock_record_errors__(transactions, n, 0);
485 i = find_nl_transaction_by_seq(transactions, n,
486 nl_msg_nlmsghdr(reply)->nlmsg_seq);
488 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32,
489 nl_msg_nlmsghdr(reply)->nlmsg_seq);
490 ofpbuf_delete(reply);
494 nl_sock_record_errors__(transactions, i, 0);
495 if (nl_msg_nlmsgerr(reply, &error)) {
496 transactions[i]->reply = NULL;
497 transactions[i]->error = error;
499 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
500 error, strerror(error));
502 ofpbuf_delete(reply);
504 transactions[i]->reply = reply;
505 transactions[i]->error = 0;
509 transactions += i + 1;
516 /* Sends the 'request' member of the 'n' transactions in 'transactions' to the
517 * kernel, in order, and waits for responses to all of them. Fills in the
518 * 'error' member of each transaction with 0 if it was successful, otherwise
519 * with a positive errno value. 'reply' will be NULL on error or if the
520 * transaction was successful but had no reply beyond an indication of success.
521 * For a successful transaction that did have a more detailed reply, 'reply'
522 * will be set to the reply message.
524 * The caller is responsible for destroying each request and reply, and the
525 * transactions array itself.
527 * Before sending each message, this function will finalize nlmsg_len in each
528 * 'request' to match the ofpbuf's size, and set nlmsg_pid to 'sock''s pid.
530 * Bare Netlink is an unreliable transport protocol. This function layers
531 * reliable delivery and reply semantics on top of bare Netlink. See
532 * nl_sock_transact() for some caveats.
535 nl_sock_transact_multiple(struct nl_sock *sock,
536 struct nl_transaction **transactions, size_t n)
545 error = nl_sock_cow__(sock);
547 nl_sock_record_errors__(transactions, n, error);
551 /* In theory, every request could have a 64 kB reply. But the default and
552 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
553 * be a bit below 128 kB, so that would only allow a single message in a
554 * "batch". So we assume that replies average (at most) 4 kB, which allows
555 * a good deal of batching.
557 * In practice, most of the requests that we batch either have no reply at
558 * all or a brief reply. */
559 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
560 max_batch_count = MIN(max_batch_count, max_iovs);
566 /* Batch up to 'max_batch_count' transactions. But cap it at about a
567 * page of requests total because big skbuffs are expensive to
568 * allocate in the kernel. */
569 #if defined(PAGESIZE)
570 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
572 enum { MAX_BATCH_BYTES = 4096 - 512 };
574 bytes = transactions[0]->request->size;
575 for (count = 1; count < n && count < max_batch_count; count++) {
576 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
579 bytes += transactions[count]->request->size;
582 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
583 transactions += done;
586 if (error == ENOBUFS) {
587 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
589 VLOG_ERR_RL(&rl, "transaction error (%s)", strerror(error));
590 nl_sock_record_errors__(transactions, n, error);
595 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
596 * successful, returns 0. On failure, returns a positive errno value.
598 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
599 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
600 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
601 * reply, if any, is discarded.
603 * Before the message is sent, nlmsg_len in 'request' will be finalized to
604 * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
605 * be initialized, NLM_F_ACK will be set in nlmsg_flags.
607 * The caller is responsible for destroying 'request'.
609 * Bare Netlink is an unreliable transport protocol. This function layers
610 * reliable delivery and reply semantics on top of bare Netlink.
612 * In Netlink, sending a request to the kernel is reliable enough, because the
613 * kernel will tell us if the message cannot be queued (and we will in that
614 * case put it on the transmit queue and wait until it can be delivered).
616 * Receiving the reply is the real problem: if the socket buffer is full when
617 * the kernel tries to send the reply, the reply will be dropped. However, the
618 * kernel sets a flag that a reply has been dropped. The next call to recv
619 * then returns ENOBUFS. We can then re-send the request.
623 * 1. Netlink depends on sequence numbers to match up requests and
624 * replies. The sender of a request supplies a sequence number, and
625 * the reply echos back that sequence number.
627 * This is fine, but (1) some kernel netlink implementations are
628 * broken, in that they fail to echo sequence numbers and (2) this
629 * function will drop packets with non-matching sequence numbers, so
630 * that only a single request can be usefully transacted at a time.
632 * 2. Resending the request causes it to be re-executed, so the request
633 * needs to be idempotent.
636 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
637 struct ofpbuf **replyp)
639 struct nl_transaction *transactionp;
640 struct nl_transaction transaction;
642 transaction.request = (struct ofpbuf *) request;
643 transactionp = &transaction;
644 nl_sock_transact_multiple(sock, &transactionp, 1);
646 *replyp = transaction.reply;
648 ofpbuf_delete(transaction.reply);
650 return transaction.error;
653 /* Drain all the messages currently in 'sock''s receive queue. */
655 nl_sock_drain(struct nl_sock *sock)
657 int error = nl_sock_cow__(sock);
661 return drain_rcvbuf(sock->fd);
664 /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
665 * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
666 * old fd over to the dump. */
668 nl_sock_cow__(struct nl_sock *sock)
670 struct nl_sock *copy;
679 error = nl_sock_clone(sock, ©);
689 sock->pid = copy->pid;
692 sock->dump->sock = copy;
698 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
699 * 'sock', and initializes 'dump' to reflect the state of the operation.
701 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
702 * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
703 * NLM_F_ACK will be set in nlmsg_flags.
705 * This Netlink socket library is designed to ensure that the dump is reliable
706 * and that it will not interfere with other operations on 'sock', including
707 * destroying or sending and receiving messages on 'sock'. One corner case is
710 * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
711 * whose response has not yet been received (e.g. with nl_sock_recv()).
712 * This is unusual: usually nl_sock_transact() is used to send a message
713 * and receive its reply all in one go.
715 * This function provides no status indication. An error status for the entire
716 * dump operation is provided when it is completed by calling nl_dump_done().
718 * The caller is responsible for destroying 'request'.
720 * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
724 nl_dump_start(struct nl_dump *dump,
725 struct nl_sock *sock, const struct ofpbuf *request)
729 /* 'sock' already has an ongoing dump. Clone the socket because
730 * Netlink only allows one dump at a time. */
731 dump->status = nl_sock_clone(sock, &dump->sock);
741 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
742 dump->status = nl_sock_send__(sock, request, true);
743 dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
746 /* Helper function for nl_dump_next(). */
748 nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp)
750 struct nlmsghdr *nlmsghdr;
751 struct ofpbuf *buffer;
754 retval = nl_sock_recv__(dump->sock, bufferp, true);
756 return retval == EINTR ? EAGAIN : retval;
760 nlmsghdr = nl_msg_nlmsghdr(buffer);
761 if (dump->seq != nlmsghdr->nlmsg_seq) {
762 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
763 nlmsghdr->nlmsg_seq, dump->seq);
767 if (nl_msg_nlmsgerr(buffer, &retval)) {
768 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
770 return retval && retval != EAGAIN ? retval : EPROTO;
776 /* Attempts to retrieve another reply from 'dump', which must have been
777 * initialized with nl_dump_start().
779 * If successful, returns true and points 'reply->data' and 'reply->size' to
780 * the message that was retrieved. The caller must not modify 'reply' (because
781 * it points into the middle of a larger buffer).
783 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
784 * to 0. Failure might indicate an actual error or merely the end of replies.
785 * An error status for the entire dump operation is provided when it is
786 * completed by calling nl_dump_done().
789 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
791 struct nlmsghdr *nlmsghdr;
799 if (dump->buffer && !dump->buffer->size) {
800 ofpbuf_delete(dump->buffer);
803 while (!dump->buffer) {
804 int retval = nl_dump_recv(dump, &dump->buffer);
806 ofpbuf_delete(dump->buffer);
808 if (retval != EAGAIN) {
809 dump->status = retval;
815 nlmsghdr = nl_msg_next(dump->buffer, reply);
817 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
818 dump->status = EPROTO;
820 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
828 /* Completes Netlink dump operation 'dump', which must have been initialized
829 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
830 * otherwise a positive errno value describing the problem. */
832 nl_dump_done(struct nl_dump *dump)
834 /* Drain any remaining messages that the client didn't read. Otherwise the
835 * kernel will continue to queue them up and waste buffer space. */
836 while (!dump->status) {
838 if (!nl_dump_next(dump, &reply)) {
839 assert(dump->status);
844 if (dump->sock->dump) {
845 dump->sock->dump = NULL;
847 nl_sock_destroy(dump->sock);
850 ofpbuf_delete(dump->buffer);
851 return dump->status == EOF ? 0 : dump->status;
854 /* Causes poll_block() to wake up when any of the specified 'events' (which is
855 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
857 nl_sock_wait(const struct nl_sock *sock, short int events)
859 poll_fd_wait(sock->fd, events);
862 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
863 * that can't use nl_sock_wait().
865 * It's a little tricky to use the returned fd correctly, because nl_sock does
866 * "copy on write" to allow a single nl_sock to be used for notifications,
867 * transactions, and dumps. If 'sock' is used only for notifications and
868 * transactions (and never for dump) then the usage is safe. */
870 nl_sock_fd(const struct nl_sock *sock)
875 /* Returns the PID associated with this socket. */
877 nl_sock_pid(const struct nl_sock *sock)
885 struct hmap_node hmap_node;
890 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
892 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
893 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
894 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
897 static struct genl_family *
898 find_genl_family_by_id(uint16_t id)
900 struct genl_family *family;
902 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
904 if (family->id == id) {
912 define_genl_family(uint16_t id, const char *name)
914 struct genl_family *family = find_genl_family_by_id(id);
917 if (!strcmp(family->name, name)) {
922 family = xmalloc(sizeof *family);
924 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
926 family->name = xstrdup(name);
930 genl_family_to_name(uint16_t id)
932 if (id == GENL_ID_CTRL) {
935 struct genl_family *family = find_genl_family_by_id(id);
936 return family ? family->name : "unknown";
941 do_lookup_genl_family(const char *name, struct nlattr **attrs,
942 struct ofpbuf **replyp)
944 struct nl_sock *sock;
945 struct ofpbuf request, *reply;
949 error = nl_sock_create(NETLINK_GENERIC, &sock);
954 ofpbuf_init(&request, 0);
955 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
956 CTRL_CMD_GETFAMILY, 1);
957 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
958 error = nl_sock_transact(sock, &request, &reply);
959 ofpbuf_uninit(&request);
961 nl_sock_destroy(sock);
965 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
966 family_policy, attrs, ARRAY_SIZE(family_policy))
967 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
968 nl_sock_destroy(sock);
969 ofpbuf_delete(reply);
973 nl_sock_destroy(sock);
978 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
979 * When successful, writes its result to 'multicast_group' and returns 0.
980 * Otherwise, clears 'multicast_group' and returns a positive error code.
982 * Some kernels do not support looking up a multicast group with this function.
983 * In this case, 'multicast_group' will be populated with 'fallback'. */
985 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
986 unsigned int *multicast_group, unsigned int fallback)
988 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
989 const struct nlattr *mc;
990 struct ofpbuf *reply;
994 *multicast_group = 0;
995 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1000 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1001 *multicast_group = fallback;
1002 VLOG_WARN("%s-%s: has no multicast group, using fallback %d",
1003 family_name, group_name, *multicast_group);
1008 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1009 static const struct nl_policy mc_policy[] = {
1010 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1011 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1014 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1015 const char *mc_name;
1017 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1022 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1023 if (!strcmp(group_name, mc_name)) {
1025 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1033 ofpbuf_delete(reply);
1037 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1038 * number and stores it in '*number'. If successful, returns 0 and the caller
1039 * may use '*number' as the family number. On failure, returns a positive
1040 * errno value and '*number' caches the errno value. */
1042 nl_lookup_genl_family(const char *name, int *number)
1045 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1046 struct ofpbuf *reply;
1049 error = do_lookup_genl_family(name, attrs, &reply);
1051 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1052 define_genl_family(*number, name);
1056 ofpbuf_delete(reply);
1058 assert(*number != 0);
1060 return *number > 0 ? 0 : -*number;
1064 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1066 uint32_t seq = sock->next_seq;
1068 sock->next_seq += n;
1070 /* Make it impossible for the next request for sequence numbers to wrap
1071 * around to 0. Start over with 1 to avoid ever using a sequence number of
1072 * 0, because the kernel uses sequence number 0 for notifications. */
1073 if (sock->next_seq >= UINT32_MAX / 2) {
1081 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1087 static const struct nlmsg_flag flags[] = {
1088 { NLM_F_REQUEST, "REQUEST" },
1089 { NLM_F_MULTI, "MULTI" },
1090 { NLM_F_ACK, "ACK" },
1091 { NLM_F_ECHO, "ECHO" },
1092 { NLM_F_DUMP, "DUMP" },
1093 { NLM_F_ROOT, "ROOT" },
1094 { NLM_F_MATCH, "MATCH" },
1095 { NLM_F_ATOMIC, "ATOMIC" },
1097 const struct nlmsg_flag *flag;
1098 uint16_t flags_left;
1100 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1101 h->nlmsg_len, h->nlmsg_type);
1102 if (h->nlmsg_type == NLMSG_NOOP) {
1103 ds_put_cstr(ds, "(no-op)");
1104 } else if (h->nlmsg_type == NLMSG_ERROR) {
1105 ds_put_cstr(ds, "(error)");
1106 } else if (h->nlmsg_type == NLMSG_DONE) {
1107 ds_put_cstr(ds, "(done)");
1108 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1109 ds_put_cstr(ds, "(overrun)");
1110 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1111 ds_put_cstr(ds, "(reserved)");
1112 } else if (protocol == NETLINK_GENERIC) {
1113 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1115 ds_put_cstr(ds, "(family-defined)");
1117 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1118 flags_left = h->nlmsg_flags;
1119 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1120 if ((flags_left & flag->bits) == flag->bits) {
1121 ds_put_format(ds, "[%s]", flag->name);
1122 flags_left &= ~flag->bits;
1126 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1128 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1129 h->nlmsg_seq, h->nlmsg_pid);
1133 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1135 struct ds ds = DS_EMPTY_INITIALIZER;
1136 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1138 nlmsghdr_to_string(h, protocol, &ds);
1139 if (h->nlmsg_type == NLMSG_ERROR) {
1140 const struct nlmsgerr *e;
1141 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1142 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1144 ds_put_format(&ds, " error(%d", e->error);
1146 ds_put_format(&ds, "(%s)", strerror(-e->error));
1148 ds_put_cstr(&ds, ", in-reply-to(");
1149 nlmsghdr_to_string(&e->msg, protocol, &ds);
1150 ds_put_cstr(&ds, "))");
1152 ds_put_cstr(&ds, " error(truncated)");
1154 } else if (h->nlmsg_type == NLMSG_DONE) {
1155 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1157 ds_put_format(&ds, " done(%d", *error);
1159 ds_put_format(&ds, "(%s)", strerror(-*error));
1161 ds_put_cstr(&ds, ")");
1163 ds_put_cstr(&ds, " done(truncated)");
1165 } else if (protocol == NETLINK_GENERIC) {
1166 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1168 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1169 genl->cmd, genl->version);
1173 ds_put_cstr(&ds, "nl(truncated)");
1179 log_nlmsg(const char *function, int error,
1180 const void *message, size_t size, int protocol)
1182 struct ofpbuf buffer;
1185 if (!VLOG_IS_DBG_ENABLED()) {
1189 ofpbuf_use_const(&buffer, message, size);
1190 nlmsg = nlmsg_to_string(&buffer, protocol);
1191 VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);