]>
Commit | Line | Data |
---|---|---|
2fe27d5a | 1 | /* |
db1fc210 | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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" | |
26 | #include "dynamic-string.h" | |
2ad204c8 BP |
27 | #include "hash.h" |
28 | #include "hmap.h" | |
2fe27d5a BP |
29 | #include "netlink.h" |
30 | #include "netlink-protocol.h" | |
886dd35a | 31 | #include "odp-netlink.h" |
2fe27d5a | 32 | #include "ofpbuf.h" |
0bd01224 | 33 | #include "ovs-thread.h" |
2fe27d5a | 34 | #include "poll-loop.h" |
0672776e | 35 | #include "seq.h" |
6b7c12fd | 36 | #include "socket-util.h" |
cc75061a | 37 | #include "util.h" |
2fe27d5a BP |
38 | #include "vlog.h" |
39 | ||
40 | VLOG_DEFINE_THIS_MODULE(netlink_socket); | |
41 | ||
42 | COVERAGE_DEFINE(netlink_overflow); | |
43 | COVERAGE_DEFINE(netlink_received); | |
fc999dda | 44 | COVERAGE_DEFINE(netlink_recv_jumbo); |
2fe27d5a BP |
45 | COVERAGE_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 | ||
22326ba6 AS |
52 | #ifdef _WIN32 |
53 | static struct ovs_mutex portid_mutex = OVS_MUTEX_INITIALIZER; | |
54 | static uint32_t g_last_portid = 0; | |
55 | ||
56 | /* Port IDs must be unique! */ | |
57 | static uint32_t | |
58 | portid_next(void) | |
59 | OVS_GUARDED_BY(portid_mutex) | |
60 | { | |
61 | g_last_portid++; | |
62 | return g_last_portid; | |
63 | } | |
886dd35a | 64 | #endif /* _WIN32 */ |
22326ba6 | 65 | |
2fe27d5a BP |
66 | /* A single (bad) Netlink message can in theory dump out many, many log |
67 | * messages, so the burst size is set quite high here to avoid missing useful | |
68 | * information. Also, at high logging levels we log *all* Netlink messages. */ | |
69 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600); | |
70 | ||
7d7447df | 71 | static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n); |
2fe27d5a | 72 | static void log_nlmsg(const char *function, int error, |
7041c3a9 | 73 | const void *message, size_t size, int protocol); |
886dd35a | 74 | #ifdef _WIN32 |
ebac7fb7 | 75 | static int get_sock_pid_from_kernel(struct nl_sock *sock); |
886dd35a | 76 | #endif |
2fe27d5a BP |
77 | \f |
78 | /* Netlink sockets. */ | |
79 | ||
0d121c73 | 80 | struct nl_sock { |
22326ba6 AS |
81 | #ifdef _WIN32 |
82 | HANDLE handle; | |
7fa09611 | 83 | OVERLAPPED overlapped; |
64513e68 | 84 | DWORD read_ioctl; |
22326ba6 | 85 | #else |
2fe27d5a | 86 | int fd; |
22326ba6 | 87 | #endif |
7d7447df | 88 | uint32_t next_seq; |
2fe27d5a | 89 | uint32_t pid; |
7041c3a9 | 90 | int protocol; |
cc75061a | 91 | unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */ |
2fe27d5a BP |
92 | }; |
93 | ||
cc75061a BP |
94 | /* Compile-time limit on iovecs, so that we can allocate a maximum-size array |
95 | * of iovecs on the stack. */ | |
96 | #define MAX_IOVS 128 | |
97 | ||
98 | /* Maximum number of iovecs that may be passed to sendmsg, capped at a | |
99 | * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS. | |
100 | * | |
101 | * Initialized by nl_sock_create(). */ | |
102 | static int max_iovs; | |
103 | ||
a88b4e04 BP |
104 | static int nl_pool_alloc(int protocol, struct nl_sock **sockp); |
105 | static void nl_pool_release(struct nl_sock *); | |
2fe27d5a BP |
106 | |
107 | /* Creates a new netlink socket for the given netlink 'protocol' | |
108 | * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the | |
a88b4e04 | 109 | * new socket if successful, otherwise returns a positive errno value. */ |
2fe27d5a | 110 | int |
cceb11f5 | 111 | nl_sock_create(int protocol, struct nl_sock **sockp) |
2fe27d5a | 112 | { |
0bd01224 | 113 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2fe27d5a | 114 | struct nl_sock *sock; |
22326ba6 | 115 | #ifndef _WIN32 |
2fe27d5a | 116 | struct sockaddr_nl local, remote; |
22326ba6 | 117 | #endif |
2c5a6834 | 118 | socklen_t local_size; |
d2b9f5b0 | 119 | int rcvbuf; |
2fe27d5a BP |
120 | int retval = 0; |
121 | ||
0bd01224 | 122 | if (ovsthread_once_start(&once)) { |
cc75061a BP |
123 | int save_errno = errno; |
124 | errno = 0; | |
125 | ||
126 | max_iovs = sysconf(_SC_UIO_MAXIOV); | |
127 | if (max_iovs < _XOPEN_IOV_MAX) { | |
128 | if (max_iovs == -1 && errno) { | |
10a89ef0 | 129 | VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno)); |
cc75061a BP |
130 | } |
131 | max_iovs = _XOPEN_IOV_MAX; | |
132 | } else if (max_iovs > MAX_IOVS) { | |
133 | max_iovs = MAX_IOVS; | |
134 | } | |
135 | ||
136 | errno = save_errno; | |
0bd01224 | 137 | ovsthread_once_done(&once); |
cc75061a BP |
138 | } |
139 | ||
2fe27d5a | 140 | *sockp = NULL; |
488232b7 | 141 | sock = xmalloc(sizeof *sock); |
2fe27d5a | 142 | |
22326ba6 | 143 | #ifdef _WIN32 |
7fa09611 EE |
144 | sock->handle = CreateFile(OVS_DEVICE_NAME_USER, |
145 | GENERIC_READ | GENERIC_WRITE, | |
146 | FILE_SHARE_READ | FILE_SHARE_WRITE, | |
147 | NULL, OPEN_EXISTING, | |
148 | FILE_FLAG_OVERLAPPED, NULL); | |
22326ba6 AS |
149 | |
150 | if (sock->handle == INVALID_HANDLE_VALUE) { | |
52a1540a | 151 | VLOG_ERR("fcntl: %s", ovs_lasterror_to_string()); |
7fa09611 EE |
152 | goto error; |
153 | } | |
154 | ||
155 | memset(&sock->overlapped, 0, sizeof sock->overlapped); | |
156 | sock->overlapped.hEvent = CreateEvent(NULL, FALSE, FALSE, NULL); | |
157 | if (sock->overlapped.hEvent == NULL) { | |
52a1540a | 158 | VLOG_ERR("fcntl: %s", ovs_lasterror_to_string()); |
22326ba6 AS |
159 | goto error; |
160 | } | |
64513e68 EE |
161 | /* Initialize the type/ioctl to Generic */ |
162 | sock->read_ioctl = OVS_IOCTL_READ; | |
22326ba6 | 163 | #else |
2fe27d5a BP |
164 | sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol); |
165 | if (sock->fd < 0) { | |
10a89ef0 | 166 | VLOG_ERR("fcntl: %s", ovs_strerror(errno)); |
2fe27d5a BP |
167 | goto error; |
168 | } | |
22326ba6 AS |
169 | #endif |
170 | ||
7041c3a9 | 171 | sock->protocol = protocol; |
7d7447df | 172 | sock->next_seq = 1; |
2fe27d5a | 173 | |
d2b9f5b0 | 174 | rcvbuf = 1024 * 1024; |
22326ba6 AS |
175 | #ifdef _WIN32 |
176 | sock->rcvbuf = rcvbuf; | |
ebac7fb7 | 177 | retval = get_sock_pid_from_kernel(sock); |
886dd35a NR |
178 | if (retval != 0) { |
179 | goto error; | |
180 | } | |
22326ba6 | 181 | #else |
d2b9f5b0 BP |
182 | if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE, |
183 | &rcvbuf, sizeof rcvbuf)) { | |
80af5ee5 BP |
184 | /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM. |
185 | * Warn only if the failure is therefore unexpected. */ | |
f28b6dd3 | 186 | if (errno != EPERM) { |
80af5ee5 | 187 | VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed " |
10a89ef0 | 188 | "(%s)", rcvbuf, ovs_strerror(errno)); |
80af5ee5 | 189 | } |
d2b9f5b0 BP |
190 | } |
191 | ||
cc75061a BP |
192 | retval = get_socket_rcvbuf(sock->fd); |
193 | if (retval < 0) { | |
194 | retval = -retval; | |
195 | goto error; | |
196 | } | |
197 | sock->rcvbuf = retval; | |
198 | ||
2c5a6834 | 199 | /* Connect to kernel (pid 0) as remote address. */ |
2fe27d5a BP |
200 | memset(&remote, 0, sizeof remote); |
201 | remote.nl_family = AF_NETLINK; | |
202 | remote.nl_pid = 0; | |
203 | if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) { | |
10a89ef0 | 204 | VLOG_ERR("connect(0): %s", ovs_strerror(errno)); |
2c5a6834 BP |
205 | goto error; |
206 | } | |
207 | ||
208 | /* Obtain pid assigned by kernel. */ | |
209 | local_size = sizeof local; | |
210 | if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) { | |
10a89ef0 | 211 | VLOG_ERR("getsockname: %s", ovs_strerror(errno)); |
2c5a6834 BP |
212 | goto error; |
213 | } | |
214 | if (local_size < sizeof local || local.nl_family != AF_NETLINK) { | |
215 | VLOG_ERR("getsockname returned bad Netlink name"); | |
216 | retval = EINVAL; | |
217 | goto error; | |
2fe27d5a | 218 | } |
2c5a6834 | 219 | sock->pid = local.nl_pid; |
22326ba6 | 220 | #endif |
2fe27d5a | 221 | |
2fe27d5a BP |
222 | *sockp = sock; |
223 | return 0; | |
224 | ||
2fe27d5a BP |
225 | error: |
226 | if (retval == 0) { | |
227 | retval = errno; | |
228 | if (retval == 0) { | |
229 | retval = EINVAL; | |
230 | } | |
231 | } | |
22326ba6 | 232 | #ifdef _WIN32 |
7fa09611 EE |
233 | if (sock->overlapped.hEvent) { |
234 | CloseHandle(sock->overlapped.hEvent); | |
235 | } | |
22326ba6 AS |
236 | if (sock->handle != INVALID_HANDLE_VALUE) { |
237 | CloseHandle(sock->handle); | |
238 | } | |
239 | #else | |
2fe27d5a BP |
240 | if (sock->fd >= 0) { |
241 | close(sock->fd); | |
242 | } | |
22326ba6 | 243 | #endif |
2fe27d5a BP |
244 | free(sock); |
245 | return retval; | |
246 | } | |
247 | ||
c6eab56d BP |
248 | /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and |
249 | * sets '*sockp' to the new socket if successful, otherwise returns a positive | |
250 | * errno value. */ | |
251 | int | |
252 | nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp) | |
253 | { | |
254 | return nl_sock_create(src->protocol, sockp); | |
255 | } | |
256 | ||
2fe27d5a BP |
257 | /* Destroys netlink socket 'sock'. */ |
258 | void | |
259 | nl_sock_destroy(struct nl_sock *sock) | |
260 | { | |
261 | if (sock) { | |
22326ba6 | 262 | #ifdef _WIN32 |
7fa09611 EE |
263 | if (sock->overlapped.hEvent) { |
264 | CloseHandle(sock->overlapped.hEvent); | |
265 | } | |
22326ba6 AS |
266 | CloseHandle(sock->handle); |
267 | #else | |
a88b4e04 | 268 | close(sock->fd); |
22326ba6 | 269 | #endif |
a88b4e04 | 270 | free(sock); |
2fe27d5a BP |
271 | } |
272 | } | |
273 | ||
886dd35a NR |
274 | #ifdef _WIN32 |
275 | /* Reads the pid for 'sock' generated in the kernel datapath. The function | |
276 | * follows a transaction semantic. Eventually this function should call into | |
277 | * nl_transact. */ | |
278 | static int | |
ebac7fb7 | 279 | get_sock_pid_from_kernel(struct nl_sock *sock) |
886dd35a NR |
280 | { |
281 | struct nl_transaction txn; | |
282 | struct ofpbuf request; | |
283 | uint64_t request_stub[128]; | |
284 | struct ofpbuf reply; | |
285 | uint64_t reply_stub[128]; | |
286 | struct ovs_header *ovs_header; | |
287 | struct nlmsghdr *nlmsg; | |
288 | uint32_t seq; | |
289 | int retval; | |
290 | DWORD bytes; | |
291 | int ovs_msg_size = sizeof (struct nlmsghdr) + sizeof (struct genlmsghdr) + | |
292 | sizeof (struct ovs_header); | |
293 | ||
294 | ofpbuf_use_stub(&request, request_stub, sizeof request_stub); | |
295 | txn.request = &request; | |
296 | ofpbuf_use_stub(&reply, reply_stub, sizeof reply_stub); | |
297 | txn.reply = &reply; | |
298 | ||
299 | seq = nl_sock_allocate_seq(sock, 1); | |
300 | nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0, | |
301 | OVS_CTRL_CMD_WIN_GET_PID, OVS_WIN_CONTROL_VERSION); | |
302 | nlmsg = nl_msg_nlmsghdr(txn.request); | |
303 | nlmsg->nlmsg_seq = seq; | |
304 | ||
305 | ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header); | |
306 | ovs_header->dp_ifindex = 0; | |
307 | ovs_header = ofpbuf_put_uninit(&reply, ovs_msg_size); | |
308 | ||
309 | if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT, | |
310 | ofpbuf_data(txn.request), ofpbuf_size(txn.request), | |
311 | ofpbuf_data(txn.reply), ofpbuf_size(txn.reply), | |
312 | &bytes, NULL)) { | |
313 | retval = EINVAL; | |
314 | goto done; | |
315 | } else { | |
316 | if (bytes < ovs_msg_size) { | |
317 | retval = EINVAL; | |
318 | goto done; | |
319 | } | |
320 | ||
ebac7fb7 | 321 | nlmsg = nl_msg_nlmsghdr(txn.reply); |
886dd35a NR |
322 | if (nlmsg->nlmsg_seq != seq) { |
323 | retval = EINVAL; | |
324 | goto done; | |
325 | } | |
ebac7fb7 | 326 | sock->pid = nlmsg->nlmsg_pid; |
886dd35a NR |
327 | } |
328 | retval = 0; | |
329 | ||
330 | done: | |
331 | ofpbuf_uninit(&request); | |
332 | ofpbuf_uninit(&reply); | |
333 | return retval; | |
334 | } | |
335 | #endif /* _WIN32 */ | |
336 | ||
b8f958ea EE |
337 | #ifdef _WIN32 |
338 | static int __inline | |
339 | nl_sock_mcgroup(struct nl_sock *sock, unsigned int multicast_group, bool join) | |
340 | { | |
341 | struct ofpbuf request; | |
342 | uint64_t request_stub[128]; | |
343 | struct ovs_header *ovs_header; | |
344 | struct nlmsghdr *nlmsg; | |
345 | int error; | |
346 | ||
347 | ofpbuf_use_stub(&request, request_stub, sizeof request_stub); | |
348 | ||
349 | nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0, | |
350 | OVS_CTRL_CMD_MC_SUBSCRIBE_REQ, | |
351 | OVS_WIN_CONTROL_VERSION); | |
352 | ||
353 | ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header); | |
354 | ovs_header->dp_ifindex = 0; | |
355 | ||
356 | nl_msg_put_u32(&request, OVS_NL_ATTR_MCAST_GRP, multicast_group); | |
357 | nl_msg_put_u8(&request, OVS_NL_ATTR_MCAST_JOIN, join ? 1 : 0); | |
358 | ||
359 | error = nl_sock_send(sock, &request, true); | |
360 | ofpbuf_uninit(&request); | |
361 | return error; | |
362 | } | |
363 | #endif | |
cceb11f5 BP |
364 | /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if |
365 | * successful, otherwise a positive errno value. | |
366 | * | |
a838c4fe BP |
367 | * A socket that is subscribed to a multicast group that receives asynchronous |
368 | * notifications must not be used for Netlink transactions or dumps, because | |
369 | * transactions and dumps can cause notifications to be lost. | |
370 | * | |
cceb11f5 BP |
371 | * Multicast group numbers are always positive. |
372 | * | |
373 | * It is not an error to attempt to join a multicast group to which a socket | |
374 | * already belongs. */ | |
375 | int | |
376 | nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
377 | { | |
22326ba6 | 378 | #ifdef _WIN32 |
64513e68 EE |
379 | /* Set the socket type as a "multicast" socket */ |
380 | sock->read_ioctl = OVS_IOCTL_READ_EVENT; | |
b8f958ea EE |
381 | int error = nl_sock_mcgroup(sock, multicast_group, true); |
382 | if (error) { | |
64513e68 | 383 | sock->read_ioctl = OVS_IOCTL_READ; |
b8f958ea | 384 | VLOG_WARN("could not join multicast group %u (%s)", |
15fd9052 NR |
385 | multicast_group, ovs_strerror(error)); |
386 | return error; | |
b8f958ea | 387 | } |
22326ba6 | 388 | #else |
cceb11f5 BP |
389 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, |
390 | &multicast_group, sizeof multicast_group) < 0) { | |
391 | VLOG_WARN("could not join multicast group %u (%s)", | |
10a89ef0 | 392 | multicast_group, ovs_strerror(errno)); |
cceb11f5 BP |
393 | return errno; |
394 | } | |
22326ba6 | 395 | #endif |
cceb11f5 BP |
396 | return 0; |
397 | } | |
398 | ||
36791e21 NR |
399 | #ifdef _WIN32 |
400 | int | |
401 | nl_sock_subscribe_packets(struct nl_sock *sock) | |
402 | { | |
403 | int error; | |
404 | ||
405 | if (sock->read_ioctl != OVS_IOCTL_READ) { | |
406 | return EINVAL; | |
407 | } | |
408 | ||
409 | error = nl_sock_subscribe_packet__(sock, true); | |
410 | if (error) { | |
411 | VLOG_WARN("could not unsubscribe packets (%s)", | |
412 | ovs_strerror(errno)); | |
413 | return error; | |
414 | } | |
415 | sock->read_ioctl = OVS_IOCTL_READ_PACKET; | |
416 | ||
417 | return 0; | |
418 | } | |
419 | ||
420 | int | |
421 | nl_sock_unsubscribe_packets(struct nl_sock *sock) | |
422 | { | |
423 | ovs_assert(sock->read_ioctl == OVS_IOCTL_READ_PACKET); | |
424 | ||
425 | int error = nl_sock_subscribe_packet__(sock, false); | |
426 | if (error) { | |
427 | VLOG_WARN("could not subscribe to packets (%s)", | |
428 | ovs_strerror(errno)); | |
429 | return error; | |
430 | } | |
431 | ||
432 | sock->read_ioctl = OVS_IOCTL_READ; | |
433 | return 0; | |
434 | } | |
435 | ||
436 | int | |
437 | nl_sock_subscribe_packet__(struct nl_sock *sock, bool subscribe) | |
438 | { | |
439 | struct ofpbuf request; | |
440 | uint64_t request_stub[128]; | |
441 | struct ovs_header *ovs_header; | |
442 | struct nlmsghdr *nlmsg; | |
443 | int error; | |
444 | ||
445 | ofpbuf_use_stub(&request, request_stub, sizeof request_stub); | |
446 | nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0, | |
447 | OVS_CTRL_CMD_PACKET_SUBSCRIBE_REQ, | |
448 | OVS_WIN_CONTROL_VERSION); | |
449 | ||
450 | ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header); | |
451 | ovs_header->dp_ifindex = 0; | |
452 | nl_msg_put_u8(&request, OVS_NL_ATTR_PACKET_SUBSCRIBE, subscribe ? 1 : 0); | |
453 | nl_msg_put_u32(&request, OVS_NL_ATTR_PACKET_PID, sock->pid); | |
454 | ||
455 | error = nl_sock_send(sock, &request, true); | |
456 | ofpbuf_uninit(&request); | |
457 | return error; | |
458 | } | |
459 | #endif | |
460 | ||
cceb11f5 BP |
461 | /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if |
462 | * successful, otherwise a positive errno value. | |
463 | * | |
464 | * Multicast group numbers are always positive. | |
465 | * | |
466 | * It is not an error to attempt to leave a multicast group to which a socket | |
467 | * does not belong. | |
468 | * | |
469 | * On success, reading from 'sock' will still return any messages that were | |
470 | * received on 'multicast_group' before the group was left. */ | |
471 | int | |
472 | nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
473 | { | |
22326ba6 | 474 | #ifdef _WIN32 |
b8f958ea EE |
475 | int error = nl_sock_mcgroup(sock, multicast_group, false); |
476 | if (error) { | |
477 | VLOG_WARN("could not leave multicast group %u (%s)", | |
15fd9052 NR |
478 | multicast_group, ovs_strerror(error)); |
479 | return error; | |
b8f958ea | 480 | } |
64513e68 | 481 | sock->read_ioctl = OVS_IOCTL_READ; |
22326ba6 | 482 | #else |
cceb11f5 BP |
483 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP, |
484 | &multicast_group, sizeof multicast_group) < 0) { | |
485 | VLOG_WARN("could not leave multicast group %u (%s)", | |
10a89ef0 | 486 | multicast_group, ovs_strerror(errno)); |
cceb11f5 BP |
487 | return errno; |
488 | } | |
22326ba6 | 489 | #endif |
cceb11f5 BP |
490 | return 0; |
491 | } | |
492 | ||
c6eab56d | 493 | static int |
ff459dd6 BP |
494 | nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, |
495 | uint32_t nlmsg_seq, bool wait) | |
2fe27d5a BP |
496 | { |
497 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg); | |
498 | int error; | |
499 | ||
1f317cb5 | 500 | nlmsg->nlmsg_len = ofpbuf_size(msg); |
ff459dd6 | 501 | nlmsg->nlmsg_seq = nlmsg_seq; |
2fe27d5a BP |
502 | nlmsg->nlmsg_pid = sock->pid; |
503 | do { | |
504 | int retval; | |
22326ba6 | 505 | #ifdef _WIN32 |
fd972eb8 NR |
506 | DWORD bytes; |
507 | ||
508 | if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE, | |
0fd22ae2 NR |
509 | ofpbuf_data(msg), ofpbuf_size(msg), NULL, 0, |
510 | &bytes, NULL)) { | |
22326ba6 | 511 | retval = -1; |
fd972eb8 NR |
512 | /* XXX: Map to a more appropriate error based on GetLastError(). */ |
513 | errno = EINVAL; | |
514 | } else { | |
515 | retval = ofpbuf_size(msg); | |
22326ba6 AS |
516 | } |
517 | #else | |
fd972eb8 NR |
518 | retval = send(sock->fd, ofpbuf_data(msg), ofpbuf_size(msg), |
519 | wait ? 0 : MSG_DONTWAIT); | |
22326ba6 | 520 | #endif |
2fe27d5a BP |
521 | error = retval < 0 ? errno : 0; |
522 | } while (error == EINTR); | |
1f317cb5 | 523 | log_nlmsg(__func__, error, ofpbuf_data(msg), ofpbuf_size(msg), sock->protocol); |
2fe27d5a BP |
524 | if (!error) { |
525 | COVERAGE_INC(netlink_sent); | |
526 | } | |
527 | return error; | |
528 | } | |
529 | ||
c6eab56d | 530 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on |
1f317cb5 | 531 | * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid |
ff459dd6 BP |
532 | * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh |
533 | * sequence number, before the message is sent. | |
c6eab56d BP |
534 | * |
535 | * Returns 0 if successful, otherwise a positive errno value. If | |
536 | * 'wait' is true, then the send will wait until buffer space is ready; | |
537 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. */ | |
538 | int | |
539 | nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait) | |
ff459dd6 BP |
540 | { |
541 | return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait); | |
542 | } | |
543 | ||
544 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on | |
1f317cb5 | 545 | * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid |
ff459dd6 BP |
546 | * will be set to 'sock''s pid, and nlmsg_seq will be initialized to |
547 | * 'nlmsg_seq', before the message is sent. | |
548 | * | |
549 | * Returns 0 if successful, otherwise a positive errno value. If | |
550 | * 'wait' is true, then the send will wait until buffer space is ready; | |
551 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. | |
552 | * | |
553 | * This function is suitable for sending a reply to a request that was received | |
554 | * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */ | |
555 | int | |
556 | nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg, | |
557 | uint32_t nlmsg_seq, bool wait) | |
c6eab56d | 558 | { |
ff459dd6 | 559 | return nl_sock_send__(sock, msg, nlmsg_seq, wait); |
c6eab56d BP |
560 | } |
561 | ||
c6eab56d | 562 | static int |
72d32ac0 | 563 | nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait) |
2fe27d5a | 564 | { |
72d32ac0 BP |
565 | /* We can't accurately predict the size of the data to be received. The |
566 | * caller is supposed to have allocated enough space in 'buf' to handle the | |
567 | * "typical" case. To handle exceptions, we make available enough space in | |
568 | * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable | |
569 | * figure since that's the maximum length of a Netlink attribute). */ | |
2fe27d5a | 570 | struct nlmsghdr *nlmsghdr; |
72d32ac0 | 571 | uint8_t tail[65536]; |
fc999dda | 572 | struct iovec iov[2]; |
fc999dda BP |
573 | struct msghdr msg; |
574 | ssize_t retval; | |
8f20fd98 | 575 | int error; |
fc999dda | 576 | |
cb22974d | 577 | ovs_assert(buf->allocated >= sizeof *nlmsghdr); |
72d32ac0 | 578 | ofpbuf_clear(buf); |
2fe27d5a | 579 | |
1f317cb5 | 580 | iov[0].iov_base = ofpbuf_base(buf); |
72d32ac0 | 581 | iov[0].iov_len = buf->allocated; |
fc999dda | 582 | iov[1].iov_base = tail; |
72d32ac0 | 583 | iov[1].iov_len = sizeof tail; |
fc999dda BP |
584 | |
585 | memset(&msg, 0, sizeof msg); | |
586 | msg.msg_iov = iov; | |
587 | msg.msg_iovlen = 2; | |
588 | ||
8f20fd98 BP |
589 | /* Receive a Netlink message from the kernel. |
590 | * | |
591 | * This works around a kernel bug in which the kernel returns an error code | |
592 | * as if it were the number of bytes read. It doesn't actually modify | |
593 | * anything in the receive buffer in that case, so we can initialize the | |
594 | * Netlink header with an impossible message length and then, upon success, | |
595 | * check whether it changed. */ | |
596 | nlmsghdr = ofpbuf_base(buf); | |
2fe27d5a | 597 | do { |
8f20fd98 | 598 | nlmsghdr->nlmsg_len = UINT32_MAX; |
22326ba6 | 599 | #ifdef _WIN32 |
fd972eb8 | 600 | DWORD bytes; |
64513e68 | 601 | if (!DeviceIoControl(sock->handle, sock->read_ioctl, |
fd972eb8 | 602 | NULL, 0, tail, sizeof tail, &bytes, NULL)) { |
22326ba6 | 603 | retval = -1; |
fd972eb8 | 604 | errno = EINVAL; |
22326ba6 | 605 | } else { |
fd972eb8 NR |
606 | retval = bytes; |
607 | if (retval == 0) { | |
608 | retval = -1; | |
609 | errno = EAGAIN; | |
610 | } else { | |
611 | if (retval >= buf->allocated) { | |
612 | ofpbuf_reinit(buf, retval); | |
15fd9052 NR |
613 | nlmsghdr = ofpbuf_base(buf); |
614 | nlmsghdr->nlmsg_len = UINT32_MAX; | |
fd972eb8 NR |
615 | } |
616 | memcpy(ofpbuf_data(buf), tail, retval); | |
617 | ofpbuf_set_size(buf, retval); | |
618 | } | |
22326ba6 AS |
619 | } |
620 | #else | |
fc999dda | 621 | retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT); |
22326ba6 | 622 | #endif |
8f20fd98 BP |
623 | error = (retval < 0 ? errno |
624 | : retval == 0 ? ECONNRESET /* not possible? */ | |
625 | : nlmsghdr->nlmsg_len != UINT32_MAX ? 0 | |
7f8e2646 | 626 | : retval); |
8f20fd98 BP |
627 | } while (error == EINTR); |
628 | if (error) { | |
fc999dda BP |
629 | if (error == ENOBUFS) { |
630 | /* Socket receive buffer overflow dropped one or more messages that | |
631 | * the kernel tried to send to us. */ | |
632 | COVERAGE_INC(netlink_overflow); | |
633 | } | |
fc999dda | 634 | return error; |
2fe27d5a | 635 | } |
fc999dda | 636 | |
2fe27d5a | 637 | if (msg.msg_flags & MSG_TRUNC) { |
34582733 | 638 | VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)", |
72d32ac0 | 639 | sizeof tail); |
fc999dda | 640 | return E2BIG; |
2fe27d5a | 641 | } |
2fe27d5a | 642 | |
fc999dda | 643 | if (retval < sizeof *nlmsghdr |
2fe27d5a | 644 | || nlmsghdr->nlmsg_len < sizeof *nlmsghdr |
fc999dda | 645 | || nlmsghdr->nlmsg_len > retval) { |
e5e4b47c | 646 | VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE" bytes < %"PRIuSIZE")", |
72d32ac0 | 647 | retval, sizeof *nlmsghdr); |
2fe27d5a BP |
648 | return EPROTO; |
649 | } | |
22326ba6 | 650 | #ifndef _WIN32 |
1f317cb5 | 651 | ofpbuf_set_size(buf, MIN(retval, buf->allocated)); |
72d32ac0 BP |
652 | if (retval > buf->allocated) { |
653 | COVERAGE_INC(netlink_recv_jumbo); | |
654 | ofpbuf_put(buf, tail, retval - buf->allocated); | |
655 | } | |
22326ba6 | 656 | #endif |
72d32ac0 | 657 | |
1f317cb5 | 658 | log_nlmsg(__func__, 0, ofpbuf_data(buf), ofpbuf_size(buf), sock->protocol); |
2fe27d5a BP |
659 | COVERAGE_INC(netlink_received); |
660 | ||
661 | return 0; | |
662 | } | |
663 | ||
72d32ac0 BP |
664 | /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If |
665 | * 'wait' is true, waits for a message to be ready. Otherwise, fails with | |
666 | * EAGAIN if the 'sock' receive buffer is empty. | |
667 | * | |
668 | * The caller must have initialized 'buf' with an allocation of at least | |
669 | * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough | |
670 | * space for a "typical" message. | |
671 | * | |
672 | * On success, returns 0 and replaces 'buf''s previous content by the received | |
673 | * message. This function expands 'buf''s allocated memory, as necessary, to | |
674 | * hold the actual size of the received message. | |
c6eab56d | 675 | * |
72d32ac0 BP |
676 | * On failure, returns a positive errno value and clears 'buf' to zero length. |
677 | * 'buf' retains its previous memory allocation. | |
678 | * | |
679 | * Regardless of success or failure, this function resets 'buf''s headroom to | |
680 | * 0. */ | |
c6eab56d | 681 | int |
72d32ac0 | 682 | nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait) |
c6eab56d | 683 | { |
72d32ac0 | 684 | return nl_sock_recv__(sock, buf, wait); |
cc75061a BP |
685 | } |
686 | ||
687 | static void | |
688 | nl_sock_record_errors__(struct nl_transaction **transactions, size_t n, | |
689 | int error) | |
690 | { | |
691 | size_t i; | |
692 | ||
693 | for (i = 0; i < n; i++) { | |
72d32ac0 BP |
694 | struct nl_transaction *txn = transactions[i]; |
695 | ||
696 | txn->error = error; | |
697 | if (txn->reply) { | |
698 | ofpbuf_clear(txn->reply); | |
699 | } | |
cc75061a BP |
700 | } |
701 | } | |
702 | ||
703 | static int | |
704 | nl_sock_transact_multiple__(struct nl_sock *sock, | |
705 | struct nl_transaction **transactions, size_t n, | |
706 | size_t *done) | |
707 | { | |
72d32ac0 BP |
708 | uint64_t tmp_reply_stub[1024 / 8]; |
709 | struct nl_transaction tmp_txn; | |
710 | struct ofpbuf tmp_reply; | |
711 | ||
712 | uint32_t base_seq; | |
cc75061a BP |
713 | struct iovec iovs[MAX_IOVS]; |
714 | struct msghdr msg; | |
715 | int error; | |
716 | int i; | |
717 | ||
72d32ac0 | 718 | base_seq = nl_sock_allocate_seq(sock, n); |
cc75061a BP |
719 | *done = 0; |
720 | for (i = 0; i < n; i++) { | |
72d32ac0 BP |
721 | struct nl_transaction *txn = transactions[i]; |
722 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request); | |
cc75061a | 723 | |
1f317cb5 | 724 | nlmsg->nlmsg_len = ofpbuf_size(txn->request); |
72d32ac0 | 725 | nlmsg->nlmsg_seq = base_seq + i; |
cc75061a | 726 | nlmsg->nlmsg_pid = sock->pid; |
cc75061a | 727 | |
1f317cb5 PS |
728 | iovs[i].iov_base = ofpbuf_data(txn->request); |
729 | iovs[i].iov_len = ofpbuf_size(txn->request); | |
cc75061a BP |
730 | } |
731 | ||
0fd22ae2 | 732 | #ifndef _WIN32 |
cc75061a BP |
733 | memset(&msg, 0, sizeof msg); |
734 | msg.msg_iov = iovs; | |
735 | msg.msg_iovlen = n; | |
736 | do { | |
737 | error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0; | |
738 | } while (error == EINTR); | |
739 | ||
740 | for (i = 0; i < n; i++) { | |
72d32ac0 | 741 | struct nl_transaction *txn = transactions[i]; |
cc75061a | 742 | |
0fd22ae2 NR |
743 | log_nlmsg(__func__, error, ofpbuf_data(txn->request), |
744 | ofpbuf_size(txn->request), sock->protocol); | |
cc75061a BP |
745 | } |
746 | if (!error) { | |
747 | COVERAGE_ADD(netlink_sent, n); | |
748 | } | |
749 | ||
750 | if (error) { | |
751 | return error; | |
752 | } | |
753 | ||
72d32ac0 BP |
754 | ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub); |
755 | tmp_txn.request = NULL; | |
756 | tmp_txn.reply = &tmp_reply; | |
757 | tmp_txn.error = 0; | |
cc75061a | 758 | while (n > 0) { |
72d32ac0 BP |
759 | struct nl_transaction *buf_txn, *txn; |
760 | uint32_t seq; | |
761 | ||
762 | /* Find a transaction whose buffer we can use for receiving a reply. | |
763 | * If no such transaction is left, use tmp_txn. */ | |
764 | buf_txn = &tmp_txn; | |
765 | for (i = 0; i < n; i++) { | |
766 | if (transactions[i]->reply) { | |
767 | buf_txn = transactions[i]; | |
768 | break; | |
769 | } | |
770 | } | |
cc75061a | 771 | |
72d32ac0 BP |
772 | /* Receive a reply. */ |
773 | error = nl_sock_recv__(sock, buf_txn->reply, false); | |
774 | if (error) { | |
775 | if (error == EAGAIN) { | |
776 | nl_sock_record_errors__(transactions, n, 0); | |
777 | *done += n; | |
778 | error = 0; | |
779 | } | |
780 | break; | |
cc75061a BP |
781 | } |
782 | ||
72d32ac0 BP |
783 | /* Match the reply up with a transaction. */ |
784 | seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq; | |
785 | if (seq < base_seq || seq >= base_seq + n) { | |
786 | VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq); | |
cc75061a BP |
787 | continue; |
788 | } | |
72d32ac0 BP |
789 | i = seq - base_seq; |
790 | txn = transactions[i]; | |
cc75061a | 791 | |
72d32ac0 BP |
792 | /* Fill in the results for 'txn'. */ |
793 | if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) { | |
794 | if (txn->reply) { | |
795 | ofpbuf_clear(txn->reply); | |
796 | } | |
797 | if (txn->error) { | |
cc75061a | 798 | VLOG_DBG_RL(&rl, "received NAK error=%d (%s)", |
10a89ef0 | 799 | error, ovs_strerror(txn->error)); |
cc75061a | 800 | } |
cc75061a | 801 | } else { |
72d32ac0 BP |
802 | txn->error = 0; |
803 | if (txn->reply && txn != buf_txn) { | |
804 | /* Swap buffers. */ | |
805 | struct ofpbuf *reply = buf_txn->reply; | |
806 | buf_txn->reply = txn->reply; | |
807 | txn->reply = reply; | |
808 | } | |
cc75061a BP |
809 | } |
810 | ||
72d32ac0 BP |
811 | /* Fill in the results for transactions before 'txn'. (We have to do |
812 | * this after the results for 'txn' itself because of the buffer swap | |
813 | * above.) */ | |
814 | nl_sock_record_errors__(transactions, i, 0); | |
815 | ||
816 | /* Advance. */ | |
cc75061a BP |
817 | *done += i + 1; |
818 | transactions += i + 1; | |
819 | n -= i + 1; | |
72d32ac0 | 820 | base_seq += i + 1; |
cc75061a | 821 | } |
72d32ac0 | 822 | ofpbuf_uninit(&tmp_reply); |
0fd22ae2 NR |
823 | #else |
824 | error = 0; | |
9189184d | 825 | uint8_t reply_buf[65536]; |
0fd22ae2 NR |
826 | for (i = 0; i < n; i++) { |
827 | DWORD reply_len; | |
0fd22ae2 NR |
828 | struct nl_transaction *txn = transactions[i]; |
829 | struct nlmsghdr *request_nlmsg, *reply_nlmsg; | |
830 | ||
831 | if (!DeviceIoControl(sock->handle, OVS_IOCTL_TRANSACT, | |
832 | ofpbuf_data(txn->request), | |
833 | ofpbuf_size(txn->request), | |
9189184d | 834 | reply_buf, sizeof reply_buf, |
0fd22ae2 NR |
835 | &reply_len, NULL)) { |
836 | /* XXX: Map to a more appropriate error. */ | |
837 | error = EINVAL; | |
838 | break; | |
839 | } | |
840 | ||
9189184d NR |
841 | if (reply_len < sizeof *reply_nlmsg) { |
842 | nl_sock_record_errors__(transactions, n, 0); | |
843 | VLOG_DBG_RL(&rl, "insufficient length of reply %#"PRIu32 | |
844 | " for seq: %#"PRIx32, reply_len, request_nlmsg->nlmsg_seq); | |
845 | break; | |
846 | } | |
0fd22ae2 | 847 | |
9189184d NR |
848 | /* Validate the sequence number in the reply. */ |
849 | request_nlmsg = nl_msg_nlmsghdr(txn->request); | |
850 | reply_nlmsg = (struct nlmsghdr *)reply_buf; | |
0fd22ae2 | 851 | |
9189184d NR |
852 | if (request_nlmsg->nlmsg_seq != reply_nlmsg->nlmsg_seq) { |
853 | ovs_assert(request_nlmsg->nlmsg_seq == reply_nlmsg->nlmsg_seq); | |
854 | VLOG_DBG_RL(&rl, "mismatched seq request %#"PRIx32 | |
855 | ", reply %#"PRIx32, request_nlmsg->nlmsg_seq, | |
856 | reply_nlmsg->nlmsg_seq); | |
857 | break; | |
858 | } | |
0fd22ae2 | 859 | |
9189184d NR |
860 | /* Handle errors embedded within the netlink message. */ |
861 | ofpbuf_use_stub(&tmp_reply, reply_buf, sizeof reply_buf); | |
862 | ofpbuf_set_size(&tmp_reply, sizeof reply_buf); | |
863 | if (nl_msg_nlmsgerr(&tmp_reply, &txn->error)) { | |
864 | if (txn->reply) { | |
865 | ofpbuf_clear(txn->reply); | |
0fd22ae2 | 866 | } |
9189184d NR |
867 | if (txn->error) { |
868 | VLOG_DBG_RL(&rl, "received NAK error=%d (%s)", | |
869 | error, ovs_strerror(txn->error)); | |
0fd22ae2 | 870 | } |
9189184d NR |
871 | } else { |
872 | txn->error = 0; | |
873 | if (txn->reply) { | |
874 | /* Copy the reply to the buffer specified by the caller. */ | |
875 | if (reply_len > txn->reply->allocated) { | |
876 | ofpbuf_reinit(txn->reply, reply_len); | |
0fd22ae2 | 877 | } |
9189184d NR |
878 | memcpy(ofpbuf_data(txn->reply), reply_buf, reply_len); |
879 | ofpbuf_set_size(txn->reply, reply_len); | |
0fd22ae2 NR |
880 | } |
881 | } | |
9189184d | 882 | ofpbuf_uninit(&tmp_reply); |
0fd22ae2 NR |
883 | |
884 | /* Count the number of successful transactions. */ | |
885 | (*done)++; | |
9189184d | 886 | |
0fd22ae2 NR |
887 | } |
888 | ||
889 | if (!error) { | |
890 | COVERAGE_ADD(netlink_sent, n); | |
891 | } | |
892 | #endif | |
cc75061a | 893 | |
72d32ac0 | 894 | return error; |
cc75061a BP |
895 | } |
896 | ||
022ad2b9 | 897 | static void |
cc75061a BP |
898 | nl_sock_transact_multiple(struct nl_sock *sock, |
899 | struct nl_transaction **transactions, size_t n) | |
900 | { | |
901 | int max_batch_count; | |
902 | int error; | |
903 | ||
904 | if (!n) { | |
905 | return; | |
906 | } | |
907 | ||
cc75061a BP |
908 | /* In theory, every request could have a 64 kB reply. But the default and |
909 | * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to | |
910 | * be a bit below 128 kB, so that would only allow a single message in a | |
911 | * "batch". So we assume that replies average (at most) 4 kB, which allows | |
912 | * a good deal of batching. | |
913 | * | |
914 | * In practice, most of the requests that we batch either have no reply at | |
915 | * all or a brief reply. */ | |
916 | max_batch_count = MAX(sock->rcvbuf / 4096, 1); | |
917 | max_batch_count = MIN(max_batch_count, max_iovs); | |
918 | ||
919 | while (n > 0) { | |
920 | size_t count, bytes; | |
921 | size_t done; | |
922 | ||
923 | /* Batch up to 'max_batch_count' transactions. But cap it at about a | |
924 | * page of requests total because big skbuffs are expensive to | |
925 | * allocate in the kernel. */ | |
926 | #if defined(PAGESIZE) | |
927 | enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) }; | |
928 | #else | |
929 | enum { MAX_BATCH_BYTES = 4096 - 512 }; | |
930 | #endif | |
1f317cb5 | 931 | bytes = ofpbuf_size(transactions[0]->request); |
cc75061a | 932 | for (count = 1; count < n && count < max_batch_count; count++) { |
1f317cb5 | 933 | if (bytes + ofpbuf_size(transactions[count]->request) > MAX_BATCH_BYTES) { |
cc75061a BP |
934 | break; |
935 | } | |
1f317cb5 | 936 | bytes += ofpbuf_size(transactions[count]->request); |
cc75061a BP |
937 | } |
938 | ||
939 | error = nl_sock_transact_multiple__(sock, transactions, count, &done); | |
940 | transactions += done; | |
941 | n -= done; | |
942 | ||
943 | if (error == ENOBUFS) { | |
944 | VLOG_DBG_RL(&rl, "receive buffer overflow, resending request"); | |
945 | } else if (error) { | |
10a89ef0 | 946 | VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error)); |
cc75061a BP |
947 | nl_sock_record_errors__(transactions, n, error); |
948 | } | |
949 | } | |
950 | } | |
951 | ||
022ad2b9 | 952 | static int |
cc75061a BP |
953 | nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request, |
954 | struct ofpbuf **replyp) | |
2fe27d5a | 955 | { |
cc75061a BP |
956 | struct nl_transaction *transactionp; |
957 | struct nl_transaction transaction; | |
2fe27d5a | 958 | |
ebc56baa | 959 | transaction.request = CONST_CAST(struct ofpbuf *, request); |
72d32ac0 | 960 | transaction.reply = replyp ? ofpbuf_new(1024) : NULL; |
cc75061a | 961 | transactionp = &transaction; |
72d32ac0 | 962 | |
cc75061a | 963 | nl_sock_transact_multiple(sock, &transactionp, 1); |
72d32ac0 | 964 | |
2fe27d5a | 965 | if (replyp) { |
72d32ac0 BP |
966 | if (transaction.error) { |
967 | ofpbuf_delete(transaction.reply); | |
968 | *replyp = NULL; | |
969 | } else { | |
970 | *replyp = transaction.reply; | |
971 | } | |
2fe27d5a | 972 | } |
72d32ac0 | 973 | |
cc75061a | 974 | return transaction.error; |
2fe27d5a BP |
975 | } |
976 | ||
6b7c12fd BP |
977 | /* Drain all the messages currently in 'sock''s receive queue. */ |
978 | int | |
979 | nl_sock_drain(struct nl_sock *sock) | |
980 | { | |
22326ba6 AS |
981 | #ifdef _WIN32 |
982 | return 0; | |
983 | #else | |
6b7c12fd | 984 | return drain_rcvbuf(sock->fd); |
22326ba6 | 985 | #endif |
6b7c12fd BP |
986 | } |
987 | ||
a88b4e04 BP |
988 | /* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a |
989 | * Netlink socket created with the given 'protocol', and initializes 'dump' to | |
990 | * reflect the state of the operation. | |
2fe27d5a | 991 | * |
db1fc210 JS |
992 | * 'request' must contain a Netlink message. Before sending the message, |
993 | * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be | |
994 | * set to the Netlink socket's pid. NLM_F_DUMP and NLM_F_ACK will be set in | |
995 | * nlmsg_flags. | |
2fe27d5a | 996 | * |
a88b4e04 | 997 | * The design of this Netlink socket library ensures that the dump is reliable. |
2fe27d5a | 998 | * |
db1fc210 JS |
999 | * This function provides no status indication. nl_dump_done() provides an |
1000 | * error status for the entire dump operation. | |
2fe27d5a | 1001 | * |
db1fc210 | 1002 | * The caller must eventually destroy 'request'. |
2fe27d5a BP |
1003 | */ |
1004 | void | |
a88b4e04 | 1005 | nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request) |
2fe27d5a | 1006 | { |
7d7447df | 1007 | nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK; |
93295354 BP |
1008 | |
1009 | ovs_mutex_init(&dump->mutex); | |
1010 | ovs_mutex_lock(&dump->mutex); | |
1011 | dump->status = nl_pool_alloc(protocol, &dump->sock); | |
1012 | if (!dump->status) { | |
1013 | dump->status = nl_sock_send__(dump->sock, request, | |
1014 | nl_sock_allocate_seq(dump->sock, 1), | |
1015 | true); | |
b2d1c78a | 1016 | } |
9c8ad495 | 1017 | dump->nl_seq = nl_msg_nlmsghdr(request)->nlmsg_seq; |
93295354 BP |
1018 | ovs_mutex_unlock(&dump->mutex); |
1019 | } | |
1020 | ||
1021 | static int | |
1022 | nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer) | |
1023 | OVS_REQUIRES(dump->mutex) | |
1024 | { | |
1025 | struct nlmsghdr *nlmsghdr; | |
1026 | int error; | |
1027 | ||
1028 | while (!ofpbuf_size(buffer)) { | |
1738803a | 1029 | error = nl_sock_recv__(dump->sock, buffer, false); |
93295354 | 1030 | if (error) { |
1738803a AW |
1031 | /* The kernel never blocks providing the results of a dump, so |
1032 | * error == EAGAIN means that we've read the whole thing, and | |
1033 | * therefore transform it into EOF. (The kernel always provides | |
1034 | * NLMSG_DONE as a sentinel. Some other thread must have received | |
1035 | * that already but not yet signaled it in 'status'.) | |
1036 | * | |
1037 | * Any other error is just an error. */ | |
93295354 BP |
1038 | return error == EAGAIN ? EOF : error; |
1039 | } | |
1040 | ||
1041 | nlmsghdr = nl_msg_nlmsghdr(buffer); | |
1042 | if (dump->nl_seq != nlmsghdr->nlmsg_seq) { | |
1043 | VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32, | |
1044 | nlmsghdr->nlmsg_seq, dump->nl_seq); | |
1045 | ofpbuf_clear(buffer); | |
1046 | } | |
1047 | } | |
1048 | ||
1049 | if (nl_msg_nlmsgerr(buffer, &error) && error) { | |
1050 | VLOG_INFO_RL(&rl, "netlink dump request error (%s)", | |
1051 | ovs_strerror(error)); | |
1052 | ofpbuf_clear(buffer); | |
1053 | return error; | |
1054 | } | |
1055 | ||
1056 | return 0; | |
1057 | } | |
1058 | ||
1059 | static int | |
1060 | nl_dump_next__(struct ofpbuf *reply, struct ofpbuf *buffer) | |
1061 | { | |
1062 | struct nlmsghdr *nlmsghdr = nl_msg_next(buffer, reply); | |
1063 | if (!nlmsghdr) { | |
1064 | VLOG_WARN_RL(&rl, "netlink dump contains message fragment"); | |
1065 | return EPROTO; | |
1066 | } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) { | |
1067 | return EOF; | |
1068 | } else { | |
1069 | return 0; | |
1070 | } | |
2fe27d5a BP |
1071 | } |
1072 | ||
d57695d7 JS |
1073 | /* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must |
1074 | * have been initialized with nl_dump_start(), and 'buffer' must have been | |
1075 | * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long. | |
2fe27d5a | 1076 | * |
19aa20a0 BP |
1077 | * If successful, returns true and points 'reply->data' and |
1078 | * 'ofpbuf_size(reply)' to the message that was retrieved. The caller must not | |
1079 | * modify 'reply' (because it points within 'buffer', which will be used by | |
1080 | * future calls to this function). | |
1081 | * | |
1082 | * On failure, returns false and sets 'reply->data' to NULL and | |
1083 | * 'ofpbuf_size(reply)' to 0. Failure might indicate an actual error or merely | |
1084 | * the end of replies. An error status for the entire dump operation is | |
1085 | * provided when it is completed by calling nl_dump_done(). | |
0672776e JS |
1086 | * |
1087 | * Multiple threads may call this function, passing the same nl_dump, however | |
1088 | * each must provide independent buffers. This function may cache multiple | |
1089 | * replies in the buffer, and these will be processed before more replies are | |
1090 | * fetched. When this function returns false, other threads may continue to | |
1091 | * process replies in their buffers, but they will not fetch more replies. | |
2fe27d5a BP |
1092 | */ |
1093 | bool | |
d57695d7 | 1094 | nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer) |
2fe27d5a | 1095 | { |
93295354 | 1096 | int retval = 0; |
0672776e | 1097 | |
93295354 BP |
1098 | /* If the buffer is empty, refill it. |
1099 | * | |
1100 | * If the buffer is not empty, we don't check the dump's status. | |
1101 | * Otherwise, we could end up skipping some of the dump results if thread A | |
1102 | * hits EOF while thread B is in the midst of processing a batch. */ | |
1103 | if (!ofpbuf_size(buffer)) { | |
0791315e | 1104 | ovs_mutex_lock(&dump->mutex); |
93295354 BP |
1105 | if (!dump->status) { |
1106 | /* Take the mutex here to avoid an in-kernel race. If two threads | |
1107 | * try to read from a Netlink dump socket at once, then the socket | |
1108 | * error can be set to EINVAL, which will be encountered on the | |
1109 | * next recv on that socket, which could be anywhere due to the way | |
1110 | * that we pool Netlink sockets. Serializing the recv calls avoids | |
1111 | * the issue. */ | |
1112 | dump->status = nl_dump_refill(dump, buffer); | |
1113 | } | |
1114 | retval = dump->status; | |
0791315e | 1115 | ovs_mutex_unlock(&dump->mutex); |
93295354 | 1116 | } |
0791315e | 1117 | |
93295354 BP |
1118 | /* Fetch the next message from the buffer. */ |
1119 | if (!retval) { | |
1120 | retval = nl_dump_next__(reply, buffer); | |
2fe27d5a | 1121 | if (retval) { |
93295354 BP |
1122 | /* Record 'retval' as the dump status, but don't overwrite an error |
1123 | * with EOF. */ | |
1124 | ovs_mutex_lock(&dump->mutex); | |
1125 | if (dump->status <= 0) { | |
1126 | dump->status = retval; | |
2fe27d5a | 1127 | } |
93295354 | 1128 | ovs_mutex_unlock(&dump->mutex); |
2fe27d5a | 1129 | } |
2fe27d5a BP |
1130 | } |
1131 | ||
93295354 BP |
1132 | if (retval) { |
1133 | ofpbuf_set_data(reply, NULL); | |
1134 | ofpbuf_set_size(reply, 0); | |
0672776e | 1135 | } |
93295354 | 1136 | return !retval; |
2fe27d5a BP |
1137 | } |
1138 | ||
1139 | /* Completes Netlink dump operation 'dump', which must have been initialized | |
1140 | * with nl_dump_start(). Returns 0 if the dump operation was error-free, | |
1141 | * otherwise a positive errno value describing the problem. */ | |
1142 | int | |
1143 | nl_dump_done(struct nl_dump *dump) | |
1144 | { | |
0672776e | 1145 | int status; |
d57695d7 | 1146 | |
93295354 BP |
1147 | ovs_mutex_lock(&dump->mutex); |
1148 | status = dump->status; | |
1149 | ovs_mutex_unlock(&dump->mutex); | |
1150 | ||
2fe27d5a | 1151 | /* Drain any remaining messages that the client didn't read. Otherwise the |
a88b4e04 BP |
1152 | * kernel will continue to queue them up and waste buffer space. |
1153 | * | |
1154 | * XXX We could just destroy and discard the socket in this case. */ | |
0672776e JS |
1155 | if (!status) { |
1156 | uint64_t tmp_reply_stub[NL_DUMP_BUFSIZE / 8]; | |
1157 | struct ofpbuf reply, buf; | |
1158 | ||
1159 | ofpbuf_use_stub(&buf, tmp_reply_stub, sizeof tmp_reply_stub); | |
1160 | while (nl_dump_next(dump, &reply, &buf)) { | |
1161 | /* Nothing to do. */ | |
2fe27d5a | 1162 | } |
0672776e | 1163 | ofpbuf_uninit(&buf); |
93295354 BP |
1164 | |
1165 | ovs_mutex_lock(&dump->mutex); | |
1166 | status = dump->status; | |
1167 | ovs_mutex_unlock(&dump->mutex); | |
1168 | ovs_assert(status); | |
2fe27d5a | 1169 | } |
93295354 | 1170 | |
a88b4e04 | 1171 | nl_pool_release(dump->sock); |
0791315e | 1172 | ovs_mutex_destroy(&dump->mutex); |
93295354 BP |
1173 | |
1174 | return status == EOF ? 0 : status; | |
2fe27d5a BP |
1175 | } |
1176 | ||
7fa09611 EE |
1177 | #ifdef _WIN32 |
1178 | /* Pend an I/O request in the driver. The driver completes the I/O whenever | |
1179 | * an event or a packet is ready to be read. Once the I/O is completed | |
1180 | * the overlapped structure event associated with the pending I/O will be set | |
1181 | */ | |
1182 | static int | |
8341662d | 1183 | pend_io_request(struct nl_sock *sock) |
7fa09611 EE |
1184 | { |
1185 | struct ofpbuf request; | |
1186 | uint64_t request_stub[128]; | |
1187 | struct ovs_header *ovs_header; | |
1188 | struct nlmsghdr *nlmsg; | |
1189 | uint32_t seq; | |
1190 | int retval; | |
1191 | int error; | |
1192 | DWORD bytes; | |
64513e68 | 1193 | OVERLAPPED *overlapped = CONST_CAST(OVERLAPPED *, &sock->overlapped); |
7fa09611 EE |
1194 | |
1195 | int ovs_msg_size = sizeof (struct nlmsghdr) + sizeof (struct genlmsghdr) + | |
1196 | sizeof (struct ovs_header); | |
1197 | ||
1198 | ofpbuf_use_stub(&request, request_stub, sizeof request_stub); | |
1199 | ||
1200 | seq = nl_sock_allocate_seq(sock, 1); | |
1201 | nl_msg_put_genlmsghdr(&request, 0, OVS_WIN_NL_CTRL_FAMILY_ID, 0, | |
1202 | OVS_CTRL_CMD_WIN_PEND_REQ, OVS_WIN_CONTROL_VERSION); | |
1203 | nlmsg = nl_msg_nlmsghdr(&request); | |
1204 | nlmsg->nlmsg_seq = seq; | |
1205 | ||
1206 | ovs_header = ofpbuf_put_uninit(&request, sizeof *ovs_header); | |
1207 | ovs_header->dp_ifindex = 0; | |
1208 | ||
1209 | if (!DeviceIoControl(sock->handle, OVS_IOCTL_WRITE, | |
1210 | ofpbuf_data(&request), ofpbuf_size(&request), | |
1211 | NULL, 0, &bytes, overlapped)) { | |
1212 | error = GetLastError(); | |
1213 | /* Check if the I/O got pended */ | |
1214 | if (error != ERROR_IO_INCOMPLETE && error != ERROR_IO_PENDING) { | |
1215 | VLOG_ERR("nl_sock_wait failed - %s\n", ovs_format_message(error)); | |
1216 | retval = EINVAL; | |
1217 | goto done; | |
1218 | } | |
1219 | } else { | |
1220 | /* The I/O was completed synchronously */ | |
1221 | poll_immediate_wake(); | |
1222 | } | |
1223 | retval = 0; | |
1224 | ||
1225 | done: | |
1226 | ofpbuf_uninit(&request); | |
1227 | return retval; | |
1228 | } | |
1229 | #endif /* _WIN32 */ | |
1230 | ||
2fe27d5a | 1231 | /* Causes poll_block() to wake up when any of the specified 'events' (which is |
8341662d NR |
1232 | * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. |
1233 | * On Windows, 'sock' is not treated as const, and may be modified. */ | |
2fe27d5a BP |
1234 | void |
1235 | nl_sock_wait(const struct nl_sock *sock, short int events) | |
1236 | { | |
22326ba6 | 1237 | #ifdef _WIN32 |
7fa09611 | 1238 | if (sock->overlapped.Internal != STATUS_PENDING) { |
8341662d | 1239 | pend_io_request(CONST_CAST(struct nl_sock *, sock)); |
2c26eabf | 1240 | /* XXX: poll_wevent_wait(sock->overlapped.hEvent); */ |
7fa09611 | 1241 | } |
2c26eabf | 1242 | poll_immediate_wake(); /* XXX: temporary. */ |
22326ba6 | 1243 | #else |
2fe27d5a | 1244 | poll_fd_wait(sock->fd, events); |
22326ba6 | 1245 | #endif |
2fe27d5a | 1246 | } |
50802adb | 1247 | |
8522ba09 BP |
1248 | /* Returns the underlying fd for 'sock', for use in "poll()"-like operations |
1249 | * that can't use nl_sock_wait(). | |
1250 | * | |
1251 | * It's a little tricky to use the returned fd correctly, because nl_sock does | |
1252 | * "copy on write" to allow a single nl_sock to be used for notifications, | |
1253 | * transactions, and dumps. If 'sock' is used only for notifications and | |
1254 | * transactions (and never for dump) then the usage is safe. */ | |
1255 | int | |
1256 | nl_sock_fd(const struct nl_sock *sock) | |
1257 | { | |
22326ba6 | 1258 | #ifdef _WIN32 |
8341662d NR |
1259 | BUILD_ASSERT_DECL(sizeof sock->handle == sizeof(int)); |
1260 | return (int)sock->handle; | |
22326ba6 | 1261 | #else |
8522ba09 | 1262 | return sock->fd; |
22326ba6 | 1263 | #endif |
8522ba09 BP |
1264 | } |
1265 | ||
50802adb JG |
1266 | /* Returns the PID associated with this socket. */ |
1267 | uint32_t | |
1268 | nl_sock_pid(const struct nl_sock *sock) | |
1269 | { | |
1270 | return sock->pid; | |
1271 | } | |
2fe27d5a BP |
1272 | \f |
1273 | /* Miscellaneous. */ | |
1274 | ||
2ad204c8 BP |
1275 | struct genl_family { |
1276 | struct hmap_node hmap_node; | |
1277 | uint16_t id; | |
1278 | char *name; | |
1279 | }; | |
1280 | ||
1281 | static struct hmap genl_families = HMAP_INITIALIZER(&genl_families); | |
1282 | ||
2fe27d5a BP |
1283 | static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = { |
1284 | [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16}, | |
213a13ed | 1285 | [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true}, |
2fe27d5a BP |
1286 | }; |
1287 | ||
2ad204c8 BP |
1288 | static struct genl_family * |
1289 | find_genl_family_by_id(uint16_t id) | |
1290 | { | |
1291 | struct genl_family *family; | |
1292 | ||
1293 | HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0), | |
1294 | &genl_families) { | |
1295 | if (family->id == id) { | |
1296 | return family; | |
1297 | } | |
1298 | } | |
1299 | return NULL; | |
1300 | } | |
1301 | ||
1302 | static void | |
1303 | define_genl_family(uint16_t id, const char *name) | |
1304 | { | |
1305 | struct genl_family *family = find_genl_family_by_id(id); | |
1306 | ||
1307 | if (family) { | |
1308 | if (!strcmp(family->name, name)) { | |
1309 | return; | |
1310 | } | |
1311 | free(family->name); | |
1312 | } else { | |
1313 | family = xmalloc(sizeof *family); | |
1314 | family->id = id; | |
1315 | hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0)); | |
1316 | } | |
1317 | family->name = xstrdup(name); | |
1318 | } | |
1319 | ||
1320 | static const char * | |
1321 | genl_family_to_name(uint16_t id) | |
1322 | { | |
1323 | if (id == GENL_ID_CTRL) { | |
1324 | return "control"; | |
1325 | } else { | |
1326 | struct genl_family *family = find_genl_family_by_id(id); | |
1327 | return family ? family->name : "unknown"; | |
1328 | } | |
1329 | } | |
1330 | ||
b3fca8a8 | 1331 | #ifndef _WIN32 |
e408762f | 1332 | static int |
2a477244 BP |
1333 | do_lookup_genl_family(const char *name, struct nlattr **attrs, |
1334 | struct ofpbuf **replyp) | |
2fe27d5a BP |
1335 | { |
1336 | struct nl_sock *sock; | |
1337 | struct ofpbuf request, *reply; | |
2a477244 | 1338 | int error; |
2fe27d5a | 1339 | |
2a477244 BP |
1340 | *replyp = NULL; |
1341 | error = nl_sock_create(NETLINK_GENERIC, &sock); | |
1342 | if (error) { | |
1343 | return error; | |
2fe27d5a BP |
1344 | } |
1345 | ||
1346 | ofpbuf_init(&request, 0); | |
1347 | nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST, | |
1348 | CTRL_CMD_GETFAMILY, 1); | |
1349 | nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name); | |
2a477244 | 1350 | error = nl_sock_transact(sock, &request, &reply); |
2fe27d5a | 1351 | ofpbuf_uninit(&request); |
2a477244 | 1352 | if (error) { |
2fe27d5a | 1353 | nl_sock_destroy(sock); |
2a477244 | 1354 | return error; |
2fe27d5a BP |
1355 | } |
1356 | ||
1357 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN, | |
2a477244 BP |
1358 | family_policy, attrs, ARRAY_SIZE(family_policy)) |
1359 | || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) { | |
2fe27d5a BP |
1360 | nl_sock_destroy(sock); |
1361 | ofpbuf_delete(reply); | |
2a477244 | 1362 | return EPROTO; |
2fe27d5a BP |
1363 | } |
1364 | ||
2fe27d5a | 1365 | nl_sock_destroy(sock); |
2a477244 BP |
1366 | *replyp = reply; |
1367 | return 0; | |
2fe27d5a | 1368 | } |
b3fca8a8 NR |
1369 | #else |
1370 | static int | |
1371 | do_lookup_genl_family(const char *name, struct nlattr **attrs, | |
1372 | struct ofpbuf **replyp) | |
1373 | { | |
4c484aca | 1374 | struct nlmsghdr *nlmsg; |
b3fca8a8 NR |
1375 | struct ofpbuf *reply; |
1376 | int error; | |
1377 | uint16_t family_id; | |
1378 | const char *family_name; | |
1379 | uint32_t family_version; | |
1380 | uint32_t family_attrmax; | |
4c484aca NR |
1381 | uint32_t mcgrp_id = OVS_WIN_NL_INVALID_MCGRP_ID; |
1382 | const char *mcgrp_name = NULL; | |
b3fca8a8 NR |
1383 | |
1384 | *replyp = NULL; | |
1385 | reply = ofpbuf_new(1024); | |
1386 | ||
4c484aca | 1387 | /* CTRL_ATTR_MCAST_GROUPS is supported only for VPORT family. */ |
b3fca8a8 NR |
1388 | if (!strcmp(name, OVS_WIN_CONTROL_FAMILY)) { |
1389 | family_id = OVS_WIN_NL_CTRL_FAMILY_ID; | |
1390 | family_name = OVS_WIN_CONTROL_FAMILY; | |
1391 | family_version = OVS_WIN_CONTROL_VERSION; | |
1392 | family_attrmax = OVS_WIN_CONTROL_ATTR_MAX; | |
1393 | } else if (!strcmp(name, OVS_DATAPATH_FAMILY)) { | |
1394 | family_id = OVS_WIN_NL_DATAPATH_FAMILY_ID; | |
1395 | family_name = OVS_DATAPATH_FAMILY; | |
1396 | family_version = OVS_DATAPATH_VERSION; | |
1397 | family_attrmax = OVS_DP_ATTR_MAX; | |
1398 | } else if (!strcmp(name, OVS_PACKET_FAMILY)) { | |
1399 | family_id = OVS_WIN_NL_PACKET_FAMILY_ID; | |
1400 | family_name = OVS_PACKET_FAMILY; | |
1401 | family_version = OVS_PACKET_VERSION; | |
1402 | family_attrmax = OVS_PACKET_ATTR_MAX; | |
1403 | } else if (!strcmp(name, OVS_VPORT_FAMILY)) { | |
1404 | family_id = OVS_WIN_NL_VPORT_FAMILY_ID; | |
1405 | family_name = OVS_VPORT_FAMILY; | |
1406 | family_version = OVS_VPORT_VERSION; | |
1407 | family_attrmax = OVS_VPORT_ATTR_MAX; | |
4c484aca NR |
1408 | mcgrp_id = OVS_WIN_NL_VPORT_MCGRP_ID; |
1409 | mcgrp_name = OVS_VPORT_MCGROUP; | |
b3fca8a8 NR |
1410 | } else if (!strcmp(name, OVS_FLOW_FAMILY)) { |
1411 | family_id = OVS_WIN_NL_FLOW_FAMILY_ID; | |
1412 | family_name = OVS_FLOW_FAMILY; | |
1413 | family_version = OVS_FLOW_VERSION; | |
1414 | family_attrmax = OVS_FLOW_ATTR_MAX; | |
83cc9d56 NR |
1415 | } else if (!strcmp(name, OVS_WIN_NETDEV_FAMILY)) { |
1416 | family_id = OVS_WIN_NL_NETDEV_FAMILY_ID; | |
1417 | family_name = OVS_WIN_NETDEV_FAMILY; | |
1418 | family_version = OVS_WIN_NETDEV_VERSION; | |
1419 | family_attrmax = OVS_WIN_NETDEV_ATTR_MAX; | |
b3fca8a8 NR |
1420 | } else { |
1421 | ofpbuf_delete(reply); | |
1422 | return EINVAL; | |
1423 | } | |
1424 | ||
1425 | nl_msg_put_genlmsghdr(reply, 0, GENL_ID_CTRL, 0, | |
1426 | CTRL_CMD_NEWFAMILY, family_version); | |
1427 | /* CTRL_ATTR_HDRSIZE and CTRL_ATTR_OPS are not populated, but the | |
1428 | * callers do not seem to need them. */ | |
1429 | nl_msg_put_u16(reply, CTRL_ATTR_FAMILY_ID, family_id); | |
1430 | nl_msg_put_string(reply, CTRL_ATTR_FAMILY_NAME, family_name); | |
1431 | nl_msg_put_u32(reply, CTRL_ATTR_VERSION, family_version); | |
1432 | nl_msg_put_u32(reply, CTRL_ATTR_MAXATTR, family_attrmax); | |
1433 | ||
4c484aca NR |
1434 | if (mcgrp_id != OVS_WIN_NL_INVALID_MCGRP_ID) { |
1435 | size_t mcgrp_ofs1 = nl_msg_start_nested(reply, CTRL_ATTR_MCAST_GROUPS); | |
1436 | size_t mcgrp_ofs2= nl_msg_start_nested(reply, | |
1437 | OVS_WIN_NL_VPORT_MCGRP_ID - OVS_WIN_NL_MCGRP_START_ID); | |
1438 | nl_msg_put_u32(reply, CTRL_ATTR_MCAST_GRP_ID, mcgrp_id); | |
1439 | ovs_assert(mcgrp_name != NULL); | |
1440 | nl_msg_put_string(reply, CTRL_ATTR_MCAST_GRP_NAME, mcgrp_name); | |
1441 | nl_msg_end_nested(reply, mcgrp_ofs2); | |
1442 | nl_msg_end_nested(reply, mcgrp_ofs1); | |
1443 | } | |
1444 | ||
1445 | /* Set the total length of the netlink message. */ | |
1446 | nlmsg = nl_msg_nlmsghdr(reply); | |
1447 | nlmsg->nlmsg_len = ofpbuf_size(reply); | |
1448 | ||
b3fca8a8 NR |
1449 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN, |
1450 | family_policy, attrs, ARRAY_SIZE(family_policy)) | |
1451 | || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) { | |
b3fca8a8 NR |
1452 | ofpbuf_delete(reply); |
1453 | return EPROTO; | |
1454 | } | |
1455 | ||
1456 | *replyp = reply; | |
1457 | return 0; | |
1458 | } | |
1459 | #endif | |
2fe27d5a | 1460 | |
e408762f EJ |
1461 | /* Finds the multicast group called 'group_name' in genl family 'family_name'. |
1462 | * When successful, writes its result to 'multicast_group' and returns 0. | |
213a13ed | 1463 | * Otherwise, clears 'multicast_group' and returns a positive error code. |
b3dcb73c | 1464 | */ |
e408762f EJ |
1465 | int |
1466 | nl_lookup_genl_mcgroup(const char *family_name, const char *group_name, | |
b3dcb73c | 1467 | unsigned int *multicast_group) |
e408762f EJ |
1468 | { |
1469 | struct nlattr *family_attrs[ARRAY_SIZE(family_policy)]; | |
6d23c6f4 | 1470 | const struct nlattr *mc; |
2a477244 | 1471 | struct ofpbuf *reply; |
e408762f | 1472 | unsigned int left; |
2a477244 | 1473 | int error; |
e408762f EJ |
1474 | |
1475 | *multicast_group = 0; | |
2a477244 BP |
1476 | error = do_lookup_genl_family(family_name, family_attrs, &reply); |
1477 | if (error) { | |
1478 | return error; | |
e408762f EJ |
1479 | } |
1480 | ||
213a13ed | 1481 | if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) { |
b3dcb73c | 1482 | error = EPROTO; |
213a13ed EJ |
1483 | goto exit; |
1484 | } | |
1485 | ||
6d23c6f4 | 1486 | NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) { |
e408762f EJ |
1487 | static const struct nl_policy mc_policy[] = { |
1488 | [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32}, | |
1489 | [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING}, | |
1490 | }; | |
1491 | ||
1492 | struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)]; | |
1493 | const char *mc_name; | |
1494 | ||
1495 | if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) { | |
2a477244 BP |
1496 | error = EPROTO; |
1497 | goto exit; | |
e408762f EJ |
1498 | } |
1499 | ||
1500 | mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]); | |
1501 | if (!strcmp(group_name, mc_name)) { | |
1502 | *multicast_group = | |
1503 | nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]); | |
2a477244 BP |
1504 | error = 0; |
1505 | goto exit; | |
e408762f EJ |
1506 | } |
1507 | } | |
2a477244 | 1508 | error = EPROTO; |
e408762f | 1509 | |
2a477244 BP |
1510 | exit: |
1511 | ofpbuf_delete(reply); | |
1512 | return error; | |
e408762f EJ |
1513 | } |
1514 | ||
2fe27d5a BP |
1515 | /* If '*number' is 0, translates the given Generic Netlink family 'name' to a |
1516 | * number and stores it in '*number'. If successful, returns 0 and the caller | |
1517 | * may use '*number' as the family number. On failure, returns a positive | |
1518 | * errno value and '*number' caches the errno value. */ | |
1519 | int | |
1520 | nl_lookup_genl_family(const char *name, int *number) | |
1521 | { | |
1522 | if (*number == 0) { | |
2a477244 BP |
1523 | struct nlattr *attrs[ARRAY_SIZE(family_policy)]; |
1524 | struct ofpbuf *reply; | |
1525 | int error; | |
1526 | ||
1527 | error = do_lookup_genl_family(name, attrs, &reply); | |
1528 | if (!error) { | |
1529 | *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]); | |
1530 | define_genl_family(*number, name); | |
1531 | } else { | |
1532 | *number = -error; | |
1533 | } | |
1534 | ofpbuf_delete(reply); | |
1535 | ||
cb22974d | 1536 | ovs_assert(*number != 0); |
2fe27d5a BP |
1537 | } |
1538 | return *number > 0 ? 0 : -*number; | |
1539 | } | |
a88b4e04 BP |
1540 | \f |
1541 | struct nl_pool { | |
1542 | struct nl_sock *socks[16]; | |
1543 | int n; | |
1544 | }; | |
1545 | ||
834d6caf | 1546 | static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER; |
97be1538 | 1547 | static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex); |
a88b4e04 BP |
1548 | |
1549 | static int | |
1550 | nl_pool_alloc(int protocol, struct nl_sock **sockp) | |
1551 | { | |
0bd01224 | 1552 | struct nl_sock *sock = NULL; |
a88b4e04 BP |
1553 | struct nl_pool *pool; |
1554 | ||
1555 | ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools)); | |
1556 | ||
97be1538 | 1557 | ovs_mutex_lock(&pool_mutex); |
a88b4e04 BP |
1558 | pool = &pools[protocol]; |
1559 | if (pool->n > 0) { | |
0bd01224 BP |
1560 | sock = pool->socks[--pool->n]; |
1561 | } | |
97be1538 | 1562 | ovs_mutex_unlock(&pool_mutex); |
0bd01224 BP |
1563 | |
1564 | if (sock) { | |
1565 | *sockp = sock; | |
a88b4e04 BP |
1566 | return 0; |
1567 | } else { | |
1568 | return nl_sock_create(protocol, sockp); | |
1569 | } | |
1570 | } | |
1571 | ||
1572 | static void | |
1573 | nl_pool_release(struct nl_sock *sock) | |
1574 | { | |
1575 | if (sock) { | |
1576 | struct nl_pool *pool = &pools[sock->protocol]; | |
1577 | ||
97be1538 | 1578 | ovs_mutex_lock(&pool_mutex); |
a88b4e04 BP |
1579 | if (pool->n < ARRAY_SIZE(pool->socks)) { |
1580 | pool->socks[pool->n++] = sock; | |
0bd01224 | 1581 | sock = NULL; |
a88b4e04 | 1582 | } |
97be1538 | 1583 | ovs_mutex_unlock(&pool_mutex); |
0bd01224 BP |
1584 | |
1585 | nl_sock_destroy(sock); | |
a88b4e04 BP |
1586 | } |
1587 | } | |
1588 | ||
022ad2b9 BP |
1589 | /* Sends 'request' to the kernel on a Netlink socket for the given 'protocol' |
1590 | * (e.g. NETLINK_ROUTE or NETLINK_GENERIC) and waits for a response. If | |
1591 | * successful, returns 0. On failure, returns a positive errno value. | |
1592 | * | |
1593 | * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's | |
1594 | * reply, which the caller is responsible for freeing with ofpbuf_delete(), and | |
1595 | * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's | |
1596 | * reply, if any, is discarded. | |
1597 | * | |
1598 | * Before the message is sent, nlmsg_len in 'request' will be finalized to | |
1599 | * match ofpbuf_size(msg), nlmsg_pid will be set to the pid of the socket used | |
1600 | * for sending the request, and nlmsg_seq will be initialized. | |
1601 | * | |
1602 | * The caller is responsible for destroying 'request'. | |
1603 | * | |
1604 | * Bare Netlink is an unreliable transport protocol. This function layers | |
1605 | * reliable delivery and reply semantics on top of bare Netlink. | |
1606 | * | |
1607 | * In Netlink, sending a request to the kernel is reliable enough, because the | |
1608 | * kernel will tell us if the message cannot be queued (and we will in that | |
1609 | * case put it on the transmit queue and wait until it can be delivered). | |
1610 | * | |
1611 | * Receiving the reply is the real problem: if the socket buffer is full when | |
1612 | * the kernel tries to send the reply, the reply will be dropped. However, the | |
1613 | * kernel sets a flag that a reply has been dropped. The next call to recv | |
1614 | * then returns ENOBUFS. We can then re-send the request. | |
1615 | * | |
1616 | * Caveats: | |
1617 | * | |
1618 | * 1. Netlink depends on sequence numbers to match up requests and | |
1619 | * replies. The sender of a request supplies a sequence number, and | |
1620 | * the reply echos back that sequence number. | |
1621 | * | |
1622 | * This is fine, but (1) some kernel netlink implementations are | |
1623 | * broken, in that they fail to echo sequence numbers and (2) this | |
1624 | * function will drop packets with non-matching sequence numbers, so | |
1625 | * that only a single request can be usefully transacted at a time. | |
1626 | * | |
1627 | * 2. Resending the request causes it to be re-executed, so the request | |
1628 | * needs to be idempotent. | |
1629 | */ | |
a88b4e04 BP |
1630 | int |
1631 | nl_transact(int protocol, const struct ofpbuf *request, | |
1632 | struct ofpbuf **replyp) | |
1633 | { | |
1634 | struct nl_sock *sock; | |
1635 | int error; | |
1636 | ||
1637 | error = nl_pool_alloc(protocol, &sock); | |
1638 | if (error) { | |
1639 | *replyp = NULL; | |
1640 | return error; | |
1641 | } | |
1642 | ||
1643 | error = nl_sock_transact(sock, request, replyp); | |
1644 | ||
1645 | nl_pool_release(sock); | |
1646 | return error; | |
1647 | } | |
1648 | ||
022ad2b9 BP |
1649 | /* Sends the 'request' member of the 'n' transactions in 'transactions' on a |
1650 | * Netlink socket for the given 'protocol' (e.g. NETLINK_ROUTE or | |
1651 | * NETLINK_GENERIC), in order, and receives responses to all of them. Fills in | |
1652 | * the 'error' member of each transaction with 0 if it was successful, | |
1653 | * otherwise with a positive errno value. If 'reply' is nonnull, then it will | |
1654 | * be filled with the reply if the message receives a detailed reply. In other | |
1655 | * cases, i.e. where the request failed or had no reply beyond an indication of | |
1656 | * success, 'reply' will be cleared if it is nonnull. | |
1657 | * | |
1658 | * The caller is responsible for destroying each request and reply, and the | |
1659 | * transactions array itself. | |
1660 | * | |
1661 | * Before sending each message, this function will finalize nlmsg_len in each | |
1662 | * 'request' to match the ofpbuf's size, set nlmsg_pid to the pid of the socket | |
1663 | * used for the transaction, and initialize nlmsg_seq. | |
1664 | * | |
1665 | * Bare Netlink is an unreliable transport protocol. This function layers | |
1666 | * reliable delivery and reply semantics on top of bare Netlink. See | |
1667 | * nl_transact() for some caveats. | |
1668 | */ | |
a88b4e04 BP |
1669 | void |
1670 | nl_transact_multiple(int protocol, | |
1671 | struct nl_transaction **transactions, size_t n) | |
1672 | { | |
1673 | struct nl_sock *sock; | |
1674 | int error; | |
1675 | ||
1676 | error = nl_pool_alloc(protocol, &sock); | |
1677 | if (!error) { | |
1678 | nl_sock_transact_multiple(sock, transactions, n); | |
1679 | nl_pool_release(sock); | |
1680 | } else { | |
1681 | nl_sock_record_errors__(transactions, n, error); | |
1682 | } | |
1683 | } | |
1684 | ||
2fe27d5a | 1685 | \f |
7d7447df BP |
1686 | static uint32_t |
1687 | nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n) | |
1688 | { | |
1689 | uint32_t seq = sock->next_seq; | |
1690 | ||
1691 | sock->next_seq += n; | |
1692 | ||
1693 | /* Make it impossible for the next request for sequence numbers to wrap | |
1694 | * around to 0. Start over with 1 to avoid ever using a sequence number of | |
1695 | * 0, because the kernel uses sequence number 0 for notifications. */ | |
1696 | if (sock->next_seq >= UINT32_MAX / 2) { | |
1697 | sock->next_seq = 1; | |
1698 | } | |
1699 | ||
1700 | return seq; | |
1701 | } | |
1702 | ||
2fe27d5a | 1703 | static void |
2ad204c8 | 1704 | nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds) |
2fe27d5a BP |
1705 | { |
1706 | struct nlmsg_flag { | |
1707 | unsigned int bits; | |
1708 | const char *name; | |
1709 | }; | |
1710 | static const struct nlmsg_flag flags[] = { | |
1711 | { NLM_F_REQUEST, "REQUEST" }, | |
1712 | { NLM_F_MULTI, "MULTI" }, | |
1713 | { NLM_F_ACK, "ACK" }, | |
1714 | { NLM_F_ECHO, "ECHO" }, | |
1715 | { NLM_F_DUMP, "DUMP" }, | |
1716 | { NLM_F_ROOT, "ROOT" }, | |
1717 | { NLM_F_MATCH, "MATCH" }, | |
1718 | { NLM_F_ATOMIC, "ATOMIC" }, | |
1719 | }; | |
1720 | const struct nlmsg_flag *flag; | |
1721 | uint16_t flags_left; | |
1722 | ||
1723 | ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16, | |
1724 | h->nlmsg_len, h->nlmsg_type); | |
1725 | if (h->nlmsg_type == NLMSG_NOOP) { | |
1726 | ds_put_cstr(ds, "(no-op)"); | |
1727 | } else if (h->nlmsg_type == NLMSG_ERROR) { | |
1728 | ds_put_cstr(ds, "(error)"); | |
1729 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1730 | ds_put_cstr(ds, "(done)"); | |
1731 | } else if (h->nlmsg_type == NLMSG_OVERRUN) { | |
1732 | ds_put_cstr(ds, "(overrun)"); | |
1733 | } else if (h->nlmsg_type < NLMSG_MIN_TYPE) { | |
1734 | ds_put_cstr(ds, "(reserved)"); | |
2ad204c8 BP |
1735 | } else if (protocol == NETLINK_GENERIC) { |
1736 | ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type)); | |
2fe27d5a BP |
1737 | } else { |
1738 | ds_put_cstr(ds, "(family-defined)"); | |
1739 | } | |
1740 | ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags); | |
1741 | flags_left = h->nlmsg_flags; | |
1742 | for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) { | |
1743 | if ((flags_left & flag->bits) == flag->bits) { | |
1744 | ds_put_format(ds, "[%s]", flag->name); | |
1745 | flags_left &= ~flag->bits; | |
1746 | } | |
1747 | } | |
1748 | if (flags_left) { | |
1749 | ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left); | |
1750 | } | |
2c5a6834 BP |
1751 | ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32, |
1752 | h->nlmsg_seq, h->nlmsg_pid); | |
2fe27d5a BP |
1753 | } |
1754 | ||
1755 | static char * | |
7041c3a9 | 1756 | nlmsg_to_string(const struct ofpbuf *buffer, int protocol) |
2fe27d5a BP |
1757 | { |
1758 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1759 | const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN); | |
1760 | if (h) { | |
2ad204c8 | 1761 | nlmsghdr_to_string(h, protocol, &ds); |
2fe27d5a BP |
1762 | if (h->nlmsg_type == NLMSG_ERROR) { |
1763 | const struct nlmsgerr *e; | |
1764 | e = ofpbuf_at(buffer, NLMSG_HDRLEN, | |
1765 | NLMSG_ALIGN(sizeof(struct nlmsgerr))); | |
1766 | if (e) { | |
1767 | ds_put_format(&ds, " error(%d", e->error); | |
1768 | if (e->error < 0) { | |
10a89ef0 | 1769 | ds_put_format(&ds, "(%s)", ovs_strerror(-e->error)); |
2fe27d5a BP |
1770 | } |
1771 | ds_put_cstr(&ds, ", in-reply-to("); | |
2ad204c8 | 1772 | nlmsghdr_to_string(&e->msg, protocol, &ds); |
2fe27d5a BP |
1773 | ds_put_cstr(&ds, "))"); |
1774 | } else { | |
1775 | ds_put_cstr(&ds, " error(truncated)"); | |
1776 | } | |
1777 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1778 | int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error); | |
1779 | if (error) { | |
1780 | ds_put_format(&ds, " done(%d", *error); | |
1781 | if (*error < 0) { | |
10a89ef0 | 1782 | ds_put_format(&ds, "(%s)", ovs_strerror(-*error)); |
2fe27d5a BP |
1783 | } |
1784 | ds_put_cstr(&ds, ")"); | |
1785 | } else { | |
1786 | ds_put_cstr(&ds, " done(truncated)"); | |
1787 | } | |
7041c3a9 BP |
1788 | } else if (protocol == NETLINK_GENERIC) { |
1789 | struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer); | |
1790 | if (genl) { | |
1791 | ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")", | |
1792 | genl->cmd, genl->version); | |
1793 | } | |
2fe27d5a BP |
1794 | } |
1795 | } else { | |
1796 | ds_put_cstr(&ds, "nl(truncated)"); | |
1797 | } | |
1798 | return ds.string; | |
1799 | } | |
1800 | ||
1801 | static void | |
1802 | log_nlmsg(const char *function, int error, | |
7041c3a9 | 1803 | const void *message, size_t size, int protocol) |
2fe27d5a BP |
1804 | { |
1805 | struct ofpbuf buffer; | |
1806 | char *nlmsg; | |
1807 | ||
1808 | if (!VLOG_IS_DBG_ENABLED()) { | |
1809 | return; | |
1810 | } | |
1811 | ||
1812 | ofpbuf_use_const(&buffer, message, size); | |
7041c3a9 | 1813 | nlmsg = nlmsg_to_string(&buffer, protocol); |
10a89ef0 | 1814 | VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg); |
2fe27d5a BP |
1815 | free(nlmsg); |
1816 | } |