2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
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"
22 #include <sys/types.h>
26 #include "dynamic-string.h"
30 #include "netlink-protocol.h"
32 #include "poll-loop.h"
33 #include "socket-util.h"
37 VLOG_DEFINE_THIS_MODULE(netlink_socket);
39 COVERAGE_DEFINE(netlink_overflow);
40 COVERAGE_DEFINE(netlink_received);
41 COVERAGE_DEFINE(netlink_recv_jumbo);
42 COVERAGE_DEFINE(netlink_send);
43 COVERAGE_DEFINE(netlink_sent);
45 /* Linux header file confusion causes this to be undefined. */
47 #define SOL_NETLINK 270
50 /* A single (bad) Netlink message can in theory dump out many, many log
51 * messages, so the burst size is set quite high here to avoid missing useful
52 * information. Also, at high logging levels we log *all* Netlink messages. */
53 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
55 static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
56 static void log_nlmsg(const char *function, int error,
57 const void *message, size_t size, int protocol);
59 /* Netlink sockets. */
67 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
70 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
71 * of iovecs on the stack. */
74 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
75 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
77 * Initialized by nl_sock_create(). */
80 static int nl_sock_cow__(struct nl_sock *);
82 /* Creates a new netlink socket for the given netlink 'protocol'
83 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
84 * new socket if successful, otherwise returns a positive errno value. */
86 nl_sock_create(int protocol, struct nl_sock **sockp)
89 struct sockaddr_nl local, remote;
95 int save_errno = errno;
98 max_iovs = sysconf(_SC_UIO_MAXIOV);
99 if (max_iovs < _XOPEN_IOV_MAX) {
100 if (max_iovs == -1 && errno) {
101 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno));
103 max_iovs = _XOPEN_IOV_MAX;
104 } else if (max_iovs > MAX_IOVS) {
112 sock = xmalloc(sizeof *sock);
114 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
116 VLOG_ERR("fcntl: %s", ovs_strerror(errno));
119 sock->protocol = protocol;
123 rcvbuf = 1024 * 1024;
124 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
125 &rcvbuf, sizeof rcvbuf)) {
126 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
127 * Warn only if the failure is therefore unexpected. */
128 if (errno != EPERM) {
129 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
130 "(%s)", rcvbuf, ovs_strerror(errno));
134 retval = get_socket_rcvbuf(sock->fd);
139 sock->rcvbuf = retval;
141 /* Connect to kernel (pid 0) as remote address. */
142 memset(&remote, 0, sizeof remote);
143 remote.nl_family = AF_NETLINK;
145 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
146 VLOG_ERR("connect(0): %s", ovs_strerror(errno));
150 /* Obtain pid assigned by kernel. */
151 local_size = sizeof local;
152 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
153 VLOG_ERR("getsockname: %s", ovs_strerror(errno));
156 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
157 VLOG_ERR("getsockname returned bad Netlink name");
161 sock->pid = local.nl_pid;
180 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
181 * sets '*sockp' to the new socket if successful, otherwise returns a positive
184 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
186 return nl_sock_create(src->protocol, sockp);
189 /* Destroys netlink socket 'sock'. */
191 nl_sock_destroy(struct nl_sock *sock)
203 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
204 * successful, otherwise a positive errno value.
206 * A socket that is subscribed to a multicast group that receives asynchronous
207 * notifications must not be used for Netlink transactions or dumps, because
208 * transactions and dumps can cause notifications to be lost.
210 * Multicast group numbers are always positive.
212 * It is not an error to attempt to join a multicast group to which a socket
213 * already belongs. */
215 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
217 int error = nl_sock_cow__(sock);
221 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
222 &multicast_group, sizeof multicast_group) < 0) {
223 VLOG_WARN("could not join multicast group %u (%s)",
224 multicast_group, ovs_strerror(errno));
230 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
231 * successful, otherwise a positive errno value.
233 * Multicast group numbers are always positive.
235 * It is not an error to attempt to leave a multicast group to which a socket
238 * On success, reading from 'sock' will still return any messages that were
239 * received on 'multicast_group' before the group was left. */
241 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
243 ovs_assert(!sock->dump);
244 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
245 &multicast_group, sizeof multicast_group) < 0) {
246 VLOG_WARN("could not leave multicast group %u (%s)",
247 multicast_group, ovs_strerror(errno));
254 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
255 uint32_t nlmsg_seq, bool wait)
257 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
260 nlmsg->nlmsg_len = msg->size;
261 nlmsg->nlmsg_seq = nlmsg_seq;
262 nlmsg->nlmsg_pid = sock->pid;
265 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
266 error = retval < 0 ? errno : 0;
267 } while (error == EINTR);
268 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
270 COVERAGE_INC(netlink_sent);
275 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
276 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
277 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
278 * sequence number, 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 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
289 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
290 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
291 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
292 * 'nlmsg_seq', before the message is sent.
294 * Returns 0 if successful, otherwise a positive errno value. If
295 * 'wait' is true, then the send will wait until buffer space is ready;
296 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
298 * This function is suitable for sending a reply to a request that was received
299 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
301 nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
302 uint32_t nlmsg_seq, bool wait)
304 int error = nl_sock_cow__(sock);
308 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
312 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
314 /* We can't accurately predict the size of the data to be received. The
315 * caller is supposed to have allocated enough space in 'buf' to handle the
316 * "typical" case. To handle exceptions, we make available enough space in
317 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
318 * figure since that's the maximum length of a Netlink attribute). */
319 struct nlmsghdr *nlmsghdr;
325 ovs_assert(buf->allocated >= sizeof *nlmsghdr);
328 iov[0].iov_base = buf->base;
329 iov[0].iov_len = buf->allocated;
330 iov[1].iov_base = tail;
331 iov[1].iov_len = sizeof tail;
333 memset(&msg, 0, sizeof msg);
338 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
339 } while (retval < 0 && errno == EINTR);
343 if (error == ENOBUFS) {
344 /* Socket receive buffer overflow dropped one or more messages that
345 * the kernel tried to send to us. */
346 COVERAGE_INC(netlink_overflow);
351 if (msg.msg_flags & MSG_TRUNC) {
352 VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)",
357 nlmsghdr = buf->data;
358 if (retval < sizeof *nlmsghdr
359 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
360 || nlmsghdr->nlmsg_len > retval) {
361 VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)",
362 retval, sizeof *nlmsghdr);
366 buf->size = MIN(retval, buf->allocated);
367 if (retval > buf->allocated) {
368 COVERAGE_INC(netlink_recv_jumbo);
369 ofpbuf_put(buf, tail, retval - buf->allocated);
372 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
373 COVERAGE_INC(netlink_received);
378 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
379 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
380 * EAGAIN if the 'sock' receive buffer is empty.
382 * The caller must have initialized 'buf' with an allocation of at least
383 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
384 * space for a "typical" message.
386 * On success, returns 0 and replaces 'buf''s previous content by the received
387 * message. This function expands 'buf''s allocated memory, as necessary, to
388 * hold the actual size of the received message.
390 * On failure, returns a positive errno value and clears 'buf' to zero length.
391 * 'buf' retains its previous memory allocation.
393 * Regardless of success or failure, this function resets 'buf''s headroom to
396 nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
398 int error = nl_sock_cow__(sock);
402 return nl_sock_recv__(sock, buf, wait);
406 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
411 for (i = 0; i < n; i++) {
412 struct nl_transaction *txn = transactions[i];
416 ofpbuf_clear(txn->reply);
422 nl_sock_transact_multiple__(struct nl_sock *sock,
423 struct nl_transaction **transactions, size_t n,
426 uint64_t tmp_reply_stub[1024 / 8];
427 struct nl_transaction tmp_txn;
428 struct ofpbuf tmp_reply;
431 struct iovec iovs[MAX_IOVS];
436 base_seq = nl_sock_allocate_seq(sock, n);
438 for (i = 0; i < n; i++) {
439 struct nl_transaction *txn = transactions[i];
440 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
442 nlmsg->nlmsg_len = txn->request->size;
443 nlmsg->nlmsg_seq = base_seq + i;
444 nlmsg->nlmsg_pid = sock->pid;
446 iovs[i].iov_base = txn->request->data;
447 iovs[i].iov_len = txn->request->size;
450 memset(&msg, 0, sizeof msg);
454 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
455 } while (error == EINTR);
457 for (i = 0; i < n; i++) {
458 struct nl_transaction *txn = transactions[i];
460 log_nlmsg(__func__, error, txn->request->data, txn->request->size,
464 COVERAGE_ADD(netlink_sent, n);
471 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
472 tmp_txn.request = NULL;
473 tmp_txn.reply = &tmp_reply;
476 struct nl_transaction *buf_txn, *txn;
479 /* Find a transaction whose buffer we can use for receiving a reply.
480 * If no such transaction is left, use tmp_txn. */
482 for (i = 0; i < n; i++) {
483 if (transactions[i]->reply) {
484 buf_txn = transactions[i];
489 /* Receive a reply. */
490 error = nl_sock_recv__(sock, buf_txn->reply, false);
492 if (error == EAGAIN) {
493 nl_sock_record_errors__(transactions, n, 0);
500 /* Match the reply up with a transaction. */
501 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
502 if (seq < base_seq || seq >= base_seq + n) {
503 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
507 txn = transactions[i];
509 /* Fill in the results for 'txn'. */
510 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
512 ofpbuf_clear(txn->reply);
515 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
516 error, ovs_strerror(txn->error));
520 if (txn->reply && txn != buf_txn) {
522 struct ofpbuf *reply = buf_txn->reply;
523 buf_txn->reply = txn->reply;
528 /* Fill in the results for transactions before 'txn'. (We have to do
529 * this after the results for 'txn' itself because of the buffer swap
531 nl_sock_record_errors__(transactions, i, 0);
535 transactions += i + 1;
539 ofpbuf_uninit(&tmp_reply);
544 /* Sends the 'request' member of the 'n' transactions in 'transactions' on
545 * 'sock', in order, and receives responses to all of them. Fills in the
546 * 'error' member of each transaction with 0 if it was successful, otherwise
547 * with a positive errno value. If 'reply' is nonnull, then it will be filled
548 * with the reply if the message receives a detailed reply. In other cases,
549 * i.e. where the request failed or had no reply beyond an indication of
550 * success, 'reply' will be cleared if it is nonnull.
552 * The caller is responsible for destroying each request and reply, and the
553 * transactions array itself.
555 * Before sending each message, this function will finalize nlmsg_len in each
556 * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
557 * initialize nlmsg_seq.
559 * Bare Netlink is an unreliable transport protocol. This function layers
560 * reliable delivery and reply semantics on top of bare Netlink. See
561 * nl_sock_transact() for some caveats.
564 nl_sock_transact_multiple(struct nl_sock *sock,
565 struct nl_transaction **transactions, size_t n)
574 error = nl_sock_cow__(sock);
576 nl_sock_record_errors__(transactions, n, error);
580 /* In theory, every request could have a 64 kB reply. But the default and
581 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
582 * be a bit below 128 kB, so that would only allow a single message in a
583 * "batch". So we assume that replies average (at most) 4 kB, which allows
584 * a good deal of batching.
586 * In practice, most of the requests that we batch either have no reply at
587 * all or a brief reply. */
588 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
589 max_batch_count = MIN(max_batch_count, max_iovs);
595 /* Batch up to 'max_batch_count' transactions. But cap it at about a
596 * page of requests total because big skbuffs are expensive to
597 * allocate in the kernel. */
598 #if defined(PAGESIZE)
599 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
601 enum { MAX_BATCH_BYTES = 4096 - 512 };
603 bytes = transactions[0]->request->size;
604 for (count = 1; count < n && count < max_batch_count; count++) {
605 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
608 bytes += transactions[count]->request->size;
611 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
612 transactions += done;
615 if (error == ENOBUFS) {
616 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
618 VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error));
619 nl_sock_record_errors__(transactions, n, error);
624 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
625 * successful, returns 0. On failure, returns a positive errno value.
627 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
628 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
629 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
630 * reply, if any, is discarded.
632 * Before the message is sent, nlmsg_len in 'request' will be finalized to
633 * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
634 * be initialized, NLM_F_ACK will be set in nlmsg_flags.
636 * The caller is responsible for destroying 'request'.
638 * Bare Netlink is an unreliable transport protocol. This function layers
639 * reliable delivery and reply semantics on top of bare Netlink.
641 * In Netlink, sending a request to the kernel is reliable enough, because the
642 * kernel will tell us if the message cannot be queued (and we will in that
643 * case put it on the transmit queue and wait until it can be delivered).
645 * Receiving the reply is the real problem: if the socket buffer is full when
646 * the kernel tries to send the reply, the reply will be dropped. However, the
647 * kernel sets a flag that a reply has been dropped. The next call to recv
648 * then returns ENOBUFS. We can then re-send the request.
652 * 1. Netlink depends on sequence numbers to match up requests and
653 * replies. The sender of a request supplies a sequence number, and
654 * the reply echos back that sequence number.
656 * This is fine, but (1) some kernel netlink implementations are
657 * broken, in that they fail to echo sequence numbers and (2) this
658 * function will drop packets with non-matching sequence numbers, so
659 * that only a single request can be usefully transacted at a time.
661 * 2. Resending the request causes it to be re-executed, so the request
662 * needs to be idempotent.
665 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
666 struct ofpbuf **replyp)
668 struct nl_transaction *transactionp;
669 struct nl_transaction transaction;
671 transaction.request = CONST_CAST(struct ofpbuf *, request);
672 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
673 transactionp = &transaction;
675 nl_sock_transact_multiple(sock, &transactionp, 1);
678 if (transaction.error) {
679 ofpbuf_delete(transaction.reply);
682 *replyp = transaction.reply;
686 return transaction.error;
689 /* Drain all the messages currently in 'sock''s receive queue. */
691 nl_sock_drain(struct nl_sock *sock)
693 int error = nl_sock_cow__(sock);
697 return drain_rcvbuf(sock->fd);
700 /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
701 * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
702 * old fd over to the dump. */
704 nl_sock_cow__(struct nl_sock *sock)
706 struct nl_sock *copy;
715 error = nl_sock_clone(sock, ©);
725 sock->pid = copy->pid;
728 sock->dump->sock = copy;
734 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
735 * 'sock', and initializes 'dump' to reflect the state of the operation.
737 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
738 * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
739 * NLM_F_ACK will be set in nlmsg_flags.
741 * This Netlink socket library is designed to ensure that the dump is reliable
742 * and that it will not interfere with other operations on 'sock', including
743 * destroying or sending and receiving messages on 'sock'. One corner case is
746 * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
747 * whose response has not yet been received (e.g. with nl_sock_recv()).
748 * This is unusual: usually nl_sock_transact() is used to send a message
749 * and receive its reply all in one go.
751 * This function provides no status indication. An error status for the entire
752 * dump operation is provided when it is completed by calling nl_dump_done().
754 * The caller is responsible for destroying 'request'.
756 * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
760 nl_dump_start(struct nl_dump *dump,
761 struct nl_sock *sock, const struct ofpbuf *request)
763 ofpbuf_init(&dump->buffer, 4096);
765 /* 'sock' already has an ongoing dump. Clone the socket because
766 * Netlink only allows one dump at a time. */
767 dump->status = nl_sock_clone(sock, &dump->sock);
777 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
778 dump->status = nl_sock_send__(sock, request, nl_sock_allocate_seq(sock, 1),
780 dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
783 /* Helper function for nl_dump_next(). */
785 nl_dump_recv(struct nl_dump *dump)
787 struct nlmsghdr *nlmsghdr;
790 retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
792 return retval == EINTR ? EAGAIN : retval;
795 nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
796 if (dump->seq != nlmsghdr->nlmsg_seq) {
797 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
798 nlmsghdr->nlmsg_seq, dump->seq);
802 if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
803 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
804 ovs_strerror(retval));
805 return retval && retval != EAGAIN ? retval : EPROTO;
811 /* Attempts to retrieve another reply from 'dump', which must have been
812 * initialized with nl_dump_start().
814 * If successful, returns true and points 'reply->data' and 'reply->size' to
815 * the message that was retrieved. The caller must not modify 'reply' (because
816 * it points into the middle of a larger buffer).
818 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
819 * to 0. Failure might indicate an actual error or merely the end of replies.
820 * An error status for the entire dump operation is provided when it is
821 * completed by calling nl_dump_done().
824 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
826 struct nlmsghdr *nlmsghdr;
834 while (!dump->buffer.size) {
835 int retval = nl_dump_recv(dump);
837 ofpbuf_clear(&dump->buffer);
838 if (retval != EAGAIN) {
839 dump->status = retval;
845 nlmsghdr = nl_msg_next(&dump->buffer, reply);
847 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
848 dump->status = EPROTO;
850 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
858 /* Completes Netlink dump operation 'dump', which must have been initialized
859 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
860 * otherwise a positive errno value describing the problem. */
862 nl_dump_done(struct nl_dump *dump)
864 /* Drain any remaining messages that the client didn't read. Otherwise the
865 * kernel will continue to queue them up and waste buffer space. */
866 while (!dump->status) {
868 if (!nl_dump_next(dump, &reply)) {
869 ovs_assert(dump->status);
874 if (dump->sock->dump) {
875 dump->sock->dump = NULL;
877 nl_sock_destroy(dump->sock);
880 ofpbuf_uninit(&dump->buffer);
881 return dump->status == EOF ? 0 : dump->status;
884 /* Causes poll_block() to wake up when any of the specified 'events' (which is
885 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
887 nl_sock_wait(const struct nl_sock *sock, short int events)
889 poll_fd_wait(sock->fd, events);
892 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
893 * that can't use nl_sock_wait().
895 * It's a little tricky to use the returned fd correctly, because nl_sock does
896 * "copy on write" to allow a single nl_sock to be used for notifications,
897 * transactions, and dumps. If 'sock' is used only for notifications and
898 * transactions (and never for dump) then the usage is safe. */
900 nl_sock_fd(const struct nl_sock *sock)
905 /* Returns the PID associated with this socket. */
907 nl_sock_pid(const struct nl_sock *sock)
915 struct hmap_node hmap_node;
920 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
922 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
923 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
924 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
927 static struct genl_family *
928 find_genl_family_by_id(uint16_t id)
930 struct genl_family *family;
932 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
934 if (family->id == id) {
942 define_genl_family(uint16_t id, const char *name)
944 struct genl_family *family = find_genl_family_by_id(id);
947 if (!strcmp(family->name, name)) {
952 family = xmalloc(sizeof *family);
954 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
956 family->name = xstrdup(name);
960 genl_family_to_name(uint16_t id)
962 if (id == GENL_ID_CTRL) {
965 struct genl_family *family = find_genl_family_by_id(id);
966 return family ? family->name : "unknown";
971 do_lookup_genl_family(const char *name, struct nlattr **attrs,
972 struct ofpbuf **replyp)
974 struct nl_sock *sock;
975 struct ofpbuf request, *reply;
979 error = nl_sock_create(NETLINK_GENERIC, &sock);
984 ofpbuf_init(&request, 0);
985 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
986 CTRL_CMD_GETFAMILY, 1);
987 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
988 error = nl_sock_transact(sock, &request, &reply);
989 ofpbuf_uninit(&request);
991 nl_sock_destroy(sock);
995 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
996 family_policy, attrs, ARRAY_SIZE(family_policy))
997 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
998 nl_sock_destroy(sock);
999 ofpbuf_delete(reply);
1003 nl_sock_destroy(sock);
1008 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
1009 * When successful, writes its result to 'multicast_group' and returns 0.
1010 * Otherwise, clears 'multicast_group' and returns a positive error code.
1012 * Some kernels do not support looking up a multicast group with this function.
1013 * In this case, 'multicast_group' will be populated with 'fallback'. */
1015 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
1016 unsigned int *multicast_group, unsigned int fallback)
1018 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
1019 const struct nlattr *mc;
1020 struct ofpbuf *reply;
1024 *multicast_group = 0;
1025 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1030 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1031 *multicast_group = fallback;
1032 VLOG_WARN("%s-%s: has no multicast group, using fallback %d",
1033 family_name, group_name, *multicast_group);
1038 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1039 static const struct nl_policy mc_policy[] = {
1040 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1041 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1044 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1045 const char *mc_name;
1047 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1052 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1053 if (!strcmp(group_name, mc_name)) {
1055 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1063 ofpbuf_delete(reply);
1067 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1068 * number and stores it in '*number'. If successful, returns 0 and the caller
1069 * may use '*number' as the family number. On failure, returns a positive
1070 * errno value and '*number' caches the errno value. */
1072 nl_lookup_genl_family(const char *name, int *number)
1075 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1076 struct ofpbuf *reply;
1079 error = do_lookup_genl_family(name, attrs, &reply);
1081 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1082 define_genl_family(*number, name);
1086 ofpbuf_delete(reply);
1088 ovs_assert(*number != 0);
1090 return *number > 0 ? 0 : -*number;
1094 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1096 uint32_t seq = sock->next_seq;
1098 sock->next_seq += n;
1100 /* Make it impossible for the next request for sequence numbers to wrap
1101 * around to 0. Start over with 1 to avoid ever using a sequence number of
1102 * 0, because the kernel uses sequence number 0 for notifications. */
1103 if (sock->next_seq >= UINT32_MAX / 2) {
1111 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1117 static const struct nlmsg_flag flags[] = {
1118 { NLM_F_REQUEST, "REQUEST" },
1119 { NLM_F_MULTI, "MULTI" },
1120 { NLM_F_ACK, "ACK" },
1121 { NLM_F_ECHO, "ECHO" },
1122 { NLM_F_DUMP, "DUMP" },
1123 { NLM_F_ROOT, "ROOT" },
1124 { NLM_F_MATCH, "MATCH" },
1125 { NLM_F_ATOMIC, "ATOMIC" },
1127 const struct nlmsg_flag *flag;
1128 uint16_t flags_left;
1130 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1131 h->nlmsg_len, h->nlmsg_type);
1132 if (h->nlmsg_type == NLMSG_NOOP) {
1133 ds_put_cstr(ds, "(no-op)");
1134 } else if (h->nlmsg_type == NLMSG_ERROR) {
1135 ds_put_cstr(ds, "(error)");
1136 } else if (h->nlmsg_type == NLMSG_DONE) {
1137 ds_put_cstr(ds, "(done)");
1138 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1139 ds_put_cstr(ds, "(overrun)");
1140 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1141 ds_put_cstr(ds, "(reserved)");
1142 } else if (protocol == NETLINK_GENERIC) {
1143 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1145 ds_put_cstr(ds, "(family-defined)");
1147 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1148 flags_left = h->nlmsg_flags;
1149 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1150 if ((flags_left & flag->bits) == flag->bits) {
1151 ds_put_format(ds, "[%s]", flag->name);
1152 flags_left &= ~flag->bits;
1156 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1158 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1159 h->nlmsg_seq, h->nlmsg_pid);
1163 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1165 struct ds ds = DS_EMPTY_INITIALIZER;
1166 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1168 nlmsghdr_to_string(h, protocol, &ds);
1169 if (h->nlmsg_type == NLMSG_ERROR) {
1170 const struct nlmsgerr *e;
1171 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1172 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1174 ds_put_format(&ds, " error(%d", e->error);
1176 ds_put_format(&ds, "(%s)", ovs_strerror(-e->error));
1178 ds_put_cstr(&ds, ", in-reply-to(");
1179 nlmsghdr_to_string(&e->msg, protocol, &ds);
1180 ds_put_cstr(&ds, "))");
1182 ds_put_cstr(&ds, " error(truncated)");
1184 } else if (h->nlmsg_type == NLMSG_DONE) {
1185 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1187 ds_put_format(&ds, " done(%d", *error);
1189 ds_put_format(&ds, "(%s)", ovs_strerror(-*error));
1191 ds_put_cstr(&ds, ")");
1193 ds_put_cstr(&ds, " done(truncated)");
1195 } else if (protocol == NETLINK_GENERIC) {
1196 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1198 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1199 genl->cmd, genl->version);
1203 ds_put_cstr(&ds, "nl(truncated)");
1209 log_nlmsg(const char *function, int error,
1210 const void *message, size_t size, int protocol)
1212 struct ofpbuf buffer;
1215 if (!VLOG_IS_DBG_ENABLED()) {
1219 ofpbuf_use_const(&buffer, message, size);
1220 nlmsg = nlmsg_to_string(&buffer, protocol);
1221 VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg);