2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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 "ovs-thread.h"
33 #include "poll-loop.h"
35 #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_sent);
46 /* Linux header file confusion causes this to be undefined. */
48 #define SOL_NETLINK 270
52 static struct ovs_mutex portid_mutex = OVS_MUTEX_INITIALIZER;
53 static uint32_t g_last_portid = 0;
55 /* Port IDs must be unique! */
58 OVS_GUARDED_BY(portid_mutex)
65 set_sock_pid_in_kernel(HANDLE handle, uint32_t pid)
66 OVS_GUARDED_BY(portid_mutex)
68 struct nlmsghdr msg = { 0 };
70 msg.nlmsg_len = sizeof(struct nlmsghdr);
71 msg.nlmsg_type = 80; /* target = set file pid */
76 WriteFile(handle, &msg, sizeof(struct nlmsghdr), NULL, NULL);
80 /* A single (bad) Netlink message can in theory dump out many, many log
81 * messages, so the burst size is set quite high here to avoid missing useful
82 * information. Also, at high logging levels we log *all* Netlink messages. */
83 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
85 static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
86 static void log_nlmsg(const char *function, int error,
87 const void *message, size_t size, int protocol);
89 /* Netlink sockets. */
100 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
103 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
104 * of iovecs on the stack. */
107 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
108 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
110 * Initialized by nl_sock_create(). */
113 static int nl_pool_alloc(int protocol, struct nl_sock **sockp);
114 static void nl_pool_release(struct nl_sock *);
116 /* Creates a new netlink socket for the given netlink 'protocol'
117 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
118 * new socket if successful, otherwise returns a positive errno value. */
120 nl_sock_create(int protocol, struct nl_sock **sockp)
122 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
123 struct nl_sock *sock;
125 struct sockaddr_nl local, remote;
127 socklen_t local_size;
131 if (ovsthread_once_start(&once)) {
132 int save_errno = errno;
135 max_iovs = sysconf(_SC_UIO_MAXIOV);
136 if (max_iovs < _XOPEN_IOV_MAX) {
137 if (max_iovs == -1 && errno) {
138 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno));
140 max_iovs = _XOPEN_IOV_MAX;
141 } else if (max_iovs > MAX_IOVS) {
146 ovsthread_once_done(&once);
150 sock = xmalloc(sizeof *sock);
153 sock->handle = CreateFileA("\\\\.\\OpenVSwitchDevice",
154 GENERIC_READ | GENERIC_WRITE,
155 FILE_SHARE_READ | FILE_SHARE_WRITE,
157 FILE_ATTRIBUTE_NORMAL, NULL);
159 int last_error = GetLastError();
161 if (sock->handle == INVALID_HANDLE_VALUE) {
162 VLOG_ERR("fcntl: %s", ovs_strerror(last_error));
166 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
168 VLOG_ERR("fcntl: %s", ovs_strerror(errno));
173 sock->protocol = protocol;
176 rcvbuf = 1024 * 1024;
178 sock->rcvbuf = rcvbuf;
179 ovs_mutex_lock(&portid_mutex);
180 sock->pid = portid_next();
181 set_sock_pid_in_kernel(sock->handle, sock->pid);
182 ovs_mutex_unlock(&portid_mutex);
184 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
185 &rcvbuf, sizeof rcvbuf)) {
186 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
187 * Warn only if the failure is therefore unexpected. */
188 if (errno != EPERM) {
189 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
190 "(%s)", rcvbuf, ovs_strerror(errno));
194 retval = get_socket_rcvbuf(sock->fd);
199 sock->rcvbuf = retval;
201 /* Connect to kernel (pid 0) as remote address. */
202 memset(&remote, 0, sizeof remote);
203 remote.nl_family = AF_NETLINK;
205 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
206 VLOG_ERR("connect(0): %s", ovs_strerror(errno));
210 /* Obtain pid assigned by kernel. */
211 local_size = sizeof local;
212 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
213 VLOG_ERR("getsockname: %s", ovs_strerror(errno));
216 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
217 VLOG_ERR("getsockname returned bad Netlink name");
221 sock->pid = local.nl_pid;
235 if (sock->handle != INVALID_HANDLE_VALUE) {
236 CloseHandle(sock->handle);
247 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
248 * sets '*sockp' to the new socket if successful, otherwise returns a positive
251 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
253 return nl_sock_create(src->protocol, sockp);
256 /* Destroys netlink socket 'sock'. */
258 nl_sock_destroy(struct nl_sock *sock)
262 CloseHandle(sock->handle);
270 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
271 * successful, otherwise a positive errno value.
273 * A socket that is subscribed to a multicast group that receives asynchronous
274 * notifications must not be used for Netlink transactions or dumps, because
275 * transactions and dumps can cause notifications to be lost.
277 * Multicast group numbers are always positive.
279 * It is not an error to attempt to join a multicast group to which a socket
280 * already belongs. */
282 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
285 #define OVS_VPORT_MCGROUP_FALLBACK_ID 33
286 struct ofpbuf msg_buf;
287 struct message_multicast
290 /* if true, join; if else, leave */
292 unsigned int groupId;
295 struct message_multicast msg = { 0 };
297 msg.nlmsg_len = sizeof(struct message_multicast);
298 msg.nlmsg_type = OVS_VPORT_MCGROUP_FALLBACK_ID;
301 msg.nlmsg_pid = sock->pid;
304 msg.groupId = multicast_group;
305 msg_buf.base_ = &msg;
306 msg_buf.data_ = &msg;
307 msg_buf.size_ = msg.nlmsg_len;
309 nl_sock_send__(sock, &msg_buf, msg.nlmsg_seq, 0);
311 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
312 &multicast_group, sizeof multicast_group) < 0) {
313 VLOG_WARN("could not join multicast group %u (%s)",
314 multicast_group, ovs_strerror(errno));
321 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
322 * successful, otherwise a positive errno value.
324 * Multicast group numbers are always positive.
326 * It is not an error to attempt to leave a multicast group to which a socket
329 * On success, reading from 'sock' will still return any messages that were
330 * received on 'multicast_group' before the group was left. */
332 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
335 struct ofpbuf msg_buf;
336 struct message_multicast
339 /* if true, join; if else, leave*/
343 struct message_multicast msg = { 0 };
344 nl_msg_put_nlmsghdr(&msg, sizeof(struct message_multicast),
348 msg_buf.base_ = &msg;
349 msg_buf.data_ = &msg;
350 msg_buf.size_ = msg.nlmsg_len;
352 nl_sock_send__(sock, &msg_buf, msg.nlmsg_seq, 0);
354 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
355 &multicast_group, sizeof multicast_group) < 0) {
356 VLOG_WARN("could not leave multicast group %u (%s)",
357 multicast_group, ovs_strerror(errno));
365 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
366 uint32_t nlmsg_seq, bool wait)
368 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
371 nlmsg->nlmsg_len = ofpbuf_size(msg);
372 nlmsg->nlmsg_seq = nlmsg_seq;
373 nlmsg->nlmsg_pid = sock->pid;
378 DWORD last_error = 0;
379 result = WriteFile(sock->handle, ofpbuf_data(msg), ofpbuf_size(msg),
381 last_error = GetLastError();
382 if (last_error != ERROR_SUCCESS && !result) {
387 retval = send(sock->fd, ofpbuf_data(msg), ofpbuf_size(msg), wait ? 0 : MSG_DONTWAIT);
389 error = retval < 0 ? errno : 0;
390 } while (error == EINTR);
391 log_nlmsg(__func__, error, ofpbuf_data(msg), ofpbuf_size(msg), sock->protocol);
393 COVERAGE_INC(netlink_sent);
398 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
399 * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid
400 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
401 * sequence number, before the message is sent.
403 * Returns 0 if successful, otherwise a positive errno value. If
404 * 'wait' is true, then the send will wait until buffer space is ready;
405 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
407 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
409 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
412 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
413 * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid
414 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
415 * 'nlmsg_seq', before the message is sent.
417 * Returns 0 if successful, otherwise a positive errno value. If
418 * 'wait' is true, then the send will wait until buffer space is ready;
419 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
421 * This function is suitable for sending a reply to a request that was received
422 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
424 nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
425 uint32_t nlmsg_seq, bool wait)
427 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
431 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
433 /* We can't accurately predict the size of the data to be received. The
434 * caller is supposed to have allocated enough space in 'buf' to handle the
435 * "typical" case. To handle exceptions, we make available enough space in
436 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
437 * figure since that's the maximum length of a Netlink attribute). */
438 struct nlmsghdr *nlmsghdr;
440 #define MAX_STACK_LENGTH 81920
441 uint8_t tail[MAX_STACK_LENGTH];
450 ovs_assert(buf->allocated >= sizeof *nlmsghdr);
453 iov[0].iov_base = ofpbuf_base(buf);
454 iov[0].iov_len = buf->allocated;
455 iov[1].iov_base = tail;
456 iov[1].iov_len = sizeof tail;
458 memset(&msg, 0, sizeof msg);
462 /* Receive a Netlink message from the kernel.
464 * This works around a kernel bug in which the kernel returns an error code
465 * as if it were the number of bytes read. It doesn't actually modify
466 * anything in the receive buffer in that case, so we can initialize the
467 * Netlink header with an impossible message length and then, upon success,
468 * check whether it changed. */
469 nlmsghdr = ofpbuf_base(buf);
471 nlmsghdr->nlmsg_len = UINT32_MAX;
473 boolean result = false;
474 DWORD last_error = 0;
475 result = ReadFile(sock->handle, tail, MAX_STACK_LENGTH, &retval, NULL);
476 last_error = GetLastError();
477 if (last_error != ERROR_SUCCESS && !result) {
481 ofpbuf_put(buf, tail, retval);
484 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
486 error = (retval < 0 ? errno
487 : retval == 0 ? ECONNRESET /* not possible? */
488 : nlmsghdr->nlmsg_len != UINT32_MAX ? 0
490 } while (error == EINTR);
492 if (error == ENOBUFS) {
493 /* Socket receive buffer overflow dropped one or more messages that
494 * the kernel tried to send to us. */
495 COVERAGE_INC(netlink_overflow);
500 if (msg.msg_flags & MSG_TRUNC) {
501 VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)",
506 if (retval < sizeof *nlmsghdr
507 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
508 || nlmsghdr->nlmsg_len > retval) {
509 VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE" bytes < %"PRIuSIZE")",
510 retval, sizeof *nlmsghdr);
514 ofpbuf_set_size(buf, MIN(retval, buf->allocated));
515 if (retval > buf->allocated) {
516 COVERAGE_INC(netlink_recv_jumbo);
517 ofpbuf_put(buf, tail, retval - buf->allocated);
521 log_nlmsg(__func__, 0, ofpbuf_data(buf), ofpbuf_size(buf), sock->protocol);
522 COVERAGE_INC(netlink_received);
527 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
528 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
529 * EAGAIN if the 'sock' receive buffer is empty.
531 * The caller must have initialized 'buf' with an allocation of at least
532 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
533 * space for a "typical" message.
535 * On success, returns 0 and replaces 'buf''s previous content by the received
536 * message. This function expands 'buf''s allocated memory, as necessary, to
537 * hold the actual size of the received message.
539 * On failure, returns a positive errno value and clears 'buf' to zero length.
540 * 'buf' retains its previous memory allocation.
542 * Regardless of success or failure, this function resets 'buf''s headroom to
545 nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
547 return nl_sock_recv__(sock, buf, wait);
551 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
556 for (i = 0; i < n; i++) {
557 struct nl_transaction *txn = transactions[i];
561 ofpbuf_clear(txn->reply);
567 nl_sock_transact_multiple__(struct nl_sock *sock,
568 struct nl_transaction **transactions, size_t n,
571 uint64_t tmp_reply_stub[1024 / 8];
572 struct nl_transaction tmp_txn;
573 struct ofpbuf tmp_reply;
576 struct iovec iovs[MAX_IOVS];
581 base_seq = nl_sock_allocate_seq(sock, n);
583 for (i = 0; i < n; i++) {
584 struct nl_transaction *txn = transactions[i];
585 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
587 nlmsg->nlmsg_len = ofpbuf_size(txn->request);
588 nlmsg->nlmsg_seq = base_seq + i;
589 nlmsg->nlmsg_pid = sock->pid;
591 iovs[i].iov_base = ofpbuf_data(txn->request);
592 iovs[i].iov_len = ofpbuf_size(txn->request);
595 memset(&msg, 0, sizeof msg);
600 DWORD last_error = 0;
602 for (i = 0; i < n; i++) {
603 result = WriteFile((HANDLE)sock->handle, iovs[i].iov_base, iovs[i].iov_len,
605 last_error = GetLastError();
606 if (last_error != ERROR_SUCCESS && !result) {
614 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
616 } while (error == EINTR);
618 for (i = 0; i < n; i++) {
619 struct nl_transaction *txn = transactions[i];
621 log_nlmsg(__func__, error, ofpbuf_data(txn->request), ofpbuf_size(txn->request),
625 COVERAGE_ADD(netlink_sent, n);
632 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
633 tmp_txn.request = NULL;
634 tmp_txn.reply = &tmp_reply;
637 struct nl_transaction *buf_txn, *txn;
640 /* Find a transaction whose buffer we can use for receiving a reply.
641 * If no such transaction is left, use tmp_txn. */
643 for (i = 0; i < n; i++) {
644 if (transactions[i]->reply) {
645 buf_txn = transactions[i];
650 /* Receive a reply. */
651 error = nl_sock_recv__(sock, buf_txn->reply, false);
653 if (error == EAGAIN) {
654 nl_sock_record_errors__(transactions, n, 0);
661 /* Match the reply up with a transaction. */
662 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
663 if (seq < base_seq || seq >= base_seq + n) {
664 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
668 txn = transactions[i];
670 /* Fill in the results for 'txn'. */
671 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
673 ofpbuf_clear(txn->reply);
676 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
677 error, ovs_strerror(txn->error));
681 if (txn->reply && txn != buf_txn) {
683 struct ofpbuf *reply = buf_txn->reply;
684 buf_txn->reply = txn->reply;
689 /* Fill in the results for transactions before 'txn'. (We have to do
690 * this after the results for 'txn' itself because of the buffer swap
692 nl_sock_record_errors__(transactions, i, 0);
696 transactions += i + 1;
700 ofpbuf_uninit(&tmp_reply);
706 nl_sock_transact_multiple(struct nl_sock *sock,
707 struct nl_transaction **transactions, size_t n)
716 /* In theory, every request could have a 64 kB reply. But the default and
717 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
718 * be a bit below 128 kB, so that would only allow a single message in a
719 * "batch". So we assume that replies average (at most) 4 kB, which allows
720 * a good deal of batching.
722 * In practice, most of the requests that we batch either have no reply at
723 * all or a brief reply. */
724 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
725 max_batch_count = MIN(max_batch_count, max_iovs);
731 /* Batch up to 'max_batch_count' transactions. But cap it at about a
732 * page of requests total because big skbuffs are expensive to
733 * allocate in the kernel. */
734 #if defined(PAGESIZE)
735 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
737 enum { MAX_BATCH_BYTES = 4096 - 512 };
739 bytes = ofpbuf_size(transactions[0]->request);
740 for (count = 1; count < n && count < max_batch_count; count++) {
741 if (bytes + ofpbuf_size(transactions[count]->request) > MAX_BATCH_BYTES) {
744 bytes += ofpbuf_size(transactions[count]->request);
747 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
748 transactions += done;
751 if (error == ENOBUFS) {
752 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
754 VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error));
755 nl_sock_record_errors__(transactions, n, error);
761 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
762 struct ofpbuf **replyp)
764 struct nl_transaction *transactionp;
765 struct nl_transaction transaction;
767 transaction.request = CONST_CAST(struct ofpbuf *, request);
768 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
769 transactionp = &transaction;
771 nl_sock_transact_multiple(sock, &transactionp, 1);
774 if (transaction.error) {
775 ofpbuf_delete(transaction.reply);
778 *replyp = transaction.reply;
782 return transaction.error;
785 /* Drain all the messages currently in 'sock''s receive queue. */
787 nl_sock_drain(struct nl_sock *sock)
792 return drain_rcvbuf(sock->fd);
796 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a
797 * Netlink socket created with the given 'protocol', and initializes 'dump' to
798 * reflect the state of the operation.
800 * 'request' must contain a Netlink message. Before sending the message,
801 * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be
802 * set to the Netlink socket's pid. NLM_F_DUMP and NLM_F_ACK will be set in
805 * The design of this Netlink socket library ensures that the dump is reliable.
807 * This function provides no status indication. nl_dump_done() provides an
808 * error status for the entire dump operation.
810 * The caller must eventually destroy 'request'.
813 nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request)
815 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
817 ovs_mutex_init(&dump->mutex);
818 ovs_mutex_lock(&dump->mutex);
819 dump->status = nl_pool_alloc(protocol, &dump->sock);
821 dump->status = nl_sock_send__(dump->sock, request,
822 nl_sock_allocate_seq(dump->sock, 1),
825 dump->nl_seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
826 ovs_mutex_unlock(&dump->mutex);
830 nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer)
831 OVS_REQUIRES(dump->mutex)
833 struct nlmsghdr *nlmsghdr;
836 while (!ofpbuf_size(buffer)) {
837 error = nl_sock_recv__(dump->sock, buffer, false);
839 /* The kernel never blocks providing the results of a dump, so
840 * error == EAGAIN means that we've read the whole thing, and
841 * therefore transform it into EOF. (The kernel always provides
842 * NLMSG_DONE as a sentinel. Some other thread must have received
843 * that already but not yet signaled it in 'status'.)
845 * Any other error is just an error. */
846 return error == EAGAIN ? EOF : error;
849 nlmsghdr = nl_msg_nlmsghdr(buffer);
850 if (dump->nl_seq != nlmsghdr->nlmsg_seq) {
851 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
852 nlmsghdr->nlmsg_seq, dump->nl_seq);
853 ofpbuf_clear(buffer);
857 if (nl_msg_nlmsgerr(buffer, &error) && error) {
858 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
859 ovs_strerror(error));
860 ofpbuf_clear(buffer);
868 nl_dump_next__(struct ofpbuf *reply, struct ofpbuf *buffer)
870 struct nlmsghdr *nlmsghdr = nl_msg_next(buffer, reply);
872 VLOG_WARN_RL(&rl, "netlink dump contains message fragment");
874 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
881 /* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must
882 * have been initialized with nl_dump_start(), and 'buffer' must have been
883 * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long.
885 * If successful, returns true and points 'reply->data' and
886 * 'ofpbuf_size(reply)' to the message that was retrieved. The caller must not
887 * modify 'reply' (because it points within 'buffer', which will be used by
888 * future calls to this function).
890 * On failure, returns false and sets 'reply->data' to NULL and
891 * 'ofpbuf_size(reply)' to 0. Failure might indicate an actual error or merely
892 * the end of replies. An error status for the entire dump operation is
893 * provided when it is completed by calling nl_dump_done().
895 * Multiple threads may call this function, passing the same nl_dump, however
896 * each must provide independent buffers. This function may cache multiple
897 * replies in the buffer, and these will be processed before more replies are
898 * fetched. When this function returns false, other threads may continue to
899 * process replies in their buffers, but they will not fetch more replies.
902 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer)
906 /* If the buffer is empty, refill it.
908 * If the buffer is not empty, we don't check the dump's status.
909 * Otherwise, we could end up skipping some of the dump results if thread A
910 * hits EOF while thread B is in the midst of processing a batch. */
911 if (!ofpbuf_size(buffer)) {
912 ovs_mutex_lock(&dump->mutex);
914 /* Take the mutex here to avoid an in-kernel race. If two threads
915 * try to read from a Netlink dump socket at once, then the socket
916 * error can be set to EINVAL, which will be encountered on the
917 * next recv on that socket, which could be anywhere due to the way
918 * that we pool Netlink sockets. Serializing the recv calls avoids
920 dump->status = nl_dump_refill(dump, buffer);
922 retval = dump->status;
923 ovs_mutex_unlock(&dump->mutex);
926 /* Fetch the next message from the buffer. */
928 retval = nl_dump_next__(reply, buffer);
930 /* Record 'retval' as the dump status, but don't overwrite an error
932 ovs_mutex_lock(&dump->mutex);
933 if (dump->status <= 0) {
934 dump->status = retval;
936 ovs_mutex_unlock(&dump->mutex);
941 ofpbuf_set_data(reply, NULL);
942 ofpbuf_set_size(reply, 0);
947 /* Completes Netlink dump operation 'dump', which must have been initialized
948 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
949 * otherwise a positive errno value describing the problem. */
951 nl_dump_done(struct nl_dump *dump)
955 ovs_mutex_lock(&dump->mutex);
956 status = dump->status;
957 ovs_mutex_unlock(&dump->mutex);
959 /* Drain any remaining messages that the client didn't read. Otherwise the
960 * kernel will continue to queue them up and waste buffer space.
962 * XXX We could just destroy and discard the socket in this case. */
964 uint64_t tmp_reply_stub[NL_DUMP_BUFSIZE / 8];
965 struct ofpbuf reply, buf;
967 ofpbuf_use_stub(&buf, tmp_reply_stub, sizeof tmp_reply_stub);
968 while (nl_dump_next(dump, &reply, &buf)) {
973 ovs_mutex_lock(&dump->mutex);
974 status = dump->status;
975 ovs_mutex_unlock(&dump->mutex);
979 nl_pool_release(dump->sock);
980 ovs_mutex_destroy(&dump->mutex);
982 return status == EOF ? 0 : status;
985 /* Causes poll_block() to wake up when any of the specified 'events' (which is
986 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
988 nl_sock_wait(const struct nl_sock *sock, short int events)
991 poll_fd_wait(sock->handle, events);
993 poll_fd_wait(sock->fd, events);
997 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
998 * that can't use nl_sock_wait().
1000 * It's a little tricky to use the returned fd correctly, because nl_sock does
1001 * "copy on write" to allow a single nl_sock to be used for notifications,
1002 * transactions, and dumps. If 'sock' is used only for notifications and
1003 * transactions (and never for dump) then the usage is safe. */
1005 nl_sock_fd(const struct nl_sock *sock)
1008 return sock->handle;
1014 /* Returns the PID associated with this socket. */
1016 nl_sock_pid(const struct nl_sock *sock)
1021 /* Miscellaneous. */
1023 struct genl_family {
1024 struct hmap_node hmap_node;
1029 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
1031 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
1032 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
1033 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
1036 static struct genl_family *
1037 find_genl_family_by_id(uint16_t id)
1039 struct genl_family *family;
1041 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
1043 if (family->id == id) {
1051 define_genl_family(uint16_t id, const char *name)
1053 struct genl_family *family = find_genl_family_by_id(id);
1056 if (!strcmp(family->name, name)) {
1061 family = xmalloc(sizeof *family);
1063 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
1065 family->name = xstrdup(name);
1069 genl_family_to_name(uint16_t id)
1071 if (id == GENL_ID_CTRL) {
1074 struct genl_family *family = find_genl_family_by_id(id);
1075 return family ? family->name : "unknown";
1080 do_lookup_genl_family(const char *name, struct nlattr **attrs,
1081 struct ofpbuf **replyp)
1083 struct nl_sock *sock;
1084 struct ofpbuf request, *reply;
1088 error = nl_sock_create(NETLINK_GENERIC, &sock);
1093 ofpbuf_init(&request, 0);
1094 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
1095 CTRL_CMD_GETFAMILY, 1);
1096 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
1097 error = nl_sock_transact(sock, &request, &reply);
1098 ofpbuf_uninit(&request);
1100 nl_sock_destroy(sock);
1104 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1105 family_policy, attrs, ARRAY_SIZE(family_policy))
1106 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
1107 nl_sock_destroy(sock);
1108 ofpbuf_delete(reply);
1112 nl_sock_destroy(sock);
1117 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
1118 * When successful, writes its result to 'multicast_group' and returns 0.
1119 * Otherwise, clears 'multicast_group' and returns a positive error code.
1122 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
1123 unsigned int *multicast_group)
1125 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
1126 const struct nlattr *mc;
1127 struct ofpbuf *reply;
1131 *multicast_group = 0;
1132 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1137 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1142 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
1143 static const struct nl_policy mc_policy[] = {
1144 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1145 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1148 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1149 const char *mc_name;
1151 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
1156 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1157 if (!strcmp(group_name, mc_name)) {
1159 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
1167 ofpbuf_delete(reply);
1171 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1172 * number and stores it in '*number'. If successful, returns 0 and the caller
1173 * may use '*number' as the family number. On failure, returns a positive
1174 * errno value and '*number' caches the errno value. */
1176 nl_lookup_genl_family(const char *name, int *number)
1179 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1180 struct ofpbuf *reply;
1183 error = do_lookup_genl_family(name, attrs, &reply);
1185 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1186 define_genl_family(*number, name);
1190 ofpbuf_delete(reply);
1192 ovs_assert(*number != 0);
1194 return *number > 0 ? 0 : -*number;
1198 struct nl_sock *socks[16];
1202 static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER;
1203 static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex);
1206 nl_pool_alloc(int protocol, struct nl_sock **sockp)
1208 struct nl_sock *sock = NULL;
1209 struct nl_pool *pool;
1211 ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools));
1213 ovs_mutex_lock(&pool_mutex);
1214 pool = &pools[protocol];
1216 sock = pool->socks[--pool->n];
1218 ovs_mutex_unlock(&pool_mutex);
1224 return nl_sock_create(protocol, sockp);
1229 nl_pool_release(struct nl_sock *sock)
1232 struct nl_pool *pool = &pools[sock->protocol];
1234 ovs_mutex_lock(&pool_mutex);
1235 if (pool->n < ARRAY_SIZE(pool->socks)) {
1236 pool->socks[pool->n++] = sock;
1239 ovs_mutex_unlock(&pool_mutex);
1241 nl_sock_destroy(sock);
1245 /* Sends 'request' to the kernel on a Netlink socket for the given 'protocol'
1246 * (e.g. NETLINK_ROUTE or NETLINK_GENERIC) and waits for a response. If
1247 * successful, returns 0. On failure, returns a positive errno value.
1249 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
1250 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
1251 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
1252 * reply, if any, is discarded.
1254 * Before the message is sent, nlmsg_len in 'request' will be finalized to
1255 * match ofpbuf_size(msg), nlmsg_pid will be set to the pid of the socket used
1256 * for sending the request, and nlmsg_seq will be initialized.
1258 * The caller is responsible for destroying 'request'.
1260 * Bare Netlink is an unreliable transport protocol. This function layers
1261 * reliable delivery and reply semantics on top of bare Netlink.
1263 * In Netlink, sending a request to the kernel is reliable enough, because the
1264 * kernel will tell us if the message cannot be queued (and we will in that
1265 * case put it on the transmit queue and wait until it can be delivered).
1267 * Receiving the reply is the real problem: if the socket buffer is full when
1268 * the kernel tries to send the reply, the reply will be dropped. However, the
1269 * kernel sets a flag that a reply has been dropped. The next call to recv
1270 * then returns ENOBUFS. We can then re-send the request.
1274 * 1. Netlink depends on sequence numbers to match up requests and
1275 * replies. The sender of a request supplies a sequence number, and
1276 * the reply echos back that sequence number.
1278 * This is fine, but (1) some kernel netlink implementations are
1279 * broken, in that they fail to echo sequence numbers and (2) this
1280 * function will drop packets with non-matching sequence numbers, so
1281 * that only a single request can be usefully transacted at a time.
1283 * 2. Resending the request causes it to be re-executed, so the request
1284 * needs to be idempotent.
1287 nl_transact(int protocol, const struct ofpbuf *request,
1288 struct ofpbuf **replyp)
1290 struct nl_sock *sock;
1293 error = nl_pool_alloc(protocol, &sock);
1299 error = nl_sock_transact(sock, request, replyp);
1301 nl_pool_release(sock);
1305 /* Sends the 'request' member of the 'n' transactions in 'transactions' on a
1306 * Netlink socket for the given 'protocol' (e.g. NETLINK_ROUTE or
1307 * NETLINK_GENERIC), in order, and receives responses to all of them. Fills in
1308 * the 'error' member of each transaction with 0 if it was successful,
1309 * otherwise with a positive errno value. If 'reply' is nonnull, then it will
1310 * be filled with the reply if the message receives a detailed reply. In other
1311 * cases, i.e. where the request failed or had no reply beyond an indication of
1312 * success, 'reply' will be cleared if it is nonnull.
1314 * The caller is responsible for destroying each request and reply, and the
1315 * transactions array itself.
1317 * Before sending each message, this function will finalize nlmsg_len in each
1318 * 'request' to match the ofpbuf's size, set nlmsg_pid to the pid of the socket
1319 * used for the transaction, and initialize nlmsg_seq.
1321 * Bare Netlink is an unreliable transport protocol. This function layers
1322 * reliable delivery and reply semantics on top of bare Netlink. See
1323 * nl_transact() for some caveats.
1326 nl_transact_multiple(int protocol,
1327 struct nl_transaction **transactions, size_t n)
1329 struct nl_sock *sock;
1332 error = nl_pool_alloc(protocol, &sock);
1334 nl_sock_transact_multiple(sock, transactions, n);
1335 nl_pool_release(sock);
1337 nl_sock_record_errors__(transactions, n, error);
1343 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1345 uint32_t seq = sock->next_seq;
1347 sock->next_seq += n;
1349 /* Make it impossible for the next request for sequence numbers to wrap
1350 * around to 0. Start over with 1 to avoid ever using a sequence number of
1351 * 0, because the kernel uses sequence number 0 for notifications. */
1352 if (sock->next_seq >= UINT32_MAX / 2) {
1360 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1366 static const struct nlmsg_flag flags[] = {
1367 { NLM_F_REQUEST, "REQUEST" },
1368 { NLM_F_MULTI, "MULTI" },
1369 { NLM_F_ACK, "ACK" },
1370 { NLM_F_ECHO, "ECHO" },
1371 { NLM_F_DUMP, "DUMP" },
1372 { NLM_F_ROOT, "ROOT" },
1373 { NLM_F_MATCH, "MATCH" },
1374 { NLM_F_ATOMIC, "ATOMIC" },
1376 const struct nlmsg_flag *flag;
1377 uint16_t flags_left;
1379 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1380 h->nlmsg_len, h->nlmsg_type);
1381 if (h->nlmsg_type == NLMSG_NOOP) {
1382 ds_put_cstr(ds, "(no-op)");
1383 } else if (h->nlmsg_type == NLMSG_ERROR) {
1384 ds_put_cstr(ds, "(error)");
1385 } else if (h->nlmsg_type == NLMSG_DONE) {
1386 ds_put_cstr(ds, "(done)");
1387 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1388 ds_put_cstr(ds, "(overrun)");
1389 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1390 ds_put_cstr(ds, "(reserved)");
1391 } else if (protocol == NETLINK_GENERIC) {
1392 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1394 ds_put_cstr(ds, "(family-defined)");
1396 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1397 flags_left = h->nlmsg_flags;
1398 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1399 if ((flags_left & flag->bits) == flag->bits) {
1400 ds_put_format(ds, "[%s]", flag->name);
1401 flags_left &= ~flag->bits;
1405 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1407 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1408 h->nlmsg_seq, h->nlmsg_pid);
1412 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1414 struct ds ds = DS_EMPTY_INITIALIZER;
1415 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1417 nlmsghdr_to_string(h, protocol, &ds);
1418 if (h->nlmsg_type == NLMSG_ERROR) {
1419 const struct nlmsgerr *e;
1420 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1421 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1423 ds_put_format(&ds, " error(%d", e->error);
1425 ds_put_format(&ds, "(%s)", ovs_strerror(-e->error));
1427 ds_put_cstr(&ds, ", in-reply-to(");
1428 nlmsghdr_to_string(&e->msg, protocol, &ds);
1429 ds_put_cstr(&ds, "))");
1431 ds_put_cstr(&ds, " error(truncated)");
1433 } else if (h->nlmsg_type == NLMSG_DONE) {
1434 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1436 ds_put_format(&ds, " done(%d", *error);
1438 ds_put_format(&ds, "(%s)", ovs_strerror(-*error));
1440 ds_put_cstr(&ds, ")");
1442 ds_put_cstr(&ds, " done(truncated)");
1444 } else if (protocol == NETLINK_GENERIC) {
1445 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1447 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1448 genl->cmd, genl->version);
1452 ds_put_cstr(&ds, "nl(truncated)");
1458 log_nlmsg(const char *function, int error,
1459 const void *message, size_t size, int protocol)
1461 struct ofpbuf buffer;
1464 if (!VLOG_IS_DBG_ENABLED()) {
1468 ofpbuf_use_const(&buffer, message, size);
1469 nlmsg = nlmsg_to_string(&buffer, protocol);
1470 VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg);