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 = xmalloc(sizeof *sock);
115 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
117 VLOG_ERR("fcntl: %s", strerror(errno));
120 sock->protocol = protocol;
124 rcvbuf = 1024 * 1024;
125 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
126 &rcvbuf, sizeof rcvbuf)) {
127 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
128 * Warn only if the failure is therefore unexpected. */
129 if (errno != EPERM) {
130 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
131 "(%s)", rcvbuf, strerror(errno));
135 retval = get_socket_rcvbuf(sock->fd);
140 sock->rcvbuf = retval;
142 /* Connect to kernel (pid 0) as remote address. */
143 memset(&remote, 0, sizeof remote);
144 remote.nl_family = AF_NETLINK;
146 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
147 VLOG_ERR("connect(0): %s", strerror(errno));
151 /* Obtain pid assigned by kernel. */
152 local_size = sizeof local;
153 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
154 VLOG_ERR("getsockname: %s", strerror(errno));
157 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
158 VLOG_ERR("getsockname returned bad Netlink name");
162 sock->pid = local.nl_pid;
181 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
182 * sets '*sockp' to the new socket if successful, otherwise returns a positive
185 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
187 return nl_sock_create(src->protocol, sockp);
190 /* Destroys netlink socket 'sock'. */
192 nl_sock_destroy(struct nl_sock *sock)
204 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
205 * successful, otherwise a positive errno value.
207 * A socket that is subscribed to a multicast group that receives asynchronous
208 * notifications must not be used for Netlink transactions or dumps, because
209 * transactions and dumps can cause notifications to be lost.
211 * Multicast group numbers are always positive.
213 * It is not an error to attempt to join a multicast group to which a socket
214 * already belongs. */
216 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
218 int error = nl_sock_cow__(sock);
222 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
223 &multicast_group, sizeof multicast_group) < 0) {
224 VLOG_WARN("could not join multicast group %u (%s)",
225 multicast_group, strerror(errno));
231 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
232 * successful, otherwise a positive errno value.
234 * Multicast group numbers are always positive.
236 * It is not an error to attempt to leave a multicast group to which a socket
239 * On success, reading from 'sock' will still return any messages that were
240 * received on 'multicast_group' before the group was left. */
242 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
244 ovs_assert(!sock->dump);
245 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
246 &multicast_group, sizeof multicast_group) < 0) {
247 VLOG_WARN("could not leave multicast group %u (%s)",
248 multicast_group, strerror(errno));
255 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
256 uint32_t nlmsg_seq, bool wait)
258 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
261 nlmsg->nlmsg_len = msg->size;
262 nlmsg->nlmsg_seq = nlmsg_seq;
263 nlmsg->nlmsg_pid = sock->pid;
266 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
267 error = retval < 0 ? errno : 0;
268 } while (error == EINTR);
269 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
271 COVERAGE_INC(netlink_sent);
276 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
277 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
278 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
279 * sequence number, before the message is sent.
281 * Returns 0 if successful, otherwise a positive errno value. If
282 * 'wait' is true, then the send will wait until buffer space is ready;
283 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
285 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
287 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
290 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
291 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
292 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
293 * 'nlmsg_seq', before the message is sent.
295 * Returns 0 if successful, otherwise a positive errno value. If
296 * 'wait' is true, then the send will wait until buffer space is ready;
297 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
299 * This function is suitable for sending a reply to a request that was received
300 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
302 nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
303 uint32_t nlmsg_seq, bool wait)
305 int error = nl_sock_cow__(sock);
309 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
312 /* This stress option is useful for testing that OVS properly tolerates
313 * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
314 * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
315 * reply to a request. They can also occur if messages arrive on a multicast
316 * channel faster than OVS can process them. */
318 netlink_overflow, "simulate netlink socket receive buffer overflow",
322 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
324 /* We can't accurately predict the size of the data to be received. The
325 * caller is supposed to have allocated enough space in 'buf' to handle the
326 * "typical" case. To handle exceptions, we make available enough space in
327 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
328 * figure since that's the maximum length of a Netlink attribute). */
329 struct nlmsghdr *nlmsghdr;
335 ovs_assert(buf->allocated >= sizeof *nlmsghdr);
338 iov[0].iov_base = buf->base;
339 iov[0].iov_len = buf->allocated;
340 iov[1].iov_base = tail;
341 iov[1].iov_len = sizeof tail;
343 memset(&msg, 0, sizeof msg);
348 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
349 } while (retval < 0 && errno == EINTR);
353 if (error == ENOBUFS) {
354 /* Socket receive buffer overflow dropped one or more messages that
355 * the kernel tried to send to us. */
356 COVERAGE_INC(netlink_overflow);
361 if (msg.msg_flags & MSG_TRUNC) {
362 VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)",
367 nlmsghdr = buf->data;
368 if (retval < sizeof *nlmsghdr
369 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
370 || nlmsghdr->nlmsg_len > retval) {
371 VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)",
372 retval, sizeof *nlmsghdr);
376 if (STRESS(netlink_overflow)) {
380 buf->size = MIN(retval, buf->allocated);
381 if (retval > buf->allocated) {
382 COVERAGE_INC(netlink_recv_jumbo);
383 ofpbuf_put(buf, tail, retval - buf->allocated);
386 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
387 COVERAGE_INC(netlink_received);
392 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
393 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
394 * EAGAIN if the 'sock' receive buffer is empty.
396 * The caller must have initialized 'buf' with an allocation of at least
397 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
398 * space for a "typical" message.
400 * On success, returns 0 and replaces 'buf''s previous content by the received
401 * message. This function expands 'buf''s allocated memory, as necessary, to
402 * hold the actual size of the received message.
404 * On failure, returns a positive errno value and clears 'buf' to zero length.
405 * 'buf' retains its previous memory allocation.
407 * Regardless of success or failure, this function resets 'buf''s headroom to
410 nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
412 int error = nl_sock_cow__(sock);
416 return nl_sock_recv__(sock, buf, wait);
420 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
425 for (i = 0; i < n; i++) {
426 struct nl_transaction *txn = transactions[i];
430 ofpbuf_clear(txn->reply);
436 nl_sock_transact_multiple__(struct nl_sock *sock,
437 struct nl_transaction **transactions, size_t n,
440 uint64_t tmp_reply_stub[1024 / 8];
441 struct nl_transaction tmp_txn;
442 struct ofpbuf tmp_reply;
445 struct iovec iovs[MAX_IOVS];
450 base_seq = nl_sock_allocate_seq(sock, n);
452 for (i = 0; i < n; i++) {
453 struct nl_transaction *txn = transactions[i];
454 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
456 nlmsg->nlmsg_len = txn->request->size;
457 nlmsg->nlmsg_seq = base_seq + i;
458 nlmsg->nlmsg_pid = sock->pid;
460 iovs[i].iov_base = txn->request->data;
461 iovs[i].iov_len = txn->request->size;
464 memset(&msg, 0, sizeof msg);
468 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
469 } while (error == EINTR);
471 for (i = 0; i < n; i++) {
472 struct nl_transaction *txn = transactions[i];
474 log_nlmsg(__func__, error, txn->request->data, txn->request->size,
478 COVERAGE_ADD(netlink_sent, n);
485 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
486 tmp_txn.request = NULL;
487 tmp_txn.reply = &tmp_reply;
490 struct nl_transaction *buf_txn, *txn;
493 /* Find a transaction whose buffer we can use for receiving a reply.
494 * If no such transaction is left, use tmp_txn. */
496 for (i = 0; i < n; i++) {
497 if (transactions[i]->reply) {
498 buf_txn = transactions[i];
503 /* Receive a reply. */
504 error = nl_sock_recv__(sock, buf_txn->reply, false);
506 if (error == EAGAIN) {
507 nl_sock_record_errors__(transactions, n, 0);
514 /* Match the reply up with a transaction. */
515 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
516 if (seq < base_seq || seq >= base_seq + n) {
517 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
521 txn = transactions[i];
523 /* Fill in the results for 'txn'. */
524 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
526 ofpbuf_clear(txn->reply);
529 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
530 error, strerror(txn->error));
534 if (txn->reply && txn != buf_txn) {
536 struct ofpbuf *reply = buf_txn->reply;
537 buf_txn->reply = txn->reply;
542 /* Fill in the results for transactions before 'txn'. (We have to do
543 * this after the results for 'txn' itself because of the buffer swap
545 nl_sock_record_errors__(transactions, i, 0);
549 transactions += i + 1;
553 ofpbuf_uninit(&tmp_reply);
558 /* Sends the 'request' member of the 'n' transactions in 'transactions' on
559 * 'sock', in order, and receives responses to all of them. Fills in the
560 * 'error' member of each transaction with 0 if it was successful, otherwise
561 * with a positive errno value. If 'reply' is nonnull, then it will be filled
562 * with the reply if the message receives a detailed reply. In other cases,
563 * i.e. where the request failed or had no reply beyond an indication of
564 * success, 'reply' will be cleared if it is nonnull.
566 * The caller is responsible for destroying each request and reply, and the
567 * transactions array itself.
569 * Before sending each message, this function will finalize nlmsg_len in each
570 * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
571 * initialize nlmsg_seq.
573 * Bare Netlink is an unreliable transport protocol. This function layers
574 * reliable delivery and reply semantics on top of bare Netlink. See
575 * nl_sock_transact() for some caveats.
578 nl_sock_transact_multiple(struct nl_sock *sock,
579 struct nl_transaction **transactions, size_t n)
588 error = nl_sock_cow__(sock);
590 nl_sock_record_errors__(transactions, n, error);
594 /* In theory, every request could have a 64 kB reply. But the default and
595 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
596 * be a bit below 128 kB, so that would only allow a single message in a
597 * "batch". So we assume that replies average (at most) 4 kB, which allows
598 * a good deal of batching.
600 * In practice, most of the requests that we batch either have no reply at
601 * all or a brief reply. */
602 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
603 max_batch_count = MIN(max_batch_count, max_iovs);
609 /* Batch up to 'max_batch_count' transactions. But cap it at about a
610 * page of requests total because big skbuffs are expensive to
611 * allocate in the kernel. */
612 #if defined(PAGESIZE)
613 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
615 enum { MAX_BATCH_BYTES = 4096 - 512 };
617 bytes = transactions[0]->request->size;
618 for (count = 1; count < n && count < max_batch_count; count++) {
619 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
622 bytes += transactions[count]->request->size;
625 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
626 transactions += done;
629 if (error == ENOBUFS) {
630 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
632 VLOG_ERR_RL(&rl, "transaction error (%s)", strerror(error));
633 nl_sock_record_errors__(transactions, n, error);
638 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
639 * successful, returns 0. On failure, returns a positive errno value.
641 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
642 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
643 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
644 * reply, if any, is discarded.
646 * Before the message is sent, nlmsg_len in 'request' will be finalized to
647 * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
648 * be initialized, NLM_F_ACK will be set in nlmsg_flags.
650 * The caller is responsible for destroying 'request'.
652 * Bare Netlink is an unreliable transport protocol. This function layers
653 * reliable delivery and reply semantics on top of bare Netlink.
655 * In Netlink, sending a request to the kernel is reliable enough, because the
656 * kernel will tell us if the message cannot be queued (and we will in that
657 * case put it on the transmit queue and wait until it can be delivered).
659 * Receiving the reply is the real problem: if the socket buffer is full when
660 * the kernel tries to send the reply, the reply will be dropped. However, the
661 * kernel sets a flag that a reply has been dropped. The next call to recv
662 * then returns ENOBUFS. We can then re-send the request.
666 * 1. Netlink depends on sequence numbers to match up requests and
667 * replies. The sender of a request supplies a sequence number, and
668 * the reply echos back that sequence number.
670 * This is fine, but (1) some kernel netlink implementations are
671 * broken, in that they fail to echo sequence numbers and (2) this
672 * function will drop packets with non-matching sequence numbers, so
673 * that only a single request can be usefully transacted at a time.
675 * 2. Resending the request causes it to be re-executed, so the request
676 * needs to be idempotent.
679 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
680 struct ofpbuf **replyp)
682 struct nl_transaction *transactionp;
683 struct nl_transaction transaction;
685 transaction.request = CONST_CAST(struct ofpbuf *, request);
686 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
687 transactionp = &transaction;
689 nl_sock_transact_multiple(sock, &transactionp, 1);
692 if (transaction.error) {
693 ofpbuf_delete(transaction.reply);
696 *replyp = transaction.reply;
700 return transaction.error;
703 /* Drain all the messages currently in 'sock''s receive queue. */
705 nl_sock_drain(struct nl_sock *sock)
707 int error = nl_sock_cow__(sock);
711 return drain_rcvbuf(sock->fd);
714 /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
715 * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
716 * old fd over to the dump. */
718 nl_sock_cow__(struct nl_sock *sock)
720 struct nl_sock *copy;
729 error = nl_sock_clone(sock, ©);
739 sock->pid = copy->pid;
742 sock->dump->sock = copy;
748 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
749 * 'sock', and initializes 'dump' to reflect the state of the operation.
751 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
752 * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
753 * NLM_F_ACK will be set in nlmsg_flags.
755 * This Netlink socket library is designed to ensure that the dump is reliable
756 * and that it will not interfere with other operations on 'sock', including
757 * destroying or sending and receiving messages on 'sock'. One corner case is
760 * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
761 * whose response has not yet been received (e.g. with nl_sock_recv()).
762 * This is unusual: usually nl_sock_transact() is used to send a message
763 * and receive its reply all in one go.
765 * This function provides no status indication. An error status for the entire
766 * dump operation is provided when it is completed by calling nl_dump_done().
768 * The caller is responsible for destroying 'request'.
770 * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
774 nl_dump_start(struct nl_dump *dump,
775 struct nl_sock *sock, const struct ofpbuf *request)
777 ofpbuf_init(&dump->buffer, 4096);
779 /* 'sock' already has an ongoing dump. Clone the socket because
780 * Netlink only allows one dump at a time. */
781 dump->status = nl_sock_clone(sock, &dump->sock);
791 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
792 dump->status = nl_sock_send__(sock, request, nl_sock_allocate_seq(sock, 1),
794 dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
797 /* Helper function for nl_dump_next(). */
799 nl_dump_recv(struct nl_dump *dump)
801 struct nlmsghdr *nlmsghdr;
804 retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
806 return retval == EINTR ? EAGAIN : retval;
809 nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
810 if (dump->seq != nlmsghdr->nlmsg_seq) {
811 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
812 nlmsghdr->nlmsg_seq, dump->seq);
816 if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
817 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
819 return retval && retval != EAGAIN ? retval : EPROTO;
825 /* Attempts to retrieve another reply from 'dump', which must have been
826 * initialized with nl_dump_start().
828 * If successful, returns true and points 'reply->data' and 'reply->size' to
829 * the message that was retrieved. The caller must not modify 'reply' (because
830 * it points into the middle of a larger buffer).
832 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
833 * to 0. Failure might indicate an actual error or merely the end of replies.
834 * An error status for the entire dump operation is provided when it is
835 * completed by calling nl_dump_done().
838 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
840 struct nlmsghdr *nlmsghdr;
848 while (!dump->buffer.size) {
849 int retval = nl_dump_recv(dump);
851 ofpbuf_clear(&dump->buffer);
852 if (retval != EAGAIN) {
853 dump->status = retval;
859 nlmsghdr = nl_msg_next(&dump->buffer, reply);
861 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
862 dump->status = EPROTO;
864 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
872 /* Completes Netlink dump operation 'dump', which must have been initialized
873 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
874 * otherwise a positive errno value describing the problem. */
876 nl_dump_done(struct nl_dump *dump)
878 /* Drain any remaining messages that the client didn't read. Otherwise the
879 * kernel will continue to queue them up and waste buffer space. */
880 while (!dump->status) {
882 if (!nl_dump_next(dump, &reply)) {
883 ovs_assert(dump->status);
888 if (dump->sock->dump) {
889 dump->sock->dump = NULL;
891 nl_sock_destroy(dump->sock);
894 ofpbuf_uninit(&dump->buffer);
895 return dump->status == EOF ? 0 : dump->status;
898 /* Causes poll_block() to wake up when any of the specified 'events' (which is
899 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
901 nl_sock_wait(const struct nl_sock *sock, short int events)
903 poll_fd_wait(sock->fd, events);
906 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
907 * that can't use nl_sock_wait().
909 * It's a little tricky to use the returned fd correctly, because nl_sock does
910 * "copy on write" to allow a single nl_sock to be used for notifications,
911 * transactions, and dumps. If 'sock' is used only for notifications and
912 * transactions (and never for dump) then the usage is safe. */
914 nl_sock_fd(const struct nl_sock *sock)
919 /* Returns the PID associated with this socket. */
921 nl_sock_pid(const struct nl_sock *sock)
929 struct hmap_node hmap_node;
934 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
936 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
937 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
938 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
941 static struct genl_family *
942 find_genl_family_by_id(uint16_t id)
944 struct genl_family *family;
946 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
948 if (family->id == id) {
956 define_genl_family(uint16_t id, const char *name)
958 struct genl_family *family = find_genl_family_by_id(id);
961 if (!strcmp(family->name, name)) {
966 family = xmalloc(sizeof *family);
968 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
970 family->name = xstrdup(name);
974 genl_family_to_name(uint16_t id)
976 if (id == GENL_ID_CTRL) {
979 struct genl_family *family = find_genl_family_by_id(id);
980 return family ? family->name : "unknown";
985 do_lookup_genl_family(const char *name, struct nlattr **attrs,
986 struct ofpbuf **replyp)
988 struct nl_sock *sock;
989 struct ofpbuf request, *reply;
993 error = nl_sock_create(NETLINK_GENERIC, &sock);
998 ofpbuf_init(&request, 0);
999 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
1000 CTRL_CMD_GETFAMILY, 1);
1001 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
1002 error = nl_sock_transact(sock, &request, &reply);
1003 ofpbuf_uninit(&request);
1005 nl_sock_destroy(sock);
1009 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1010 family_policy, attrs, ARRAY_SIZE(family_policy))
1011 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
1012 nl_sock_destroy(sock);
1013 ofpbuf_delete(reply);
1017 nl_sock_destroy(sock);
1022 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
1023 * When successful, writes its result to 'multicast_group' and returns 0.
1024 * Otherwise, clears 'multicast_group' and returns a positive error code.
1026 * Some kernels do not support looking up a multicast group with this function.
1027 * In this case, 'multicast_group' will be populated with 'fallback'. */
1029 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
1030 unsigned int *multicast_group, unsigned int fallback)
1032 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
1033 const struct nlattr *mc;
1034 struct ofpbuf *reply;
1038 *multicast_group = 0;
1039 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1044 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1045 *multicast_group = fallback;
1046 VLOG_WARN("%s-%s: has no multicast group, using fallback %d",
1047 family_name, group_name, *multicast_group);
1052 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1053 static const struct nl_policy mc_policy[] = {
1054 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1055 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1058 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1059 const char *mc_name;
1061 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1066 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1067 if (!strcmp(group_name, mc_name)) {
1069 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1077 ofpbuf_delete(reply);
1081 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1082 * number and stores it in '*number'. If successful, returns 0 and the caller
1083 * may use '*number' as the family number. On failure, returns a positive
1084 * errno value and '*number' caches the errno value. */
1086 nl_lookup_genl_family(const char *name, int *number)
1089 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1090 struct ofpbuf *reply;
1093 error = do_lookup_genl_family(name, attrs, &reply);
1095 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1096 define_genl_family(*number, name);
1100 ofpbuf_delete(reply);
1102 ovs_assert(*number != 0);
1104 return *number > 0 ? 0 : -*number;
1108 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1110 uint32_t seq = sock->next_seq;
1112 sock->next_seq += n;
1114 /* Make it impossible for the next request for sequence numbers to wrap
1115 * around to 0. Start over with 1 to avoid ever using a sequence number of
1116 * 0, because the kernel uses sequence number 0 for notifications. */
1117 if (sock->next_seq >= UINT32_MAX / 2) {
1125 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1131 static const struct nlmsg_flag flags[] = {
1132 { NLM_F_REQUEST, "REQUEST" },
1133 { NLM_F_MULTI, "MULTI" },
1134 { NLM_F_ACK, "ACK" },
1135 { NLM_F_ECHO, "ECHO" },
1136 { NLM_F_DUMP, "DUMP" },
1137 { NLM_F_ROOT, "ROOT" },
1138 { NLM_F_MATCH, "MATCH" },
1139 { NLM_F_ATOMIC, "ATOMIC" },
1141 const struct nlmsg_flag *flag;
1142 uint16_t flags_left;
1144 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1145 h->nlmsg_len, h->nlmsg_type);
1146 if (h->nlmsg_type == NLMSG_NOOP) {
1147 ds_put_cstr(ds, "(no-op)");
1148 } else if (h->nlmsg_type == NLMSG_ERROR) {
1149 ds_put_cstr(ds, "(error)");
1150 } else if (h->nlmsg_type == NLMSG_DONE) {
1151 ds_put_cstr(ds, "(done)");
1152 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1153 ds_put_cstr(ds, "(overrun)");
1154 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1155 ds_put_cstr(ds, "(reserved)");
1156 } else if (protocol == NETLINK_GENERIC) {
1157 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1159 ds_put_cstr(ds, "(family-defined)");
1161 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1162 flags_left = h->nlmsg_flags;
1163 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1164 if ((flags_left & flag->bits) == flag->bits) {
1165 ds_put_format(ds, "[%s]", flag->name);
1166 flags_left &= ~flag->bits;
1170 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1172 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1173 h->nlmsg_seq, h->nlmsg_pid);
1177 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1179 struct ds ds = DS_EMPTY_INITIALIZER;
1180 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1182 nlmsghdr_to_string(h, protocol, &ds);
1183 if (h->nlmsg_type == NLMSG_ERROR) {
1184 const struct nlmsgerr *e;
1185 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1186 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1188 ds_put_format(&ds, " error(%d", e->error);
1190 ds_put_format(&ds, "(%s)", strerror(-e->error));
1192 ds_put_cstr(&ds, ", in-reply-to(");
1193 nlmsghdr_to_string(&e->msg, protocol, &ds);
1194 ds_put_cstr(&ds, "))");
1196 ds_put_cstr(&ds, " error(truncated)");
1198 } else if (h->nlmsg_type == NLMSG_DONE) {
1199 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1201 ds_put_format(&ds, " done(%d", *error);
1203 ds_put_format(&ds, "(%s)", strerror(-*error));
1205 ds_put_cstr(&ds, ")");
1207 ds_put_cstr(&ds, " done(truncated)");
1209 } else if (protocol == NETLINK_GENERIC) {
1210 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1212 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1213 genl->cmd, genl->version);
1217 ds_put_cstr(&ds, "nl(truncated)");
1223 log_nlmsg(const char *function, int error,
1224 const void *message, size_t size, int protocol)
1226 struct ofpbuf buffer;
1229 if (!VLOG_IS_DBG_ENABLED()) {
1233 ofpbuf_use_const(&buffer, message, size);
1234 nlmsg = nlmsg_to_string(&buffer, protocol);
1235 VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);