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"
38 VLOG_DEFINE_THIS_MODULE(netlink_socket);
40 COVERAGE_DEFINE(netlink_overflow);
41 COVERAGE_DEFINE(netlink_received);
42 COVERAGE_DEFINE(netlink_recv_jumbo);
43 COVERAGE_DEFINE(netlink_send);
44 COVERAGE_DEFINE(netlink_sent);
46 /* Linux header file confusion causes this to be undefined. */
48 #define SOL_NETLINK 270
51 /* A single (bad) Netlink message can in theory dump out many, many log
52 * messages, so the burst size is set quite high here to avoid missing useful
53 * information. Also, at high logging levels we log *all* Netlink messages. */
54 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
56 static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
57 static void log_nlmsg(const char *function, int error,
58 const void *message, size_t size, int protocol);
60 /* Netlink sockets. */
69 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
72 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
73 * of iovecs on the stack. */
76 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
77 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
79 * Initialized by nl_sock_create(). */
82 static int nl_sock_cow__(struct nl_sock *);
84 /* Creates a new netlink socket for the given netlink 'protocol'
85 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
86 * new socket if successful, otherwise returns a positive errno value. */
88 nl_sock_create(int protocol, struct nl_sock **sockp)
91 struct sockaddr_nl local, remote;
97 int save_errno = errno;
100 max_iovs = sysconf(_SC_UIO_MAXIOV);
101 if (max_iovs < _XOPEN_IOV_MAX) {
102 if (max_iovs == -1 && errno) {
103 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", strerror(errno));
105 max_iovs = _XOPEN_IOV_MAX;
106 } else if (max_iovs > MAX_IOVS) {
114 sock = malloc(sizeof *sock);
119 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
121 VLOG_ERR("fcntl: %s", strerror(errno));
124 sock->protocol = protocol;
128 rcvbuf = 1024 * 1024;
129 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
130 &rcvbuf, sizeof rcvbuf)) {
131 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
132 * Warn only if the failure is therefore unexpected. */
133 if (errno != EPERM || !getuid()) {
134 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
135 "(%s)", rcvbuf, strerror(errno));
139 retval = get_socket_rcvbuf(sock->fd);
144 sock->rcvbuf = retval;
146 /* Connect to kernel (pid 0) as remote address. */
147 memset(&remote, 0, sizeof remote);
148 remote.nl_family = AF_NETLINK;
150 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
151 VLOG_ERR("connect(0): %s", strerror(errno));
155 /* Obtain pid assigned by kernel. */
156 local_size = sizeof local;
157 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
158 VLOG_ERR("getsockname: %s", strerror(errno));
161 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
162 VLOG_ERR("getsockname returned bad Netlink name");
166 sock->pid = local.nl_pid;
185 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
186 * sets '*sockp' to the new socket if successful, otherwise returns a positive
189 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
191 return nl_sock_create(src->protocol, sockp);
194 /* Destroys netlink socket 'sock'. */
196 nl_sock_destroy(struct nl_sock *sock)
208 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
209 * successful, otherwise a positive errno value.
211 * A socket that is subscribed to a multicast group that receives asynchronous
212 * notifications must not be used for Netlink transactions or dumps, because
213 * transactions and dumps can cause notifications to be lost.
215 * Multicast group numbers are always positive.
217 * It is not an error to attempt to join a multicast group to which a socket
218 * already belongs. */
220 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
222 int error = nl_sock_cow__(sock);
226 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
227 &multicast_group, sizeof multicast_group) < 0) {
228 VLOG_WARN("could not join multicast group %u (%s)",
229 multicast_group, strerror(errno));
235 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
236 * successful, otherwise a positive errno value.
238 * Multicast group numbers are always positive.
240 * It is not an error to attempt to leave a multicast group to which a socket
243 * On success, reading from 'sock' will still return any messages that were
244 * received on 'multicast_group' before the group was left. */
246 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
248 ovs_assert(!sock->dump);
249 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
250 &multicast_group, sizeof multicast_group) < 0) {
251 VLOG_WARN("could not leave multicast group %u (%s)",
252 multicast_group, strerror(errno));
259 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
260 uint32_t nlmsg_seq, bool wait)
262 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
265 nlmsg->nlmsg_len = msg->size;
266 nlmsg->nlmsg_seq = nlmsg_seq;
267 nlmsg->nlmsg_pid = sock->pid;
270 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
271 error = retval < 0 ? errno : 0;
272 } while (error == EINTR);
273 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
275 COVERAGE_INC(netlink_sent);
280 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
281 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
282 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
283 * sequence number, before the message is sent.
285 * Returns 0 if successful, otherwise a positive errno value. If
286 * 'wait' is true, then the send will wait until buffer space is ready;
287 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
289 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
291 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
294 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
295 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
296 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
297 * 'nlmsg_seq', before the message is sent.
299 * Returns 0 if successful, otherwise a positive errno value. If
300 * 'wait' is true, then the send will wait until buffer space is ready;
301 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
303 * This function is suitable for sending a reply to a request that was received
304 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
306 nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
307 uint32_t nlmsg_seq, bool wait)
309 int error = nl_sock_cow__(sock);
313 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
316 /* This stress option is useful for testing that OVS properly tolerates
317 * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
318 * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
319 * reply to a request. They can also occur if messages arrive on a multicast
320 * channel faster than OVS can process them. */
322 netlink_overflow, "simulate netlink socket receive buffer overflow",
326 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
328 /* We can't accurately predict the size of the data to be received. The
329 * caller is supposed to have allocated enough space in 'buf' to handle the
330 * "typical" case. To handle exceptions, we make available enough space in
331 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
332 * figure since that's the maximum length of a Netlink attribute). */
333 struct nlmsghdr *nlmsghdr;
339 ovs_assert(buf->allocated >= sizeof *nlmsghdr);
342 iov[0].iov_base = buf->base;
343 iov[0].iov_len = buf->allocated;
344 iov[1].iov_base = tail;
345 iov[1].iov_len = sizeof tail;
347 memset(&msg, 0, sizeof msg);
352 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
353 } while (retval < 0 && errno == EINTR);
357 if (error == ENOBUFS) {
358 /* Socket receive buffer overflow dropped one or more messages that
359 * the kernel tried to send to us. */
360 COVERAGE_INC(netlink_overflow);
365 if (msg.msg_flags & MSG_TRUNC) {
366 VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)",
371 nlmsghdr = buf->data;
372 if (retval < sizeof *nlmsghdr
373 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
374 || nlmsghdr->nlmsg_len > retval) {
375 VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)",
376 retval, sizeof *nlmsghdr);
380 if (STRESS(netlink_overflow)) {
384 buf->size = MIN(retval, buf->allocated);
385 if (retval > buf->allocated) {
386 COVERAGE_INC(netlink_recv_jumbo);
387 ofpbuf_put(buf, tail, retval - buf->allocated);
390 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
391 COVERAGE_INC(netlink_received);
396 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
397 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
398 * EAGAIN if the 'sock' receive buffer is empty.
400 * The caller must have initialized 'buf' with an allocation of at least
401 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
402 * space for a "typical" message.
404 * On success, returns 0 and replaces 'buf''s previous content by the received
405 * message. This function expands 'buf''s allocated memory, as necessary, to
406 * hold the actual size of the received message.
408 * On failure, returns a positive errno value and clears 'buf' to zero length.
409 * 'buf' retains its previous memory allocation.
411 * Regardless of success or failure, this function resets 'buf''s headroom to
414 nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
416 int error = nl_sock_cow__(sock);
420 return nl_sock_recv__(sock, buf, wait);
424 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
429 for (i = 0; i < n; i++) {
430 struct nl_transaction *txn = transactions[i];
434 ofpbuf_clear(txn->reply);
440 nl_sock_transact_multiple__(struct nl_sock *sock,
441 struct nl_transaction **transactions, size_t n,
444 uint64_t tmp_reply_stub[1024 / 8];
445 struct nl_transaction tmp_txn;
446 struct ofpbuf tmp_reply;
449 struct iovec iovs[MAX_IOVS];
454 base_seq = nl_sock_allocate_seq(sock, n);
456 for (i = 0; i < n; i++) {
457 struct nl_transaction *txn = transactions[i];
458 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
460 nlmsg->nlmsg_len = txn->request->size;
461 nlmsg->nlmsg_seq = base_seq + i;
462 nlmsg->nlmsg_pid = sock->pid;
464 iovs[i].iov_base = txn->request->data;
465 iovs[i].iov_len = txn->request->size;
468 memset(&msg, 0, sizeof msg);
472 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
473 } while (error == EINTR);
475 for (i = 0; i < n; i++) {
476 struct nl_transaction *txn = transactions[i];
478 log_nlmsg(__func__, error, txn->request->data, txn->request->size,
482 COVERAGE_ADD(netlink_sent, n);
489 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
490 tmp_txn.request = NULL;
491 tmp_txn.reply = &tmp_reply;
494 struct nl_transaction *buf_txn, *txn;
497 /* Find a transaction whose buffer we can use for receiving a reply.
498 * If no such transaction is left, use tmp_txn. */
500 for (i = 0; i < n; i++) {
501 if (transactions[i]->reply) {
502 buf_txn = transactions[i];
507 /* Receive a reply. */
508 error = nl_sock_recv__(sock, buf_txn->reply, false);
510 if (error == EAGAIN) {
511 nl_sock_record_errors__(transactions, n, 0);
518 /* Match the reply up with a transaction. */
519 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
520 if (seq < base_seq || seq >= base_seq + n) {
521 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
525 txn = transactions[i];
527 /* Fill in the results for 'txn'. */
528 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
530 ofpbuf_clear(txn->reply);
533 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
534 error, strerror(txn->error));
538 if (txn->reply && txn != buf_txn) {
540 struct ofpbuf *reply = buf_txn->reply;
541 buf_txn->reply = txn->reply;
546 /* Fill in the results for transactions before 'txn'. (We have to do
547 * this after the results for 'txn' itself because of the buffer swap
549 nl_sock_record_errors__(transactions, i, 0);
553 transactions += i + 1;
557 ofpbuf_uninit(&tmp_reply);
562 /* Sends the 'request' member of the 'n' transactions in 'transactions' on
563 * 'sock', in order, and receives responses to all of them. Fills in the
564 * 'error' member of each transaction with 0 if it was successful, otherwise
565 * with a positive errno value. If 'reply' is nonnull, then it will be filled
566 * with the reply if the message receives a detailed reply. In other cases,
567 * i.e. where the request failed or had no reply beyond an indication of
568 * success, 'reply' will be cleared if it is nonnull.
570 * The caller is responsible for destroying each request and reply, and the
571 * transactions array itself.
573 * Before sending each message, this function will finalize nlmsg_len in each
574 * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
575 * initialize nlmsg_seq.
577 * Bare Netlink is an unreliable transport protocol. This function layers
578 * reliable delivery and reply semantics on top of bare Netlink. See
579 * nl_sock_transact() for some caveats.
582 nl_sock_transact_multiple(struct nl_sock *sock,
583 struct nl_transaction **transactions, size_t n)
592 error = nl_sock_cow__(sock);
594 nl_sock_record_errors__(transactions, n, error);
598 /* In theory, every request could have a 64 kB reply. But the default and
599 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
600 * be a bit below 128 kB, so that would only allow a single message in a
601 * "batch". So we assume that replies average (at most) 4 kB, which allows
602 * a good deal of batching.
604 * In practice, most of the requests that we batch either have no reply at
605 * all or a brief reply. */
606 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
607 max_batch_count = MIN(max_batch_count, max_iovs);
613 /* Batch up to 'max_batch_count' transactions. But cap it at about a
614 * page of requests total because big skbuffs are expensive to
615 * allocate in the kernel. */
616 #if defined(PAGESIZE)
617 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
619 enum { MAX_BATCH_BYTES = 4096 - 512 };
621 bytes = transactions[0]->request->size;
622 for (count = 1; count < n && count < max_batch_count; count++) {
623 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
626 bytes += transactions[count]->request->size;
629 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
630 transactions += done;
633 if (error == ENOBUFS) {
634 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
636 VLOG_ERR_RL(&rl, "transaction error (%s)", strerror(error));
637 nl_sock_record_errors__(transactions, n, error);
642 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
643 * successful, returns 0. On failure, returns a positive errno value.
645 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
646 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
647 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
648 * reply, if any, is discarded.
650 * Before the message is sent, nlmsg_len in 'request' will be finalized to
651 * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
652 * be initialized, NLM_F_ACK will be set in nlmsg_flags.
654 * The caller is responsible for destroying 'request'.
656 * Bare Netlink is an unreliable transport protocol. This function layers
657 * reliable delivery and reply semantics on top of bare Netlink.
659 * In Netlink, sending a request to the kernel is reliable enough, because the
660 * kernel will tell us if the message cannot be queued (and we will in that
661 * case put it on the transmit queue and wait until it can be delivered).
663 * Receiving the reply is the real problem: if the socket buffer is full when
664 * the kernel tries to send the reply, the reply will be dropped. However, the
665 * kernel sets a flag that a reply has been dropped. The next call to recv
666 * then returns ENOBUFS. We can then re-send the request.
670 * 1. Netlink depends on sequence numbers to match up requests and
671 * replies. The sender of a request supplies a sequence number, and
672 * the reply echos back that sequence number.
674 * This is fine, but (1) some kernel netlink implementations are
675 * broken, in that they fail to echo sequence numbers and (2) this
676 * function will drop packets with non-matching sequence numbers, so
677 * that only a single request can be usefully transacted at a time.
679 * 2. Resending the request causes it to be re-executed, so the request
680 * needs to be idempotent.
683 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
684 struct ofpbuf **replyp)
686 struct nl_transaction *transactionp;
687 struct nl_transaction transaction;
689 transaction.request = CONST_CAST(struct ofpbuf *, request);
690 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
691 transactionp = &transaction;
693 nl_sock_transact_multiple(sock, &transactionp, 1);
696 if (transaction.error) {
697 ofpbuf_delete(transaction.reply);
700 *replyp = transaction.reply;
704 return transaction.error;
707 /* Drain all the messages currently in 'sock''s receive queue. */
709 nl_sock_drain(struct nl_sock *sock)
711 int error = nl_sock_cow__(sock);
715 return drain_rcvbuf(sock->fd);
718 /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
719 * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
720 * old fd over to the dump. */
722 nl_sock_cow__(struct nl_sock *sock)
724 struct nl_sock *copy;
733 error = nl_sock_clone(sock, ©);
743 sock->pid = copy->pid;
746 sock->dump->sock = copy;
752 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
753 * 'sock', and initializes 'dump' to reflect the state of the operation.
755 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
756 * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
757 * NLM_F_ACK will be set in nlmsg_flags.
759 * This Netlink socket library is designed to ensure that the dump is reliable
760 * and that it will not interfere with other operations on 'sock', including
761 * destroying or sending and receiving messages on 'sock'. One corner case is
764 * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
765 * whose response has not yet been received (e.g. with nl_sock_recv()).
766 * This is unusual: usually nl_sock_transact() is used to send a message
767 * and receive its reply all in one go.
769 * This function provides no status indication. An error status for the entire
770 * dump operation is provided when it is completed by calling nl_dump_done().
772 * The caller is responsible for destroying 'request'.
774 * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
778 nl_dump_start(struct nl_dump *dump,
779 struct nl_sock *sock, const struct ofpbuf *request)
781 ofpbuf_init(&dump->buffer, 4096);
783 /* 'sock' already has an ongoing dump. Clone the socket because
784 * Netlink only allows one dump at a time. */
785 dump->status = nl_sock_clone(sock, &dump->sock);
795 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
796 dump->status = nl_sock_send__(sock, request, nl_sock_allocate_seq(sock, 1),
798 dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
801 /* Helper function for nl_dump_next(). */
803 nl_dump_recv(struct nl_dump *dump)
805 struct nlmsghdr *nlmsghdr;
808 retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
810 return retval == EINTR ? EAGAIN : retval;
813 nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
814 if (dump->seq != nlmsghdr->nlmsg_seq) {
815 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
816 nlmsghdr->nlmsg_seq, dump->seq);
820 if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
821 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
823 return retval && retval != EAGAIN ? retval : EPROTO;
829 /* Attempts to retrieve another reply from 'dump', which must have been
830 * initialized with nl_dump_start().
832 * If successful, returns true and points 'reply->data' and 'reply->size' to
833 * the message that was retrieved. The caller must not modify 'reply' (because
834 * it points into the middle of a larger buffer).
836 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
837 * to 0. Failure might indicate an actual error or merely the end of replies.
838 * An error status for the entire dump operation is provided when it is
839 * completed by calling nl_dump_done().
842 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
844 struct nlmsghdr *nlmsghdr;
852 while (!dump->buffer.size) {
853 int retval = nl_dump_recv(dump);
855 ofpbuf_clear(&dump->buffer);
856 if (retval != EAGAIN) {
857 dump->status = retval;
863 nlmsghdr = nl_msg_next(&dump->buffer, reply);
865 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
866 dump->status = EPROTO;
868 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
876 /* Completes Netlink dump operation 'dump', which must have been initialized
877 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
878 * otherwise a positive errno value describing the problem. */
880 nl_dump_done(struct nl_dump *dump)
882 /* Drain any remaining messages that the client didn't read. Otherwise the
883 * kernel will continue to queue them up and waste buffer space. */
884 while (!dump->status) {
886 if (!nl_dump_next(dump, &reply)) {
887 ovs_assert(dump->status);
892 if (dump->sock->dump) {
893 dump->sock->dump = NULL;
895 nl_sock_destroy(dump->sock);
898 ofpbuf_uninit(&dump->buffer);
899 return dump->status == EOF ? 0 : dump->status;
902 /* Causes poll_block() to wake up when any of the specified 'events' (which is
903 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
905 nl_sock_wait(const struct nl_sock *sock, short int events)
907 poll_fd_wait(sock->fd, events);
910 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
911 * that can't use nl_sock_wait().
913 * It's a little tricky to use the returned fd correctly, because nl_sock does
914 * "copy on write" to allow a single nl_sock to be used for notifications,
915 * transactions, and dumps. If 'sock' is used only for notifications and
916 * transactions (and never for dump) then the usage is safe. */
918 nl_sock_fd(const struct nl_sock *sock)
923 /* Returns the PID associated with this socket. */
925 nl_sock_pid(const struct nl_sock *sock)
933 struct hmap_node hmap_node;
938 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
940 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
941 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
942 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
945 static struct genl_family *
946 find_genl_family_by_id(uint16_t id)
948 struct genl_family *family;
950 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
952 if (family->id == id) {
960 define_genl_family(uint16_t id, const char *name)
962 struct genl_family *family = find_genl_family_by_id(id);
965 if (!strcmp(family->name, name)) {
970 family = xmalloc(sizeof *family);
972 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
974 family->name = xstrdup(name);
978 genl_family_to_name(uint16_t id)
980 if (id == GENL_ID_CTRL) {
983 struct genl_family *family = find_genl_family_by_id(id);
984 return family ? family->name : "unknown";
989 do_lookup_genl_family(const char *name, struct nlattr **attrs,
990 struct ofpbuf **replyp)
992 struct nl_sock *sock;
993 struct ofpbuf request, *reply;
997 error = nl_sock_create(NETLINK_GENERIC, &sock);
1002 ofpbuf_init(&request, 0);
1003 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
1004 CTRL_CMD_GETFAMILY, 1);
1005 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
1006 error = nl_sock_transact(sock, &request, &reply);
1007 ofpbuf_uninit(&request);
1009 nl_sock_destroy(sock);
1013 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1014 family_policy, attrs, ARRAY_SIZE(family_policy))
1015 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
1016 nl_sock_destroy(sock);
1017 ofpbuf_delete(reply);
1021 nl_sock_destroy(sock);
1026 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
1027 * When successful, writes its result to 'multicast_group' and returns 0.
1028 * Otherwise, clears 'multicast_group' and returns a positive error code.
1030 * Some kernels do not support looking up a multicast group with this function.
1031 * In this case, 'multicast_group' will be populated with 'fallback'. */
1033 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
1034 unsigned int *multicast_group, unsigned int fallback)
1036 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
1037 const struct nlattr *mc;
1038 struct ofpbuf *reply;
1042 *multicast_group = 0;
1043 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1048 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1049 *multicast_group = fallback;
1050 VLOG_WARN("%s-%s: has no multicast group, using fallback %d",
1051 family_name, group_name, *multicast_group);
1056 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1057 static const struct nl_policy mc_policy[] = {
1058 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1059 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1062 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1063 const char *mc_name;
1065 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1070 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1071 if (!strcmp(group_name, mc_name)) {
1073 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1081 ofpbuf_delete(reply);
1085 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1086 * number and stores it in '*number'. If successful, returns 0 and the caller
1087 * may use '*number' as the family number. On failure, returns a positive
1088 * errno value and '*number' caches the errno value. */
1090 nl_lookup_genl_family(const char *name, int *number)
1093 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1094 struct ofpbuf *reply;
1097 error = do_lookup_genl_family(name, attrs, &reply);
1099 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1100 define_genl_family(*number, name);
1104 ofpbuf_delete(reply);
1106 ovs_assert(*number != 0);
1108 return *number > 0 ? 0 : -*number;
1112 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1114 uint32_t seq = sock->next_seq;
1116 sock->next_seq += n;
1118 /* Make it impossible for the next request for sequence numbers to wrap
1119 * around to 0. Start over with 1 to avoid ever using a sequence number of
1120 * 0, because the kernel uses sequence number 0 for notifications. */
1121 if (sock->next_seq >= UINT32_MAX / 2) {
1129 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1135 static const struct nlmsg_flag flags[] = {
1136 { NLM_F_REQUEST, "REQUEST" },
1137 { NLM_F_MULTI, "MULTI" },
1138 { NLM_F_ACK, "ACK" },
1139 { NLM_F_ECHO, "ECHO" },
1140 { NLM_F_DUMP, "DUMP" },
1141 { NLM_F_ROOT, "ROOT" },
1142 { NLM_F_MATCH, "MATCH" },
1143 { NLM_F_ATOMIC, "ATOMIC" },
1145 const struct nlmsg_flag *flag;
1146 uint16_t flags_left;
1148 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1149 h->nlmsg_len, h->nlmsg_type);
1150 if (h->nlmsg_type == NLMSG_NOOP) {
1151 ds_put_cstr(ds, "(no-op)");
1152 } else if (h->nlmsg_type == NLMSG_ERROR) {
1153 ds_put_cstr(ds, "(error)");
1154 } else if (h->nlmsg_type == NLMSG_DONE) {
1155 ds_put_cstr(ds, "(done)");
1156 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1157 ds_put_cstr(ds, "(overrun)");
1158 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1159 ds_put_cstr(ds, "(reserved)");
1160 } else if (protocol == NETLINK_GENERIC) {
1161 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1163 ds_put_cstr(ds, "(family-defined)");
1165 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1166 flags_left = h->nlmsg_flags;
1167 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1168 if ((flags_left & flag->bits) == flag->bits) {
1169 ds_put_format(ds, "[%s]", flag->name);
1170 flags_left &= ~flag->bits;
1174 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1176 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1177 h->nlmsg_seq, h->nlmsg_pid);
1181 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1183 struct ds ds = DS_EMPTY_INITIALIZER;
1184 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1186 nlmsghdr_to_string(h, protocol, &ds);
1187 if (h->nlmsg_type == NLMSG_ERROR) {
1188 const struct nlmsgerr *e;
1189 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1190 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1192 ds_put_format(&ds, " error(%d", e->error);
1194 ds_put_format(&ds, "(%s)", strerror(-e->error));
1196 ds_put_cstr(&ds, ", in-reply-to(");
1197 nlmsghdr_to_string(&e->msg, protocol, &ds);
1198 ds_put_cstr(&ds, "))");
1200 ds_put_cstr(&ds, " error(truncated)");
1202 } else if (h->nlmsg_type == NLMSG_DONE) {
1203 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1205 ds_put_format(&ds, " done(%d", *error);
1207 ds_put_format(&ds, "(%s)", strerror(-*error));
1209 ds_put_cstr(&ds, ")");
1211 ds_put_cstr(&ds, " done(truncated)");
1213 } else if (protocol == NETLINK_GENERIC) {
1214 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1216 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1217 genl->cmd, genl->version);
1221 ds_put_cstr(&ds, "nl(truncated)");
1227 log_nlmsg(const char *function, int error,
1228 const void *message, size_t size, int protocol)
1230 struct ofpbuf buffer;
1233 if (!VLOG_IS_DBG_ENABLED()) {
1237 ofpbuf_use_const(&buffer, message, size);
1238 nlmsg = nlmsg_to_string(&buffer, protocol);
1239 VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);