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. */
68 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
71 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
72 * of iovecs on the stack. */
75 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
76 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
78 * Initialized by nl_sock_create(). */
81 static int nl_sock_cow__(struct nl_sock *);
83 /* Creates a new netlink socket for the given netlink 'protocol'
84 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
85 * new socket if successful, otherwise returns a positive errno value. */
87 nl_sock_create(int protocol, struct nl_sock **sockp)
90 struct sockaddr_nl local, remote;
96 int save_errno = errno;
99 max_iovs = sysconf(_SC_UIO_MAXIOV);
100 if (max_iovs < _XOPEN_IOV_MAX) {
101 if (max_iovs == -1 && errno) {
102 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", strerror(errno));
104 max_iovs = _XOPEN_IOV_MAX;
105 } else if (max_iovs > MAX_IOVS) {
113 sock = malloc(sizeof *sock);
118 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
120 VLOG_ERR("fcntl: %s", strerror(errno));
123 sock->protocol = protocol;
127 rcvbuf = 1024 * 1024;
128 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
129 &rcvbuf, sizeof rcvbuf)) {
130 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
131 * Warn only if the failure is therefore unexpected. */
132 if (errno != EPERM) {
133 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
134 "(%s)", rcvbuf, strerror(errno));
138 retval = get_socket_rcvbuf(sock->fd);
143 sock->rcvbuf = retval;
145 /* Connect to kernel (pid 0) as remote address. */
146 memset(&remote, 0, sizeof remote);
147 remote.nl_family = AF_NETLINK;
149 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
150 VLOG_ERR("connect(0): %s", strerror(errno));
154 /* Obtain pid assigned by kernel. */
155 local_size = sizeof local;
156 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
157 VLOG_ERR("getsockname: %s", strerror(errno));
160 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
161 VLOG_ERR("getsockname returned bad Netlink name");
165 sock->pid = local.nl_pid;
184 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
185 * sets '*sockp' to the new socket if successful, otherwise returns a positive
188 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
190 return nl_sock_create(src->protocol, sockp);
193 /* Destroys netlink socket 'sock'. */
195 nl_sock_destroy(struct nl_sock *sock)
207 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
208 * successful, otherwise a positive errno value.
210 * A socket that is subscribed to a multicast group that receives asynchronous
211 * notifications must not be used for Netlink transactions or dumps, because
212 * transactions and dumps can cause notifications to be lost.
214 * Multicast group numbers are always positive.
216 * It is not an error to attempt to join a multicast group to which a socket
217 * already belongs. */
219 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
221 int error = nl_sock_cow__(sock);
225 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
226 &multicast_group, sizeof multicast_group) < 0) {
227 VLOG_WARN("could not join multicast group %u (%s)",
228 multicast_group, strerror(errno));
234 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
235 * successful, otherwise a positive errno value.
237 * Multicast group numbers are always positive.
239 * It is not an error to attempt to leave a multicast group to which a socket
242 * On success, reading from 'sock' will still return any messages that were
243 * received on 'multicast_group' before the group was left. */
245 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
247 ovs_assert(!sock->dump);
248 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
249 &multicast_group, sizeof multicast_group) < 0) {
250 VLOG_WARN("could not leave multicast group %u (%s)",
251 multicast_group, strerror(errno));
258 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
259 uint32_t nlmsg_seq, bool wait)
261 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
264 nlmsg->nlmsg_len = msg->size;
265 nlmsg->nlmsg_seq = nlmsg_seq;
266 nlmsg->nlmsg_pid = sock->pid;
269 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
270 error = retval < 0 ? errno : 0;
271 } while (error == EINTR);
272 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
274 COVERAGE_INC(netlink_sent);
279 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
280 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
281 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
282 * sequence number, before the message is sent.
284 * Returns 0 if successful, otherwise a positive errno value. If
285 * 'wait' is true, then the send will wait until buffer space is ready;
286 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
288 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
290 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
293 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
294 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
295 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
296 * 'nlmsg_seq', before the message is sent.
298 * Returns 0 if successful, otherwise a positive errno value. If
299 * 'wait' is true, then the send will wait until buffer space is ready;
300 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
302 * This function is suitable for sending a reply to a request that was received
303 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
305 nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
306 uint32_t nlmsg_seq, bool wait)
308 int error = nl_sock_cow__(sock);
312 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
315 /* This stress option is useful for testing that OVS properly tolerates
316 * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
317 * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
318 * reply to a request. They can also occur if messages arrive on a multicast
319 * channel faster than OVS can process them. */
321 netlink_overflow, "simulate netlink socket receive buffer overflow",
325 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
327 /* We can't accurately predict the size of the data to be received. The
328 * caller is supposed to have allocated enough space in 'buf' to handle the
329 * "typical" case. To handle exceptions, we make available enough space in
330 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
331 * figure since that's the maximum length of a Netlink attribute). */
332 struct nlmsghdr *nlmsghdr;
338 ovs_assert(buf->allocated >= sizeof *nlmsghdr);
341 iov[0].iov_base = buf->base;
342 iov[0].iov_len = buf->allocated;
343 iov[1].iov_base = tail;
344 iov[1].iov_len = sizeof tail;
346 memset(&msg, 0, sizeof msg);
351 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
352 } while (retval < 0 && errno == EINTR);
356 if (error == ENOBUFS) {
357 /* Socket receive buffer overflow dropped one or more messages that
358 * the kernel tried to send to us. */
359 COVERAGE_INC(netlink_overflow);
364 if (msg.msg_flags & MSG_TRUNC) {
365 VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)",
370 nlmsghdr = buf->data;
371 if (retval < sizeof *nlmsghdr
372 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
373 || nlmsghdr->nlmsg_len > retval) {
374 VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)",
375 retval, sizeof *nlmsghdr);
379 if (STRESS(netlink_overflow)) {
383 buf->size = MIN(retval, buf->allocated);
384 if (retval > buf->allocated) {
385 COVERAGE_INC(netlink_recv_jumbo);
386 ofpbuf_put(buf, tail, retval - buf->allocated);
389 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
390 COVERAGE_INC(netlink_received);
395 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
396 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
397 * EAGAIN if the 'sock' receive buffer is empty.
399 * The caller must have initialized 'buf' with an allocation of at least
400 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
401 * space for a "typical" message.
403 * On success, returns 0 and replaces 'buf''s previous content by the received
404 * message. This function expands 'buf''s allocated memory, as necessary, to
405 * hold the actual size of the received message.
407 * On failure, returns a positive errno value and clears 'buf' to zero length.
408 * 'buf' retains its previous memory allocation.
410 * Regardless of success or failure, this function resets 'buf''s headroom to
413 nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
415 int error = nl_sock_cow__(sock);
419 return nl_sock_recv__(sock, buf, wait);
423 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
428 for (i = 0; i < n; i++) {
429 struct nl_transaction *txn = transactions[i];
433 ofpbuf_clear(txn->reply);
439 nl_sock_transact_multiple__(struct nl_sock *sock,
440 struct nl_transaction **transactions, size_t n,
443 uint64_t tmp_reply_stub[1024 / 8];
444 struct nl_transaction tmp_txn;
445 struct ofpbuf tmp_reply;
448 struct iovec iovs[MAX_IOVS];
453 base_seq = nl_sock_allocate_seq(sock, n);
455 for (i = 0; i < n; i++) {
456 struct nl_transaction *txn = transactions[i];
457 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
459 nlmsg->nlmsg_len = txn->request->size;
460 nlmsg->nlmsg_seq = base_seq + i;
461 nlmsg->nlmsg_pid = sock->pid;
463 iovs[i].iov_base = txn->request->data;
464 iovs[i].iov_len = txn->request->size;
467 memset(&msg, 0, sizeof msg);
471 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
472 } while (error == EINTR);
474 for (i = 0; i < n; i++) {
475 struct nl_transaction *txn = transactions[i];
477 log_nlmsg(__func__, error, txn->request->data, txn->request->size,
481 COVERAGE_ADD(netlink_sent, n);
488 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
489 tmp_txn.request = NULL;
490 tmp_txn.reply = &tmp_reply;
493 struct nl_transaction *buf_txn, *txn;
496 /* Find a transaction whose buffer we can use for receiving a reply.
497 * If no such transaction is left, use tmp_txn. */
499 for (i = 0; i < n; i++) {
500 if (transactions[i]->reply) {
501 buf_txn = transactions[i];
506 /* Receive a reply. */
507 error = nl_sock_recv__(sock, buf_txn->reply, false);
509 if (error == EAGAIN) {
510 nl_sock_record_errors__(transactions, n, 0);
517 /* Match the reply up with a transaction. */
518 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
519 if (seq < base_seq || seq >= base_seq + n) {
520 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
524 txn = transactions[i];
526 /* Fill in the results for 'txn'. */
527 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
529 ofpbuf_clear(txn->reply);
532 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
533 error, strerror(txn->error));
537 if (txn->reply && txn != buf_txn) {
539 struct ofpbuf *reply = buf_txn->reply;
540 buf_txn->reply = txn->reply;
545 /* Fill in the results for transactions before 'txn'. (We have to do
546 * this after the results for 'txn' itself because of the buffer swap
548 nl_sock_record_errors__(transactions, i, 0);
552 transactions += i + 1;
556 ofpbuf_uninit(&tmp_reply);
561 /* Sends the 'request' member of the 'n' transactions in 'transactions' on
562 * 'sock', in order, and receives responses to all of them. Fills in the
563 * 'error' member of each transaction with 0 if it was successful, otherwise
564 * with a positive errno value. If 'reply' is nonnull, then it will be filled
565 * with the reply if the message receives a detailed reply. In other cases,
566 * i.e. where the request failed or had no reply beyond an indication of
567 * success, 'reply' will be cleared if it is nonnull.
569 * The caller is responsible for destroying each request and reply, and the
570 * transactions array itself.
572 * Before sending each message, this function will finalize nlmsg_len in each
573 * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
574 * initialize nlmsg_seq.
576 * Bare Netlink is an unreliable transport protocol. This function layers
577 * reliable delivery and reply semantics on top of bare Netlink. See
578 * nl_sock_transact() for some caveats.
581 nl_sock_transact_multiple(struct nl_sock *sock,
582 struct nl_transaction **transactions, size_t n)
591 error = nl_sock_cow__(sock);
593 nl_sock_record_errors__(transactions, n, error);
597 /* In theory, every request could have a 64 kB reply. But the default and
598 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
599 * be a bit below 128 kB, so that would only allow a single message in a
600 * "batch". So we assume that replies average (at most) 4 kB, which allows
601 * a good deal of batching.
603 * In practice, most of the requests that we batch either have no reply at
604 * all or a brief reply. */
605 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
606 max_batch_count = MIN(max_batch_count, max_iovs);
612 /* Batch up to 'max_batch_count' transactions. But cap it at about a
613 * page of requests total because big skbuffs are expensive to
614 * allocate in the kernel. */
615 #if defined(PAGESIZE)
616 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
618 enum { MAX_BATCH_BYTES = 4096 - 512 };
620 bytes = transactions[0]->request->size;
621 for (count = 1; count < n && count < max_batch_count; count++) {
622 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
625 bytes += transactions[count]->request->size;
628 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
629 transactions += done;
632 if (error == ENOBUFS) {
633 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
635 VLOG_ERR_RL(&rl, "transaction error (%s)", strerror(error));
636 nl_sock_record_errors__(transactions, n, error);
641 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
642 * successful, returns 0. On failure, returns a positive errno value.
644 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
645 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
646 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
647 * reply, if any, is discarded.
649 * Before the message is sent, nlmsg_len in 'request' will be finalized to
650 * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
651 * be initialized, NLM_F_ACK will be set in nlmsg_flags.
653 * The caller is responsible for destroying 'request'.
655 * Bare Netlink is an unreliable transport protocol. This function layers
656 * reliable delivery and reply semantics on top of bare Netlink.
658 * In Netlink, sending a request to the kernel is reliable enough, because the
659 * kernel will tell us if the message cannot be queued (and we will in that
660 * case put it on the transmit queue and wait until it can be delivered).
662 * Receiving the reply is the real problem: if the socket buffer is full when
663 * the kernel tries to send the reply, the reply will be dropped. However, the
664 * kernel sets a flag that a reply has been dropped. The next call to recv
665 * then returns ENOBUFS. We can then re-send the request.
669 * 1. Netlink depends on sequence numbers to match up requests and
670 * replies. The sender of a request supplies a sequence number, and
671 * the reply echos back that sequence number.
673 * This is fine, but (1) some kernel netlink implementations are
674 * broken, in that they fail to echo sequence numbers and (2) this
675 * function will drop packets with non-matching sequence numbers, so
676 * that only a single request can be usefully transacted at a time.
678 * 2. Resending the request causes it to be re-executed, so the request
679 * needs to be idempotent.
682 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
683 struct ofpbuf **replyp)
685 struct nl_transaction *transactionp;
686 struct nl_transaction transaction;
688 transaction.request = CONST_CAST(struct ofpbuf *, request);
689 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
690 transactionp = &transaction;
692 nl_sock_transact_multiple(sock, &transactionp, 1);
695 if (transaction.error) {
696 ofpbuf_delete(transaction.reply);
699 *replyp = transaction.reply;
703 return transaction.error;
706 /* Drain all the messages currently in 'sock''s receive queue. */
708 nl_sock_drain(struct nl_sock *sock)
710 int error = nl_sock_cow__(sock);
714 return drain_rcvbuf(sock->fd);
717 /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
718 * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
719 * old fd over to the dump. */
721 nl_sock_cow__(struct nl_sock *sock)
723 struct nl_sock *copy;
732 error = nl_sock_clone(sock, ©);
742 sock->pid = copy->pid;
745 sock->dump->sock = copy;
751 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
752 * 'sock', and initializes 'dump' to reflect the state of the operation.
754 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
755 * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
756 * NLM_F_ACK will be set in nlmsg_flags.
758 * This Netlink socket library is designed to ensure that the dump is reliable
759 * and that it will not interfere with other operations on 'sock', including
760 * destroying or sending and receiving messages on 'sock'. One corner case is
763 * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
764 * whose response has not yet been received (e.g. with nl_sock_recv()).
765 * This is unusual: usually nl_sock_transact() is used to send a message
766 * and receive its reply all in one go.
768 * This function provides no status indication. An error status for the entire
769 * dump operation is provided when it is completed by calling nl_dump_done().
771 * The caller is responsible for destroying 'request'.
773 * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
777 nl_dump_start(struct nl_dump *dump,
778 struct nl_sock *sock, const struct ofpbuf *request)
780 ofpbuf_init(&dump->buffer, 4096);
782 /* 'sock' already has an ongoing dump. Clone the socket because
783 * Netlink only allows one dump at a time. */
784 dump->status = nl_sock_clone(sock, &dump->sock);
794 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
795 dump->status = nl_sock_send__(sock, request, nl_sock_allocate_seq(sock, 1),
797 dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
800 /* Helper function for nl_dump_next(). */
802 nl_dump_recv(struct nl_dump *dump)
804 struct nlmsghdr *nlmsghdr;
807 retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
809 return retval == EINTR ? EAGAIN : retval;
812 nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
813 if (dump->seq != nlmsghdr->nlmsg_seq) {
814 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
815 nlmsghdr->nlmsg_seq, dump->seq);
819 if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
820 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
822 return retval && retval != EAGAIN ? retval : EPROTO;
828 /* Attempts to retrieve another reply from 'dump', which must have been
829 * initialized with nl_dump_start().
831 * If successful, returns true and points 'reply->data' and 'reply->size' to
832 * the message that was retrieved. The caller must not modify 'reply' (because
833 * it points into the middle of a larger buffer).
835 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
836 * to 0. Failure might indicate an actual error or merely the end of replies.
837 * An error status for the entire dump operation is provided when it is
838 * completed by calling nl_dump_done().
841 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
843 struct nlmsghdr *nlmsghdr;
851 while (!dump->buffer.size) {
852 int retval = nl_dump_recv(dump);
854 ofpbuf_clear(&dump->buffer);
855 if (retval != EAGAIN) {
856 dump->status = retval;
862 nlmsghdr = nl_msg_next(&dump->buffer, reply);
864 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
865 dump->status = EPROTO;
867 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
875 /* Completes Netlink dump operation 'dump', which must have been initialized
876 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
877 * otherwise a positive errno value describing the problem. */
879 nl_dump_done(struct nl_dump *dump)
881 /* Drain any remaining messages that the client didn't read. Otherwise the
882 * kernel will continue to queue them up and waste buffer space. */
883 while (!dump->status) {
885 if (!nl_dump_next(dump, &reply)) {
886 ovs_assert(dump->status);
891 if (dump->sock->dump) {
892 dump->sock->dump = NULL;
894 nl_sock_destroy(dump->sock);
897 ofpbuf_uninit(&dump->buffer);
898 return dump->status == EOF ? 0 : dump->status;
901 /* Causes poll_block() to wake up when any of the specified 'events' (which is
902 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
904 nl_sock_wait(const struct nl_sock *sock, short int events)
906 poll_fd_wait(sock->fd, events);
909 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
910 * that can't use nl_sock_wait().
912 * It's a little tricky to use the returned fd correctly, because nl_sock does
913 * "copy on write" to allow a single nl_sock to be used for notifications,
914 * transactions, and dumps. If 'sock' is used only for notifications and
915 * transactions (and never for dump) then the usage is safe. */
917 nl_sock_fd(const struct nl_sock *sock)
922 /* Returns the PID associated with this socket. */
924 nl_sock_pid(const struct nl_sock *sock)
932 struct hmap_node hmap_node;
937 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
939 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
940 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
941 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
944 static struct genl_family *
945 find_genl_family_by_id(uint16_t id)
947 struct genl_family *family;
949 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
951 if (family->id == id) {
959 define_genl_family(uint16_t id, const char *name)
961 struct genl_family *family = find_genl_family_by_id(id);
964 if (!strcmp(family->name, name)) {
969 family = xmalloc(sizeof *family);
971 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
973 family->name = xstrdup(name);
977 genl_family_to_name(uint16_t id)
979 if (id == GENL_ID_CTRL) {
982 struct genl_family *family = find_genl_family_by_id(id);
983 return family ? family->name : "unknown";
988 do_lookup_genl_family(const char *name, struct nlattr **attrs,
989 struct ofpbuf **replyp)
991 struct nl_sock *sock;
992 struct ofpbuf request, *reply;
996 error = nl_sock_create(NETLINK_GENERIC, &sock);
1001 ofpbuf_init(&request, 0);
1002 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
1003 CTRL_CMD_GETFAMILY, 1);
1004 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
1005 error = nl_sock_transact(sock, &request, &reply);
1006 ofpbuf_uninit(&request);
1008 nl_sock_destroy(sock);
1012 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1013 family_policy, attrs, ARRAY_SIZE(family_policy))
1014 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
1015 nl_sock_destroy(sock);
1016 ofpbuf_delete(reply);
1020 nl_sock_destroy(sock);
1025 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
1026 * When successful, writes its result to 'multicast_group' and returns 0.
1027 * Otherwise, clears 'multicast_group' and returns a positive error code.
1029 * Some kernels do not support looking up a multicast group with this function.
1030 * In this case, 'multicast_group' will be populated with 'fallback'. */
1032 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
1033 unsigned int *multicast_group, unsigned int fallback)
1035 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
1036 const struct nlattr *mc;
1037 struct ofpbuf *reply;
1041 *multicast_group = 0;
1042 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1047 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1048 *multicast_group = fallback;
1049 VLOG_WARN("%s-%s: has no multicast group, using fallback %d",
1050 family_name, group_name, *multicast_group);
1055 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1056 static const struct nl_policy mc_policy[] = {
1057 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1058 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1061 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1062 const char *mc_name;
1064 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1069 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1070 if (!strcmp(group_name, mc_name)) {
1072 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1080 ofpbuf_delete(reply);
1084 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1085 * number and stores it in '*number'. If successful, returns 0 and the caller
1086 * may use '*number' as the family number. On failure, returns a positive
1087 * errno value and '*number' caches the errno value. */
1089 nl_lookup_genl_family(const char *name, int *number)
1092 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1093 struct ofpbuf *reply;
1096 error = do_lookup_genl_family(name, attrs, &reply);
1098 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1099 define_genl_family(*number, name);
1103 ofpbuf_delete(reply);
1105 ovs_assert(*number != 0);
1107 return *number > 0 ? 0 : -*number;
1111 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1113 uint32_t seq = sock->next_seq;
1115 sock->next_seq += n;
1117 /* Make it impossible for the next request for sequence numbers to wrap
1118 * around to 0. Start over with 1 to avoid ever using a sequence number of
1119 * 0, because the kernel uses sequence number 0 for notifications. */
1120 if (sock->next_seq >= UINT32_MAX / 2) {
1128 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1134 static const struct nlmsg_flag flags[] = {
1135 { NLM_F_REQUEST, "REQUEST" },
1136 { NLM_F_MULTI, "MULTI" },
1137 { NLM_F_ACK, "ACK" },
1138 { NLM_F_ECHO, "ECHO" },
1139 { NLM_F_DUMP, "DUMP" },
1140 { NLM_F_ROOT, "ROOT" },
1141 { NLM_F_MATCH, "MATCH" },
1142 { NLM_F_ATOMIC, "ATOMIC" },
1144 const struct nlmsg_flag *flag;
1145 uint16_t flags_left;
1147 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1148 h->nlmsg_len, h->nlmsg_type);
1149 if (h->nlmsg_type == NLMSG_NOOP) {
1150 ds_put_cstr(ds, "(no-op)");
1151 } else if (h->nlmsg_type == NLMSG_ERROR) {
1152 ds_put_cstr(ds, "(error)");
1153 } else if (h->nlmsg_type == NLMSG_DONE) {
1154 ds_put_cstr(ds, "(done)");
1155 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1156 ds_put_cstr(ds, "(overrun)");
1157 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1158 ds_put_cstr(ds, "(reserved)");
1159 } else if (protocol == NETLINK_GENERIC) {
1160 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1162 ds_put_cstr(ds, "(family-defined)");
1164 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1165 flags_left = h->nlmsg_flags;
1166 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1167 if ((flags_left & flag->bits) == flag->bits) {
1168 ds_put_format(ds, "[%s]", flag->name);
1169 flags_left &= ~flag->bits;
1173 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1175 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1176 h->nlmsg_seq, h->nlmsg_pid);
1180 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1182 struct ds ds = DS_EMPTY_INITIALIZER;
1183 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1185 nlmsghdr_to_string(h, protocol, &ds);
1186 if (h->nlmsg_type == NLMSG_ERROR) {
1187 const struct nlmsgerr *e;
1188 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1189 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1191 ds_put_format(&ds, " error(%d", e->error);
1193 ds_put_format(&ds, "(%s)", strerror(-e->error));
1195 ds_put_cstr(&ds, ", in-reply-to(");
1196 nlmsghdr_to_string(&e->msg, protocol, &ds);
1197 ds_put_cstr(&ds, "))");
1199 ds_put_cstr(&ds, " error(truncated)");
1201 } else if (h->nlmsg_type == NLMSG_DONE) {
1202 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1204 ds_put_format(&ds, " done(%d", *error);
1206 ds_put_format(&ds, "(%s)", strerror(-*error));
1208 ds_put_cstr(&ds, ")");
1210 ds_put_cstr(&ds, " done(truncated)");
1212 } else if (protocol == NETLINK_GENERIC) {
1213 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1215 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1216 genl->cmd, genl->version);
1220 ds_put_cstr(&ds, "nl(truncated)");
1226 log_nlmsg(const char *function, int error,
1227 const void *message, size_t size, int protocol)
1229 struct ofpbuf buffer;
1232 if (!VLOG_IS_DBG_ENABLED()) {
1236 ofpbuf_use_const(&buffer, message, size);
1237 nlmsg = nlmsg_to_string(&buffer, protocol);
1238 VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);