]> git.proxmox.com Git - mirror_ovs.git/blame - lib/netlink-socket.c
Use new default nb and sb dbs for sandbox northd:
[mirror_ovs.git] / lib / netlink-socket.c
CommitLineData
2fe27d5a 1/*
0a2869d5 2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2016 Nicira, Inc.
2fe27d5a
BP
3 *
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:
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
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.
15 */
16
17#include <config.h>
18#include "netlink-socket.h"
2fe27d5a
BP
19#include <errno.h>
20#include <inttypes.h>
21#include <stdlib.h>
22#include <sys/types.h>
cc75061a 23#include <sys/uio.h>
2fe27d5a
BP
24#include <unistd.h>
25#include "coverage.h"
3e8a2ad1 26#include "openvswitch/dynamic-string.h"
2ad204c8 27#include "hash.h"
ee89ea7b 28#include "openvswitch/hmap.h"
2fe27d5a
BP
29#include "netlink.h"
30#include "netlink-protocol.h"
886dd35a 31#include "odp-netlink.h"
64c96779 32#include "openvswitch/ofpbuf.h"
0bd01224 33#include "ovs-thread.h"
fd016ae3 34#include "openvswitch/poll-loop.h"
0672776e 35#include "seq.h"
6b7c12fd 36#include "socket-util.h"
cc75061a 37#include "util.h"
e6211adc 38#include "openvswitch/vlog.h"
2fe27d5a
BP
39
40VLOG_DEFINE_THIS_MODULE(netlink_socket);
41
42COVERAGE_DEFINE(netlink_overflow);
43COVERAGE_DEFINE(netlink_received);
fc999dda 44COVERAGE_DEFINE(netlink_recv_jumbo);
2fe27d5a
BP
45COVERAGE_DEFINE(netlink_sent);
46
47/* Linux header file confusion causes this to be undefined. */
48#ifndef SOL_NETLINK
49#define SOL_NETLINK 270
50#endif
51
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. */
55static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
56
7d7447df 57static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
2fe27d5a 58static void log_nlmsg(const char *function, int error,
7041c3a9 59 const void *message, size_t size, int protocol);
886dd35a 60#ifdef _WIN32
ebac7fb7 61static int get_sock_pid_from_kernel(struct nl_sock *sock);
e70f55ed 62static int set_sock_property(struct nl_sock *sock);
74804b21
AS
63static int nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
64 struct ofpbuf **replyp);
d553dd54
AS
65
66/* In the case DeviceIoControl failed and GetLastError returns with
67 * ERROR_NOT_FOUND means we lost communication with the kernel device.
68 * CloseHandle will fail because the handle in 'theory' does not exist.
69 * The only remaining option is to crash and allow the service to be restarted
70 * via service manager. This is the only way to close the handle from both
71 * userspace and kernel. */
72void
73lost_communication(DWORD last_err)
74{
75 if (last_err == ERROR_NOT_FOUND) {
76 ovs_abort(0, "lost communication with the kernel device");
77 }
78}
886dd35a 79#endif
2fe27d5a
BP
80\f
81/* Netlink sockets. */
82
0d121c73 83struct nl_sock {
22326ba6
AS
84#ifdef _WIN32
85 HANDLE handle;
7fa09611 86 OVERLAPPED overlapped;
64513e68 87 DWORD read_ioctl;
22326ba6 88#else
2fe27d5a 89 int fd;
22326ba6 90#endif
7d7447df 91 uint32_t next_seq;
2fe27d5a 92 uint32_t pid;
7041c3a9 93 int protocol;
cc75061a 94 unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
2fe27d5a
BP
95};
96
cc75061a
BP
97/* Compile-time limit on iovecs, so that we can allocate a maximum-size array
98 * of iovecs on the stack. */
99#define MAX_IOVS 128
100
101/* Maximum number of iovecs that may be passed to sendmsg, capped at a
102 * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
103 *
104 * Initialized by nl_sock_create(). */
105static int max_iovs;
106
a88b4e04
BP
107static int nl_pool_alloc(int protocol, struct nl_sock **sockp);
108static void nl_pool_release(struct nl_sock *);
2fe27d5a
BP
109
110/* Creates a new netlink socket for the given netlink 'protocol'
111 * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the
a88b4e04 112 * new socket if successful, otherwise returns a positive errno value. */
2fe27d5a 113int
cceb11f5 114nl_sock_create(int protocol, struct nl_sock **sockp)
2fe27d5a 115{
0bd01224 116 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
2fe27d5a 117 struct nl_sock *sock;
22326ba6 118#ifndef _WIN32
2fe27d5a 119 struct sockaddr_nl local, remote;
22326ba6 120#endif
2c5a6834 121 socklen_t local_size;
d2b9f5b0 122 int rcvbuf;
2fe27d5a
BP
123 int retval = 0;
124
0bd01224 125 if (ovsthread_once_start(&once)) {
cc75061a
BP
126 int save_errno = errno;
127 errno = 0;
128
129 max_iovs = sysconf(_SC_UIO_MAXIOV);
130 if (max_iovs < _XOPEN_IOV_MAX) {
131 if (max_iovs == -1 && errno) {
10a89ef0 132 VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno));
cc75061a
BP
133 }
134 max_iovs = _XOPEN_IOV_MAX;
135 } else if (max_iovs > MAX_IOVS) {
136 max_iovs = MAX_IOVS;
137 }
138
139 errno = save_errno;
0bd01224 140 ovsthread_once_done(&once);
cc75061a
BP
141 }
142
2fe27d5a 143 *sockp = NULL;
488232b7 144 sock = xmalloc(sizeof *sock);
2fe27d5a 145
22326ba6 146#ifdef _WIN32
e6b298ef 147 sock->overlapped.hEvent = NULL;
7fa09611
EE
148 sock->handle = CreateFile(OVS_DEVICE_NAME_USER,
149 GENERIC_READ | GENERIC_WRITE,
150 FILE_SHARE_READ | FILE_SHARE_WRITE,
151 NULL, OPEN_EXISTING,
152 FILE_FLAG_OVERLAPPED, NULL);
22326ba6
AS
153
154 if (sock->handle == INVALID_HANDLE_VALUE) {
52a1540a 155 VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
7fa09611
EE
156 goto error;
157 }
158
159 memset(&sock->overlapped, 0, sizeof sock->overlapped);
160 sock->overlapped.hEvent = CreateEvent(NULL, FALSE, FALSE, NULL);
161 if (sock->overlapped.hEvent == NULL) {
52a1540a 162 VLOG_ERR("fcntl: %s", ovs_lasterror_to_string());
22326ba6
AS
163 goto error;
164 }
64513e68
EE
165 /* Initialize the type/ioctl to Generic */
166 sock->read_ioctl = OVS_IOCTL_READ;
22326ba6 167#else
2fe27d5a
BP
168 sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol);
169 if (sock->fd < 0) {
10a89ef0 170 VLOG_ERR("fcntl: %s", ovs_strerror(errno));
2fe27d5a
BP
171 goto error;
172 }
22326ba6
AS
173#endif
174
7041c3a9 175 sock->protocol = protocol;
7d7447df 176 sock->next_seq = 1;
2fe27d5a 177
d2b9f5b0 178 rcvbuf = 1024 * 1024;
22326ba6
AS
179#ifdef _WIN32
180 sock->rcvbuf = rcvbuf;
ebac7fb7 181 retval = get_sock_pid_from_kernel(sock);
886dd35a
NR
182 if (retval != 0) {
183 goto error;
184 }
e70f55ed
SV
185 retval = set_sock_property(sock);
186 if (retval != 0) {
187 goto error;
188 }
22326ba6 189#else
d2b9f5b0
BP
190 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
191 &rcvbuf, sizeof rcvbuf)) {
80af5ee5
BP
192 /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM.
193 * Warn only if the failure is therefore unexpected. */
f28b6dd3 194 if (errno != EPERM) {
80af5ee5 195 VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed "
10a89ef0 196 "(%s)", rcvbuf, ovs_strerror(errno));
80af5ee5 197 }
d2b9f5b0
BP
198 }
199
cc75061a
BP
200 retval = get_socket_rcvbuf(sock->fd);
201 if (retval < 0) {
202 retval = -retval;
203 goto error;
204 }
205 sock->rcvbuf = retval;
2e467950 206 retval = 0;
cc75061a 207
2c5a6834 208 /* Connect to kernel (pid 0) as remote address. */
2fe27d5a
BP
209 memset(&remote, 0, sizeof remote);
210 remote.nl_family = AF_NETLINK;
211 remote.nl_pid = 0;
212 if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
10a89ef0 213 VLOG_ERR("connect(0): %s", ovs_strerror(errno));
2c5a6834
BP
214 goto error;
215 }
216
217 /* Obtain pid assigned by kernel. */
218 local_size = sizeof local;
219 if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
10a89ef0 220 VLOG_ERR("getsockname: %s", ovs_strerror(errno));
2c5a6834
BP
221 goto error;
222 }
223 if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
224 VLOG_ERR("getsockname returned bad Netlink name");
225 retval = EINVAL;
226 goto error;
2fe27d5a 227 }
2c5a6834 228 sock->pid = local.nl_pid;
22326ba6 229#endif
2fe27d5a 230
2fe27d5a
BP
231 *sockp = sock;
232 return 0;
233
2fe27d5a
BP
234error:
235 if (retval == 0) {
236 retval = errno;
237 if (retval == 0) {
238 retval = EINVAL;
239 }
240 }
22326ba6 241#ifdef _WIN32
7fa09611
EE
242 if (sock->overlapped.hEvent) {
243 CloseHandle(sock->overlapped.hEvent);
244 }
22326ba6
AS
245 if (sock->handle != INVALID_HANDLE_VALUE) {
246 CloseHandle(sock->handle);
247 }
248#else
2fe27d5a
BP
249 if (sock->fd >= 0) {
250 close(sock->fd);
251 }
22326ba6 252#endif
2fe27d5a
BP
253 free(sock);
254 return retval;
255}
256
c6eab56d
BP
257/* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and
258 * sets '*sockp' to the new socket if successful, otherwise returns a positive
259 * errno value. */
260int
261nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp)
262{
263 return nl_sock_create(src->protocol, sockp);
264}
265
2fe27d5a
BP
266/* Destroys netlink socket 'sock'. */
267void
268nl_sock_destroy(struct nl_sock *sock)
269{
270 if (sock) {
22326ba6 271#ifdef _WIN32
7fa09611
EE
272 if (sock->overlapped.hEvent) {
273 CloseHandle(sock->overlapped.hEvent);
274 }
22326ba6
AS
275 CloseHandle(sock->handle);
276#else
a88b4e04 277 close(sock->fd);
22326ba6 278#endif
a88b4e04 279 free(sock);
2fe27d5a
BP
280 }
281}
282
886dd35a
NR
283#ifdef _WIN32
284/* Reads the pid for 'sock' generated in the kernel datapath. The function
190cf533
SV
285 * uses a separate IOCTL instead of a transaction semantic to avoid unnecessary
286 * message overhead. */
886dd35a 287static int
ebac7fb7 288get_sock_pid_from_kernel(struct nl_sock *sock)
886dd35a 289{
190cf533
SV
290 uint32_t pid = 0;
291 int retval = 0;
292 DWORD bytes = 0;
886dd35a 293
190cf533
SV
294 if (!DeviceIoControl(sock->handle, OVS_IOCTL_GET_PID,
295 NULL, 0, &pid, sizeof(pid),
886dd35a 296 &bytes, NULL)) {
d553dd54 297 lost_communication(GetLastError());
886dd35a 298 retval = EINVAL;
886dd35a 299 } else {
190cf533 300 if (bytes < sizeof(pid)) {
886dd35a 301 retval = EINVAL;
190cf533
SV
302 } else {
303 sock->pid = pid;
886dd35a 304 }
886dd35a 305 }
886dd35a 306
886dd35a
NR
307 return retval;
308}
e70f55ed
SV
309
310/* Used for setting and managing socket properties in userspace and kernel.
311 * Currently two attributes are tracked - pid and protocol
312 * protocol - supplied by userspace based on the netlink family. Windows uses
313 * this property to set the value in kernel datapath.
314 * eg: (NETLINK_GENERIC/ NETLINK_NETFILTER)
315 * pid - generated by windows kernel and set in userspace. The property
316 * is not modified.
317 * Also verify if Protocol and PID in Kernel reflects the values in userspace
318 * */
319static int
320set_sock_property(struct nl_sock *sock)
321{
322 static const struct nl_policy ovs_socket_policy[] = {
323 [OVS_NL_ATTR_SOCK_PROTO] = { .type = NL_A_BE32, .optional = true },
324 [OVS_NL_ATTR_SOCK_PID] = { .type = NL_A_BE32, .optional = true }
325 };
326
327 struct ofpbuf request, *reply;
328 struct ovs_header *ovs_header;
329 struct nlattr *attrs[ARRAY_SIZE(ovs_socket_policy)];
330 int retval = 0;
331 int error;
332
333 ofpbuf_init(&request, 0);
334 nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
335 OVS_CTRL_CMD_SOCK_PROP, OVS_WIN_CONTROL_VERSION);
336 ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
337 ovs_header->dp_ifindex = 0;
338
339 nl_msg_put_be32(&request, OVS_NL_ATTR_SOCK_PROTO, sock->protocol);
340 /* pid is already set as part of get_sock_pid_from_kernel()
341 * This is added to maintain consistency
342 */
343 nl_msg_put_be32(&request, OVS_NL_ATTR_SOCK_PID, sock->pid);
344
345 error = nl_sock_transact(sock, &request, &reply);
346 ofpbuf_uninit(&request);
347 if (error) {
348 retval = EINVAL;
349 }
350
351 if (!nl_policy_parse(reply,
352 NLMSG_HDRLEN + GENL_HDRLEN + sizeof *ovs_header,
353 ovs_socket_policy, attrs,
354 ARRAY_SIZE(ovs_socket_policy))) {
355 ofpbuf_delete(reply);
356 retval = EINVAL;
357 }
358 /* Verify if the properties are setup properly */
359 if (attrs[OVS_NL_ATTR_SOCK_PROTO]) {
360 int protocol = nl_attr_get_be32(attrs[OVS_NL_ATTR_SOCK_PROTO]);
361 if (protocol != sock->protocol) {
362 VLOG_ERR("Invalid protocol returned:%d expected:%d",
363 protocol, sock->protocol);
364 retval = EINVAL;
365 }
366 }
367
368 if (attrs[OVS_NL_ATTR_SOCK_PID]) {
369 int pid = nl_attr_get_be32(attrs[OVS_NL_ATTR_SOCK_PID]);
370 if (pid != sock->pid) {
371 VLOG_ERR("Invalid pid returned:%d expected:%d",
372 pid, sock->pid);
373 retval = EINVAL;
374 }
375 }
376
377 return retval;
378}
886dd35a
NR
379#endif /* _WIN32 */
380
b8f958ea
EE
381#ifdef _WIN32
382static int __inline
383nl_sock_mcgroup(struct nl_sock *sock, unsigned int multicast_group, bool join)
384{
385 struct ofpbuf request;
386 uint64_t request_stub[128];
387 struct ovs_header *ovs_header;
388 struct nlmsghdr *nlmsg;
389 int error;
390
391 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
392
393 nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
394 OVS_CTRL_CMD_MC_SUBSCRIBE_REQ,
395 OVS_WIN_CONTROL_VERSION);
396
397 ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
398 ovs_header->dp_ifindex = 0;
399
400 nl_msg_put_u32(&request, OVS_NL_ATTR_MCAST_GRP, multicast_group);
401 nl_msg_put_u8(&request, OVS_NL_ATTR_MCAST_JOIN, join ? 1 : 0);
402
403 error = nl_sock_send(sock, &request, true);
404 ofpbuf_uninit(&request);
405 return error;
406}
407#endif
cceb11f5
BP
408/* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if
409 * successful, otherwise a positive errno value.
410 *
a838c4fe
BP
411 * A socket that is subscribed to a multicast group that receives asynchronous
412 * notifications must not be used for Netlink transactions or dumps, because
413 * transactions and dumps can cause notifications to be lost.
414 *
cceb11f5
BP
415 * Multicast group numbers are always positive.
416 *
417 * It is not an error to attempt to join a multicast group to which a socket
418 * already belongs. */
419int
420nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
421{
22326ba6 422#ifdef _WIN32
64513e68
EE
423 /* Set the socket type as a "multicast" socket */
424 sock->read_ioctl = OVS_IOCTL_READ_EVENT;
b8f958ea
EE
425 int error = nl_sock_mcgroup(sock, multicast_group, true);
426 if (error) {
64513e68 427 sock->read_ioctl = OVS_IOCTL_READ;
b8f958ea 428 VLOG_WARN("could not join multicast group %u (%s)",
15fd9052
NR
429 multicast_group, ovs_strerror(error));
430 return error;
b8f958ea 431 }
22326ba6 432#else
cceb11f5
BP
433 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP,
434 &multicast_group, sizeof multicast_group) < 0) {
435 VLOG_WARN("could not join multicast group %u (%s)",
10a89ef0 436 multicast_group, ovs_strerror(errno));
cceb11f5
BP
437 return errno;
438 }
22326ba6 439#endif
cceb11f5
BP
440 return 0;
441}
442
36791e21 443#ifdef _WIN32
74804b21
AS
444int
445nl_sock_subscribe_packet__(struct nl_sock *sock, bool subscribe)
446{
447 struct ofpbuf request;
448 uint64_t request_stub[128];
449 struct ovs_header *ovs_header;
450 struct nlmsghdr *nlmsg;
451 int error;
452
453 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
454 nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
455 OVS_CTRL_CMD_PACKET_SUBSCRIBE_REQ,
456 OVS_WIN_CONTROL_VERSION);
457
458 ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
459 ovs_header->dp_ifindex = 0;
460 nl_msg_put_u8(&request, OVS_NL_ATTR_PACKET_SUBSCRIBE, subscribe ? 1 : 0);
461 nl_msg_put_u32(&request, OVS_NL_ATTR_PACKET_PID, sock->pid);
462
463 error = nl_sock_send(sock, &request, true);
464 ofpbuf_uninit(&request);
465 return error;
466}
467
36791e21
NR
468int
469nl_sock_subscribe_packets(struct nl_sock *sock)
470{
471 int error;
472
473 if (sock->read_ioctl != OVS_IOCTL_READ) {
474 return EINVAL;
475 }
476
477 error = nl_sock_subscribe_packet__(sock, true);
478 if (error) {
d8d1ef2f
AS
479 VLOG_WARN("could not subscribe packets (%s)",
480 ovs_strerror(error));
36791e21
NR
481 return error;
482 }
483 sock->read_ioctl = OVS_IOCTL_READ_PACKET;
484
485 return 0;
486}
487
488int
489nl_sock_unsubscribe_packets(struct nl_sock *sock)
490{
491 ovs_assert(sock->read_ioctl == OVS_IOCTL_READ_PACKET);
492
493 int error = nl_sock_subscribe_packet__(sock, false);
494 if (error) {
d8d1ef2f
AS
495 VLOG_WARN("could not unsubscribe to packets (%s)",
496 ovs_strerror(error));
36791e21
NR
497 return error;
498 }
499
500 sock->read_ioctl = OVS_IOCTL_READ;
501 return 0;
502}
36791e21
NR
503#endif
504
cceb11f5
BP
505/* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if
506 * successful, otherwise a positive errno value.
507 *
508 * Multicast group numbers are always positive.
509 *
510 * It is not an error to attempt to leave a multicast group to which a socket
511 * does not belong.
512 *
513 * On success, reading from 'sock' will still return any messages that were
514 * received on 'multicast_group' before the group was left. */
515int
516nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group)
517{
22326ba6 518#ifdef _WIN32
b8f958ea
EE
519 int error = nl_sock_mcgroup(sock, multicast_group, false);
520 if (error) {
521 VLOG_WARN("could not leave multicast group %u (%s)",
15fd9052
NR
522 multicast_group, ovs_strerror(error));
523 return error;
b8f958ea 524 }
64513e68 525 sock->read_ioctl = OVS_IOCTL_READ;
22326ba6 526#else
cceb11f5
BP
527 if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP,
528 &multicast_group, sizeof multicast_group) < 0) {
529 VLOG_WARN("could not leave multicast group %u (%s)",
10a89ef0 530 multicast_group, ovs_strerror(errno));
cceb11f5
BP
531 return errno;
532 }
22326ba6 533#endif
cceb11f5
BP
534 return 0;
535}
536
c6eab56d 537static int
ff459dd6
BP
538nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
539 uint32_t nlmsg_seq, bool wait)
2fe27d5a
BP
540{
541 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
542 int error;
543
6fd6ed71 544 nlmsg->nlmsg_len = msg->size;
ff459dd6 545 nlmsg->nlmsg_seq = nlmsg_seq;
2fe27d5a
BP
546 nlmsg->nlmsg_pid = sock->pid;
547 do {
548 int retval;
22326ba6 549#ifdef _WIN32
fd972eb8
NR
550 DWORD bytes;
551
552 if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
6fd6ed71 553 msg->data, msg->size, NULL, 0,
0fd22ae2 554 &bytes, NULL)) {
d553dd54 555 lost_communication(GetLastError());
22326ba6 556 retval = -1;
fd972eb8
NR
557 /* XXX: Map to a more appropriate error based on GetLastError(). */
558 errno = EINVAL;
a51a5086 559 VLOG_DBG_RL(&rl, "fatal driver failure in write: %s",
d553dd54 560 ovs_lasterror_to_string());
fd972eb8 561 } else {
6fd6ed71 562 retval = msg->size;
22326ba6
AS
563 }
564#else
6fd6ed71 565 retval = send(sock->fd, msg->data, msg->size,
fd972eb8 566 wait ? 0 : MSG_DONTWAIT);
22326ba6 567#endif
2fe27d5a
BP
568 error = retval < 0 ? errno : 0;
569 } while (error == EINTR);
6fd6ed71 570 log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol);
2fe27d5a
BP
571 if (!error) {
572 COVERAGE_INC(netlink_sent);
573 }
574 return error;
575}
576
c6eab56d 577/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
6fd6ed71 578 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
ff459dd6
BP
579 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
580 * sequence number, before the message is sent.
c6eab56d
BP
581 *
582 * Returns 0 if successful, otherwise a positive errno value. If
583 * 'wait' is true, then the send will wait until buffer space is ready;
584 * otherwise, returns EAGAIN if the 'sock' send buffer is full. */
585int
586nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
ff459dd6
BP
587{
588 return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
589}
590
591/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
6fd6ed71 592 * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
ff459dd6
BP
593 * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
594 * 'nlmsg_seq', before the message is sent.
595 *
596 * Returns 0 if successful, otherwise a positive errno value. If
597 * 'wait' is true, then the send will wait until buffer space is ready;
598 * otherwise, returns EAGAIN if the 'sock' send buffer is full.
599 *
600 * This function is suitable for sending a reply to a request that was received
601 * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
602int
603nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
604 uint32_t nlmsg_seq, bool wait)
c6eab56d 605{
ff459dd6 606 return nl_sock_send__(sock, msg, nlmsg_seq, wait);
c6eab56d
BP
607}
608
c6eab56d 609static int
72d32ac0 610nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
2fe27d5a 611{
72d32ac0
BP
612 /* We can't accurately predict the size of the data to be received. The
613 * caller is supposed to have allocated enough space in 'buf' to handle the
614 * "typical" case. To handle exceptions, we make available enough space in
615 * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
616 * figure since that's the maximum length of a Netlink attribute). */
2fe27d5a 617 struct nlmsghdr *nlmsghdr;
72d32ac0 618 uint8_t tail[65536];
fc999dda 619 struct iovec iov[2];
fc999dda
BP
620 struct msghdr msg;
621 ssize_t retval;
8f20fd98 622 int error;
fc999dda 623
cb22974d 624 ovs_assert(buf->allocated >= sizeof *nlmsghdr);
72d32ac0 625 ofpbuf_clear(buf);
2fe27d5a 626
6fd6ed71 627 iov[0].iov_base = buf->base;
72d32ac0 628 iov[0].iov_len = buf->allocated;
fc999dda 629 iov[1].iov_base = tail;
72d32ac0 630 iov[1].iov_len = sizeof tail;
fc999dda
BP
631
632 memset(&msg, 0, sizeof msg);
633 msg.msg_iov = iov;
634 msg.msg_iovlen = 2;
635
8f20fd98
BP
636 /* Receive a Netlink message from the kernel.
637 *
638 * This works around a kernel bug in which the kernel returns an error code
639 * as if it were the number of bytes read. It doesn't actually modify
640 * anything in the receive buffer in that case, so we can initialize the
641 * Netlink header with an impossible message length and then, upon success,
642 * check whether it changed. */
6fd6ed71 643 nlmsghdr = buf->base;
2fe27d5a 644 do {
8f20fd98 645 nlmsghdr->nlmsg_len = UINT32_MAX;
22326ba6 646#ifdef _WIN32
fd972eb8 647 DWORD bytes;
64513e68 648 if (!DeviceIoControl(sock->handle, sock->read_ioctl,
fd972eb8 649 NULL, 0, tail, sizeof tail, &bytes, NULL)) {
d553dd54 650 lost_communication(GetLastError());
a51a5086 651 VLOG_DBG_RL(&rl, "fatal driver failure in transact: %s",
d71c423e 652 ovs_lasterror_to_string());
22326ba6 653 retval = -1;
a51a5086 654 /* XXX: Map to a more appropriate error. */
fd972eb8 655 errno = EINVAL;
22326ba6 656 } else {
fd972eb8
NR
657 retval = bytes;
658 if (retval == 0) {
659 retval = -1;
660 errno = EAGAIN;
661 } else {
662 if (retval >= buf->allocated) {
663 ofpbuf_reinit(buf, retval);
6fd6ed71 664 nlmsghdr = buf->base;
15fd9052 665 nlmsghdr->nlmsg_len = UINT32_MAX;
fd972eb8 666 }
6fd6ed71
PS
667 memcpy(buf->data, tail, retval);
668 buf->size = retval;
fd972eb8 669 }
22326ba6
AS
670 }
671#else
fc999dda 672 retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
22326ba6 673#endif
8f20fd98
BP
674 error = (retval < 0 ? errno
675 : retval == 0 ? ECONNRESET /* not possible? */
676 : nlmsghdr->nlmsg_len != UINT32_MAX ? 0
7f8e2646 677 : retval);
8f20fd98
BP
678 } while (error == EINTR);
679 if (error) {
fc999dda
BP
680 if (error == ENOBUFS) {
681 /* Socket receive buffer overflow dropped one or more messages that
682 * the kernel tried to send to us. */
683 COVERAGE_INC(netlink_overflow);
684 }
fc999dda 685 return error;
2fe27d5a 686 }
fc999dda 687
2fe27d5a 688 if (msg.msg_flags & MSG_TRUNC) {
34582733 689 VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)",
72d32ac0 690 sizeof tail);
fc999dda 691 return E2BIG;
2fe27d5a 692 }
2fe27d5a 693
fc999dda 694 if (retval < sizeof *nlmsghdr
2fe27d5a 695 || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
fc999dda 696 || nlmsghdr->nlmsg_len > retval) {
e5e4b47c 697 VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE" bytes < %"PRIuSIZE")",
72d32ac0 698 retval, sizeof *nlmsghdr);
2fe27d5a
BP
699 return EPROTO;
700 }
22326ba6 701#ifndef _WIN32
6fd6ed71 702 buf->size = MIN(retval, buf->allocated);
72d32ac0
BP
703 if (retval > buf->allocated) {
704 COVERAGE_INC(netlink_recv_jumbo);
705 ofpbuf_put(buf, tail, retval - buf->allocated);
706 }
22326ba6 707#endif
72d32ac0 708
6fd6ed71 709 log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
2fe27d5a
BP
710 COVERAGE_INC(netlink_received);
711
712 return 0;
713}
714
72d32ac0
BP
715/* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
716 * 'wait' is true, waits for a message to be ready. Otherwise, fails with
717 * EAGAIN if the 'sock' receive buffer is empty.
718 *
719 * The caller must have initialized 'buf' with an allocation of at least
720 * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
721 * space for a "typical" message.
722 *
723 * On success, returns 0 and replaces 'buf''s previous content by the received
724 * message. This function expands 'buf''s allocated memory, as necessary, to
725 * hold the actual size of the received message.
c6eab56d 726 *
72d32ac0
BP
727 * On failure, returns a positive errno value and clears 'buf' to zero length.
728 * 'buf' retains its previous memory allocation.
729 *
730 * Regardless of success or failure, this function resets 'buf''s headroom to
731 * 0. */
c6eab56d 732int
72d32ac0 733nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
c6eab56d 734{
72d32ac0 735 return nl_sock_recv__(sock, buf, wait);
cc75061a
BP
736}
737
738static void
739nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
740 int error)
741{
742 size_t i;
743
744 for (i = 0; i < n; i++) {
72d32ac0
BP
745 struct nl_transaction *txn = transactions[i];
746
747 txn->error = error;
748 if (txn->reply) {
749 ofpbuf_clear(txn->reply);
750 }
cc75061a
BP
751 }
752}
753
754static int
755nl_sock_transact_multiple__(struct nl_sock *sock,
756 struct nl_transaction **transactions, size_t n,
757 size_t *done)
758{
72d32ac0
BP
759 uint64_t tmp_reply_stub[1024 / 8];
760 struct nl_transaction tmp_txn;
761 struct ofpbuf tmp_reply;
762
763 uint32_t base_seq;
cc75061a
BP
764 struct iovec iovs[MAX_IOVS];
765 struct msghdr msg;
766 int error;
767 int i;
768
72d32ac0 769 base_seq = nl_sock_allocate_seq(sock, n);
cc75061a
BP
770 *done = 0;
771 for (i = 0; i < n; i++) {
72d32ac0
BP
772 struct nl_transaction *txn = transactions[i];
773 struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
cc75061a 774
6fd6ed71 775 nlmsg->nlmsg_len = txn->request->size;
72d32ac0 776 nlmsg->nlmsg_seq = base_seq + i;
cc75061a 777 nlmsg->nlmsg_pid = sock->pid;
cc75061a 778
6fd6ed71
PS
779 iovs[i].iov_base = txn->request->data;
780 iovs[i].iov_len = txn->request->size;
cc75061a
BP
781 }
782
0fd22ae2 783#ifndef _WIN32
cc75061a
BP
784 memset(&msg, 0, sizeof msg);
785 msg.msg_iov = iovs;
786 msg.msg_iovlen = n;
787 do {
788 error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
789 } while (error == EINTR);
790
791 for (i = 0; i < n; i++) {
72d32ac0 792 struct nl_transaction *txn = transactions[i];
cc75061a 793
6fd6ed71
PS
794 log_nlmsg(__func__, error, txn->request->data,
795 txn->request->size, sock->protocol);
cc75061a
BP
796 }
797 if (!error) {
798 COVERAGE_ADD(netlink_sent, n);
799 }
800
801 if (error) {
802 return error;
803 }
804
72d32ac0
BP
805 ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
806 tmp_txn.request = NULL;
807 tmp_txn.reply = &tmp_reply;
808 tmp_txn.error = 0;
cc75061a 809 while (n > 0) {
72d32ac0
BP
810 struct nl_transaction *buf_txn, *txn;
811 uint32_t seq;
812
813 /* Find a transaction whose buffer we can use for receiving a reply.
814 * If no such transaction is left, use tmp_txn. */
815 buf_txn = &tmp_txn;
816 for (i = 0; i < n; i++) {
817 if (transactions[i]->reply) {
818 buf_txn = transactions[i];
819 break;
820 }
821 }
cc75061a 822
72d32ac0
BP
823 /* Receive a reply. */
824 error = nl_sock_recv__(sock, buf_txn->reply, false);
825 if (error) {
826 if (error == EAGAIN) {
827 nl_sock_record_errors__(transactions, n, 0);
828 *done += n;
829 error = 0;
830 }
831 break;
cc75061a
BP
832 }
833
72d32ac0
BP
834 /* Match the reply up with a transaction. */
835 seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
836 if (seq < base_seq || seq >= base_seq + n) {
837 VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
cc75061a
BP
838 continue;
839 }
72d32ac0
BP
840 i = seq - base_seq;
841 txn = transactions[i];
cc75061a 842
72d32ac0
BP
843 /* Fill in the results for 'txn'. */
844 if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
845 if (txn->reply) {
846 ofpbuf_clear(txn->reply);
847 }
848 if (txn->error) {
cc75061a 849 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
10a89ef0 850 error, ovs_strerror(txn->error));
cc75061a 851 }
cc75061a 852 } else {
72d32ac0
BP
853 txn->error = 0;
854 if (txn->reply && txn != buf_txn) {
855 /* Swap buffers. */
856 struct ofpbuf *reply = buf_txn->reply;
857 buf_txn->reply = txn->reply;
858 txn->reply = reply;
859 }
cc75061a
BP
860 }
861
72d32ac0
BP
862 /* Fill in the results for transactions before 'txn'. (We have to do
863 * this after the results for 'txn' itself because of the buffer swap
864 * above.) */
865 nl_sock_record_errors__(transactions, i, 0);
866
867 /* Advance. */
cc75061a
BP
868 *done += i + 1;
869 transactions += i + 1;
870 n -= i + 1;
72d32ac0 871 base_seq += i + 1;
cc75061a 872 }
72d32ac0 873 ofpbuf_uninit(&tmp_reply);
0fd22ae2
NR
874#else
875 error = 0;
9189184d 876 uint8_t reply_buf[65536];
0fd22ae2
NR
877 for (i = 0; i < n; i++) {
878 DWORD reply_len;
a51a5086 879 bool ret;
0fd22ae2
NR
880 struct nl_transaction *txn = transactions[i];
881 struct nlmsghdr *request_nlmsg, *reply_nlmsg;
882
a51a5086
NR
883 ret = DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT,
884 txn->request->data,
885 txn->request->size,
886 reply_buf, sizeof reply_buf,
887 &reply_len, NULL);
888
889 if (ret && reply_len == 0) {
890 /*
891 * The current transaction did not produce any data to read and that
892 * is not an error as such. Continue with the remainder of the
893 * transactions.
894 */
895 txn->error = 0;
896 if (txn->reply) {
897 ofpbuf_clear(txn->reply);
898 }
899 } else if (!ret) {
0fd22ae2 900 /* XXX: Map to a more appropriate error. */
d553dd54 901 lost_communication(GetLastError());
0fd22ae2 902 error = EINVAL;
b937e116
BP
903 VLOG_DBG_RL(&rl, "fatal driver failure: %s",
904 ovs_lasterror_to_string());
0fd22ae2
NR
905 break;
906 }
907
a51a5086 908 if (reply_len != 0) {
7e86fe82
AS
909 request_nlmsg = nl_msg_nlmsghdr(txn->request);
910
a51a5086
NR
911 if (reply_len < sizeof *reply_nlmsg) {
912 nl_sock_record_errors__(transactions, n, 0);
913 VLOG_DBG_RL(&rl, "insufficient length of reply %#"PRIu32
914 " for seq: %#"PRIx32, reply_len, request_nlmsg->nlmsg_seq);
915 break;
916 }
0fd22ae2 917
a51a5086 918 /* Validate the sequence number in the reply. */
a51a5086 919 reply_nlmsg = (struct nlmsghdr *)reply_buf;
0fd22ae2 920
a51a5086
NR
921 if (request_nlmsg->nlmsg_seq != reply_nlmsg->nlmsg_seq) {
922 ovs_assert(request_nlmsg->nlmsg_seq == reply_nlmsg->nlmsg_seq);
923 VLOG_DBG_RL(&rl, "mismatched seq request %#"PRIx32
924 ", reply %#"PRIx32, request_nlmsg->nlmsg_seq,
925 reply_nlmsg->nlmsg_seq);
926 break;
0fd22ae2 927 }
a51a5086
NR
928
929 /* Handle errors embedded within the netlink message. */
930 ofpbuf_use_stub(&tmp_reply, reply_buf, sizeof reply_buf);
931 tmp_reply.size = sizeof reply_buf;
932 if (nl_msg_nlmsgerr(&tmp_reply, &txn->error)) {
933 if (txn->reply) {
934 ofpbuf_clear(txn->reply);
935 }
936 if (txn->error) {
937 VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
938 error, ovs_strerror(txn->error));
939 }
940 } else {
941 txn->error = 0;
942 if (txn->reply) {
943 /* Copy the reply to the buffer specified by the caller. */
944 if (reply_len > txn->reply->allocated) {
945 ofpbuf_reinit(txn->reply, reply_len);
946 }
947 memcpy(txn->reply->data, reply_buf, reply_len);
948 txn->reply->size = reply_len;
0fd22ae2 949 }
0fd22ae2 950 }
a51a5086 951 ofpbuf_uninit(&tmp_reply);
0fd22ae2
NR
952 }
953
954 /* Count the number of successful transactions. */
955 (*done)++;
9189184d 956
0fd22ae2
NR
957 }
958
959 if (!error) {
960 COVERAGE_ADD(netlink_sent, n);
961 }
962#endif
cc75061a 963
72d32ac0 964 return error;
cc75061a
BP
965}
966
022ad2b9 967static void
cc75061a
BP
968nl_sock_transact_multiple(struct nl_sock *sock,
969 struct nl_transaction **transactions, size_t n)
970{
971 int max_batch_count;
972 int error;
973
974 if (!n) {
975 return;
976 }
977
cc75061a
BP
978 /* In theory, every request could have a 64 kB reply. But the default and
979 * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
980 * be a bit below 128 kB, so that would only allow a single message in a
981 * "batch". So we assume that replies average (at most) 4 kB, which allows
982 * a good deal of batching.
983 *
984 * In practice, most of the requests that we batch either have no reply at
985 * all or a brief reply. */
986 max_batch_count = MAX(sock->rcvbuf / 4096, 1);
987 max_batch_count = MIN(max_batch_count, max_iovs);
988
989 while (n > 0) {
990 size_t count, bytes;
991 size_t done;
992
993 /* Batch up to 'max_batch_count' transactions. But cap it at about a
994 * page of requests total because big skbuffs are expensive to
995 * allocate in the kernel. */
996#if defined(PAGESIZE)
997 enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
998#else
999 enum { MAX_BATCH_BYTES = 4096 - 512 };
1000#endif
6fd6ed71 1001 bytes = transactions[0]->request->size;
cc75061a 1002 for (count = 1; count < n && count < max_batch_count; count++) {
6fd6ed71 1003 if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
cc75061a
BP
1004 break;
1005 }
6fd6ed71 1006 bytes += transactions[count]->request->size;
cc75061a
BP
1007 }
1008
1009 error = nl_sock_transact_multiple__(sock, transactions, count, &done);
1010 transactions += done;
1011 n -= done;
1012
1013 if (error == ENOBUFS) {
1014 VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
1015 } else if (error) {
10a89ef0 1016 VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error));
cc75061a 1017 nl_sock_record_errors__(transactions, n, error);
b937e116
BP
1018 if (error != EAGAIN) {
1019 /* A fatal error has occurred. Abort the rest of
1020 * transactions. */
1021 break;
1022 }
cc75061a
BP
1023 }
1024 }
1025}
1026
022ad2b9 1027static int
cc75061a
BP
1028nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
1029 struct ofpbuf **replyp)
2fe27d5a 1030{
cc75061a
BP
1031 struct nl_transaction *transactionp;
1032 struct nl_transaction transaction;
2fe27d5a 1033
ebc56baa 1034 transaction.request = CONST_CAST(struct ofpbuf *, request);
72d32ac0 1035 transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
cc75061a 1036 transactionp = &transaction;
72d32ac0 1037
cc75061a 1038 nl_sock_transact_multiple(sock, &transactionp, 1);
72d32ac0 1039
2fe27d5a 1040 if (replyp) {
72d32ac0
BP
1041 if (transaction.error) {
1042 ofpbuf_delete(transaction.reply);
1043 *replyp = NULL;
1044 } else {
1045 *replyp = transaction.reply;
1046 }
2fe27d5a 1047 }
72d32ac0 1048
cc75061a 1049 return transaction.error;
2fe27d5a
BP
1050}
1051
6b7c12fd
BP
1052/* Drain all the messages currently in 'sock''s receive queue. */
1053int
1054nl_sock_drain(struct nl_sock *sock)
1055{
22326ba6
AS
1056#ifdef _WIN32
1057 return 0;
1058#else
6b7c12fd 1059 return drain_rcvbuf(sock->fd);
22326ba6 1060#endif
6b7c12fd
BP
1061}
1062
a88b4e04
BP
1063/* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a
1064 * Netlink socket created with the given 'protocol', and initializes 'dump' to
1065 * reflect the state of the operation.
2fe27d5a 1066 *
db1fc210
JS
1067 * 'request' must contain a Netlink message. Before sending the message,
1068 * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be
1069 * set to the Netlink socket's pid. NLM_F_DUMP and NLM_F_ACK will be set in
1070 * nlmsg_flags.
2fe27d5a 1071 *
a88b4e04 1072 * The design of this Netlink socket library ensures that the dump is reliable.
2fe27d5a 1073 *
db1fc210
JS
1074 * This function provides no status indication. nl_dump_done() provides an
1075 * error status for the entire dump operation.
2fe27d5a 1076 *
db1fc210 1077 * The caller must eventually destroy 'request'.
2fe27d5a
BP
1078 */
1079void
a88b4e04 1080nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request)
2fe27d5a 1081{
7d7447df 1082 nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
93295354
BP
1083
1084 ovs_mutex_init(&dump->mutex);
1085 ovs_mutex_lock(&dump->mutex);
1086 dump->status = nl_pool_alloc(protocol, &dump->sock);
1087 if (!dump->status) {
1088 dump->status = nl_sock_send__(dump->sock, request,
1089 nl_sock_allocate_seq(dump->sock, 1),
1090 true);
b2d1c78a 1091 }
9c8ad495 1092 dump->nl_seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
93295354
BP
1093 ovs_mutex_unlock(&dump->mutex);
1094}
1095
1096static int
1097nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer)
1098 OVS_REQUIRES(dump->mutex)
1099{
1100 struct nlmsghdr *nlmsghdr;
1101 int error;
1102
6fd6ed71 1103 while (!buffer->size) {
1738803a 1104 error = nl_sock_recv__(dump->sock, buffer, false);
93295354 1105 if (error) {
1738803a
AW
1106 /* The kernel never blocks providing the results of a dump, so
1107 * error == EAGAIN means that we've read the whole thing, and
1108 * therefore transform it into EOF. (The kernel always provides
1109 * NLMSG_DONE as a sentinel. Some other thread must have received
1110 * that already but not yet signaled it in 'status'.)
1111 *
1112 * Any other error is just an error. */
93295354
BP
1113 return error == EAGAIN ? EOF : error;
1114 }
1115
1116 nlmsghdr = nl_msg_nlmsghdr(buffer);
1117 if (dump->nl_seq != nlmsghdr->nlmsg_seq) {
1118 VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
1119 nlmsghdr->nlmsg_seq, dump->nl_seq);
1120 ofpbuf_clear(buffer);
1121 }
1122 }
1123
1124 if (nl_msg_nlmsgerr(buffer, &error) && error) {
1125 VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
1126 ovs_strerror(error));
1127 ofpbuf_clear(buffer);
1128 return error;
1129 }
1130
1131 return 0;
1132}
1133
1134static int
1135nl_dump_next__(struct ofpbuf *reply, struct ofpbuf *buffer)
1136{
1137 struct nlmsghdr *nlmsghdr = nl_msg_next(buffer, reply);
1138 if (!nlmsghdr) {
1139 VLOG_WARN_RL(&rl, "netlink dump contains message fragment");
1140 return EPROTO;
1141 } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
1142 return EOF;
1143 } else {
1144 return 0;
1145 }
2fe27d5a
BP
1146}
1147
d57695d7
JS
1148/* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must
1149 * have been initialized with nl_dump_start(), and 'buffer' must have been
1150 * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long.
2fe27d5a 1151 *
19aa20a0 1152 * If successful, returns true and points 'reply->data' and
6fd6ed71 1153 * 'reply->size' to the message that was retrieved. The caller must not
19aa20a0
BP
1154 * modify 'reply' (because it points within 'buffer', which will be used by
1155 * future calls to this function).
1156 *
1157 * On failure, returns false and sets 'reply->data' to NULL and
6fd6ed71 1158 * 'reply->size' to 0. Failure might indicate an actual error or merely
19aa20a0
BP
1159 * the end of replies. An error status for the entire dump operation is
1160 * provided when it is completed by calling nl_dump_done().
0672776e
JS
1161 *
1162 * Multiple threads may call this function, passing the same nl_dump, however
1163 * each must provide independent buffers. This function may cache multiple
1164 * replies in the buffer, and these will be processed before more replies are
1165 * fetched. When this function returns false, other threads may continue to
1166 * process replies in their buffers, but they will not fetch more replies.
2fe27d5a
BP
1167 */
1168bool
d57695d7 1169nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer)
2fe27d5a 1170{
93295354 1171 int retval = 0;
0672776e 1172
93295354
BP
1173 /* If the buffer is empty, refill it.
1174 *
1175 * If the buffer is not empty, we don't check the dump's status.
1176 * Otherwise, we could end up skipping some of the dump results if thread A
1177 * hits EOF while thread B is in the midst of processing a batch. */
6fd6ed71 1178 if (!buffer->size) {
0791315e 1179 ovs_mutex_lock(&dump->mutex);
93295354
BP
1180 if (!dump->status) {
1181 /* Take the mutex here to avoid an in-kernel race. If two threads
1182 * try to read from a Netlink dump socket at once, then the socket
1183 * error can be set to EINVAL, which will be encountered on the
1184 * next recv on that socket, which could be anywhere due to the way
1185 * that we pool Netlink sockets. Serializing the recv calls avoids
1186 * the issue. */
1187 dump->status = nl_dump_refill(dump, buffer);
1188 }
1189 retval = dump->status;
0791315e 1190 ovs_mutex_unlock(&dump->mutex);
93295354 1191 }
0791315e 1192
93295354
BP
1193 /* Fetch the next message from the buffer. */
1194 if (!retval) {
1195 retval = nl_dump_next__(reply, buffer);
2fe27d5a 1196 if (retval) {
93295354
BP
1197 /* Record 'retval' as the dump status, but don't overwrite an error
1198 * with EOF. */
1199 ovs_mutex_lock(&dump->mutex);
1200 if (dump->status <= 0) {
1201 dump->status = retval;
2fe27d5a 1202 }
93295354 1203 ovs_mutex_unlock(&dump->mutex);
2fe27d5a 1204 }
2fe27d5a
BP
1205 }
1206
93295354 1207 if (retval) {
6fd6ed71
PS
1208 reply->data = NULL;
1209 reply->size = 0;
0672776e 1210 }
93295354 1211 return !retval;
2fe27d5a
BP
1212}
1213
1214/* Completes Netlink dump operation 'dump', which must have been initialized
1215 * with nl_dump_start(). Returns 0 if the dump operation was error-free,
1216 * otherwise a positive errno value describing the problem. */
1217int
1218nl_dump_done(struct nl_dump *dump)
1219{
0672776e 1220 int status;
d57695d7 1221
93295354
BP
1222 ovs_mutex_lock(&dump->mutex);
1223 status = dump->status;
1224 ovs_mutex_unlock(&dump->mutex);
1225
2fe27d5a 1226 /* Drain any remaining messages that the client didn't read. Otherwise the
a88b4e04
BP
1227 * kernel will continue to queue them up and waste buffer space.
1228 *
1229 * XXX We could just destroy and discard the socket in this case. */
0672776e
JS
1230 if (!status) {
1231 uint64_t tmp_reply_stub[NL_DUMP_BUFSIZE / 8];
1232 struct ofpbuf reply, buf;
1233
1234 ofpbuf_use_stub(&buf, tmp_reply_stub, sizeof tmp_reply_stub);
1235 while (nl_dump_next(dump, &reply, &buf)) {
1236 /* Nothing to do. */
2fe27d5a 1237 }
0672776e 1238 ofpbuf_uninit(&buf);
93295354
BP
1239
1240 ovs_mutex_lock(&dump->mutex);
1241 status = dump->status;
1242 ovs_mutex_unlock(&dump->mutex);
1243 ovs_assert(status);
2fe27d5a 1244 }
93295354 1245
a88b4e04 1246 nl_pool_release(dump->sock);
0791315e 1247 ovs_mutex_destroy(&dump->mutex);
93295354
BP
1248
1249 return status == EOF ? 0 : status;
2fe27d5a
BP
1250}
1251
7fa09611
EE
1252#ifdef _WIN32
1253/* Pend an I/O request in the driver. The driver completes the I/O whenever
1254 * an event or a packet is ready to be read. Once the I/O is completed
1255 * the overlapped structure event associated with the pending I/O will be set
1256 */
1257static int
8341662d 1258pend_io_request(struct nl_sock *sock)
7fa09611
EE
1259{
1260 struct ofpbuf request;
1261 uint64_t request_stub[128];
1262 struct ovs_header *ovs_header;
1263 struct nlmsghdr *nlmsg;
1264 uint32_t seq;
b91d3d03 1265 int retval = 0;
7fa09611
EE
1266 int error;
1267 DWORD bytes;
64513e68 1268 OVERLAPPED *overlapped = CONST_CAST(OVERLAPPED *, &sock->overlapped);
b91d3d03
NR
1269 uint16_t cmd = OVS_CTRL_CMD_WIN_PEND_PACKET_REQ;
1270
1271 ovs_assert(sock->read_ioctl == OVS_IOCTL_READ_PACKET ||
1272 sock->read_ioctl == OVS_IOCTL_READ_EVENT);
1273 if (sock->read_ioctl == OVS_IOCTL_READ_EVENT) {
1274 cmd = OVS_CTRL_CMD_WIN_PEND_REQ;
1275 }
7fa09611
EE
1276
1277 int ovs_msg_size = sizeof (struct nlmsghdr) + sizeof (struct genlmsghdr) +
1278 sizeof (struct ovs_header);
1279
1280 ofpbuf_use_stub(&request, request_stub, sizeof request_stub);
1281
1282 seq = nl_sock_allocate_seq(sock, 1);
1283 nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0,
b91d3d03 1284 cmd, OVS_WIN_CONTROL_VERSION);
7fa09611
EE
1285 nlmsg = nl_msg_nlmsghdr(&request);
1286 nlmsg->nlmsg_seq = seq;
92a5068f 1287 nlmsg->nlmsg_pid = sock->pid;
7fa09611
EE
1288
1289 ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header);
1290 ovs_header->dp_ifindex = 0;
e6b298ef 1291 nlmsg->nlmsg_len = request.size;
7fa09611
EE
1292
1293 if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE,
6fd6ed71 1294 request.data, request.size,
7fa09611
EE
1295 NULL, 0, &bytes, overlapped)) {
1296 error = GetLastError();
1297 /* Check if the I/O got pended */
1298 if (error != ERROR_IO_INCOMPLETE && error != ERROR_IO_PENDING) {
d553dd54 1299 lost_communication(error);
7fa09611
EE
1300 VLOG_ERR("nl_sock_wait failed - %s\n", ovs_format_message(error));
1301 retval = EINVAL;
7fa09611
EE
1302 }
1303 } else {
b91d3d03 1304 retval = EAGAIN;
7fa09611 1305 }
7fa09611
EE
1306
1307done:
1308 ofpbuf_uninit(&request);
1309 return retval;
1310}
1311#endif /* _WIN32 */
1312
2fe27d5a 1313/* Causes poll_block() to wake up when any of the specified 'events' (which is
8341662d
NR
1314 * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'.
1315 * On Windows, 'sock' is not treated as const, and may be modified. */
2fe27d5a
BP
1316void
1317nl_sock_wait(const struct nl_sock *sock, short int events)
1318{
22326ba6 1319#ifdef _WIN32
7fa09611 1320 if (sock->overlapped.Internal != STATUS_PENDING) {
b91d3d03
NR
1321 int ret = pend_io_request(CONST_CAST(struct nl_sock *, sock));
1322 if (ret == 0) {
1323 poll_wevent_wait(sock->overlapped.hEvent);
1324 } else {
1325 poll_immediate_wake();
1326 }
1327 } else {
1328 poll_wevent_wait(sock->overlapped.hEvent);
7fa09611 1329 }
22326ba6 1330#else
2fe27d5a 1331 poll_fd_wait(sock->fd, events);
22326ba6 1332#endif
2fe27d5a 1333}
50802adb 1334
9667de98 1335#ifndef _WIN32
8522ba09
BP
1336/* Returns the underlying fd for 'sock', for use in "poll()"-like operations
1337 * that can't use nl_sock_wait().
1338 *
1339 * It's a little tricky to use the returned fd correctly, because nl_sock does
1340 * "copy on write" to allow a single nl_sock to be used for notifications,
1341 * transactions, and dumps. If 'sock' is used only for notifications and
1342 * transactions (and never for dump) then the usage is safe. */
1343int
1344nl_sock_fd(const struct nl_sock *sock)
1345{
1346 return sock->fd;
1347}
9667de98 1348#endif
8522ba09 1349
50802adb
JG
1350/* Returns the PID associated with this socket. */
1351uint32_t
1352nl_sock_pid(const struct nl_sock *sock)
1353{
1354 return sock->pid;
1355}
2fe27d5a
BP
1356\f
1357/* Miscellaneous. */
1358
2ad204c8
BP
1359struct genl_family {
1360 struct hmap_node hmap_node;
1361 uint16_t id;
1362 char *name;
1363};
1364
1365static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
1366
2fe27d5a
BP
1367static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
1368 [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
213a13ed 1369 [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true},
2fe27d5a
BP
1370};
1371
2ad204c8
BP
1372static struct genl_family *
1373find_genl_family_by_id(uint16_t id)
1374{
1375 struct genl_family *family;
1376
1377 HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
1378 &genl_families) {
1379 if (family->id == id) {
1380 return family;
1381 }
1382 }
1383 return NULL;
1384}
1385
1386static void
1387define_genl_family(uint16_t id, const char *name)
1388{
1389 struct genl_family *family = find_genl_family_by_id(id);
1390
1391 if (family) {
1392 if (!strcmp(family->name, name)) {
1393 return;
1394 }
1395 free(family->name);
1396 } else {
1397 family = xmalloc(sizeof *family);
1398 family->id = id;
1399 hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
1400 }
1401 family->name = xstrdup(name);
1402}
1403
1404static const char *
1405genl_family_to_name(uint16_t id)
1406{
1407 if (id == GENL_ID_CTRL) {
1408 return "control";
1409 } else {
1410 struct genl_family *family = find_genl_family_by_id(id);
1411 return family ? family->name : "unknown";
1412 }
1413}
1414
b3fca8a8 1415#ifndef _WIN32
e408762f 1416static int
2a477244
BP
1417do_lookup_genl_family(const char *name, struct nlattr **attrs,
1418 struct ofpbuf **replyp)
2fe27d5a
BP
1419{
1420 struct nl_sock *sock;
1421 struct ofpbuf request, *reply;
2a477244 1422 int error;
2fe27d5a 1423
2a477244
BP
1424 *replyp = NULL;
1425 error = nl_sock_create(NETLINK_GENERIC, &sock);
1426 if (error) {
1427 return error;
2fe27d5a
BP
1428 }
1429
1430 ofpbuf_init(&request, 0);
1431 nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST,
1432 CTRL_CMD_GETFAMILY, 1);
1433 nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name);
2a477244 1434 error = nl_sock_transact(sock, &request, &reply);
2fe27d5a 1435 ofpbuf_uninit(&request);
2a477244 1436 if (error) {
2fe27d5a 1437 nl_sock_destroy(sock);
2a477244 1438 return error;
2fe27d5a
BP
1439 }
1440
1441 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
2a477244
BP
1442 family_policy, attrs, ARRAY_SIZE(family_policy))
1443 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
2fe27d5a
BP
1444 nl_sock_destroy(sock);
1445 ofpbuf_delete(reply);
2a477244 1446 return EPROTO;
2fe27d5a
BP
1447 }
1448
2fe27d5a 1449 nl_sock_destroy(sock);
2a477244
BP
1450 *replyp = reply;
1451 return 0;
2fe27d5a 1452}
b3fca8a8
NR
1453#else
1454static int
1455do_lookup_genl_family(const char *name, struct nlattr **attrs,
1456 struct ofpbuf **replyp)
1457{
4c484aca 1458 struct nlmsghdr *nlmsg;
b3fca8a8
NR
1459 struct ofpbuf *reply;
1460 int error;
1461 uint16_t family_id;
1462 const char *family_name;
1463 uint32_t family_version;
1464 uint32_t family_attrmax;
4c484aca
NR
1465 uint32_t mcgrp_id = OVS_WIN_NL_INVALID_MCGRP_ID;
1466 const char *mcgrp_name = NULL;
b3fca8a8
NR
1467
1468 *replyp = NULL;
1469 reply = ofpbuf_new(1024);
1470
4c484aca 1471 /* CTRL_ATTR_MCAST_GROUPS is supported only for VPORT family. */
b3fca8a8
NR
1472 if (!strcmp(name, OVS_WIN_CONTROL_FAMILY)) {
1473 family_id = OVS_WIN_NL_CTRL_FAMILY_ID;
1474 family_name = OVS_WIN_CONTROL_FAMILY;
1475 family_version = OVS_WIN_CONTROL_VERSION;
1476 family_attrmax = OVS_WIN_CONTROL_ATTR_MAX;
1477 } else if (!strcmp(name, OVS_DATAPATH_FAMILY)) {
1478 family_id = OVS_WIN_NL_DATAPATH_FAMILY_ID;
1479 family_name = OVS_DATAPATH_FAMILY;
1480 family_version = OVS_DATAPATH_VERSION;
1481 family_attrmax = OVS_DP_ATTR_MAX;
1482 } else if (!strcmp(name, OVS_PACKET_FAMILY)) {
1483 family_id = OVS_WIN_NL_PACKET_FAMILY_ID;
1484 family_name = OVS_PACKET_FAMILY;
1485 family_version = OVS_PACKET_VERSION;
1486 family_attrmax = OVS_PACKET_ATTR_MAX;
1487 } else if (!strcmp(name, OVS_VPORT_FAMILY)) {
1488 family_id = OVS_WIN_NL_VPORT_FAMILY_ID;
1489 family_name = OVS_VPORT_FAMILY;
1490 family_version = OVS_VPORT_VERSION;
1491 family_attrmax = OVS_VPORT_ATTR_MAX;
4c484aca
NR
1492 mcgrp_id = OVS_WIN_NL_VPORT_MCGRP_ID;
1493 mcgrp_name = OVS_VPORT_MCGROUP;
b3fca8a8
NR
1494 } else if (!strcmp(name, OVS_FLOW_FAMILY)) {
1495 family_id = OVS_WIN_NL_FLOW_FAMILY_ID;
1496 family_name = OVS_FLOW_FAMILY;
1497 family_version = OVS_FLOW_VERSION;
1498 family_attrmax = OVS_FLOW_ATTR_MAX;
83cc9d56
NR
1499 } else if (!strcmp(name, OVS_WIN_NETDEV_FAMILY)) {
1500 family_id = OVS_WIN_NL_NETDEV_FAMILY_ID;
1501 family_name = OVS_WIN_NETDEV_FAMILY;
1502 family_version = OVS_WIN_NETDEV_VERSION;
1503 family_attrmax = OVS_WIN_NETDEV_ATTR_MAX;
b3fca8a8
NR
1504 } else {
1505 ofpbuf_delete(reply);
1506 return EINVAL;
1507 }
1508
1509 nl_msg_put_genlmsghdr(reply, 0, GENL_ID_CTRL, 0,
1510 CTRL_CMD_NEWFAMILY, family_version);
1511 /* CTRL_ATTR_HDRSIZE and CTRL_ATTR_OPS are not populated, but the
1512 * callers do not seem to need them. */
1513 nl_msg_put_u16(reply, CTRL_ATTR_FAMILY_ID, family_id);
1514 nl_msg_put_string(reply, CTRL_ATTR_FAMILY_NAME, family_name);
1515 nl_msg_put_u32(reply, CTRL_ATTR_VERSION, family_version);
1516 nl_msg_put_u32(reply, CTRL_ATTR_MAXATTR, family_attrmax);
1517
4c484aca
NR
1518 if (mcgrp_id != OVS_WIN_NL_INVALID_MCGRP_ID) {
1519 size_t mcgrp_ofs1 = nl_msg_start_nested(reply, CTRL_ATTR_MCAST_GROUPS);
1520 size_t mcgrp_ofs2= nl_msg_start_nested(reply,
1521 OVS_WIN_NL_VPORT_MCGRP_ID - OVS_WIN_NL_MCGRP_START_ID);
1522 nl_msg_put_u32(reply, CTRL_ATTR_MCAST_GRP_ID, mcgrp_id);
1523 ovs_assert(mcgrp_name != NULL);
1524 nl_msg_put_string(reply, CTRL_ATTR_MCAST_GRP_NAME, mcgrp_name);
1525 nl_msg_end_nested(reply, mcgrp_ofs2);
1526 nl_msg_end_nested(reply, mcgrp_ofs1);
1527 }
1528
1529 /* Set the total length of the netlink message. */
1530 nlmsg = nl_msg_nlmsghdr(reply);
6fd6ed71 1531 nlmsg->nlmsg_len = reply->size;
4c484aca 1532
b3fca8a8
NR
1533 if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN,
1534 family_policy, attrs, ARRAY_SIZE(family_policy))
1535 || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) {
b3fca8a8
NR
1536 ofpbuf_delete(reply);
1537 return EPROTO;
1538 }
1539
1540 *replyp = reply;
1541 return 0;
1542}
1543#endif
2fe27d5a 1544
e408762f
EJ
1545/* Finds the multicast group called 'group_name' in genl family 'family_name'.
1546 * When successful, writes its result to 'multicast_group' and returns 0.
213a13ed 1547 * Otherwise, clears 'multicast_group' and returns a positive error code.
b3dcb73c 1548 */
e408762f
EJ
1549int
1550nl_lookup_genl_mcgroup(const char *family_name, const char *group_name,
b3dcb73c 1551 unsigned int *multicast_group)
e408762f
EJ
1552{
1553 struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
6d23c6f4 1554 const struct nlattr *mc;
2a477244 1555 struct ofpbuf *reply;
e408762f 1556 unsigned int left;
2a477244 1557 int error;
e408762f
EJ
1558
1559 *multicast_group = 0;
2a477244
BP
1560 error = do_lookup_genl_family(family_name, family_attrs, &reply);
1561 if (error) {
1562 return error;
e408762f
EJ
1563 }
1564
213a13ed 1565 if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
b3dcb73c 1566 error = EPROTO;
213a13ed
EJ
1567 goto exit;
1568 }
1569
6d23c6f4 1570 NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
e408762f
EJ
1571 static const struct nl_policy mc_policy[] = {
1572 [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
1573 [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
1574 };
1575
1576 struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)];
1577 const char *mc_name;
1578
1579 if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) {
2a477244
BP
1580 error = EPROTO;
1581 goto exit;
e408762f
EJ
1582 }
1583
1584 mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]);
1585 if (!strcmp(group_name, mc_name)) {
1586 *multicast_group =
1587 nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]);
2a477244
BP
1588 error = 0;
1589 goto exit;
e408762f
EJ
1590 }
1591 }
2a477244 1592 error = EPROTO;
e408762f 1593
2a477244
BP
1594exit:
1595 ofpbuf_delete(reply);
1596 return error;
e408762f
EJ
1597}
1598
2fe27d5a
BP
1599/* If '*number' is 0, translates the given Generic Netlink family 'name' to a
1600 * number and stores it in '*number'. If successful, returns 0 and the caller
1601 * may use '*number' as the family number. On failure, returns a positive
1602 * errno value and '*number' caches the errno value. */
1603int
1604nl_lookup_genl_family(const char *name, int *number)
1605{
1606 if (*number == 0) {
2a477244
BP
1607 struct nlattr *attrs[ARRAY_SIZE(family_policy)];
1608 struct ofpbuf *reply;
1609 int error;
1610
1611 error = do_lookup_genl_family(name, attrs, &reply);
1612 if (!error) {
1613 *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
1614 define_genl_family(*number, name);
1615 } else {
1616 *number = -error;
1617 }
1618 ofpbuf_delete(reply);
1619
cb22974d 1620 ovs_assert(*number != 0);
2fe27d5a
BP
1621 }
1622 return *number > 0 ? 0 : -*number;
1623}
a88b4e04
BP
1624\f
1625struct nl_pool {
1626 struct nl_sock *socks[16];
1627 int n;
1628};
1629
834d6caf 1630static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER;
97be1538 1631static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex);
a88b4e04
BP
1632
1633static int
1634nl_pool_alloc(int protocol, struct nl_sock **sockp)
1635{
0bd01224 1636 struct nl_sock *sock = NULL;
a88b4e04
BP
1637 struct nl_pool *pool;
1638
1639 ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools));
1640
97be1538 1641 ovs_mutex_lock(&pool_mutex);
a88b4e04
BP
1642 pool = &pools[protocol];
1643 if (pool->n > 0) {
0bd01224
BP
1644 sock = pool->socks[--pool->n];
1645 }
97be1538 1646 ovs_mutex_unlock(&pool_mutex);
0bd01224
BP
1647
1648 if (sock) {
1649 *sockp = sock;
a88b4e04
BP
1650 return 0;
1651 } else {
1652 return nl_sock_create(protocol, sockp);
1653 }
1654}
1655
1656static void
1657nl_pool_release(struct nl_sock *sock)
1658{
1659 if (sock) {
1660 struct nl_pool *pool = &pools[sock->protocol];
1661
97be1538 1662 ovs_mutex_lock(&pool_mutex);
a88b4e04
BP
1663 if (pool->n < ARRAY_SIZE(pool->socks)) {
1664 pool->socks[pool->n++] = sock;
0bd01224 1665 sock = NULL;
a88b4e04 1666 }
97be1538 1667 ovs_mutex_unlock(&pool_mutex);
0bd01224
BP
1668
1669 nl_sock_destroy(sock);
a88b4e04
BP
1670 }
1671}
1672
022ad2b9
BP
1673/* Sends 'request' to the kernel on a Netlink socket for the given 'protocol'
1674 * (e.g. NETLINK_ROUTE or NETLINK_GENERIC) and waits for a response. If
1675 * successful, returns 0. On failure, returns a positive errno value.
1676 *
1677 * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's
1678 * reply, which the caller is responsible for freeing with ofpbuf_delete(), and
1679 * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
1680 * reply, if any, is discarded.
1681 *
1682 * Before the message is sent, nlmsg_len in 'request' will be finalized to
6fd6ed71 1683 * match msg->size, nlmsg_pid will be set to the pid of the socket used
022ad2b9
BP
1684 * for sending the request, and nlmsg_seq will be initialized.
1685 *
1686 * The caller is responsible for destroying 'request'.
1687 *
1688 * Bare Netlink is an unreliable transport protocol. This function layers
1689 * reliable delivery and reply semantics on top of bare Netlink.
1690 *
1691 * In Netlink, sending a request to the kernel is reliable enough, because the
1692 * kernel will tell us if the message cannot be queued (and we will in that
1693 * case put it on the transmit queue and wait until it can be delivered).
1694 *
1695 * Receiving the reply is the real problem: if the socket buffer is full when
1696 * the kernel tries to send the reply, the reply will be dropped. However, the
1697 * kernel sets a flag that a reply has been dropped. The next call to recv
1698 * then returns ENOBUFS. We can then re-send the request.
1699 *
1700 * Caveats:
1701 *
1702 * 1. Netlink depends on sequence numbers to match up requests and
1703 * replies. The sender of a request supplies a sequence number, and
1704 * the reply echos back that sequence number.
1705 *
1706 * This is fine, but (1) some kernel netlink implementations are
1707 * broken, in that they fail to echo sequence numbers and (2) this
1708 * function will drop packets with non-matching sequence numbers, so
1709 * that only a single request can be usefully transacted at a time.
1710 *
1711 * 2. Resending the request causes it to be re-executed, so the request
1712 * needs to be idempotent.
1713 */
a88b4e04
BP
1714int
1715nl_transact(int protocol, const struct ofpbuf *request,
1716 struct ofpbuf **replyp)
1717{
1718 struct nl_sock *sock;
1719 int error;
1720
1721 error = nl_pool_alloc(protocol, &sock);
1722 if (error) {
d9c194a1
RD
1723 if (replyp) {
1724 *replyp = NULL;
1725 }
a88b4e04
BP
1726 return error;
1727 }
1728
1729 error = nl_sock_transact(sock, request, replyp);
1730
1731 nl_pool_release(sock);
1732 return error;
1733}
1734
022ad2b9
BP
1735/* Sends the 'request' member of the 'n' transactions in 'transactions' on a
1736 * Netlink socket for the given 'protocol' (e.g. NETLINK_ROUTE or
1737 * NETLINK_GENERIC), in order, and receives responses to all of them. Fills in
1738 * the 'error' member of each transaction with 0 if it was successful,
1739 * otherwise with a positive errno value. If 'reply' is nonnull, then it will
1740 * be filled with the reply if the message receives a detailed reply. In other
1741 * cases, i.e. where the request failed or had no reply beyond an indication of
1742 * success, 'reply' will be cleared if it is nonnull.
1743 *
1744 * The caller is responsible for destroying each request and reply, and the
1745 * transactions array itself.
1746 *
1747 * Before sending each message, this function will finalize nlmsg_len in each
1748 * 'request' to match the ofpbuf's size, set nlmsg_pid to the pid of the socket
1749 * used for the transaction, and initialize nlmsg_seq.
1750 *
1751 * Bare Netlink is an unreliable transport protocol. This function layers
1752 * reliable delivery and reply semantics on top of bare Netlink. See
1753 * nl_transact() for some caveats.
1754 */
a88b4e04
BP
1755void
1756nl_transact_multiple(int protocol,
1757 struct nl_transaction **transactions, size_t n)
1758{
1759 struct nl_sock *sock;
1760 int error;
1761
1762 error = nl_pool_alloc(protocol, &sock);
1763 if (!error) {
1764 nl_sock_transact_multiple(sock, transactions, n);
1765 nl_pool_release(sock);
1766 } else {
1767 nl_sock_record_errors__(transactions, n, error);
1768 }
1769}
1770
2fe27d5a 1771\f
7d7447df
BP
1772static uint32_t
1773nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
1774{
1775 uint32_t seq = sock->next_seq;
1776
1777 sock->next_seq += n;
1778
1779 /* Make it impossible for the next request for sequence numbers to wrap
1780 * around to 0. Start over with 1 to avoid ever using a sequence number of
1781 * 0, because the kernel uses sequence number 0 for notifications. */
1782 if (sock->next_seq >= UINT32_MAX / 2) {
1783 sock->next_seq = 1;
1784 }
1785
1786 return seq;
1787}
1788
2fe27d5a 1789static void
2ad204c8 1790nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
2fe27d5a
BP
1791{
1792 struct nlmsg_flag {
1793 unsigned int bits;
1794 const char *name;
1795 };
1796 static const struct nlmsg_flag flags[] = {
1797 { NLM_F_REQUEST, "REQUEST" },
1798 { NLM_F_MULTI, "MULTI" },
1799 { NLM_F_ACK, "ACK" },
1800 { NLM_F_ECHO, "ECHO" },
1801 { NLM_F_DUMP, "DUMP" },
1802 { NLM_F_ROOT, "ROOT" },
1803 { NLM_F_MATCH, "MATCH" },
1804 { NLM_F_ATOMIC, "ATOMIC" },
1805 };
1806 const struct nlmsg_flag *flag;
1807 uint16_t flags_left;
1808
1809 ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16,
1810 h->nlmsg_len, h->nlmsg_type);
1811 if (h->nlmsg_type == NLMSG_NOOP) {
1812 ds_put_cstr(ds, "(no-op)");
1813 } else if (h->nlmsg_type == NLMSG_ERROR) {
1814 ds_put_cstr(ds, "(error)");
1815 } else if (h->nlmsg_type == NLMSG_DONE) {
1816 ds_put_cstr(ds, "(done)");
1817 } else if (h->nlmsg_type == NLMSG_OVERRUN) {
1818 ds_put_cstr(ds, "(overrun)");
1819 } else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
1820 ds_put_cstr(ds, "(reserved)");
2ad204c8
BP
1821 } else if (protocol == NETLINK_GENERIC) {
1822 ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
2fe27d5a
BP
1823 } else {
1824 ds_put_cstr(ds, "(family-defined)");
1825 }
1826 ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags);
1827 flags_left = h->nlmsg_flags;
1828 for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) {
1829 if ((flags_left & flag->bits) == flag->bits) {
1830 ds_put_format(ds, "[%s]", flag->name);
1831 flags_left &= ~flag->bits;
1832 }
1833 }
1834 if (flags_left) {
1835 ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
1836 }
2c5a6834
BP
1837 ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
1838 h->nlmsg_seq, h->nlmsg_pid);
2fe27d5a
BP
1839}
1840
1841static char *
7041c3a9 1842nlmsg_to_string(const struct ofpbuf *buffer, int protocol)
2fe27d5a
BP
1843{
1844 struct ds ds = DS_EMPTY_INITIALIZER;
1845 const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
1846 if (h) {
2ad204c8 1847 nlmsghdr_to_string(h, protocol, &ds);
2fe27d5a
BP
1848 if (h->nlmsg_type == NLMSG_ERROR) {
1849 const struct nlmsgerr *e;
1850 e = ofpbuf_at(buffer, NLMSG_HDRLEN,
1851 NLMSG_ALIGN(sizeof(struct nlmsgerr)));
1852 if (e) {
1853 ds_put_format(&ds, " error(%d", e->error);
1854 if (e->error < 0) {
10a89ef0 1855 ds_put_format(&ds, "(%s)", ovs_strerror(-e->error));
2fe27d5a
BP
1856 }
1857 ds_put_cstr(&ds, ", in-reply-to(");
2ad204c8 1858 nlmsghdr_to_string(&e->msg, protocol, &ds);
2fe27d5a
BP
1859 ds_put_cstr(&ds, "))");
1860 } else {
1861 ds_put_cstr(&ds, " error(truncated)");
1862 }
1863 } else if (h->nlmsg_type == NLMSG_DONE) {
1864 int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error);
1865 if (error) {
1866 ds_put_format(&ds, " done(%d", *error);
1867 if (*error < 0) {
10a89ef0 1868 ds_put_format(&ds, "(%s)", ovs_strerror(-*error));
2fe27d5a
BP
1869 }
1870 ds_put_cstr(&ds, ")");
1871 } else {
1872 ds_put_cstr(&ds, " done(truncated)");
1873 }
7041c3a9
BP
1874 } else if (protocol == NETLINK_GENERIC) {
1875 struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer);
1876 if (genl) {
1877 ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")",
1878 genl->cmd, genl->version);
1879 }
2fe27d5a
BP
1880 }
1881 } else {
1882 ds_put_cstr(&ds, "nl(truncated)");
1883 }
1884 return ds.string;
1885}
1886
1887static void
1888log_nlmsg(const char *function, int error,
7041c3a9 1889 const void *message, size_t size, int protocol)
2fe27d5a 1890{
2fe27d5a
BP
1891 if (!VLOG_IS_DBG_ENABLED()) {
1892 return;
1893 }
1894
0a2869d5
BP
1895 struct ofpbuf buffer = ofpbuf_const_initializer(message, size);
1896 char *nlmsg = nlmsg_to_string(&buffer, protocol);
10a89ef0 1897 VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg);
2fe27d5a
BP
1898 free(nlmsg);
1899}