2 * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
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>
26 #include "dynamic-string.h"
30 #include "netlink-protocol.h"
32 #include "poll-loop.h"
33 #include "socket-util.h"
37 VLOG_DEFINE_THIS_MODULE(netlink_socket);
39 COVERAGE_DEFINE(netlink_overflow);
40 COVERAGE_DEFINE(netlink_received);
41 COVERAGE_DEFINE(netlink_recv_jumbo);
42 COVERAGE_DEFINE(netlink_send);
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 void log_nlmsg(const char *function, int error,
56 const void *message, size_t size, int protocol);
58 /* Netlink sockets. */
68 static int alloc_pid(uint32_t *);
69 static void free_pid(uint32_t);
70 static int nl_sock_cow__(struct nl_sock *);
72 /* Creates a new netlink socket for the given netlink 'protocol'
73 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
74 * new socket if successful, otherwise returns a positive errno value. */
76 nl_sock_create(int protocol, struct nl_sock **sockp)
79 struct sockaddr_nl local, remote;
83 sock = malloc(sizeof *sock);
88 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
90 VLOG_ERR("fcntl: %s", strerror(errno));
93 sock->protocol = protocol;
96 retval = alloc_pid(&sock->pid);
101 /* Bind local address as our selected pid. */
102 memset(&local, 0, sizeof local);
103 local.nl_family = AF_NETLINK;
104 local.nl_pid = sock->pid;
105 if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) {
106 VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno));
110 /* Bind remote address as the kernel (pid 0). */
111 memset(&remote, 0, sizeof remote);
112 remote.nl_family = AF_NETLINK;
114 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
115 VLOG_ERR("connect(0): %s", strerror(errno));
138 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
139 * sets '*sockp' to the new socket if successful, otherwise returns a positive
142 nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
144 return nl_sock_create(src->protocol, sockp);
147 /* Destroys netlink socket 'sock'. */
149 nl_sock_destroy(struct nl_sock *sock)
162 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
163 * successful, otherwise a positive errno value.
165 * A socket that is subscribed to a multicast group that receives asynchronous
166 * notifications must not be used for Netlink transactions or dumps, because
167 * transactions and dumps can cause notifications to be lost.
169 * Multicast group numbers are always positive.
171 * It is not an error to attempt to join a multicast group to which a socket
172 * already belongs. */
174 nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
176 int error = nl_sock_cow__(sock);
180 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
181 &multicast_group, sizeof multicast_group) < 0) {
182 VLOG_WARN("could not join multicast group %u (%s)",
183 multicast_group, strerror(errno));
189 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
190 * successful, otherwise a positive errno value.
192 * Multicast group numbers are always positive.
194 * It is not an error to attempt to leave a multicast group to which a socket
197 * On success, reading from 'sock' will still return any messages that were
198 * received on 'multicast_group' before the group was left. */
200 nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
203 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
204 &multicast_group, sizeof multicast_group) < 0) {
205 VLOG_WARN("could not leave multicast group %u (%s)",
206 multicast_group, strerror(errno));
213 nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
215 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
218 nlmsg->nlmsg_len = msg->size;
219 nlmsg->nlmsg_pid = sock->pid;
222 retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT);
223 error = retval < 0 ? errno : 0;
224 } while (error == EINTR);
225 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
227 COVERAGE_INC(netlink_sent);
232 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
233 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, and
234 * nlmsg_pid will be set to 'sock''s pid, before the message is sent.
236 * Returns 0 if successful, otherwise a positive errno value. If
237 * 'wait' is true, then the send will wait until buffer space is ready;
238 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
240 nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
242 int error = nl_sock_cow__(sock);
246 return nl_sock_send__(sock, msg, wait);
249 /* This stress option is useful for testing that OVS properly tolerates
250 * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
251 * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
252 * reply to a request. They can also occur if messages arrive on a multicast
253 * channel faster than OVS can process them. */
255 netlink_overflow, "simulate netlink socket receive buffer overflow",
259 nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
261 /* We can't accurately predict the size of the data to be received. Most
262 * received data will fit in a 2 kB buffer, so we allocate that much space.
263 * In case the data is actually bigger than that, we make available enough
264 * additional space to allow Netlink messages to be up to 64 kB long (a
265 * reasonable figure since that's the maximum length of a Netlink
267 enum { MAX_SIZE = 65536 };
268 enum { HEAD_SIZE = 2048 };
269 enum { TAIL_SIZE = MAX_SIZE - HEAD_SIZE };
271 struct nlmsghdr *nlmsghdr;
272 uint8_t tail[TAIL_SIZE];
280 buf = ofpbuf_new(HEAD_SIZE);
281 iov[0].iov_base = buf->data;
282 iov[0].iov_len = HEAD_SIZE;
283 iov[1].iov_base = tail;
284 iov[1].iov_len = TAIL_SIZE;
286 memset(&msg, 0, sizeof msg);
291 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
292 } while (retval < 0 && errno == EINTR);
296 if (error == ENOBUFS) {
297 /* Socket receive buffer overflow dropped one or more messages that
298 * the kernel tried to send to us. */
299 COVERAGE_INC(netlink_overflow);
305 if (msg.msg_flags & MSG_TRUNC) {
306 VLOG_ERR_RL(&rl, "truncated message (longer than %d bytes)", MAX_SIZE);
311 ofpbuf_put_uninit(buf, MIN(retval, HEAD_SIZE));
312 if (retval > HEAD_SIZE) {
313 COVERAGE_INC(netlink_recv_jumbo);
314 ofpbuf_put(buf, tail, retval - HEAD_SIZE);
317 nlmsghdr = buf->data;
318 if (retval < sizeof *nlmsghdr
319 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
320 || nlmsghdr->nlmsg_len > retval) {
321 VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)",
322 retval, NLMSG_HDRLEN);
327 if (STRESS(netlink_overflow)) {
333 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
334 COVERAGE_INC(netlink_received);
339 /* Tries to receive a netlink message from the kernel on 'sock'. If
340 * successful, stores the received message into '*bufp' and returns 0. The
341 * caller is responsible for destroying the message with ofpbuf_delete(). On
342 * failure, returns a positive errno value and stores a null pointer into
345 * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
346 * returns EAGAIN if the 'sock' receive buffer is empty. */
348 nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
350 int error = nl_sock_cow__(sock);
354 return nl_sock_recv__(sock, bufp, wait);
357 /* Sends 'request' to the kernel via 'sock' and waits for a response. If
358 * successful, returns 0. On failure, returns a positive errno value.
360 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
361 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
362 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
363 * reply, if any, is discarded.
365 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
366 * be set to 'sock''s pid, before the message is sent. NLM_F_ACK will be set
369 * The caller is responsible for destroying 'request'.
371 * Bare Netlink is an unreliable transport protocol. This function layers
372 * reliable delivery and reply semantics on top of bare Netlink.
374 * In Netlink, sending a request to the kernel is reliable enough, because the
375 * kernel will tell us if the message cannot be queued (and we will in that
376 * case put it on the transmit queue and wait until it can be delivered).
378 * Receiving the reply is the real problem: if the socket buffer is full when
379 * the kernel tries to send the reply, the reply will be dropped. However, the
380 * kernel sets a flag that a reply has been dropped. The next call to recv
381 * then returns ENOBUFS. We can then re-send the request.
385 * 1. Netlink depends on sequence numbers to match up requests and
386 * replies. The sender of a request supplies a sequence number, and
387 * the reply echos back that sequence number.
389 * This is fine, but (1) some kernel netlink implementations are
390 * broken, in that they fail to echo sequence numbers and (2) this
391 * function will drop packets with non-matching sequence numbers, so
392 * that only a single request can be usefully transacted at a time.
394 * 2. Resending the request causes it to be re-executed, so the request
395 * needs to be idempotent.
398 nl_sock_transact(struct nl_sock *sock,
399 const struct ofpbuf *request, struct ofpbuf **replyp)
401 uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
402 struct nlmsghdr *nlmsghdr;
403 struct ofpbuf *reply;
410 /* Ensure that we get a reply even if this message doesn't ordinarily call
412 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK;
415 retval = nl_sock_send(sock, request, true);
421 retval = nl_sock_recv(sock, &reply, true);
423 if (retval == ENOBUFS) {
424 COVERAGE_INC(netlink_overflow);
425 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
431 nlmsghdr = nl_msg_nlmsghdr(reply);
432 if (seq != nlmsghdr->nlmsg_seq) {
433 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
434 nl_msg_nlmsghdr(reply)->nlmsg_seq, seq);
435 ofpbuf_delete(reply);
439 /* If the reply is an error, discard the reply and return the error code.
441 * Except: if the reply is just an acknowledgement (error code of 0), and
442 * the caller is interested in the reply (replyp != NULL), pass the reply
443 * up to the caller. Otherwise the caller will get a return value of 0
444 * and null '*replyp', which makes unwary callers likely to segfault. */
445 if (nl_msg_nlmsgerr(reply, &retval) && (retval || !replyp)) {
446 ofpbuf_delete(reply);
448 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
449 retval, strerror(retval));
451 return retval != EAGAIN ? retval : EPROTO;
457 ofpbuf_delete(reply);
462 /* Drain all the messages currently in 'sock''s receive queue. */
464 nl_sock_drain(struct nl_sock *sock)
466 int error = nl_sock_cow__(sock);
470 return drain_rcvbuf(sock->fd);
473 /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing
474 * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s
475 * old fd over to the dump. */
477 nl_sock_cow__(struct nl_sock *sock)
479 struct nl_sock *copy;
488 error = nl_sock_clone(sock, ©);
498 sock->pid = copy->pid;
501 sock->dump->sock = copy;
507 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
508 * 'sock', and initializes 'dump' to reflect the state of the operation.
510 * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
511 * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and
512 * NLM_F_ACK will be set in nlmsg_flags.
514 * This Netlink socket library is designed to ensure that the dump is reliable
515 * and that it will not interfere with other operations on 'sock', including
516 * destroying or sending and receiving messages on 'sock'. One corner case is
519 * - If 'sock' has been used to send a request (e.g. with nl_sock_send())
520 * whose response has not yet been received (e.g. with nl_sock_recv()).
521 * This is unusual: usually nl_sock_transact() is used to send a message
522 * and receive its reply all in one go.
524 * This function provides no status indication. An error status for the entire
525 * dump operation is provided when it is completed by calling nl_dump_done().
527 * The caller is responsible for destroying 'request'.
529 * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed
533 nl_dump_start(struct nl_dump *dump,
534 struct nl_sock *sock, const struct ofpbuf *request)
536 struct nlmsghdr *nlmsghdr = nl_msg_nlmsghdr(request);
537 nlmsghdr->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
538 dump->seq = nlmsghdr->nlmsg_seq;
541 /* 'sock' already has an ongoing dump. Clone the socket because
542 * Netlink only allows one dump at a time. */
543 dump->status = nl_sock_clone(sock, &dump->sock);
552 dump->status = nl_sock_send__(sock, request, true);
555 /* Helper function for nl_dump_next(). */
557 nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp)
559 struct nlmsghdr *nlmsghdr;
560 struct ofpbuf *buffer;
563 retval = nl_sock_recv__(dump->sock, bufferp, true);
565 return retval == EINTR ? EAGAIN : retval;
569 nlmsghdr = nl_msg_nlmsghdr(buffer);
570 if (dump->seq != nlmsghdr->nlmsg_seq) {
571 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
572 nlmsghdr->nlmsg_seq, dump->seq);
576 if (nl_msg_nlmsgerr(buffer, &retval)) {
577 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
579 return retval && retval != EAGAIN ? retval : EPROTO;
585 /* Attempts to retrieve another reply from 'dump', which must have been
586 * initialized with nl_dump_start().
588 * If successful, returns true and points 'reply->data' and 'reply->size' to
589 * the message that was retrieved. The caller must not modify 'reply' (because
590 * it points into the middle of a larger buffer).
592 * On failure, returns false and sets 'reply->data' to NULL and 'reply->size'
593 * to 0. Failure might indicate an actual error or merely the end of replies.
594 * An error status for the entire dump operation is provided when it is
595 * completed by calling nl_dump_done().
598 nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
600 struct nlmsghdr *nlmsghdr;
608 if (dump->buffer && !dump->buffer->size) {
609 ofpbuf_delete(dump->buffer);
612 while (!dump->buffer) {
613 int retval = nl_dump_recv(dump, &dump->buffer);
615 ofpbuf_delete(dump->buffer);
617 if (retval != EAGAIN) {
618 dump->status = retval;
624 nlmsghdr = nl_msg_next(dump->buffer, reply);
626 VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
627 dump->status = EPROTO;
629 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
637 /* Completes Netlink dump operation 'dump', which must have been initialized
638 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
639 * otherwise a positive errno value describing the problem. */
641 nl_dump_done(struct nl_dump *dump)
643 /* Drain any remaining messages that the client didn't read. Otherwise the
644 * kernel will continue to queue them up and waste buffer space. */
645 while (!dump->status) {
647 if (!nl_dump_next(dump, &reply)) {
648 assert(dump->status);
653 if (dump->sock->dump) {
654 dump->sock->dump = NULL;
656 nl_sock_destroy(dump->sock);
659 ofpbuf_delete(dump->buffer);
660 return dump->status == EOF ? 0 : dump->status;
663 /* Causes poll_block() to wake up when any of the specified 'events' (which is
664 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
666 nl_sock_wait(const struct nl_sock *sock, short int events)
668 poll_fd_wait(sock->fd, events);
671 /* Checks whether this socket caused a wakeup in the previous run of the poll
674 nl_sock_woke(const struct nl_sock *sock)
676 return poll_fd_woke(sock->fd);
679 /* Returns the PID associated with this socket. */
681 nl_sock_pid(const struct nl_sock *sock)
689 struct hmap_node hmap_node;
694 static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
696 static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
697 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
698 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
701 static struct genl_family *
702 find_genl_family_by_id(uint16_t id)
704 struct genl_family *family;
706 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
708 if (family->id == id) {
716 define_genl_family(uint16_t id, const char *name)
718 struct genl_family *family = find_genl_family_by_id(id);
721 if (!strcmp(family->name, name)) {
726 family = xmalloc(sizeof *family);
728 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
730 family->name = xstrdup(name);
734 genl_family_to_name(uint16_t id)
736 if (id == GENL_ID_CTRL) {
739 struct genl_family *family = find_genl_family_by_id(id);
740 return family ? family->name : "unknown";
745 do_lookup_genl_family(const char *name, struct nlattr **attrs,
746 struct ofpbuf **replyp)
748 struct nl_sock *sock;
749 struct ofpbuf request, *reply;
753 error = nl_sock_create(NETLINK_GENERIC, &sock);
758 ofpbuf_init(&request, 0);
759 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
760 CTRL_CMD_GETFAMILY, 1);
761 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
762 error = nl_sock_transact(sock, &request, &reply);
763 ofpbuf_uninit(&request);
765 nl_sock_destroy(sock);
769 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
770 family_policy, attrs, ARRAY_SIZE(family_policy))
771 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
772 nl_sock_destroy(sock);
773 ofpbuf_delete(reply);
777 nl_sock_destroy(sock);
782 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
783 * When successful, writes its result to 'multicast_group' and returns 0.
784 * Otherwise, clears 'multicast_group' and returns a positive error code.
786 * Some kernels do not support looking up a multicast group with this function.
787 * In this case, 'multicast_group' will be populated with 'fallback'. */
789 nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
790 unsigned int *multicast_group, unsigned int fallback)
792 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
793 struct ofpbuf all_mcs;
794 struct ofpbuf *reply;
799 *multicast_group = 0;
800 error = do_lookup_genl_family(family_name, family_attrs, &reply);
805 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
806 *multicast_group = fallback;
807 VLOG_WARN("%s-%s: has no multicast group, using fallback %d",
808 family_name, group_name, *multicast_group);
813 nl_attr_get_nested(family_attrs[CTRL_ATTR_MCAST_GROUPS], &all_mcs);
814 NL_ATTR_FOR_EACH (mc, left, all_mcs.data, all_mcs.size) {
815 static const struct nl_policy mc_policy[] = {
816 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
817 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
820 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
823 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
828 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
829 if (!strcmp(group_name, mc_name)) {
831 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
839 ofpbuf_delete(reply);
843 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
844 * number and stores it in '*number'. If successful, returns 0 and the caller
845 * may use '*number' as the family number. On failure, returns a positive
846 * errno value and '*number' caches the errno value. */
848 nl_lookup_genl_family(const char *name, int *number)
851 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
852 struct ofpbuf *reply;
855 error = do_lookup_genl_family(name, attrs, &reply);
857 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
858 define_genl_family(*number, name);
862 ofpbuf_delete(reply);
864 assert(*number != 0);
866 return *number > 0 ? 0 : -*number;
871 * Every Netlink socket must be bound to a unique 32-bit PID. By convention,
872 * programs that have a single Netlink socket use their Unix process ID as PID,
873 * and programs with multiple Netlink sockets add a unique per-socket
874 * identifier in the bits above the Unix process ID.
876 * The kernel has Netlink PID 0.
879 /* Parameters for how many bits in the PID should come from the Unix process ID
880 * and how many unique per-socket. */
881 #define SOCKET_BITS 10
882 #define MAX_SOCKETS (1u << SOCKET_BITS)
884 #define PROCESS_BITS (32 - SOCKET_BITS)
885 #define MAX_PROCESSES (1u << PROCESS_BITS)
886 #define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1))
888 /* Bit vector of unused socket identifiers. */
889 static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)];
891 /* Allocates and returns a new Netlink PID. */
893 alloc_pid(uint32_t *pid)
897 for (i = 0; i < MAX_SOCKETS; i++) {
898 if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) {
899 avail_sockets[i / 32] |= 1u << (i % 32);
900 *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS);
904 VLOG_ERR("netlink pid space exhausted");
908 /* Makes the specified 'pid' available for reuse. */
910 free_pid(uint32_t pid)
912 int sock = pid >> PROCESS_BITS;
913 assert(avail_sockets[sock / 32] & (1u << (sock % 32)));
914 avail_sockets[sock / 32] &= ~(1u << (sock % 32));
918 nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
924 static const struct nlmsg_flag flags[] = {
925 { NLM_F_REQUEST, "REQUEST" },
926 { NLM_F_MULTI, "MULTI" },
927 { NLM_F_ACK, "ACK" },
928 { NLM_F_ECHO, "ECHO" },
929 { NLM_F_DUMP, "DUMP" },
930 { NLM_F_ROOT, "ROOT" },
931 { NLM_F_MATCH, "MATCH" },
932 { NLM_F_ATOMIC, "ATOMIC" },
934 const struct nlmsg_flag *flag;
937 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
938 h->nlmsg_len, h->nlmsg_type);
939 if (h->nlmsg_type == NLMSG_NOOP) {
940 ds_put_cstr(ds, "(no-op)");
941 } else if (h->nlmsg_type == NLMSG_ERROR) {
942 ds_put_cstr(ds, "(error)");
943 } else if (h->nlmsg_type == NLMSG_DONE) {
944 ds_put_cstr(ds, "(done)");
945 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
946 ds_put_cstr(ds, "(overrun)");
947 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
948 ds_put_cstr(ds, "(reserved)");
949 } else if (protocol == NETLINK_GENERIC) {
950 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
952 ds_put_cstr(ds, "(family-defined)");
954 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
955 flags_left = h->nlmsg_flags;
956 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
957 if ((flags_left & flag->bits) == flag->bits) {
958 ds_put_format(ds, "[%s]", flag->name);
959 flags_left &= ~flag->bits;
963 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
965 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32"(%d:%d))",
966 h->nlmsg_seq, h->nlmsg_pid,
967 (int) (h->nlmsg_pid & PROCESS_MASK),
968 (int) (h->nlmsg_pid >> PROCESS_BITS));
972 nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
974 struct ds ds = DS_EMPTY_INITIALIZER;
975 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
977 nlmsghdr_to_string(h, protocol, &ds);
978 if (h->nlmsg_type == NLMSG_ERROR) {
979 const struct nlmsgerr *e;
980 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
981 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
983 ds_put_format(&ds, " error(%d", e->error);
985 ds_put_format(&ds, "(%s)", strerror(-e->error));
987 ds_put_cstr(&ds, ", in-reply-to(");
988 nlmsghdr_to_string(&e->msg, protocol, &ds);
989 ds_put_cstr(&ds, "))");
991 ds_put_cstr(&ds, " error(truncated)");
993 } else if (h->nlmsg_type == NLMSG_DONE) {
994 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
996 ds_put_format(&ds, " done(%d", *error);
998 ds_put_format(&ds, "(%s)", strerror(-*error));
1000 ds_put_cstr(&ds, ")");
1002 ds_put_cstr(&ds, " done(truncated)");
1004 } else if (protocol == NETLINK_GENERIC) {
1005 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1007 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1008 genl->cmd, genl->version);
1012 ds_put_cstr(&ds, "nl(truncated)");
1018 log_nlmsg(const char *function, int error,
1019 const void *message, size_t size, int protocol)
1021 struct ofpbuf buffer;
1024 if (!VLOG_IS_DBG_ENABLED()) {
1028 ofpbuf_use_const(&buffer, message, size);
1029 nlmsg = nlmsg_to_string(&buffer, protocol);
1030 VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);