2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2016 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 "openvswitch/dynamic-string.h"
30 #include "netlink-protocol.h"
31 #include "odp-netlink.h"
32 #include "openvswitch/ofpbuf.h"
33 #include "ovs-thread.h"
34 #include "poll-loop.h"
36 #include "socket-util.h"
38 #include "openvswitch/vlog.h"
40 VLOG_DEFINE_THIS_MODULE(netlink_socket
);
42 COVERAGE_DEFINE(netlink_overflow
);
43 COVERAGE_DEFINE(netlink_received
);
44 COVERAGE_DEFINE(netlink_recv_jumbo
);
45 COVERAGE_DEFINE(netlink_sent
);
47 /* Linux header file confusion causes this to be undefined. */
49 #define SOL_NETLINK 270
52 /* A single (bad) Netlink message can in theory dump out many, many log
53 * messages, so the burst size is set quite high here to avoid missing useful
54 * information. Also, at high logging levels we log *all* Netlink messages. */
55 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(60, 600);
57 static uint32_t nl_sock_allocate_seq(struct nl_sock
*, unsigned int n
);
58 static void log_nlmsg(const char *function
, int error
,
59 const void *message
, size_t size
, int protocol
);
61 static int get_sock_pid_from_kernel(struct nl_sock
*sock
);
62 static int set_sock_property(struct nl_sock
*sock
);
65 /* Netlink sockets. */
70 OVERLAPPED overlapped
;
78 unsigned int rcvbuf
; /* Receive buffer size (SO_RCVBUF). */
81 /* Compile-time limit on iovecs, so that we can allocate a maximum-size array
82 * of iovecs on the stack. */
85 /* Maximum number of iovecs that may be passed to sendmsg, capped at a
86 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
88 * Initialized by nl_sock_create(). */
91 static int nl_pool_alloc(int protocol
, struct nl_sock
**sockp
);
92 static void nl_pool_release(struct nl_sock
*);
94 /* Creates a new netlink socket for the given netlink 'protocol'
95 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
96 * new socket if successful, otherwise returns a positive errno value. */
98 nl_sock_create(int protocol
, struct nl_sock
**sockp
)
100 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
101 struct nl_sock
*sock
;
103 struct sockaddr_nl local
, remote
;
105 socklen_t local_size
;
109 if (ovsthread_once_start(&once
)) {
110 int save_errno
= errno
;
113 max_iovs
= sysconf(_SC_UIO_MAXIOV
);
114 if (max_iovs
< _XOPEN_IOV_MAX
) {
115 if (max_iovs
== -1 && errno
) {
116 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno
));
118 max_iovs
= _XOPEN_IOV_MAX
;
119 } else if (max_iovs
> MAX_IOVS
) {
124 ovsthread_once_done(&once
);
128 sock
= xmalloc(sizeof *sock
);
131 sock
->overlapped
.hEvent
= NULL
;
132 sock
->handle
= CreateFile(OVS_DEVICE_NAME_USER
,
133 GENERIC_READ
| GENERIC_WRITE
,
134 FILE_SHARE_READ
| FILE_SHARE_WRITE
,
136 FILE_FLAG_OVERLAPPED
, NULL
);
138 if (sock
->handle
== INVALID_HANDLE_VALUE
) {
139 VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
143 memset(&sock
->overlapped
, 0, sizeof sock
->overlapped
);
144 sock
->overlapped
.hEvent
= CreateEvent(NULL
, FALSE
, FALSE
, NULL
);
145 if (sock
->overlapped
.hEvent
== NULL
) {
146 VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
149 /* Initialize the type/ioctl to Generic */
150 sock
->read_ioctl
= OVS_IOCTL_READ
;
152 sock
->fd
= socket(AF_NETLINK
, SOCK_RAW
, protocol
);
154 VLOG_ERR("fcntl: %s", ovs_strerror(errno
));
159 sock
->protocol
= protocol
;
162 rcvbuf
= 1024 * 1024;
164 sock
->rcvbuf
= rcvbuf
;
165 retval
= get_sock_pid_from_kernel(sock
);
169 retval
= set_sock_property(sock
);
174 if (setsockopt(sock
->fd
, SOL_SOCKET
, SO_RCVBUFFORCE
,
175 &rcvbuf
, sizeof rcvbuf
)) {
176 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
177 * Warn only if the failure is therefore unexpected. */
178 if (errno
!= EPERM
) {
179 VLOG_WARN_RL(&rl
, "setting %d-byte socket receive buffer failed "
180 "(%s)", rcvbuf
, ovs_strerror(errno
));
184 retval
= get_socket_rcvbuf(sock
->fd
);
189 sock
->rcvbuf
= retval
;
192 /* Connect to kernel (pid 0) as remote address. */
193 memset(&remote
, 0, sizeof remote
);
194 remote
.nl_family
= AF_NETLINK
;
196 if (connect(sock
->fd
, (struct sockaddr
*) &remote
, sizeof remote
) < 0) {
197 VLOG_ERR("connect(0): %s", ovs_strerror(errno
));
201 /* Obtain pid assigned by kernel. */
202 local_size
= sizeof local
;
203 if (getsockname(sock
->fd
, (struct sockaddr
*) &local
, &local_size
) < 0) {
204 VLOG_ERR("getsockname: %s", ovs_strerror(errno
));
207 if (local_size
< sizeof local
|| local
.nl_family
!= AF_NETLINK
) {
208 VLOG_ERR("getsockname returned bad Netlink name");
212 sock
->pid
= local
.nl_pid
;
226 if (sock
->overlapped
.hEvent
) {
227 CloseHandle(sock
->overlapped
.hEvent
);
229 if (sock
->handle
!= INVALID_HANDLE_VALUE
) {
230 CloseHandle(sock
->handle
);
241 /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
242 * sets '*sockp' to the new socket if successful, otherwise returns a positive
245 nl_sock_clone(const struct nl_sock
*src
, struct nl_sock
**sockp
)
247 return nl_sock_create(src
->protocol
, sockp
);
250 /* Destroys netlink socket 'sock'. */
252 nl_sock_destroy(struct nl_sock
*sock
)
256 if (sock
->overlapped
.hEvent
) {
257 CloseHandle(sock
->overlapped
.hEvent
);
259 CloseHandle(sock
->handle
);
268 /* Reads the pid for 'sock' generated in the kernel datapath. The function
269 * uses a separate IOCTL instead of a transaction semantic to avoid unnecessary
270 * message overhead. */
272 get_sock_pid_from_kernel(struct nl_sock
*sock
)
278 if (!DeviceIoControl(sock
->handle
, OVS_IOCTL_GET_PID
,
279 NULL
, 0, &pid
, sizeof(pid
),
283 if (bytes
< sizeof(pid
)) {
293 /* Used for setting and managing socket properties in userspace and kernel.
294 * Currently two attributes are tracked - pid and protocol
295 * protocol - supplied by userspace based on the netlink family. Windows uses
296 * this property to set the value in kernel datapath.
297 * eg: (NETLINK_GENERIC/ NETLINK_NETFILTER)
298 * pid - generated by windows kernel and set in userspace. The property
300 * Also verify if Protocol and PID in Kernel reflects the values in userspace
303 set_sock_property(struct nl_sock
*sock
)
305 static const struct nl_policy ovs_socket_policy
[] = {
306 [OVS_NL_ATTR_SOCK_PROTO
] = { .type
= NL_A_BE32
, .optional
= true },
307 [OVS_NL_ATTR_SOCK_PID
] = { .type
= NL_A_BE32
, .optional
= true }
310 struct ofpbuf request
, *reply
;
311 struct ovs_header
*ovs_header
;
312 struct nlattr
*attrs
[ARRAY_SIZE(ovs_socket_policy
)];
316 ofpbuf_init(&request
, 0);
317 nl_msg_put_genlmsghdr(&request
, 0, OVS_WIN_NL_CTRL_FAMILY_ID
, 0,
318 OVS_CTRL_CMD_SOCK_PROP
, OVS_WIN_CONTROL_VERSION
);
319 ovs_header
= ofpbuf_put_uninit(&request
, sizeof *ovs_header
);
320 ovs_header
->dp_ifindex
= 0;
322 nl_msg_put_be32(&request
, OVS_NL_ATTR_SOCK_PROTO
, sock
->protocol
);
323 /* pid is already set as part of get_sock_pid_from_kernel()
324 * This is added to maintain consistency
326 nl_msg_put_be32(&request
, OVS_NL_ATTR_SOCK_PID
, sock
->pid
);
328 error
= nl_sock_transact(sock
, &request
, &reply
);
329 ofpbuf_uninit(&request
);
334 if (!nl_policy_parse(reply
,
335 NLMSG_HDRLEN
+ GENL_HDRLEN
+ sizeof *ovs_header
,
336 ovs_socket_policy
, attrs
,
337 ARRAY_SIZE(ovs_socket_policy
))) {
338 ofpbuf_delete(reply
);
341 /* Verify if the properties are setup properly */
342 if (attrs
[OVS_NL_ATTR_SOCK_PROTO
]) {
343 int protocol
= nl_attr_get_be32(attrs
[OVS_NL_ATTR_SOCK_PROTO
]);
344 if (protocol
!= sock
->protocol
) {
345 VLOG_ERR("Invalid protocol returned:%d expected:%d",
346 protocol
, sock
->protocol
);
351 if (attrs
[OVS_NL_ATTR_SOCK_PID
]) {
352 int pid
= nl_attr_get_be32(attrs
[OVS_NL_ATTR_SOCK_PID
]);
353 if (pid
!= sock
->pid
) {
354 VLOG_ERR("Invalid pid returned:%d expected:%d",
366 nl_sock_mcgroup(struct nl_sock
*sock
, unsigned int multicast_group
, bool join
)
368 struct ofpbuf request
;
369 uint64_t request_stub
[128];
370 struct ovs_header
*ovs_header
;
371 struct nlmsghdr
*nlmsg
;
374 ofpbuf_use_stub(&request
, request_stub
, sizeof request_stub
);
376 nl_msg_put_genlmsghdr(&request
, 0, OVS_WIN_NL_CTRL_FAMILY_ID
, 0,
377 OVS_CTRL_CMD_MC_SUBSCRIBE_REQ
,
378 OVS_WIN_CONTROL_VERSION
);
380 ovs_header
= ofpbuf_put_uninit(&request
, sizeof *ovs_header
);
381 ovs_header
->dp_ifindex
= 0;
383 nl_msg_put_u32(&request
, OVS_NL_ATTR_MCAST_GRP
, multicast_group
);
384 nl_msg_put_u8(&request
, OVS_NL_ATTR_MCAST_JOIN
, join
? 1 : 0);
386 error
= nl_sock_send(sock
, &request
, true);
387 ofpbuf_uninit(&request
);
391 /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
392 * successful, otherwise a positive errno value.
394 * A socket that is subscribed to a multicast group that receives asynchronous
395 * notifications must not be used for Netlink transactions or dumps, because
396 * transactions and dumps can cause notifications to be lost.
398 * Multicast group numbers are always positive.
400 * It is not an error to attempt to join a multicast group to which a socket
401 * already belongs. */
403 nl_sock_join_mcgroup(struct nl_sock
*sock
, unsigned int multicast_group
)
406 /* Set the socket type as a "multicast" socket */
407 sock
->read_ioctl
= OVS_IOCTL_READ_EVENT
;
408 int error
= nl_sock_mcgroup(sock
, multicast_group
, true);
410 sock
->read_ioctl
= OVS_IOCTL_READ
;
411 VLOG_WARN("could not join multicast group %u (%s)",
412 multicast_group
, ovs_strerror(error
));
416 if (setsockopt(sock
->fd
, SOL_NETLINK
, NETLINK_ADD_MEMBERSHIP
,
417 &multicast_group
, sizeof multicast_group
) < 0) {
418 VLOG_WARN("could not join multicast group %u (%s)",
419 multicast_group
, ovs_strerror(errno
));
428 nl_sock_subscribe_packets(struct nl_sock
*sock
)
432 if (sock
->read_ioctl
!= OVS_IOCTL_READ
) {
436 error
= nl_sock_subscribe_packet__(sock
, true);
438 VLOG_WARN("could not subscribe packets (%s)",
439 ovs_strerror(error
));
442 sock
->read_ioctl
= OVS_IOCTL_READ_PACKET
;
448 nl_sock_unsubscribe_packets(struct nl_sock
*sock
)
450 ovs_assert(sock
->read_ioctl
== OVS_IOCTL_READ_PACKET
);
452 int error
= nl_sock_subscribe_packet__(sock
, false);
454 VLOG_WARN("could not unsubscribe to packets (%s)",
455 ovs_strerror(error
));
459 sock
->read_ioctl
= OVS_IOCTL_READ
;
464 nl_sock_subscribe_packet__(struct nl_sock
*sock
, bool subscribe
)
466 struct ofpbuf request
;
467 uint64_t request_stub
[128];
468 struct ovs_header
*ovs_header
;
469 struct nlmsghdr
*nlmsg
;
472 ofpbuf_use_stub(&request
, request_stub
, sizeof request_stub
);
473 nl_msg_put_genlmsghdr(&request
, 0, OVS_WIN_NL_CTRL_FAMILY_ID
, 0,
474 OVS_CTRL_CMD_PACKET_SUBSCRIBE_REQ
,
475 OVS_WIN_CONTROL_VERSION
);
477 ovs_header
= ofpbuf_put_uninit(&request
, sizeof *ovs_header
);
478 ovs_header
->dp_ifindex
= 0;
479 nl_msg_put_u8(&request
, OVS_NL_ATTR_PACKET_SUBSCRIBE
, subscribe
? 1 : 0);
480 nl_msg_put_u32(&request
, OVS_NL_ATTR_PACKET_PID
, sock
->pid
);
482 error
= nl_sock_send(sock
, &request
, true);
483 ofpbuf_uninit(&request
);
488 /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
489 * successful, otherwise a positive errno value.
491 * Multicast group numbers are always positive.
493 * It is not an error to attempt to leave a multicast group to which a socket
496 * On success, reading from 'sock' will still return any messages that were
497 * received on 'multicast_group' before the group was left. */
499 nl_sock_leave_mcgroup(struct nl_sock
*sock
, unsigned int multicast_group
)
502 int error
= nl_sock_mcgroup(sock
, multicast_group
, false);
504 VLOG_WARN("could not leave multicast group %u (%s)",
505 multicast_group
, ovs_strerror(error
));
508 sock
->read_ioctl
= OVS_IOCTL_READ
;
510 if (setsockopt(sock
->fd
, SOL_NETLINK
, NETLINK_DROP_MEMBERSHIP
,
511 &multicast_group
, sizeof multicast_group
) < 0) {
512 VLOG_WARN("could not leave multicast group %u (%s)",
513 multicast_group
, ovs_strerror(errno
));
521 nl_sock_send__(struct nl_sock
*sock
, const struct ofpbuf
*msg
,
522 uint32_t nlmsg_seq
, bool wait
)
524 struct nlmsghdr
*nlmsg
= nl_msg_nlmsghdr(msg
);
527 nlmsg
->nlmsg_len
= msg
->size
;
528 nlmsg
->nlmsg_seq
= nlmsg_seq
;
529 nlmsg
->nlmsg_pid
= sock
->pid
;
535 if (!DeviceIoControl(sock
->handle
, OVS_IOCTL_WRITE
,
536 msg
->data
, msg
->size
, NULL
, 0,
539 /* XXX: Map to a more appropriate error based on GetLastError(). */
541 VLOG_DBG_RL(&rl
, "fatal driver failure in write: %s",
542 ovs_lasterror_to_string());
547 retval
= send(sock
->fd
, msg
->data
, msg
->size
,
548 wait
? 0 : MSG_DONTWAIT
);
550 error
= retval
< 0 ? errno
: 0;
551 } while (error
== EINTR
);
552 log_nlmsg(__func__
, error
, msg
->data
, msg
->size
, sock
->protocol
);
554 COVERAGE_INC(netlink_sent
);
559 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
560 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
561 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
562 * sequence number, before the message is sent.
564 * Returns 0 if successful, otherwise a positive errno value. If
565 * 'wait' is true, then the send will wait until buffer space is ready;
566 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
568 nl_sock_send(struct nl_sock
*sock
, const struct ofpbuf
*msg
, bool wait
)
570 return nl_sock_send_seq(sock
, msg
, nl_sock_allocate_seq(sock
, 1), wait
);
573 /* Tries to send 'msg', which must contain a Netlink message, to the kernel on
574 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
575 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
576 * 'nlmsg_seq', before the message is sent.
578 * Returns 0 if successful, otherwise a positive errno value. If
579 * 'wait' is true, then the send will wait until buffer space is ready;
580 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
582 * This function is suitable for sending a reply to a request that was received
583 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
585 nl_sock_send_seq(struct nl_sock
*sock
, const struct ofpbuf
*msg
,
586 uint32_t nlmsg_seq
, bool wait
)
588 return nl_sock_send__(sock
, msg
, nlmsg_seq
, wait
);
592 nl_sock_recv__(struct nl_sock
*sock
, struct ofpbuf
*buf
, bool wait
)
594 /* We can't accurately predict the size of the data to be received. The
595 * caller is supposed to have allocated enough space in 'buf' to handle the
596 * "typical" case. To handle exceptions, we make available enough space in
597 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
598 * figure since that's the maximum length of a Netlink attribute). */
599 struct nlmsghdr
*nlmsghdr
;
606 ovs_assert(buf
->allocated
>= sizeof *nlmsghdr
);
609 iov
[0].iov_base
= buf
->base
;
610 iov
[0].iov_len
= buf
->allocated
;
611 iov
[1].iov_base
= tail
;
612 iov
[1].iov_len
= sizeof tail
;
614 memset(&msg
, 0, sizeof msg
);
618 /* Receive a Netlink message from the kernel.
620 * This works around a kernel bug in which the kernel returns an error code
621 * as if it were the number of bytes read. It doesn't actually modify
622 * anything in the receive buffer in that case, so we can initialize the
623 * Netlink header with an impossible message length and then, upon success,
624 * check whether it changed. */
625 nlmsghdr
= buf
->base
;
627 nlmsghdr
->nlmsg_len
= UINT32_MAX
;
630 if (!DeviceIoControl(sock
->handle
, sock
->read_ioctl
,
631 NULL
, 0, tail
, sizeof tail
, &bytes
, NULL
)) {
632 VLOG_DBG_RL(&rl
, "fatal driver failure in transact: %s",
633 ovs_lasterror_to_string());
635 /* XXX: Map to a more appropriate error. */
643 if (retval
>= buf
->allocated
) {
644 ofpbuf_reinit(buf
, retval
);
645 nlmsghdr
= buf
->base
;
646 nlmsghdr
->nlmsg_len
= UINT32_MAX
;
648 memcpy(buf
->data
, tail
, retval
);
653 retval
= recvmsg(sock
->fd
, &msg
, wait
? 0 : MSG_DONTWAIT
);
655 error
= (retval
< 0 ? errno
656 : retval
== 0 ? ECONNRESET
/* not possible? */
657 : nlmsghdr
->nlmsg_len
!= UINT32_MAX
? 0
659 } while (error
== EINTR
);
661 if (error
== ENOBUFS
) {
662 /* Socket receive buffer overflow dropped one or more messages that
663 * the kernel tried to send to us. */
664 COVERAGE_INC(netlink_overflow
);
669 if (msg
.msg_flags
& MSG_TRUNC
) {
670 VLOG_ERR_RL(&rl
, "truncated message (longer than %"PRIuSIZE
" bytes)",
675 if (retval
< sizeof *nlmsghdr
676 || nlmsghdr
->nlmsg_len
< sizeof *nlmsghdr
677 || nlmsghdr
->nlmsg_len
> retval
) {
678 VLOG_ERR_RL(&rl
, "received invalid nlmsg (%"PRIuSIZE
" bytes < %"PRIuSIZE
")",
679 retval
, sizeof *nlmsghdr
);
683 buf
->size
= MIN(retval
, buf
->allocated
);
684 if (retval
> buf
->allocated
) {
685 COVERAGE_INC(netlink_recv_jumbo
);
686 ofpbuf_put(buf
, tail
, retval
- buf
->allocated
);
690 log_nlmsg(__func__
, 0, buf
->data
, buf
->size
, sock
->protocol
);
691 COVERAGE_INC(netlink_received
);
696 /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
697 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
698 * EAGAIN if the 'sock' receive buffer is empty.
700 * The caller must have initialized 'buf' with an allocation of at least
701 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
702 * space for a "typical" message.
704 * On success, returns 0 and replaces 'buf''s previous content by the received
705 * message. This function expands 'buf''s allocated memory, as necessary, to
706 * hold the actual size of the received message.
708 * On failure, returns a positive errno value and clears 'buf' to zero length.
709 * 'buf' retains its previous memory allocation.
711 * Regardless of success or failure, this function resets 'buf''s headroom to
714 nl_sock_recv(struct nl_sock
*sock
, struct ofpbuf
*buf
, bool wait
)
716 return nl_sock_recv__(sock
, buf
, wait
);
720 nl_sock_record_errors__(struct nl_transaction
**transactions
, size_t n
,
725 for (i
= 0; i
< n
; i
++) {
726 struct nl_transaction
*txn
= transactions
[i
];
730 ofpbuf_clear(txn
->reply
);
736 nl_sock_transact_multiple__(struct nl_sock
*sock
,
737 struct nl_transaction
**transactions
, size_t n
,
740 uint64_t tmp_reply_stub
[1024 / 8];
741 struct nl_transaction tmp_txn
;
742 struct ofpbuf tmp_reply
;
745 struct iovec iovs
[MAX_IOVS
];
750 base_seq
= nl_sock_allocate_seq(sock
, n
);
752 for (i
= 0; i
< n
; i
++) {
753 struct nl_transaction
*txn
= transactions
[i
];
754 struct nlmsghdr
*nlmsg
= nl_msg_nlmsghdr(txn
->request
);
756 nlmsg
->nlmsg_len
= txn
->request
->size
;
757 nlmsg
->nlmsg_seq
= base_seq
+ i
;
758 nlmsg
->nlmsg_pid
= sock
->pid
;
760 iovs
[i
].iov_base
= txn
->request
->data
;
761 iovs
[i
].iov_len
= txn
->request
->size
;
765 memset(&msg
, 0, sizeof msg
);
769 error
= sendmsg(sock
->fd
, &msg
, 0) < 0 ? errno
: 0;
770 } while (error
== EINTR
);
772 for (i
= 0; i
< n
; i
++) {
773 struct nl_transaction
*txn
= transactions
[i
];
775 log_nlmsg(__func__
, error
, txn
->request
->data
,
776 txn
->request
->size
, sock
->protocol
);
779 COVERAGE_ADD(netlink_sent
, n
);
786 ofpbuf_use_stub(&tmp_reply
, tmp_reply_stub
, sizeof tmp_reply_stub
);
787 tmp_txn
.request
= NULL
;
788 tmp_txn
.reply
= &tmp_reply
;
791 struct nl_transaction
*buf_txn
, *txn
;
794 /* Find a transaction whose buffer we can use for receiving a reply.
795 * If no such transaction is left, use tmp_txn. */
797 for (i
= 0; i
< n
; i
++) {
798 if (transactions
[i
]->reply
) {
799 buf_txn
= transactions
[i
];
804 /* Receive a reply. */
805 error
= nl_sock_recv__(sock
, buf_txn
->reply
, false);
807 if (error
== EAGAIN
) {
808 nl_sock_record_errors__(transactions
, n
, 0);
815 /* Match the reply up with a transaction. */
816 seq
= nl_msg_nlmsghdr(buf_txn
->reply
)->nlmsg_seq
;
817 if (seq
< base_seq
|| seq
>= base_seq
+ n
) {
818 VLOG_DBG_RL(&rl
, "ignoring unexpected seq %#"PRIx32
, seq
);
822 txn
= transactions
[i
];
824 /* Fill in the results for 'txn'. */
825 if (nl_msg_nlmsgerr(buf_txn
->reply
, &txn
->error
)) {
827 ofpbuf_clear(txn
->reply
);
830 VLOG_DBG_RL(&rl
, "received NAK error=%d (%s)",
831 error
, ovs_strerror(txn
->error
));
835 if (txn
->reply
&& txn
!= buf_txn
) {
837 struct ofpbuf
*reply
= buf_txn
->reply
;
838 buf_txn
->reply
= txn
->reply
;
843 /* Fill in the results for transactions before 'txn'. (We have to do
844 * this after the results for 'txn' itself because of the buffer swap
846 nl_sock_record_errors__(transactions
, i
, 0);
850 transactions
+= i
+ 1;
854 ofpbuf_uninit(&tmp_reply
);
857 uint8_t reply_buf
[65536];
858 for (i
= 0; i
< n
; i
++) {
861 struct nl_transaction
*txn
= transactions
[i
];
862 struct nlmsghdr
*request_nlmsg
, *reply_nlmsg
;
864 ret
= DeviceIoControl(sock
->handle
, OVS_IOCTL_TRANSACT
,
867 reply_buf
, sizeof reply_buf
,
870 if (ret
&& reply_len
== 0) {
872 * The current transaction did not produce any data to read and that
873 * is not an error as such. Continue with the remainder of the
878 ofpbuf_clear(txn
->reply
);
881 /* XXX: Map to a more appropriate error. */
883 VLOG_DBG_RL(&rl
, "fatal driver failure: %s",
884 ovs_lasterror_to_string());
888 if (reply_len
!= 0) {
889 if (reply_len
< sizeof *reply_nlmsg
) {
890 nl_sock_record_errors__(transactions
, n
, 0);
891 VLOG_DBG_RL(&rl
, "insufficient length of reply %#"PRIu32
892 " for seq: %#"PRIx32
, reply_len
, request_nlmsg
->nlmsg_seq
);
896 /* Validate the sequence number in the reply. */
897 request_nlmsg
= nl_msg_nlmsghdr(txn
->request
);
898 reply_nlmsg
= (struct nlmsghdr
*)reply_buf
;
900 if (request_nlmsg
->nlmsg_seq
!= reply_nlmsg
->nlmsg_seq
) {
901 ovs_assert(request_nlmsg
->nlmsg_seq
== reply_nlmsg
->nlmsg_seq
);
902 VLOG_DBG_RL(&rl
, "mismatched seq request %#"PRIx32
903 ", reply %#"PRIx32
, request_nlmsg
->nlmsg_seq
,
904 reply_nlmsg
->nlmsg_seq
);
908 /* Handle errors embedded within the netlink message. */
909 ofpbuf_use_stub(&tmp_reply
, reply_buf
, sizeof reply_buf
);
910 tmp_reply
.size
= sizeof reply_buf
;
911 if (nl_msg_nlmsgerr(&tmp_reply
, &txn
->error
)) {
913 ofpbuf_clear(txn
->reply
);
916 VLOG_DBG_RL(&rl
, "received NAK error=%d (%s)",
917 error
, ovs_strerror(txn
->error
));
922 /* Copy the reply to the buffer specified by the caller. */
923 if (reply_len
> txn
->reply
->allocated
) {
924 ofpbuf_reinit(txn
->reply
, reply_len
);
926 memcpy(txn
->reply
->data
, reply_buf
, reply_len
);
927 txn
->reply
->size
= reply_len
;
930 ofpbuf_uninit(&tmp_reply
);
933 /* Count the number of successful transactions. */
939 COVERAGE_ADD(netlink_sent
, n
);
947 nl_sock_transact_multiple(struct nl_sock
*sock
,
948 struct nl_transaction
**transactions
, size_t n
)
957 /* In theory, every request could have a 64 kB reply. But the default and
958 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
959 * be a bit below 128 kB, so that would only allow a single message in a
960 * "batch". So we assume that replies average (at most) 4 kB, which allows
961 * a good deal of batching.
963 * In practice, most of the requests that we batch either have no reply at
964 * all or a brief reply. */
965 max_batch_count
= MAX(sock
->rcvbuf
/ 4096, 1);
966 max_batch_count
= MIN(max_batch_count
, max_iovs
);
972 /* Batch up to 'max_batch_count' transactions. But cap it at about a
973 * page of requests total because big skbuffs are expensive to
974 * allocate in the kernel. */
975 #if defined(PAGESIZE)
976 enum { MAX_BATCH_BYTES
= MAX(1, PAGESIZE
- 512) };
978 enum { MAX_BATCH_BYTES
= 4096 - 512 };
980 bytes
= transactions
[0]->request
->size
;
981 for (count
= 1; count
< n
&& count
< max_batch_count
; count
++) {
982 if (bytes
+ transactions
[count
]->request
->size
> MAX_BATCH_BYTES
) {
985 bytes
+= transactions
[count
]->request
->size
;
988 error
= nl_sock_transact_multiple__(sock
, transactions
, count
, &done
);
989 transactions
+= done
;
992 if (error
== ENOBUFS
) {
993 VLOG_DBG_RL(&rl
, "receive buffer overflow, resending request");
995 VLOG_ERR_RL(&rl
, "transaction error (%s)", ovs_strerror(error
));
996 nl_sock_record_errors__(transactions
, n
, error
);
997 if (error
!= EAGAIN
) {
998 /* A fatal error has occurred. Abort the rest of
1007 nl_sock_transact(struct nl_sock
*sock
, const struct ofpbuf
*request
,
1008 struct ofpbuf
**replyp
)
1010 struct nl_transaction
*transactionp
;
1011 struct nl_transaction transaction
;
1013 transaction
.request
= CONST_CAST(struct ofpbuf
*, request
);
1014 transaction
.reply
= replyp
? ofpbuf_new(1024) : NULL
;
1015 transactionp
= &transaction
;
1017 nl_sock_transact_multiple(sock
, &transactionp
, 1);
1020 if (transaction
.error
) {
1021 ofpbuf_delete(transaction
.reply
);
1024 *replyp
= transaction
.reply
;
1028 return transaction
.error
;
1031 /* Drain all the messages currently in 'sock''s receive queue. */
1033 nl_sock_drain(struct nl_sock
*sock
)
1038 return drain_rcvbuf(sock
->fd
);
1042 /* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a
1043 * Netlink socket created with the given 'protocol', and initializes 'dump' to
1044 * reflect the state of the operation.
1046 * 'request' must contain a Netlink message. Before sending the message,
1047 * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be
1048 * set to the Netlink socket's pid. NLM_F_DUMP and NLM_F_ACK will be set in
1051 * The design of this Netlink socket library ensures that the dump is reliable.
1053 * This function provides no status indication. nl_dump_done() provides an
1054 * error status for the entire dump operation.
1056 * The caller must eventually destroy 'request'.
1059 nl_dump_start(struct nl_dump
*dump
, int protocol
, const struct ofpbuf
*request
)
1061 nl_msg_nlmsghdr(request
)->nlmsg_flags
|= NLM_F_DUMP
| NLM_F_ACK
;
1063 ovs_mutex_init(&dump
->mutex
);
1064 ovs_mutex_lock(&dump
->mutex
);
1065 dump
->status
= nl_pool_alloc(protocol
, &dump
->sock
);
1066 if (!dump
->status
) {
1067 dump
->status
= nl_sock_send__(dump
->sock
, request
,
1068 nl_sock_allocate_seq(dump
->sock
, 1),
1071 dump
->nl_seq
= nl_msg_nlmsghdr(request
)->nlmsg_seq
;
1072 ovs_mutex_unlock(&dump
->mutex
);
1076 nl_dump_refill(struct nl_dump
*dump
, struct ofpbuf
*buffer
)
1077 OVS_REQUIRES(dump
->mutex
)
1079 struct nlmsghdr
*nlmsghdr
;
1082 while (!buffer
->size
) {
1083 error
= nl_sock_recv__(dump
->sock
, buffer
, false);
1085 /* The kernel never blocks providing the results of a dump, so
1086 * error == EAGAIN means that we've read the whole thing, and
1087 * therefore transform it into EOF. (The kernel always provides
1088 * NLMSG_DONE as a sentinel. Some other thread must have received
1089 * that already but not yet signaled it in 'status'.)
1091 * Any other error is just an error. */
1092 return error
== EAGAIN
? EOF
: error
;
1095 nlmsghdr
= nl_msg_nlmsghdr(buffer
);
1096 if (dump
->nl_seq
!= nlmsghdr
->nlmsg_seq
) {
1097 VLOG_DBG_RL(&rl
, "ignoring seq %#"PRIx32
" != expected %#"PRIx32
,
1098 nlmsghdr
->nlmsg_seq
, dump
->nl_seq
);
1099 ofpbuf_clear(buffer
);
1103 if (nl_msg_nlmsgerr(buffer
, &error
) && error
) {
1104 VLOG_INFO_RL(&rl
, "netlink dump request error (%s)",
1105 ovs_strerror(error
));
1106 ofpbuf_clear(buffer
);
1114 nl_dump_next__(struct ofpbuf
*reply
, struct ofpbuf
*buffer
)
1116 struct nlmsghdr
*nlmsghdr
= nl_msg_next(buffer
, reply
);
1118 VLOG_WARN_RL(&rl
, "netlink dump contains message fragment");
1120 } else if (nlmsghdr
->nlmsg_type
== NLMSG_DONE
) {
1127 /* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must
1128 * have been initialized with nl_dump_start(), and 'buffer' must have been
1129 * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long.
1131 * If successful, returns true and points 'reply->data' and
1132 * 'reply->size' to the message that was retrieved. The caller must not
1133 * modify 'reply' (because it points within 'buffer', which will be used by
1134 * future calls to this function).
1136 * On failure, returns false and sets 'reply->data' to NULL and
1137 * 'reply->size' to 0. Failure might indicate an actual error or merely
1138 * the end of replies. An error status for the entire dump operation is
1139 * provided when it is completed by calling nl_dump_done().
1141 * Multiple threads may call this function, passing the same nl_dump, however
1142 * each must provide independent buffers. This function may cache multiple
1143 * replies in the buffer, and these will be processed before more replies are
1144 * fetched. When this function returns false, other threads may continue to
1145 * process replies in their buffers, but they will not fetch more replies.
1148 nl_dump_next(struct nl_dump
*dump
, struct ofpbuf
*reply
, struct ofpbuf
*buffer
)
1152 /* If the buffer is empty, refill it.
1154 * If the buffer is not empty, we don't check the dump's status.
1155 * Otherwise, we could end up skipping some of the dump results if thread A
1156 * hits EOF while thread B is in the midst of processing a batch. */
1157 if (!buffer
->size
) {
1158 ovs_mutex_lock(&dump
->mutex
);
1159 if (!dump
->status
) {
1160 /* Take the mutex here to avoid an in-kernel race. If two threads
1161 * try to read from a Netlink dump socket at once, then the socket
1162 * error can be set to EINVAL, which will be encountered on the
1163 * next recv on that socket, which could be anywhere due to the way
1164 * that we pool Netlink sockets. Serializing the recv calls avoids
1166 dump
->status
= nl_dump_refill(dump
, buffer
);
1168 retval
= dump
->status
;
1169 ovs_mutex_unlock(&dump
->mutex
);
1172 /* Fetch the next message from the buffer. */
1174 retval
= nl_dump_next__(reply
, buffer
);
1176 /* Record 'retval' as the dump status, but don't overwrite an error
1178 ovs_mutex_lock(&dump
->mutex
);
1179 if (dump
->status
<= 0) {
1180 dump
->status
= retval
;
1182 ovs_mutex_unlock(&dump
->mutex
);
1193 /* Completes Netlink dump operation 'dump', which must have been initialized
1194 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
1195 * otherwise a positive errno value describing the problem. */
1197 nl_dump_done(struct nl_dump
*dump
)
1201 ovs_mutex_lock(&dump
->mutex
);
1202 status
= dump
->status
;
1203 ovs_mutex_unlock(&dump
->mutex
);
1205 /* Drain any remaining messages that the client didn't read. Otherwise the
1206 * kernel will continue to queue them up and waste buffer space.
1208 * XXX We could just destroy and discard the socket in this case. */
1210 uint64_t tmp_reply_stub
[NL_DUMP_BUFSIZE
/ 8];
1211 struct ofpbuf reply
, buf
;
1213 ofpbuf_use_stub(&buf
, tmp_reply_stub
, sizeof tmp_reply_stub
);
1214 while (nl_dump_next(dump
, &reply
, &buf
)) {
1215 /* Nothing to do. */
1217 ofpbuf_uninit(&buf
);
1219 ovs_mutex_lock(&dump
->mutex
);
1220 status
= dump
->status
;
1221 ovs_mutex_unlock(&dump
->mutex
);
1225 nl_pool_release(dump
->sock
);
1226 ovs_mutex_destroy(&dump
->mutex
);
1228 return status
== EOF
? 0 : status
;
1232 /* Pend an I/O request in the driver. The driver completes the I/O whenever
1233 * an event or a packet is ready to be read. Once the I/O is completed
1234 * the overlapped structure event associated with the pending I/O will be set
1237 pend_io_request(struct nl_sock
*sock
)
1239 struct ofpbuf request
;
1240 uint64_t request_stub
[128];
1241 struct ovs_header
*ovs_header
;
1242 struct nlmsghdr
*nlmsg
;
1247 OVERLAPPED
*overlapped
= CONST_CAST(OVERLAPPED
*, &sock
->overlapped
);
1248 uint16_t cmd
= OVS_CTRL_CMD_WIN_PEND_PACKET_REQ
;
1250 ovs_assert(sock
->read_ioctl
== OVS_IOCTL_READ_PACKET
||
1251 sock
->read_ioctl
== OVS_IOCTL_READ_EVENT
);
1252 if (sock
->read_ioctl
== OVS_IOCTL_READ_EVENT
) {
1253 cmd
= OVS_CTRL_CMD_WIN_PEND_REQ
;
1256 int ovs_msg_size
= sizeof (struct nlmsghdr
) + sizeof (struct genlmsghdr
) +
1257 sizeof (struct ovs_header
);
1259 ofpbuf_use_stub(&request
, request_stub
, sizeof request_stub
);
1261 seq
= nl_sock_allocate_seq(sock
, 1);
1262 nl_msg_put_genlmsghdr(&request
, 0, OVS_WIN_NL_CTRL_FAMILY_ID
, 0,
1263 cmd
, OVS_WIN_CONTROL_VERSION
);
1264 nlmsg
= nl_msg_nlmsghdr(&request
);
1265 nlmsg
->nlmsg_seq
= seq
;
1266 nlmsg
->nlmsg_pid
= sock
->pid
;
1268 ovs_header
= ofpbuf_put_uninit(&request
, sizeof *ovs_header
);
1269 ovs_header
->dp_ifindex
= 0;
1270 nlmsg
->nlmsg_len
= request
.size
;
1272 if (!DeviceIoControl(sock
->handle
, OVS_IOCTL_WRITE
,
1273 request
.data
, request
.size
,
1274 NULL
, 0, &bytes
, overlapped
)) {
1275 error
= GetLastError();
1276 /* Check if the I/O got pended */
1277 if (error
!= ERROR_IO_INCOMPLETE
&& error
!= ERROR_IO_PENDING
) {
1278 VLOG_ERR("nl_sock_wait failed - %s\n", ovs_format_message(error
));
1286 ofpbuf_uninit(&request
);
1291 /* Causes poll_block() to wake up when any of the specified 'events' (which is
1292 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'.
1293 * On Windows, 'sock' is not treated as const, and may be modified. */
1295 nl_sock_wait(const struct nl_sock
*sock
, short int events
)
1298 if (sock
->overlapped
.Internal
!= STATUS_PENDING
) {
1299 int ret
= pend_io_request(CONST_CAST(struct nl_sock
*, sock
));
1301 poll_wevent_wait(sock
->overlapped
.hEvent
);
1303 poll_immediate_wake();
1306 poll_wevent_wait(sock
->overlapped
.hEvent
);
1309 poll_fd_wait(sock
->fd
, events
);
1314 /* Returns the underlying fd for 'sock', for use in "poll()"-like operations
1315 * that can't use nl_sock_wait().
1317 * It's a little tricky to use the returned fd correctly, because nl_sock does
1318 * "copy on write" to allow a single nl_sock to be used for notifications,
1319 * transactions, and dumps. If 'sock' is used only for notifications and
1320 * transactions (and never for dump) then the usage is safe. */
1322 nl_sock_fd(const struct nl_sock
*sock
)
1328 /* Returns the PID associated with this socket. */
1330 nl_sock_pid(const struct nl_sock
*sock
)
1335 /* Miscellaneous. */
1337 struct genl_family
{
1338 struct hmap_node hmap_node
;
1343 static struct hmap genl_families
= HMAP_INITIALIZER(&genl_families
);
1345 static const struct nl_policy family_policy
[CTRL_ATTR_MAX
+ 1] = {
1346 [CTRL_ATTR_FAMILY_ID
] = {.type
= NL_A_U16
},
1347 [CTRL_ATTR_MCAST_GROUPS
] = {.type
= NL_A_NESTED
, .optional
= true},
1350 static struct genl_family
*
1351 find_genl_family_by_id(uint16_t id
)
1353 struct genl_family
*family
;
1355 HMAP_FOR_EACH_IN_BUCKET (family
, hmap_node
, hash_int(id
, 0),
1357 if (family
->id
== id
) {
1365 define_genl_family(uint16_t id
, const char *name
)
1367 struct genl_family
*family
= find_genl_family_by_id(id
);
1370 if (!strcmp(family
->name
, name
)) {
1375 family
= xmalloc(sizeof *family
);
1377 hmap_insert(&genl_families
, &family
->hmap_node
, hash_int(id
, 0));
1379 family
->name
= xstrdup(name
);
1383 genl_family_to_name(uint16_t id
)
1385 if (id
== GENL_ID_CTRL
) {
1388 struct genl_family
*family
= find_genl_family_by_id(id
);
1389 return family
? family
->name
: "unknown";
1395 do_lookup_genl_family(const char *name
, struct nlattr
**attrs
,
1396 struct ofpbuf
**replyp
)
1398 struct nl_sock
*sock
;
1399 struct ofpbuf request
, *reply
;
1403 error
= nl_sock_create(NETLINK_GENERIC
, &sock
);
1408 ofpbuf_init(&request
, 0);
1409 nl_msg_put_genlmsghdr(&request
, 0, GENL_ID_CTRL
, NLM_F_REQUEST
,
1410 CTRL_CMD_GETFAMILY
, 1);
1411 nl_msg_put_string(&request
, CTRL_ATTR_FAMILY_NAME
, name
);
1412 error
= nl_sock_transact(sock
, &request
, &reply
);
1413 ofpbuf_uninit(&request
);
1415 nl_sock_destroy(sock
);
1419 if (!nl_policy_parse(reply
, NLMSG_HDRLEN
+ GENL_HDRLEN
,
1420 family_policy
, attrs
, ARRAY_SIZE(family_policy
))
1421 || nl_attr_get_u16(attrs
[CTRL_ATTR_FAMILY_ID
]) == 0) {
1422 nl_sock_destroy(sock
);
1423 ofpbuf_delete(reply
);
1427 nl_sock_destroy(sock
);
1433 do_lookup_genl_family(const char *name
, struct nlattr
**attrs
,
1434 struct ofpbuf
**replyp
)
1436 struct nlmsghdr
*nlmsg
;
1437 struct ofpbuf
*reply
;
1440 const char *family_name
;
1441 uint32_t family_version
;
1442 uint32_t family_attrmax
;
1443 uint32_t mcgrp_id
= OVS_WIN_NL_INVALID_MCGRP_ID
;
1444 const char *mcgrp_name
= NULL
;
1447 reply
= ofpbuf_new(1024);
1449 /* CTRL_ATTR_MCAST_GROUPS is supported only for VPORT family. */
1450 if (!strcmp(name
, OVS_WIN_CONTROL_FAMILY
)) {
1451 family_id
= OVS_WIN_NL_CTRL_FAMILY_ID
;
1452 family_name
= OVS_WIN_CONTROL_FAMILY
;
1453 family_version
= OVS_WIN_CONTROL_VERSION
;
1454 family_attrmax
= OVS_WIN_CONTROL_ATTR_MAX
;
1455 } else if (!strcmp(name
, OVS_DATAPATH_FAMILY
)) {
1456 family_id
= OVS_WIN_NL_DATAPATH_FAMILY_ID
;
1457 family_name
= OVS_DATAPATH_FAMILY
;
1458 family_version
= OVS_DATAPATH_VERSION
;
1459 family_attrmax
= OVS_DP_ATTR_MAX
;
1460 } else if (!strcmp(name
, OVS_PACKET_FAMILY
)) {
1461 family_id
= OVS_WIN_NL_PACKET_FAMILY_ID
;
1462 family_name
= OVS_PACKET_FAMILY
;
1463 family_version
= OVS_PACKET_VERSION
;
1464 family_attrmax
= OVS_PACKET_ATTR_MAX
;
1465 } else if (!strcmp(name
, OVS_VPORT_FAMILY
)) {
1466 family_id
= OVS_WIN_NL_VPORT_FAMILY_ID
;
1467 family_name
= OVS_VPORT_FAMILY
;
1468 family_version
= OVS_VPORT_VERSION
;
1469 family_attrmax
= OVS_VPORT_ATTR_MAX
;
1470 mcgrp_id
= OVS_WIN_NL_VPORT_MCGRP_ID
;
1471 mcgrp_name
= OVS_VPORT_MCGROUP
;
1472 } else if (!strcmp(name
, OVS_FLOW_FAMILY
)) {
1473 family_id
= OVS_WIN_NL_FLOW_FAMILY_ID
;
1474 family_name
= OVS_FLOW_FAMILY
;
1475 family_version
= OVS_FLOW_VERSION
;
1476 family_attrmax
= OVS_FLOW_ATTR_MAX
;
1477 } else if (!strcmp(name
, OVS_WIN_NETDEV_FAMILY
)) {
1478 family_id
= OVS_WIN_NL_NETDEV_FAMILY_ID
;
1479 family_name
= OVS_WIN_NETDEV_FAMILY
;
1480 family_version
= OVS_WIN_NETDEV_VERSION
;
1481 family_attrmax
= OVS_WIN_NETDEV_ATTR_MAX
;
1483 ofpbuf_delete(reply
);
1487 nl_msg_put_genlmsghdr(reply
, 0, GENL_ID_CTRL
, 0,
1488 CTRL_CMD_NEWFAMILY
, family_version
);
1489 /* CTRL_ATTR_HDRSIZE and CTRL_ATTR_OPS are not populated, but the
1490 * callers do not seem to need them. */
1491 nl_msg_put_u16(reply
, CTRL_ATTR_FAMILY_ID
, family_id
);
1492 nl_msg_put_string(reply
, CTRL_ATTR_FAMILY_NAME
, family_name
);
1493 nl_msg_put_u32(reply
, CTRL_ATTR_VERSION
, family_version
);
1494 nl_msg_put_u32(reply
, CTRL_ATTR_MAXATTR
, family_attrmax
);
1496 if (mcgrp_id
!= OVS_WIN_NL_INVALID_MCGRP_ID
) {
1497 size_t mcgrp_ofs1
= nl_msg_start_nested(reply
, CTRL_ATTR_MCAST_GROUPS
);
1498 size_t mcgrp_ofs2
= nl_msg_start_nested(reply
,
1499 OVS_WIN_NL_VPORT_MCGRP_ID
- OVS_WIN_NL_MCGRP_START_ID
);
1500 nl_msg_put_u32(reply
, CTRL_ATTR_MCAST_GRP_ID
, mcgrp_id
);
1501 ovs_assert(mcgrp_name
!= NULL
);
1502 nl_msg_put_string(reply
, CTRL_ATTR_MCAST_GRP_NAME
, mcgrp_name
);
1503 nl_msg_end_nested(reply
, mcgrp_ofs2
);
1504 nl_msg_end_nested(reply
, mcgrp_ofs1
);
1507 /* Set the total length of the netlink message. */
1508 nlmsg
= nl_msg_nlmsghdr(reply
);
1509 nlmsg
->nlmsg_len
= reply
->size
;
1511 if (!nl_policy_parse(reply
, NLMSG_HDRLEN
+ GENL_HDRLEN
,
1512 family_policy
, attrs
, ARRAY_SIZE(family_policy
))
1513 || nl_attr_get_u16(attrs
[CTRL_ATTR_FAMILY_ID
]) == 0) {
1514 ofpbuf_delete(reply
);
1523 /* Finds the multicast group called 'group_name' in genl family 'family_name'.
1524 * When successful, writes its result to 'multicast_group' and returns 0.
1525 * Otherwise, clears 'multicast_group' and returns a positive error code.
1528 nl_lookup_genl_mcgroup(const char *family_name
, const char *group_name
,
1529 unsigned int *multicast_group
)
1531 struct nlattr
*family_attrs
[ARRAY_SIZE(family_policy
)];
1532 const struct nlattr
*mc
;
1533 struct ofpbuf
*reply
;
1537 *multicast_group
= 0;
1538 error
= do_lookup_genl_family(family_name
, family_attrs
, &reply
);
1543 if (!family_attrs
[CTRL_ATTR_MCAST_GROUPS
]) {
1548 NL_NESTED_FOR_EACH (mc
, left
, family_attrs
[CTRL_ATTR_MCAST_GROUPS
]) {
1549 static const struct nl_policy mc_policy
[] = {
1550 [CTRL_ATTR_MCAST_GRP_ID
] = {.type
= NL_A_U32
},
1551 [CTRL_ATTR_MCAST_GRP_NAME
] = {.type
= NL_A_STRING
},
1554 struct nlattr
*mc_attrs
[ARRAY_SIZE(mc_policy
)];
1555 const char *mc_name
;
1557 if (!nl_parse_nested(mc
, mc_policy
, mc_attrs
, ARRAY_SIZE(mc_policy
))) {
1562 mc_name
= nl_attr_get_string(mc_attrs
[CTRL_ATTR_MCAST_GRP_NAME
]);
1563 if (!strcmp(group_name
, mc_name
)) {
1565 nl_attr_get_u32(mc_attrs
[CTRL_ATTR_MCAST_GRP_ID
]);
1573 ofpbuf_delete(reply
);
1577 /* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1578 * number and stores it in '*number'. If successful, returns 0 and the caller
1579 * may use '*number' as the family number. On failure, returns a positive
1580 * errno value and '*number' caches the errno value. */
1582 nl_lookup_genl_family(const char *name
, int *number
)
1585 struct nlattr
*attrs
[ARRAY_SIZE(family_policy
)];
1586 struct ofpbuf
*reply
;
1589 error
= do_lookup_genl_family(name
, attrs
, &reply
);
1591 *number
= nl_attr_get_u16(attrs
[CTRL_ATTR_FAMILY_ID
]);
1592 define_genl_family(*number
, name
);
1596 ofpbuf_delete(reply
);
1598 ovs_assert(*number
!= 0);
1600 return *number
> 0 ? 0 : -*number
;
1604 struct nl_sock
*socks
[16];
1608 static struct ovs_mutex pool_mutex
= OVS_MUTEX_INITIALIZER
;
1609 static struct nl_pool pools
[MAX_LINKS
] OVS_GUARDED_BY(pool_mutex
);
1612 nl_pool_alloc(int protocol
, struct nl_sock
**sockp
)
1614 struct nl_sock
*sock
= NULL
;
1615 struct nl_pool
*pool
;
1617 ovs_assert(protocol
>= 0 && protocol
< ARRAY_SIZE(pools
));
1619 ovs_mutex_lock(&pool_mutex
);
1620 pool
= &pools
[protocol
];
1622 sock
= pool
->socks
[--pool
->n
];
1624 ovs_mutex_unlock(&pool_mutex
);
1630 return nl_sock_create(protocol
, sockp
);
1635 nl_pool_release(struct nl_sock
*sock
)
1638 struct nl_pool
*pool
= &pools
[sock
->protocol
];
1640 ovs_mutex_lock(&pool_mutex
);
1641 if (pool
->n
< ARRAY_SIZE(pool
->socks
)) {
1642 pool
->socks
[pool
->n
++] = sock
;
1645 ovs_mutex_unlock(&pool_mutex
);
1647 nl_sock_destroy(sock
);
1651 /* Sends 'request' to the kernel on a Netlink socket for the given 'protocol'
1652 * (e.g. NETLINK_ROUTE or NETLINK_GENERIC) and waits for a response. If
1653 * successful, returns 0. On failure, returns a positive errno value.
1655 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
1656 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
1657 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
1658 * reply, if any, is discarded.
1660 * Before the message is sent, nlmsg_len in 'request' will be finalized to
1661 * match msg->size, nlmsg_pid will be set to the pid of the socket used
1662 * for sending the request, and nlmsg_seq will be initialized.
1664 * The caller is responsible for destroying 'request'.
1666 * Bare Netlink is an unreliable transport protocol. This function layers
1667 * reliable delivery and reply semantics on top of bare Netlink.
1669 * In Netlink, sending a request to the kernel is reliable enough, because the
1670 * kernel will tell us if the message cannot be queued (and we will in that
1671 * case put it on the transmit queue and wait until it can be delivered).
1673 * Receiving the reply is the real problem: if the socket buffer is full when
1674 * the kernel tries to send the reply, the reply will be dropped. However, the
1675 * kernel sets a flag that a reply has been dropped. The next call to recv
1676 * then returns ENOBUFS. We can then re-send the request.
1680 * 1. Netlink depends on sequence numbers to match up requests and
1681 * replies. The sender of a request supplies a sequence number, and
1682 * the reply echos back that sequence number.
1684 * This is fine, but (1) some kernel netlink implementations are
1685 * broken, in that they fail to echo sequence numbers and (2) this
1686 * function will drop packets with non-matching sequence numbers, so
1687 * that only a single request can be usefully transacted at a time.
1689 * 2. Resending the request causes it to be re-executed, so the request
1690 * needs to be idempotent.
1693 nl_transact(int protocol
, const struct ofpbuf
*request
,
1694 struct ofpbuf
**replyp
)
1696 struct nl_sock
*sock
;
1699 error
= nl_pool_alloc(protocol
, &sock
);
1705 error
= nl_sock_transact(sock
, request
, replyp
);
1707 nl_pool_release(sock
);
1711 /* Sends the 'request' member of the 'n' transactions in 'transactions' on a
1712 * Netlink socket for the given 'protocol' (e.g. NETLINK_ROUTE or
1713 * NETLINK_GENERIC), in order, and receives responses to all of them. Fills in
1714 * the 'error' member of each transaction with 0 if it was successful,
1715 * otherwise with a positive errno value. If 'reply' is nonnull, then it will
1716 * be filled with the reply if the message receives a detailed reply. In other
1717 * cases, i.e. where the request failed or had no reply beyond an indication of
1718 * success, 'reply' will be cleared if it is nonnull.
1720 * The caller is responsible for destroying each request and reply, and the
1721 * transactions array itself.
1723 * Before sending each message, this function will finalize nlmsg_len in each
1724 * 'request' to match the ofpbuf's size, set nlmsg_pid to the pid of the socket
1725 * used for the transaction, and initialize nlmsg_seq.
1727 * Bare Netlink is an unreliable transport protocol. This function layers
1728 * reliable delivery and reply semantics on top of bare Netlink. See
1729 * nl_transact() for some caveats.
1732 nl_transact_multiple(int protocol
,
1733 struct nl_transaction
**transactions
, size_t n
)
1735 struct nl_sock
*sock
;
1738 error
= nl_pool_alloc(protocol
, &sock
);
1740 nl_sock_transact_multiple(sock
, transactions
, n
);
1741 nl_pool_release(sock
);
1743 nl_sock_record_errors__(transactions
, n
, error
);
1749 nl_sock_allocate_seq(struct nl_sock
*sock
, unsigned int n
)
1751 uint32_t seq
= sock
->next_seq
;
1753 sock
->next_seq
+= n
;
1755 /* Make it impossible for the next request for sequence numbers to wrap
1756 * around to 0. Start over with 1 to avoid ever using a sequence number of
1757 * 0, because the kernel uses sequence number 0 for notifications. */
1758 if (sock
->next_seq
>= UINT32_MAX
/ 2) {
1766 nlmsghdr_to_string(const struct nlmsghdr
*h
, int protocol
, struct ds
*ds
)
1772 static const struct nlmsg_flag flags
[] = {
1773 { NLM_F_REQUEST
, "REQUEST" },
1774 { NLM_F_MULTI
, "MULTI" },
1775 { NLM_F_ACK
, "ACK" },
1776 { NLM_F_ECHO
, "ECHO" },
1777 { NLM_F_DUMP
, "DUMP" },
1778 { NLM_F_ROOT
, "ROOT" },
1779 { NLM_F_MATCH
, "MATCH" },
1780 { NLM_F_ATOMIC
, "ATOMIC" },
1782 const struct nlmsg_flag
*flag
;
1783 uint16_t flags_left
;
1785 ds_put_format(ds
, "nl(len:%"PRIu32
", type=%"PRIu16
,
1786 h
->nlmsg_len
, h
->nlmsg_type
);
1787 if (h
->nlmsg_type
== NLMSG_NOOP
) {
1788 ds_put_cstr(ds
, "(no-op)");
1789 } else if (h
->nlmsg_type
== NLMSG_ERROR
) {
1790 ds_put_cstr(ds
, "(error)");
1791 } else if (h
->nlmsg_type
== NLMSG_DONE
) {
1792 ds_put_cstr(ds
, "(done)");
1793 } else if (h
->nlmsg_type
== NLMSG_OVERRUN
) {
1794 ds_put_cstr(ds
, "(overrun)");
1795 } else if (h
->nlmsg_type
< NLMSG_MIN_TYPE
) {
1796 ds_put_cstr(ds
, "(reserved)");
1797 } else if (protocol
== NETLINK_GENERIC
) {
1798 ds_put_format(ds
, "(%s)", genl_family_to_name(h
->nlmsg_type
));
1800 ds_put_cstr(ds
, "(family-defined)");
1802 ds_put_format(ds
, ", flags=%"PRIx16
, h
->nlmsg_flags
);
1803 flags_left
= h
->nlmsg_flags
;
1804 for (flag
= flags
; flag
< &flags
[ARRAY_SIZE(flags
)]; flag
++) {
1805 if ((flags_left
& flag
->bits
) == flag
->bits
) {
1806 ds_put_format(ds
, "[%s]", flag
->name
);
1807 flags_left
&= ~flag
->bits
;
1811 ds_put_format(ds
, "[OTHER:%"PRIx16
"]", flags_left
);
1813 ds_put_format(ds
, ", seq=%"PRIx32
", pid=%"PRIu32
,
1814 h
->nlmsg_seq
, h
->nlmsg_pid
);
1818 nlmsg_to_string(const struct ofpbuf
*buffer
, int protocol
)
1820 struct ds ds
= DS_EMPTY_INITIALIZER
;
1821 const struct nlmsghdr
*h
= ofpbuf_at(buffer
, 0, NLMSG_HDRLEN
);
1823 nlmsghdr_to_string(h
, protocol
, &ds
);
1824 if (h
->nlmsg_type
== NLMSG_ERROR
) {
1825 const struct nlmsgerr
*e
;
1826 e
= ofpbuf_at(buffer
, NLMSG_HDRLEN
,
1827 NLMSG_ALIGN(sizeof(struct nlmsgerr
)));
1829 ds_put_format(&ds
, " error(%d", e
->error
);
1831 ds_put_format(&ds
, "(%s)", ovs_strerror(-e
->error
));
1833 ds_put_cstr(&ds
, ", in-reply-to(");
1834 nlmsghdr_to_string(&e
->msg
, protocol
, &ds
);
1835 ds_put_cstr(&ds
, "))");
1837 ds_put_cstr(&ds
, " error(truncated)");
1839 } else if (h
->nlmsg_type
== NLMSG_DONE
) {
1840 int *error
= ofpbuf_at(buffer
, NLMSG_HDRLEN
, sizeof *error
);
1842 ds_put_format(&ds
, " done(%d", *error
);
1844 ds_put_format(&ds
, "(%s)", ovs_strerror(-*error
));
1846 ds_put_cstr(&ds
, ")");
1848 ds_put_cstr(&ds
, " done(truncated)");
1850 } else if (protocol
== NETLINK_GENERIC
) {
1851 struct genlmsghdr
*genl
= nl_msg_genlmsghdr(buffer
);
1853 ds_put_format(&ds
, ",genl(cmd=%"PRIu8
",version=%"PRIu8
")",
1854 genl
->cmd
, genl
->version
);
1858 ds_put_cstr(&ds
, "nl(truncated)");
1864 log_nlmsg(const char *function
, int error
,
1865 const void *message
, size_t size
, int protocol
)
1867 if (!VLOG_IS_DBG_ENABLED()) {
1871 struct ofpbuf buffer
= ofpbuf_const_initializer(message
, size
);
1872 char *nlmsg
= nlmsg_to_string(&buffer
, protocol
);
1873 VLOG_DBG_RL(&rl
, "%s (%s): %s", function
, ovs_strerror(error
), nlmsg
);