2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 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"
34 #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_sent);
45 /* Linux header file confusion causes this to be undefined. */
47 #define SOL_NETLINK 270
50 /* A single (bad) Netlink message can in theory dump out many, many log
51 * messages, so the burst size is set quite high here to avoid missing useful
52 * information. Also, at high logging levels we log *all* Netlink messages. */
53 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
55 static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
56 static void log_nlmsg(const char *function, int error,
57 const void *message, size_t size, int protocol);
59 /* Netlink sockets. */
66 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
69 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
70 * of iovecs on the stack. */
73 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
74 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
76 * Initialized by nl_sock_create(). */
79 static int nl_pool_alloc(int protocol, struct nl_sock **sockp);
80 static void nl_pool_release(struct nl_sock *);
82 /* Creates a new netlink socket for the given netlink 'protocol'
83 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
84 * new socket if successful, otherwise returns a positive errno value. */
86 nl_sock_create(int protocol, struct nl_sock **sockp)
88 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
90 struct sockaddr_nl local, remote;
95 if (ovsthread_once_start(&once)) {
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", ovs_strerror(errno));
104 max_iovs = _XOPEN_IOV_MAX;
105 } else if (max_iovs > MAX_IOVS) {
110 ovsthread_once_done(&once);
114 sock = xmalloc(sizeof *sock);
116 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
118 VLOG_ERR("fcntl: %s", ovs_strerror(errno));
121 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, ovs_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", ovs_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", ovs_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)
200 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
201 * successful, otherwise a positive errno value.
203 * A socket that is subscribed to a multicast group that receives asynchronous
204 * notifications must not be used for Netlink transactions or dumps, because
205 * transactions and dumps can cause notifications to be lost.
207 * Multicast group numbers are always positive.
209 * It is not an error to attempt to join a multicast group to which a socket
210 * already belongs. */
212 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
214 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
215 &multicast_group, sizeof multicast_group) < 0) {
216 VLOG_WARN("could not join multicast group %u (%s)",
217 multicast_group, ovs_strerror(errno));
223 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
224 * successful, otherwise a positive errno value.
226 * Multicast group numbers are always positive.
228 * It is not an error to attempt to leave a multicast group to which a socket
231 * On success, reading from 'sock' will still return any messages that were
232 * received on 'multicast_group' before the group was left. */
234 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
236 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
237 &multicast_group, sizeof multicast_group) < 0) {
238 VLOG_WARN("could not leave multicast group %u (%s)",
239 multicast_group, ovs_strerror(errno));
246 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
247 uint32_t nlmsg_seq, bool wait)
249 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
252 nlmsg->nlmsg_len = msg->size;
253 nlmsg->nlmsg_seq = nlmsg_seq;
254 nlmsg->nlmsg_pid = sock->pid;
257 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
258 error = retval < 0 ? errno : 0;
259 } while (error == EINTR);
260 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
262 COVERAGE_INC(netlink_sent);
267 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
268 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
269 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
270 * sequence number, before the message is sent.
272 * Returns 0 if successful, otherwise a positive errno value. If
273 * 'wait' is true, then the send will wait until buffer space is ready;
274 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
276 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
278 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
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
284 * 'nlmsg_seq', 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 * This function is suitable for sending a reply to a request that was received
291 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
293 nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
294 uint32_t nlmsg_seq, bool wait)
296 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
300 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
302 /* We can't accurately predict the size of the data to be received. The
303 * caller is supposed to have allocated enough space in 'buf' to handle the
304 * "typical" case. To handle exceptions, we make available enough space in
305 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
306 * figure since that's the maximum length of a Netlink attribute). */
307 struct nlmsghdr *nlmsghdr;
313 ovs_assert(buf->allocated >= sizeof *nlmsghdr);
316 iov[0].iov_base = buf->base;
317 iov[0].iov_len = buf->allocated;
318 iov[1].iov_base = tail;
319 iov[1].iov_len = sizeof tail;
321 memset(&msg, 0, sizeof msg);
326 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
327 } while (retval < 0 && errno == EINTR);
331 if (error == ENOBUFS) {
332 /* Socket receive buffer overflow dropped one or more messages that
333 * the kernel tried to send to us. */
334 COVERAGE_INC(netlink_overflow);
339 if (msg.msg_flags & MSG_TRUNC) {
340 VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)",
345 nlmsghdr = buf->data;
346 if (retval < sizeof *nlmsghdr
347 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
348 || nlmsghdr->nlmsg_len > retval) {
349 VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE"d bytes < %"PRIuSIZE")",
350 retval, sizeof *nlmsghdr);
354 buf->size = MIN(retval, buf->allocated);
355 if (retval > buf->allocated) {
356 COVERAGE_INC(netlink_recv_jumbo);
357 ofpbuf_put(buf, tail, retval - buf->allocated);
360 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
361 COVERAGE_INC(netlink_received);
366 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
367 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
368 * EAGAIN if the 'sock' receive buffer is empty.
370 * The caller must have initialized 'buf' with an allocation of at least
371 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
372 * space for a "typical" message.
374 * On success, returns 0 and replaces 'buf''s previous content by the received
375 * message. This function expands 'buf''s allocated memory, as necessary, to
376 * hold the actual size of the received message.
378 * On failure, returns a positive errno value and clears 'buf' to zero length.
379 * 'buf' retains its previous memory allocation.
381 * Regardless of success or failure, this function resets 'buf''s headroom to
384 nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
386 return nl_sock_recv__(sock, buf, wait);
390 nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
395 for (i = 0; i < n; i++) {
396 struct nl_transaction *txn = transactions[i];
400 ofpbuf_clear(txn->reply);
406 nl_sock_transact_multiple__(struct nl_sock *sock,
407 struct nl_transaction **transactions, size_t n,
410 uint64_t tmp_reply_stub[1024 / 8];
411 struct nl_transaction tmp_txn;
412 struct ofpbuf tmp_reply;
415 struct iovec iovs[MAX_IOVS];
420 base_seq = nl_sock_allocate_seq(sock, n);
422 for (i = 0; i < n; i++) {
423 struct nl_transaction *txn = transactions[i];
424 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
426 nlmsg->nlmsg_len = txn->request->size;
427 nlmsg->nlmsg_seq = base_seq + i;
428 nlmsg->nlmsg_pid = sock->pid;
430 iovs[i].iov_base = txn->request->data;
431 iovs[i].iov_len = txn->request->size;
434 memset(&msg, 0, sizeof msg);
438 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
439 } while (error == EINTR);
441 for (i = 0; i < n; i++) {
442 struct nl_transaction *txn = transactions[i];
444 log_nlmsg(__func__, error, txn->request->data, txn->request->size,
448 COVERAGE_ADD(netlink_sent, n);
455 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
456 tmp_txn.request = NULL;
457 tmp_txn.reply = &tmp_reply;
460 struct nl_transaction *buf_txn, *txn;
463 /* Find a transaction whose buffer we can use for receiving a reply.
464 * If no such transaction is left, use tmp_txn. */
466 for (i = 0; i < n; i++) {
467 if (transactions[i]->reply) {
468 buf_txn = transactions[i];
473 /* Receive a reply. */
474 error = nl_sock_recv__(sock, buf_txn->reply, false);
476 if (error == EAGAIN) {
477 nl_sock_record_errors__(transactions, n, 0);
484 /* Match the reply up with a transaction. */
485 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
486 if (seq < base_seq || seq >= base_seq + n) {
487 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
491 txn = transactions[i];
493 /* Fill in the results for 'txn'. */
494 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
496 ofpbuf_clear(txn->reply);
499 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
500 error, ovs_strerror(txn->error));
504 if (txn->reply && txn != buf_txn) {
506 struct ofpbuf *reply = buf_txn->reply;
507 buf_txn->reply = txn->reply;
512 /* Fill in the results for transactions before 'txn'. (We have to do
513 * this after the results for 'txn' itself because of the buffer swap
515 nl_sock_record_errors__(transactions, i, 0);
519 transactions += i + 1;
523 ofpbuf_uninit(&tmp_reply);
528 /* Sends the 'request' member of the 'n' transactions in 'transactions' on
529 * 'sock', in order, and receives responses to all of them. Fills in the
530 * 'error' member of each transaction with 0 if it was successful, otherwise
531 * with a positive errno value. If 'reply' is nonnull, then it will be filled
532 * with the reply if the message receives a detailed reply. In other cases,
533 * i.e. where the request failed or had no reply beyond an indication of
534 * success, 'reply' will be cleared if it is nonnull.
536 * The caller is responsible for destroying each request and reply, and the
537 * transactions array itself.
539 * Before sending each message, this function will finalize nlmsg_len in each
540 * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
541 * initialize nlmsg_seq.
543 * Bare Netlink is an unreliable transport protocol. This function layers
544 * reliable delivery and reply semantics on top of bare Netlink. See
545 * nl_sock_transact() for some caveats.
548 nl_sock_transact_multiple(struct nl_sock *sock,
549 struct nl_transaction **transactions, size_t n)
558 /* In theory, every request could have a 64 kB reply. But the default and
559 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
560 * be a bit below 128 kB, so that would only allow a single message in a
561 * "batch". So we assume that replies average (at most) 4 kB, which allows
562 * a good deal of batching.
564 * In practice, most of the requests that we batch either have no reply at
565 * all or a brief reply. */
566 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
567 max_batch_count = MIN(max_batch_count, max_iovs);
573 /* Batch up to 'max_batch_count' transactions. But cap it at about a
574 * page of requests total because big skbuffs are expensive to
575 * allocate in the kernel. */
576 #if defined(PAGESIZE)
577 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
579 enum { MAX_BATCH_BYTES = 4096 - 512 };
581 bytes = transactions[0]->request->size;
582 for (count = 1; count < n && count < max_batch_count; count++) {
583 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
586 bytes += transactions[count]->request->size;
589 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
590 transactions += done;
593 if (error == ENOBUFS) {
594 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
596 VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error));
597 nl_sock_record_errors__(transactions, n, error);
602 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
603 * successful, returns 0. On failure, returns a positive errno value.
605 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
606 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
607 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
608 * reply, if any, is discarded.
610 * Before the message is sent, nlmsg_len in 'request' will be finalized to
611 * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
612 * be initialized, NLM_F_ACK will be set in nlmsg_flags.
614 * The caller is responsible for destroying 'request'.
616 * Bare Netlink is an unreliable transport protocol. This function layers
617 * reliable delivery and reply semantics on top of bare Netlink.
619 * In Netlink, sending a request to the kernel is reliable enough, because the
620 * kernel will tell us if the message cannot be queued (and we will in that
621 * case put it on the transmit queue and wait until it can be delivered).
623 * Receiving the reply is the real problem: if the socket buffer is full when
624 * the kernel tries to send the reply, the reply will be dropped. However, the
625 * kernel sets a flag that a reply has been dropped. The next call to recv
626 * then returns ENOBUFS. We can then re-send the request.
630 * 1. Netlink depends on sequence numbers to match up requests and
631 * replies. The sender of a request supplies a sequence number, and
632 * the reply echos back that sequence number.
634 * This is fine, but (1) some kernel netlink implementations are
635 * broken, in that they fail to echo sequence numbers and (2) this
636 * function will drop packets with non-matching sequence numbers, so
637 * that only a single request can be usefully transacted at a time.
639 * 2. Resending the request causes it to be re-executed, so the request
640 * needs to be idempotent.
643 nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
644 struct ofpbuf **replyp)
646 struct nl_transaction *transactionp;
647 struct nl_transaction transaction;
649 transaction.request = CONST_CAST(struct ofpbuf *, request);
650 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
651 transactionp = &transaction;
653 nl_sock_transact_multiple(sock, &transactionp, 1);
656 if (transaction.error) {
657 ofpbuf_delete(transaction.reply);
660 *replyp = transaction.reply;
664 return transaction.error;
667 /* Drain all the messages currently in 'sock''s receive queue. */
669 nl_sock_drain(struct nl_sock *sock)
671 return drain_rcvbuf(sock->fd);
674 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a
675 * Netlink socket created with the given 'protocol', and initializes 'dump' to
676 * reflect the state of the operation.
678 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
679 * be set to the Netlink socket's pid, before the message is sent. NLM_F_DUMP
680 * and NLM_F_ACK will be set in nlmsg_flags.
682 * The design of this Netlink socket library ensures that the dump is reliable.
684 * This function provides no status indication. An error status for the entire
685 * dump operation is provided when it is completed by calling nl_dump_done().
687 * The caller is responsible for destroying 'request'.
690 nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request)
692 ofpbuf_init(&dump->buffer, 4096);
693 dump->status = nl_pool_alloc(protocol, &dump->sock);
698 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
699 dump->status = nl_sock_send__(dump->sock, request,
700 nl_sock_allocate_seq(dump->sock, 1), true);
701 dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
704 /* Helper function for nl_dump_next(). */
706 nl_dump_recv(struct nl_dump *dump)
708 struct nlmsghdr *nlmsghdr;
711 retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
713 return retval == EINTR ? EAGAIN : retval;
716 nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
717 if (dump->seq != nlmsghdr->nlmsg_seq) {
718 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
719 nlmsghdr->nlmsg_seq, dump->seq);
723 if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
724 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
725 ovs_strerror(retval));
726 return retval && retval != EAGAIN ? retval : EPROTO;
732 /* Attempts to retrieve another reply from 'dump', which must have been
733 * initialized with nl_dump_start().
735 * If successful, returns true and points 'reply->data' and 'reply->size' to
736 * the message that was retrieved. The caller must not modify 'reply' (because
737 * it points into the middle of a larger buffer).
739 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
740 * to 0. Failure might indicate an actual error or merely the end of replies.
741 * An error status for the entire dump operation is provided when it is
742 * completed by calling nl_dump_done().
745 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
747 struct nlmsghdr *nlmsghdr;
755 while (!dump->buffer.size) {
756 int retval = nl_dump_recv(dump);
758 ofpbuf_clear(&dump->buffer);
759 if (retval != EAGAIN) {
760 dump->status = retval;
766 nlmsghdr = nl_msg_next(&dump->buffer, reply);
768 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
769 dump->status = EPROTO;
771 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
779 /* Completes Netlink dump operation 'dump', which must have been initialized
780 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
781 * otherwise a positive errno value describing the problem. */
783 nl_dump_done(struct nl_dump *dump)
785 /* Drain any remaining messages that the client didn't read. Otherwise the
786 * kernel will continue to queue them up and waste buffer space.
788 * XXX We could just destroy and discard the socket in this case. */
789 while (!dump->status) {
791 if (!nl_dump_next(dump, &reply)) {
792 ovs_assert(dump->status);
795 nl_pool_release(dump->sock);
796 ofpbuf_uninit(&dump->buffer);
797 return dump->status == EOF ? 0 : dump->status;
800 /* Causes poll_block() to wake up when any of the specified 'events' (which is
801 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
803 nl_sock_wait(const struct nl_sock *sock, short int events)
805 poll_fd_wait(sock->fd, events);
808 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
809 * that can't use nl_sock_wait().
811 * It's a little tricky to use the returned fd correctly, because nl_sock does
812 * "copy on write" to allow a single nl_sock to be used for notifications,
813 * transactions, and dumps. If 'sock' is used only for notifications and
814 * transactions (and never for dump) then the usage is safe. */
816 nl_sock_fd(const struct nl_sock *sock)
821 /* Returns the PID associated with this socket. */
823 nl_sock_pid(const struct nl_sock *sock)
831 struct hmap_node hmap_node;
836 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
838 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
839 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
840 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
843 static struct genl_family *
844 find_genl_family_by_id(uint16_t id)
846 struct genl_family *family;
848 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
850 if (family->id == id) {
858 define_genl_family(uint16_t id, const char *name)
860 struct genl_family *family = find_genl_family_by_id(id);
863 if (!strcmp(family->name, name)) {
868 family = xmalloc(sizeof *family);
870 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
872 family->name = xstrdup(name);
876 genl_family_to_name(uint16_t id)
878 if (id == GENL_ID_CTRL) {
881 struct genl_family *family = find_genl_family_by_id(id);
882 return family ? family->name : "unknown";
887 do_lookup_genl_family(const char *name, struct nlattr **attrs,
888 struct ofpbuf **replyp)
890 struct nl_sock *sock;
891 struct ofpbuf request, *reply;
895 error = nl_sock_create(NETLINK_GENERIC, &sock);
900 ofpbuf_init(&request, 0);
901 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
902 CTRL_CMD_GETFAMILY, 1);
903 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
904 error = nl_sock_transact(sock, &request, &reply);
905 ofpbuf_uninit(&request);
907 nl_sock_destroy(sock);
911 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
912 family_policy, attrs, ARRAY_SIZE(family_policy))
913 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
914 nl_sock_destroy(sock);
915 ofpbuf_delete(reply);
919 nl_sock_destroy(sock);
924 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
925 * When successful, writes its result to 'multicast_group' and returns 0.
926 * Otherwise, clears 'multicast_group' and returns a positive error code.
929 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
930 unsigned int *multicast_group)
932 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
933 const struct nlattr *mc;
934 struct ofpbuf *reply;
938 *multicast_group = 0;
939 error = do_lookup_genl_family(family_name, family_attrs, &reply);
944 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
949 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
950 static const struct nl_policy mc_policy[] = {
951 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
952 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
955 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
958 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
963 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
964 if (!strcmp(group_name, mc_name)) {
966 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
974 ofpbuf_delete(reply);
978 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
979 * number and stores it in '*number'. If successful, returns 0 and the caller
980 * may use '*number' as the family number. On failure, returns a positive
981 * errno value and '*number' caches the errno value. */
983 nl_lookup_genl_family(const char *name, int *number)
986 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
987 struct ofpbuf *reply;
990 error = do_lookup_genl_family(name, attrs, &reply);
992 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
993 define_genl_family(*number, name);
997 ofpbuf_delete(reply);
999 ovs_assert(*number != 0);
1001 return *number > 0 ? 0 : -*number;
1005 struct nl_sock *socks[16];
1009 static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER;
1010 static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex);
1013 nl_pool_alloc(int protocol, struct nl_sock **sockp)
1015 struct nl_sock *sock = NULL;
1016 struct nl_pool *pool;
1018 ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools));
1020 ovs_mutex_lock(&pool_mutex);
1021 pool = &pools[protocol];
1023 sock = pool->socks[--pool->n];
1025 ovs_mutex_unlock(&pool_mutex);
1031 return nl_sock_create(protocol, sockp);
1036 nl_pool_release(struct nl_sock *sock)
1039 struct nl_pool *pool = &pools[sock->protocol];
1041 ovs_mutex_lock(&pool_mutex);
1042 if (pool->n < ARRAY_SIZE(pool->socks)) {
1043 pool->socks[pool->n++] = sock;
1046 ovs_mutex_unlock(&pool_mutex);
1048 nl_sock_destroy(sock);
1053 nl_transact(int protocol, const struct ofpbuf *request,
1054 struct ofpbuf **replyp)
1056 struct nl_sock *sock;
1059 error = nl_pool_alloc(protocol, &sock);
1065 error = nl_sock_transact(sock, request, replyp);
1067 nl_pool_release(sock);
1072 nl_transact_multiple(int protocol,
1073 struct nl_transaction **transactions, size_t n)
1075 struct nl_sock *sock;
1078 error = nl_pool_alloc(protocol, &sock);
1080 nl_sock_transact_multiple(sock, transactions, n);
1081 nl_pool_release(sock);
1083 nl_sock_record_errors__(transactions, n, error);
1089 nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1091 uint32_t seq = sock->next_seq;
1093 sock->next_seq += n;
1095 /* Make it impossible for the next request for sequence numbers to wrap
1096 * around to 0. Start over with 1 to avoid ever using a sequence number of
1097 * 0, because the kernel uses sequence number 0 for notifications. */
1098 if (sock->next_seq >= UINT32_MAX / 2) {
1106 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
1112 static const struct nlmsg_flag flags[] = {
1113 { NLM_F_REQUEST, "REQUEST" },
1114 { NLM_F_MULTI, "MULTI" },
1115 { NLM_F_ACK, "ACK" },
1116 { NLM_F_ECHO, "ECHO" },
1117 { NLM_F_DUMP, "DUMP" },
1118 { NLM_F_ROOT, "ROOT" },
1119 { NLM_F_MATCH, "MATCH" },
1120 { NLM_F_ATOMIC, "ATOMIC" },
1122 const struct nlmsg_flag *flag;
1123 uint16_t flags_left;
1125 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1126 h->nlmsg_len, h->nlmsg_type);
1127 if (h->nlmsg_type == NLMSG_NOOP) {
1128 ds_put_cstr(ds, "(no-op)");
1129 } else if (h->nlmsg_type == NLMSG_ERROR) {
1130 ds_put_cstr(ds, "(error)");
1131 } else if (h->nlmsg_type == NLMSG_DONE) {
1132 ds_put_cstr(ds, "(done)");
1133 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1134 ds_put_cstr(ds, "(overrun)");
1135 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1136 ds_put_cstr(ds, "(reserved)");
1137 } else if (protocol == NETLINK_GENERIC) {
1138 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
1140 ds_put_cstr(ds, "(family-defined)");
1142 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1143 flags_left = h->nlmsg_flags;
1144 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1145 if ((flags_left & flag->bits) == flag->bits) {
1146 ds_put_format(ds, "[%s]", flag->name);
1147 flags_left &= ~flag->bits;
1151 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1153 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1154 h->nlmsg_seq, h->nlmsg_pid);
1158 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
1160 struct ds ds = DS_EMPTY_INITIALIZER;
1161 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1163 nlmsghdr_to_string(h, protocol, &ds);
1164 if (h->nlmsg_type == NLMSG_ERROR) {
1165 const struct nlmsgerr *e;
1166 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1167 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1169 ds_put_format(&ds, " error(%d", e->error);
1171 ds_put_format(&ds, "(%s)", ovs_strerror(-e->error));
1173 ds_put_cstr(&ds, ", in-reply-to(");
1174 nlmsghdr_to_string(&e->msg, protocol, &ds);
1175 ds_put_cstr(&ds, "))");
1177 ds_put_cstr(&ds, " error(truncated)");
1179 } else if (h->nlmsg_type == NLMSG_DONE) {
1180 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1182 ds_put_format(&ds, " done(%d", *error);
1184 ds_put_format(&ds, "(%s)", ovs_strerror(-*error));
1186 ds_put_cstr(&ds, ")");
1188 ds_put_cstr(&ds, " done(truncated)");
1190 } else if (protocol == NETLINK_GENERIC) {
1191 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1193 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1194 genl->cmd, genl->version);
1198 ds_put_cstr(&ds, "nl(truncated)");
1204 log_nlmsg(const char *function, int error,
1205 const void *message, size_t size, int protocol)
1207 struct ofpbuf buffer;
1210 if (!VLOG_IS_DBG_ENABLED()) {
1214 ofpbuf_use_const(&buffer, message, size);
1215 nlmsg = nlmsg_to_string(&buffer, protocol);
1216 VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg);