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"
23 #include <sys/types.h>
27 #include "dynamic-string.h"
31 #include "netlink-protocol.h"
33 #include "poll-loop.h"
34 #include "socket-util.h"
39 VLOG_DEFINE_THIS_MODULE(netlink_socket);
41 COVERAGE_DEFINE(netlink_overflow);
42 COVERAGE_DEFINE(netlink_received);
43 COVERAGE_DEFINE(netlink_recv_jumbo);
44 COVERAGE_DEFINE(netlink_send);
45 COVERAGE_DEFINE(netlink_sent);
47 /* Linux header file confusion causes this to be undefined. */
49 #define SOL_NETLINK 270
52 /* A single (bad) Netlink message can in theory dump out many, many log
53 * messages, so the burst size is set quite high here to avoid missing useful
54 * information. Also, at high logging levels we log *all* Netlink messages. */
55 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
57 static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
58 static void log_nlmsg(const char *function, int error,
59 const void *message, size_t size, int protocol);
61 /* Netlink sockets. */
70 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
73 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
74 * of iovecs on the stack. */
77 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
78 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
80 * Initialized by nl_sock_create(). */
83 static int nl_sock_cow__(struct nl_sock *);
85 /* Creates a new netlink socket for the given netlink 'protocol'
86 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
87 * new socket if successful, otherwise returns a positive errno value. */
89 nl_sock_create(int protocol, struct nl_sock **sockp)
92 struct sockaddr_nl local, remote;
98 int save_errno = errno;
101 max_iovs = sysconf(_SC_UIO_MAXIOV);
102 if (max_iovs < _XOPEN_IOV_MAX) {
103 if (max_iovs == -1 && errno) {
104 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", strerror(errno));
106 max_iovs = _XOPEN_IOV_MAX;
107 } else if (max_iovs > MAX_IOVS) {
115 sock = malloc(sizeof *sock);
120 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
122 VLOG_ERR("fcntl: %s", strerror(errno));
125 sock->protocol = protocol;
129 rcvbuf = 1024 * 1024;
130 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
131 &rcvbuf, sizeof rcvbuf)) {
132 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
133 * Warn only if the failure is therefore unexpected. */
134 if (errno != EPERM || !getuid()) {
135 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
136 "(%s)", rcvbuf, strerror(errno));
140 retval = get_socket_rcvbuf(sock->fd);
145 sock->rcvbuf = retval;
147 /* Connect to kernel (pid 0) as remote address. */
148 memset(&remote, 0, sizeof remote);
149 remote.nl_family = AF_NETLINK;
151 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
152 VLOG_ERR("connect(0): %s", strerror(errno));
156 /* Obtain pid assigned by kernel. */
157 local_size = sizeof local;
158 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
159 VLOG_ERR("getsockname: %s", strerror(errno));
162 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
163 VLOG_ERR("getsockname returned bad Netlink name");
167 sock->pid = local.nl_pid;
186 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
187 * sets '*sockp' to the new socket if successful, otherwise returns a positive
190 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
192 return nl_sock_create(src->protocol, sockp);
195 /* Destroys netlink socket 'sock'. */
197 nl_sock_destroy(struct nl_sock *sock)
209 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
210 * successful, otherwise a positive errno value.
212 * A socket that is subscribed to a multicast group that receives asynchronous
213 * notifications must not be used for Netlink transactions or dumps, because
214 * transactions and dumps can cause notifications to be lost.
216 * Multicast group numbers are always positive.
218 * It is not an error to attempt to join a multicast group to which a socket
219 * already belongs. */
221 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
223 int error = nl_sock_cow__(sock);
227 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
228 &multicast_group, sizeof multicast_group) < 0) {
229 VLOG_WARN("could not join multicast group %u (%s)",
230 multicast_group, strerror(errno));
236 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
237 * successful, otherwise a positive errno value.
239 * Multicast group numbers are always positive.
241 * It is not an error to attempt to leave a multicast group to which a socket
244 * On success, reading from 'sock' will still return any messages that were
245 * received on 'multicast_group' before the group was left. */
247 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
250 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
251 &multicast_group, sizeof multicast_group) < 0) {
252 VLOG_WARN("could not leave multicast group %u (%s)",
253 multicast_group, strerror(errno));
260 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
261 uint32_t nlmsg_seq, bool wait)
263 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
266 nlmsg->nlmsg_len = msg->size;
267 nlmsg->nlmsg_seq = nlmsg_seq;
268 nlmsg->nlmsg_pid = sock->pid;
271 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
272 error = retval < 0 ? errno : 0;
273 } while (error == EINTR);
274 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
276 COVERAGE_INC(netlink_sent);
281 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
282 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
283 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
284 * sequence number, before the message is sent.
286 * Returns 0 if successful, otherwise a positive errno value. If
287 * 'wait' is true, then the send will wait until buffer space is ready;
288 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
290 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
292 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
295 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
296 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
297 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
298 * 'nlmsg_seq', before the message is sent.
300 * Returns 0 if successful, otherwise a positive errno value. If
301 * 'wait' is true, then the send will wait until buffer space is ready;
302 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
304 * This function is suitable for sending a reply to a request that was received
305 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
307 nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
308 uint32_t nlmsg_seq, bool wait)
310 int error = nl_sock_cow__(sock);
314 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
317 /* This stress option is useful for testing that OVS properly tolerates
318 * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
319 * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
320 * reply to a request. They can also occur if messages arrive on a multicast
321 * channel faster than OVS can process them. */
323 netlink_overflow, "simulate netlink socket receive buffer overflow",
327 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
329 /* We can't accurately predict the size of the data to be received. The
330 * caller is supposed to have allocated enough space in 'buf' to handle the
331 * "typical" case. To handle exceptions, we make available enough space in
332 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
333 * figure since that's the maximum length of a Netlink attribute). */
334 struct nlmsghdr *nlmsghdr;
340 assert(buf->allocated >= sizeof *nlmsghdr);
343 iov[0].iov_base = buf->base;
344 iov[0].iov_len = buf->allocated;
345 iov[1].iov_base = tail;
346 iov[1].iov_len = sizeof tail;
348 memset(&msg, 0, sizeof msg);
353 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
354 } while (retval < 0 && errno == EINTR);
358 if (error == ENOBUFS) {
359 /* Socket receive buffer overflow dropped one or more messages that
360 * the kernel tried to send to us. */
361 COVERAGE_INC(netlink_overflow);
366 if (msg.msg_flags & MSG_TRUNC) {
367 VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)",
372 nlmsghdr = buf->data;
373 if (retval < sizeof *nlmsghdr
374 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
375 || nlmsghdr->nlmsg_len > retval) {
376 VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)",
377 retval, sizeof *nlmsghdr);
381 if (STRESS(netlink_overflow)) {
385 buf->size = MIN(retval, buf->allocated);
386 if (retval > buf->allocated) {
387 COVERAGE_INC(netlink_recv_jumbo);
388 ofpbuf_put(buf, tail, retval - buf->allocated);
391 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
392 COVERAGE_INC(netlink_received);
397 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
398 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
399 * EAGAIN if the 'sock' receive buffer is empty.
401 * The caller must have initialized 'buf' with an allocation of at least
402 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
403 * space for a "typical" message.
405 * On success, returns 0 and replaces 'buf''s previous content by the received
406 * message. This function expands 'buf''s allocated memory, as necessary, to
407 * hold the actual size of the received message.
409 * On failure, returns a positive errno value and clears 'buf' to zero length.
410 * 'buf' retains its previous memory allocation.
412 * Regardless of success or failure, this function resets 'buf''s headroom to
415 nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
417 int error = nl_sock_cow__(sock);
421 return nl_sock_recv__(sock, buf, wait);
425 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
430 for (i = 0; i < n; i++) {
431 struct nl_transaction *txn = transactions[i];
435 ofpbuf_clear(txn->reply);
441 nl_sock_transact_multiple__(struct nl_sock *sock,
442 struct nl_transaction **transactions, size_t n,
445 uint64_t tmp_reply_stub[1024 / 8];
446 struct nl_transaction tmp_txn;
447 struct ofpbuf tmp_reply;
450 struct iovec iovs[MAX_IOVS];
455 base_seq = nl_sock_allocate_seq(sock, n);
457 for (i = 0; i < n; i++) {
458 struct nl_transaction *txn = transactions[i];
459 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
461 nlmsg->nlmsg_len = txn->request->size;
462 nlmsg->nlmsg_seq = base_seq + i;
463 nlmsg->nlmsg_pid = sock->pid;
465 iovs[i].iov_base = txn->request->data;
466 iovs[i].iov_len = txn->request->size;
469 memset(&msg, 0, sizeof msg);
473 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
474 } while (error == EINTR);
476 for (i = 0; i < n; i++) {
477 struct nl_transaction *txn = transactions[i];
479 log_nlmsg(__func__, error, txn->request->data, txn->request->size,
483 COVERAGE_ADD(netlink_sent, n);
490 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
491 tmp_txn.request = NULL;
492 tmp_txn.reply = &tmp_reply;
495 struct nl_transaction *buf_txn, *txn;
498 /* Find a transaction whose buffer we can use for receiving a reply.
499 * If no such transaction is left, use tmp_txn. */
501 for (i = 0; i < n; i++) {
502 if (transactions[i]->reply) {
503 buf_txn = transactions[i];
508 /* Receive a reply. */
509 error = nl_sock_recv__(sock, buf_txn->reply, false);
511 if (error == EAGAIN) {
512 nl_sock_record_errors__(transactions, n, 0);
519 /* Match the reply up with a transaction. */
520 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
521 if (seq < base_seq || seq >= base_seq + n) {
522 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
526 txn = transactions[i];
528 /* Fill in the results for 'txn'. */
529 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
531 ofpbuf_clear(txn->reply);
534 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
535 error, strerror(txn->error));
539 if (txn->reply && txn != buf_txn) {
541 struct ofpbuf *reply = buf_txn->reply;
542 buf_txn->reply = txn->reply;
547 /* Fill in the results for transactions before 'txn'. (We have to do
548 * this after the results for 'txn' itself because of the buffer swap
550 nl_sock_record_errors__(transactions, i, 0);
554 transactions += i + 1;
558 ofpbuf_uninit(&tmp_reply);
563 /* Sends the 'request' member of the 'n' transactions in 'transactions' on
564 * 'sock', in order, and receives responses to all of them. Fills in the
565 * 'error' member of each transaction with 0 if it was successful, otherwise
566 * with a positive errno value. If 'reply' is nonnull, then it will be filled
567 * with the reply if the message receives a detailed reply. In other cases,
568 * i.e. where the request failed or had no reply beyond an indication of
569 * success, 'reply' will be cleared if it is nonnull.
571 * The caller is responsible for destroying each request and reply, and the
572 * transactions array itself.
574 * Before sending each message, this function will finalize nlmsg_len in each
575 * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
576 * initialize nlmsg_seq.
578 * Bare Netlink is an unreliable transport protocol. This function layers
579 * reliable delivery and reply semantics on top of bare Netlink. See
580 * nl_sock_transact() for some caveats.
583 nl_sock_transact_multiple(struct nl_sock *sock,
584 struct nl_transaction **transactions, size_t n)
593 error = nl_sock_cow__(sock);
595 nl_sock_record_errors__(transactions, n, error);
599 /* In theory, every request could have a 64 kB reply. But the default and
600 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
601 * be a bit below 128 kB, so that would only allow a single message in a
602 * "batch". So we assume that replies average (at most) 4 kB, which allows
603 * a good deal of batching.
605 * In practice, most of the requests that we batch either have no reply at
606 * all or a brief reply. */
607 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
608 max_batch_count = MIN(max_batch_count, max_iovs);
614 /* Batch up to 'max_batch_count' transactions. But cap it at about a
615 * page of requests total because big skbuffs are expensive to
616 * allocate in the kernel. */
617 #if defined(PAGESIZE)
618 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
620 enum { MAX_BATCH_BYTES = 4096 - 512 };
622 bytes = transactions[0]->request->size;
623 for (count = 1; count < n && count < max_batch_count; count++) {
624 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
627 bytes += transactions[count]->request->size;
630 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
631 transactions += done;
634 if (error == ENOBUFS) {
635 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
637 VLOG_ERR_RL(&rl, "transaction error (%s)", strerror(error));
638 nl_sock_record_errors__(transactions, n, error);
643 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
644 * successful, returns 0. On failure, returns a positive errno value.
646 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
647 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
648 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
649 * reply, if any, is discarded.
651 * Before the message is sent, nlmsg_len in 'request' will be finalized to
652 * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
653 * be initialized, NLM_F_ACK will be set in nlmsg_flags.
655 * The caller is responsible for destroying 'request'.
657 * Bare Netlink is an unreliable transport protocol. This function layers
658 * reliable delivery and reply semantics on top of bare Netlink.
660 * In Netlink, sending a request to the kernel is reliable enough, because the
661 * kernel will tell us if the message cannot be queued (and we will in that
662 * case put it on the transmit queue and wait until it can be delivered).
664 * Receiving the reply is the real problem: if the socket buffer is full when
665 * the kernel tries to send the reply, the reply will be dropped. However, the
666 * kernel sets a flag that a reply has been dropped. The next call to recv
667 * then returns ENOBUFS. We can then re-send the request.
671 * 1. Netlink depends on sequence numbers to match up requests and
672 * replies. The sender of a request supplies a sequence number, and
673 * the reply echos back that sequence number.
675 * This is fine, but (1) some kernel netlink implementations are
676 * broken, in that they fail to echo sequence numbers and (2) this
677 * function will drop packets with non-matching sequence numbers, so
678 * that only a single request can be usefully transacted at a time.
680 * 2. Resending the request causes it to be re-executed, so the request
681 * needs to be idempotent.
684 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
685 struct ofpbuf **replyp)
687 struct nl_transaction *transactionp;
688 struct nl_transaction transaction;
690 transaction.request = CONST_CAST(struct ofpbuf *, request);
691 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
692 transactionp = &transaction;
694 nl_sock_transact_multiple(sock, &transactionp, 1);
697 if (transaction.error) {
698 ofpbuf_delete(transaction.reply);
701 *replyp = transaction.reply;
705 return transaction.error;
708 /* Drain all the messages currently in 'sock''s receive queue. */
710 nl_sock_drain(struct nl_sock *sock)
712 int error = nl_sock_cow__(sock);
716 return drain_rcvbuf(sock->fd);
719 /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
720 * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
721 * old fd over to the dump. */
723 nl_sock_cow__(struct nl_sock *sock)
725 struct nl_sock *copy;
734 error = nl_sock_clone(sock, ©);
744 sock->pid = copy->pid;
747 sock->dump->sock = copy;
753 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
754 * 'sock', and initializes 'dump' to reflect the state of the operation.
756 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
757 * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
758 * NLM_F_ACK will be set in nlmsg_flags.
760 * This Netlink socket library is designed to ensure that the dump is reliable
761 * and that it will not interfere with other operations on 'sock', including
762 * destroying or sending and receiving messages on 'sock'. One corner case is
765 * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
766 * whose response has not yet been received (e.g. with nl_sock_recv()).
767 * This is unusual: usually nl_sock_transact() is used to send a message
768 * and receive its reply all in one go.
770 * This function provides no status indication. An error status for the entire
771 * dump operation is provided when it is completed by calling nl_dump_done().
773 * The caller is responsible for destroying 'request'.
775 * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
779 nl_dump_start(struct nl_dump *dump,
780 struct nl_sock *sock, const struct ofpbuf *request)
782 ofpbuf_init(&dump->buffer, 4096);
784 /* 'sock' already has an ongoing dump. Clone the socket because
785 * Netlink only allows one dump at a time. */
786 dump->status = nl_sock_clone(sock, &dump->sock);
796 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
797 dump->status = nl_sock_send__(sock, request, nl_sock_allocate_seq(sock, 1),
799 dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
802 /* Helper function for nl_dump_next(). */
804 nl_dump_recv(struct nl_dump *dump)
806 struct nlmsghdr *nlmsghdr;
809 retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
811 return retval == EINTR ? EAGAIN : retval;
814 nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
815 if (dump->seq != nlmsghdr->nlmsg_seq) {
816 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
817 nlmsghdr->nlmsg_seq, dump->seq);
821 if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
822 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
824 return retval && retval != EAGAIN ? retval : EPROTO;
830 /* Attempts to retrieve another reply from 'dump', which must have been
831 * initialized with nl_dump_start().
833 * If successful, returns true and points 'reply->data' and 'reply->size' to
834 * the message that was retrieved. The caller must not modify 'reply' (because
835 * it points into the middle of a larger buffer).
837 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
838 * to 0. Failure might indicate an actual error or merely the end of replies.
839 * An error status for the entire dump operation is provided when it is
840 * completed by calling nl_dump_done().
843 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
845 struct nlmsghdr *nlmsghdr;
853 while (!dump->buffer.size) {
854 int retval = nl_dump_recv(dump);
856 ofpbuf_clear(&dump->buffer);
857 if (retval != EAGAIN) {
858 dump->status = retval;
864 nlmsghdr = nl_msg_next(&dump->buffer, reply);
866 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
867 dump->status = EPROTO;
869 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
877 /* Completes Netlink dump operation 'dump', which must have been initialized
878 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
879 * otherwise a positive errno value describing the problem. */
881 nl_dump_done(struct nl_dump *dump)
883 /* Drain any remaining messages that the client didn't read. Otherwise the
884 * kernel will continue to queue them up and waste buffer space. */
885 while (!dump->status) {
887 if (!nl_dump_next(dump, &reply)) {
888 assert(dump->status);
893 if (dump->sock->dump) {
894 dump->sock->dump = NULL;
896 nl_sock_destroy(dump->sock);
899 ofpbuf_uninit(&dump->buffer);
900 return dump->status == EOF ? 0 : dump->status;
903 /* Causes poll_block() to wake up when any of the specified 'events' (which is
904 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
906 nl_sock_wait(const struct nl_sock *sock, short int events)
908 poll_fd_wait(sock->fd, events);
911 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
912 * that can't use nl_sock_wait().
914 * It's a little tricky to use the returned fd correctly, because nl_sock does
915 * "copy on write" to allow a single nl_sock to be used for notifications,
916 * transactions, and dumps. If 'sock' is used only for notifications and
917 * transactions (and never for dump) then the usage is safe. */
919 nl_sock_fd(const struct nl_sock *sock)
924 /* Returns the PID associated with this socket. */
926 nl_sock_pid(const struct nl_sock *sock)
934 struct hmap_node hmap_node;
939 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
941 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
942 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
943 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
946 static struct genl_family *
947 find_genl_family_by_id(uint16_t id)
949 struct genl_family *family;
951 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
953 if (family->id == id) {
961 define_genl_family(uint16_t id, const char *name)
963 struct genl_family *family = find_genl_family_by_id(id);
966 if (!strcmp(family->name, name)) {
971 family = xmalloc(sizeof *family);
973 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
975 family->name = xstrdup(name);
979 genl_family_to_name(uint16_t id)
981 if (id == GENL_ID_CTRL) {
984 struct genl_family *family = find_genl_family_by_id(id);
985 return family ? family->name : "unknown";
990 do_lookup_genl_family(const char *name, struct nlattr **attrs,
991 struct ofpbuf **replyp)
993 struct nl_sock *sock;
994 struct ofpbuf request, *reply;
998 error = nl_sock_create(NETLINK_GENERIC, &sock);
1003 ofpbuf_init(&request, 0);
1004 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
1005 CTRL_CMD_GETFAMILY, 1);
1006 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
1007 error = nl_sock_transact(sock, &request, &reply);
1008 ofpbuf_uninit(&request);
1010 nl_sock_destroy(sock);
1014 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1015 family_policy, attrs, ARRAY_SIZE(family_policy))
1016 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
1017 nl_sock_destroy(sock);
1018 ofpbuf_delete(reply);
1022 nl_sock_destroy(sock);
1027 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
1028 * When successful, writes its result to 'multicast_group' and returns 0.
1029 * Otherwise, clears 'multicast_group' and returns a positive error code.
1031 * Some kernels do not support looking up a multicast group with this function.
1032 * In this case, 'multicast_group' will be populated with 'fallback'. */
1034 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
1035 unsigned int *multicast_group, unsigned int fallback)
1037 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
1038 const struct nlattr *mc;
1039 struct ofpbuf *reply;
1043 *multicast_group = 0;
1044 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1049 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1050 *multicast_group = fallback;
1051 VLOG_WARN("%s-%s: has no multicast group, using fallback %d",
1052 family_name, group_name, *multicast_group);
1057 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1058 static const struct nl_policy mc_policy[] = {
1059 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1060 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1063 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1064 const char *mc_name;
1066 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1071 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1072 if (!strcmp(group_name, mc_name)) {
1074 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1082 ofpbuf_delete(reply);
1086 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1087 * number and stores it in '*number'. If successful, returns 0 and the caller
1088 * may use '*number' as the family number. On failure, returns a positive
1089 * errno value and '*number' caches the errno value. */
1091 nl_lookup_genl_family(const char *name, int *number)
1094 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1095 struct ofpbuf *reply;
1098 error = do_lookup_genl_family(name, attrs, &reply);
1100 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1101 define_genl_family(*number, name);
1105 ofpbuf_delete(reply);
1107 assert(*number != 0);
1109 return *number > 0 ? 0 : -*number;
1113 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1115 uint32_t seq = sock->next_seq;
1117 sock->next_seq += n;
1119 /* Make it impossible for the next request for sequence numbers to wrap
1120 * around to 0. Start over with 1 to avoid ever using a sequence number of
1121 * 0, because the kernel uses sequence number 0 for notifications. */
1122 if (sock->next_seq >= UINT32_MAX / 2) {
1130 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1136 static const struct nlmsg_flag flags[] = {
1137 { NLM_F_REQUEST, "REQUEST" },
1138 { NLM_F_MULTI, "MULTI" },
1139 { NLM_F_ACK, "ACK" },
1140 { NLM_F_ECHO, "ECHO" },
1141 { NLM_F_DUMP, "DUMP" },
1142 { NLM_F_ROOT, "ROOT" },
1143 { NLM_F_MATCH, "MATCH" },
1144 { NLM_F_ATOMIC, "ATOMIC" },
1146 const struct nlmsg_flag *flag;
1147 uint16_t flags_left;
1149 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1150 h->nlmsg_len, h->nlmsg_type);
1151 if (h->nlmsg_type == NLMSG_NOOP) {
1152 ds_put_cstr(ds, "(no-op)");
1153 } else if (h->nlmsg_type == NLMSG_ERROR) {
1154 ds_put_cstr(ds, "(error)");
1155 } else if (h->nlmsg_type == NLMSG_DONE) {
1156 ds_put_cstr(ds, "(done)");
1157 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1158 ds_put_cstr(ds, "(overrun)");
1159 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1160 ds_put_cstr(ds, "(reserved)");
1161 } else if (protocol == NETLINK_GENERIC) {
1162 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1164 ds_put_cstr(ds, "(family-defined)");
1166 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1167 flags_left = h->nlmsg_flags;
1168 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1169 if ((flags_left & flag->bits) == flag->bits) {
1170 ds_put_format(ds, "[%s]", flag->name);
1171 flags_left &= ~flag->bits;
1175 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1177 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1178 h->nlmsg_seq, h->nlmsg_pid);
1182 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1184 struct ds ds = DS_EMPTY_INITIALIZER;
1185 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1187 nlmsghdr_to_string(h, protocol, &ds);
1188 if (h->nlmsg_type == NLMSG_ERROR) {
1189 const struct nlmsgerr *e;
1190 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1191 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1193 ds_put_format(&ds, " error(%d", e->error);
1195 ds_put_format(&ds, "(%s)", strerror(-e->error));
1197 ds_put_cstr(&ds, ", in-reply-to(");
1198 nlmsghdr_to_string(&e->msg, protocol, &ds);
1199 ds_put_cstr(&ds, "))");
1201 ds_put_cstr(&ds, " error(truncated)");
1203 } else if (h->nlmsg_type == NLMSG_DONE) {
1204 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1206 ds_put_format(&ds, " done(%d", *error);
1208 ds_put_format(&ds, "(%s)", strerror(-*error));
1210 ds_put_cstr(&ds, ")");
1212 ds_put_cstr(&ds, " done(truncated)");
1214 } else if (protocol == NETLINK_GENERIC) {
1215 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1217 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1218 genl->cmd, genl->version);
1222 ds_put_cstr(&ds, "nl(truncated)");
1228 log_nlmsg(const char *function, int error,
1229 const void *message, size_t size, int protocol)
1231 struct ofpbuf buffer;
1234 if (!VLOG_IS_DBG_ENABLED()) {
1238 ofpbuf_use_const(&buffer, message, size);
1239 nlmsg = nlmsg_to_string(&buffer, protocol);
1240 VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);