]>
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" | |
31 | #include "ofpbuf.h" | |
0bd01224 | 32 | #include "ovs-thread.h" |
2fe27d5a | 33 | #include "poll-loop.h" |
0672776e | 34 | #include "seq.h" |
6b7c12fd | 35 | #include "socket-util.h" |
cc75061a | 36 | #include "util.h" |
2fe27d5a BP |
37 | #include "vlog.h" |
38 | ||
39 | VLOG_DEFINE_THIS_MODULE(netlink_socket); | |
40 | ||
41 | COVERAGE_DEFINE(netlink_overflow); | |
42 | COVERAGE_DEFINE(netlink_received); | |
fc999dda | 43 | COVERAGE_DEFINE(netlink_recv_jumbo); |
2fe27d5a BP |
44 | COVERAGE_DEFINE(netlink_sent); |
45 | ||
46 | /* Linux header file confusion causes this to be undefined. */ | |
47 | #ifndef SOL_NETLINK | |
48 | #define SOL_NETLINK 270 | |
49 | #endif | |
50 | ||
22326ba6 AS |
51 | #ifdef _WIN32 |
52 | static struct ovs_mutex portid_mutex = OVS_MUTEX_INITIALIZER; | |
53 | static uint32_t g_last_portid = 0; | |
54 | ||
55 | /* Port IDs must be unique! */ | |
56 | static uint32_t | |
57 | portid_next(void) | |
58 | OVS_GUARDED_BY(portid_mutex) | |
59 | { | |
60 | g_last_portid++; | |
61 | return g_last_portid; | |
62 | } | |
63 | ||
64 | static void | |
65 | set_sock_pid_in_kernel(HANDLE handle, uint32_t pid) | |
66 | OVS_GUARDED_BY(portid_mutex) | |
67 | { | |
68 | struct nlmsghdr msg = { 0 }; | |
69 | ||
70 | msg.nlmsg_len = sizeof(struct nlmsghdr); | |
71 | msg.nlmsg_type = 80; /* target = set file pid */ | |
72 | msg.nlmsg_flags = 0; | |
73 | msg.nlmsg_seq = 0; | |
74 | msg.nlmsg_pid = pid; | |
75 | ||
76 | WriteFile(handle, &msg, sizeof(struct nlmsghdr), NULL, NULL); | |
77 | } | |
78 | #endif /* _WIN32 */ | |
79 | ||
2fe27d5a BP |
80 | /* A single (bad) Netlink message can in theory dump out many, many log |
81 | * messages, so the burst size is set quite high here to avoid missing useful | |
82 | * information. Also, at high logging levels we log *all* Netlink messages. */ | |
83 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600); | |
84 | ||
7d7447df | 85 | static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n); |
2fe27d5a | 86 | static void log_nlmsg(const char *function, int error, |
7041c3a9 | 87 | const void *message, size_t size, int protocol); |
2fe27d5a BP |
88 | \f |
89 | /* Netlink sockets. */ | |
90 | ||
0d121c73 | 91 | struct nl_sock { |
22326ba6 AS |
92 | #ifdef _WIN32 |
93 | HANDLE handle; | |
94 | #else | |
2fe27d5a | 95 | int fd; |
22326ba6 | 96 | #endif |
7d7447df | 97 | uint32_t next_seq; |
2fe27d5a | 98 | uint32_t pid; |
7041c3a9 | 99 | int protocol; |
cc75061a | 100 | unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */ |
2fe27d5a BP |
101 | }; |
102 | ||
cc75061a BP |
103 | /* Compile-time limit on iovecs, so that we can allocate a maximum-size array |
104 | * of iovecs on the stack. */ | |
105 | #define MAX_IOVS 128 | |
106 | ||
107 | /* Maximum number of iovecs that may be passed to sendmsg, capped at a | |
108 | * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS. | |
109 | * | |
110 | * Initialized by nl_sock_create(). */ | |
111 | static int max_iovs; | |
112 | ||
a88b4e04 BP |
113 | static int nl_pool_alloc(int protocol, struct nl_sock **sockp); |
114 | static void nl_pool_release(struct nl_sock *); | |
2fe27d5a BP |
115 | |
116 | /* Creates a new netlink socket for the given netlink 'protocol' | |
117 | * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the | |
a88b4e04 | 118 | * new socket if successful, otherwise returns a positive errno value. */ |
2fe27d5a | 119 | int |
cceb11f5 | 120 | nl_sock_create(int protocol, struct nl_sock **sockp) |
2fe27d5a | 121 | { |
0bd01224 | 122 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2fe27d5a | 123 | struct nl_sock *sock; |
22326ba6 | 124 | #ifndef _WIN32 |
2fe27d5a | 125 | struct sockaddr_nl local, remote; |
22326ba6 | 126 | #endif |
2c5a6834 | 127 | socklen_t local_size; |
d2b9f5b0 | 128 | int rcvbuf; |
2fe27d5a BP |
129 | int retval = 0; |
130 | ||
0bd01224 | 131 | if (ovsthread_once_start(&once)) { |
cc75061a BP |
132 | int save_errno = errno; |
133 | errno = 0; | |
134 | ||
135 | max_iovs = sysconf(_SC_UIO_MAXIOV); | |
136 | if (max_iovs < _XOPEN_IOV_MAX) { | |
137 | if (max_iovs == -1 && errno) { | |
10a89ef0 | 138 | VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno)); |
cc75061a BP |
139 | } |
140 | max_iovs = _XOPEN_IOV_MAX; | |
141 | } else if (max_iovs > MAX_IOVS) { | |
142 | max_iovs = MAX_IOVS; | |
143 | } | |
144 | ||
145 | errno = save_errno; | |
0bd01224 | 146 | ovsthread_once_done(&once); |
cc75061a BP |
147 | } |
148 | ||
2fe27d5a | 149 | *sockp = NULL; |
488232b7 | 150 | sock = xmalloc(sizeof *sock); |
2fe27d5a | 151 | |
22326ba6 AS |
152 | #ifdef _WIN32 |
153 | sock->handle = CreateFileA("\\\\.\\OpenVSwitchDevice", | |
154 | GENERIC_READ | GENERIC_WRITE, | |
155 | FILE_SHARE_READ | FILE_SHARE_WRITE, | |
156 | NULL, OPEN_EXISTING, | |
157 | FILE_ATTRIBUTE_NORMAL, NULL); | |
158 | ||
159 | int last_error = GetLastError(); | |
160 | ||
161 | if (sock->handle == INVALID_HANDLE_VALUE) { | |
162 | VLOG_ERR("fcntl: %s", ovs_strerror(last_error)); | |
163 | goto error; | |
164 | } | |
165 | #else | |
2fe27d5a BP |
166 | sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol); |
167 | if (sock->fd < 0) { | |
10a89ef0 | 168 | VLOG_ERR("fcntl: %s", ovs_strerror(errno)); |
2fe27d5a BP |
169 | goto error; |
170 | } | |
22326ba6 AS |
171 | #endif |
172 | ||
7041c3a9 | 173 | sock->protocol = protocol; |
7d7447df | 174 | sock->next_seq = 1; |
2fe27d5a | 175 | |
d2b9f5b0 | 176 | rcvbuf = 1024 * 1024; |
22326ba6 AS |
177 | #ifdef _WIN32 |
178 | sock->rcvbuf = rcvbuf; | |
179 | ovs_mutex_lock(&portid_mutex); | |
180 | sock->pid = portid_next(); | |
181 | set_sock_pid_in_kernel(sock->handle, sock->pid); | |
182 | ovs_mutex_unlock(&portid_mutex); | |
183 | #else | |
d2b9f5b0 BP |
184 | if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE, |
185 | &rcvbuf, sizeof rcvbuf)) { | |
80af5ee5 BP |
186 | /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM. |
187 | * Warn only if the failure is therefore unexpected. */ | |
f28b6dd3 | 188 | if (errno != EPERM) { |
80af5ee5 | 189 | VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed " |
10a89ef0 | 190 | "(%s)", rcvbuf, ovs_strerror(errno)); |
80af5ee5 | 191 | } |
d2b9f5b0 BP |
192 | } |
193 | ||
cc75061a BP |
194 | retval = get_socket_rcvbuf(sock->fd); |
195 | if (retval < 0) { | |
196 | retval = -retval; | |
197 | goto error; | |
198 | } | |
199 | sock->rcvbuf = retval; | |
200 | ||
2c5a6834 | 201 | /* Connect to kernel (pid 0) as remote address. */ |
2fe27d5a BP |
202 | memset(&remote, 0, sizeof remote); |
203 | remote.nl_family = AF_NETLINK; | |
204 | remote.nl_pid = 0; | |
205 | if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) { | |
10a89ef0 | 206 | VLOG_ERR("connect(0): %s", ovs_strerror(errno)); |
2c5a6834 BP |
207 | goto error; |
208 | } | |
209 | ||
210 | /* Obtain pid assigned by kernel. */ | |
211 | local_size = sizeof local; | |
212 | if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) { | |
10a89ef0 | 213 | VLOG_ERR("getsockname: %s", ovs_strerror(errno)); |
2c5a6834 BP |
214 | goto error; |
215 | } | |
216 | if (local_size < sizeof local || local.nl_family != AF_NETLINK) { | |
217 | VLOG_ERR("getsockname returned bad Netlink name"); | |
218 | retval = EINVAL; | |
219 | goto error; | |
2fe27d5a | 220 | } |
2c5a6834 | 221 | sock->pid = local.nl_pid; |
22326ba6 | 222 | #endif |
2fe27d5a | 223 | |
2fe27d5a BP |
224 | *sockp = sock; |
225 | return 0; | |
226 | ||
2fe27d5a BP |
227 | error: |
228 | if (retval == 0) { | |
229 | retval = errno; | |
230 | if (retval == 0) { | |
231 | retval = EINVAL; | |
232 | } | |
233 | } | |
22326ba6 AS |
234 | #ifdef _WIN32 |
235 | if (sock->handle != INVALID_HANDLE_VALUE) { | |
236 | CloseHandle(sock->handle); | |
237 | } | |
238 | #else | |
2fe27d5a BP |
239 | if (sock->fd >= 0) { |
240 | close(sock->fd); | |
241 | } | |
22326ba6 | 242 | #endif |
2fe27d5a BP |
243 | free(sock); |
244 | return retval; | |
245 | } | |
246 | ||
c6eab56d BP |
247 | /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and |
248 | * sets '*sockp' to the new socket if successful, otherwise returns a positive | |
249 | * errno value. */ | |
250 | int | |
251 | nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp) | |
252 | { | |
253 | return nl_sock_create(src->protocol, sockp); | |
254 | } | |
255 | ||
2fe27d5a BP |
256 | /* Destroys netlink socket 'sock'. */ |
257 | void | |
258 | nl_sock_destroy(struct nl_sock *sock) | |
259 | { | |
260 | if (sock) { | |
22326ba6 AS |
261 | #ifdef _WIN32 |
262 | CloseHandle(sock->handle); | |
263 | #else | |
a88b4e04 | 264 | close(sock->fd); |
22326ba6 | 265 | #endif |
a88b4e04 | 266 | free(sock); |
2fe27d5a BP |
267 | } |
268 | } | |
269 | ||
cceb11f5 BP |
270 | /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if |
271 | * successful, otherwise a positive errno value. | |
272 | * | |
a838c4fe BP |
273 | * A socket that is subscribed to a multicast group that receives asynchronous |
274 | * notifications must not be used for Netlink transactions or dumps, because | |
275 | * transactions and dumps can cause notifications to be lost. | |
276 | * | |
cceb11f5 BP |
277 | * Multicast group numbers are always positive. |
278 | * | |
279 | * It is not an error to attempt to join a multicast group to which a socket | |
280 | * already belongs. */ | |
281 | int | |
282 | nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
283 | { | |
22326ba6 AS |
284 | #ifdef _WIN32 |
285 | #define OVS_VPORT_MCGROUP_FALLBACK_ID 33 | |
286 | struct ofpbuf msg_buf; | |
287 | struct message_multicast | |
288 | { | |
289 | struct nlmsghdr; | |
290 | /* if true, join; if else, leave */ | |
291 | unsigned char join; | |
292 | unsigned int groupId; | |
293 | }; | |
294 | ||
295 | struct message_multicast msg = { 0 }; | |
296 | ||
297 | msg.nlmsg_len = sizeof(struct message_multicast); | |
298 | msg.nlmsg_type = OVS_VPORT_MCGROUP_FALLBACK_ID; | |
299 | msg.nlmsg_flags = 0; | |
300 | msg.nlmsg_seq = 0; | |
301 | msg.nlmsg_pid = sock->pid; | |
302 | ||
303 | msg.join = 1; | |
304 | msg.groupId = multicast_group; | |
305 | msg_buf.base_ = &msg; | |
306 | msg_buf.data_ = &msg; | |
307 | msg_buf.size_ = msg.nlmsg_len; | |
308 | ||
309 | nl_sock_send__(sock, &msg_buf, msg.nlmsg_seq, 0); | |
310 | #else | |
cceb11f5 BP |
311 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, |
312 | &multicast_group, sizeof multicast_group) < 0) { | |
313 | VLOG_WARN("could not join multicast group %u (%s)", | |
10a89ef0 | 314 | multicast_group, ovs_strerror(errno)); |
cceb11f5 BP |
315 | return errno; |
316 | } | |
22326ba6 | 317 | #endif |
cceb11f5 BP |
318 | return 0; |
319 | } | |
320 | ||
321 | /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if | |
322 | * successful, otherwise a positive errno value. | |
323 | * | |
324 | * Multicast group numbers are always positive. | |
325 | * | |
326 | * It is not an error to attempt to leave a multicast group to which a socket | |
327 | * does not belong. | |
328 | * | |
329 | * On success, reading from 'sock' will still return any messages that were | |
330 | * received on 'multicast_group' before the group was left. */ | |
331 | int | |
332 | nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
333 | { | |
22326ba6 AS |
334 | #ifdef _WIN32 |
335 | struct ofpbuf msg_buf; | |
336 | struct message_multicast | |
337 | { | |
338 | struct nlmsghdr; | |
339 | /* if true, join; if else, leave*/ | |
340 | unsigned char join; | |
341 | }; | |
342 | ||
343 | struct message_multicast msg = { 0 }; | |
344 | nl_msg_put_nlmsghdr(&msg, sizeof(struct message_multicast), | |
345 | multicast_group, 0); | |
346 | msg.join = 0; | |
347 | ||
348 | msg_buf.base_ = &msg; | |
349 | msg_buf.data_ = &msg; | |
350 | msg_buf.size_ = msg.nlmsg_len; | |
351 | ||
352 | nl_sock_send__(sock, &msg_buf, msg.nlmsg_seq, 0); | |
353 | #else | |
cceb11f5 BP |
354 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP, |
355 | &multicast_group, sizeof multicast_group) < 0) { | |
356 | VLOG_WARN("could not leave multicast group %u (%s)", | |
10a89ef0 | 357 | multicast_group, ovs_strerror(errno)); |
cceb11f5 BP |
358 | return errno; |
359 | } | |
22326ba6 | 360 | #endif |
cceb11f5 BP |
361 | return 0; |
362 | } | |
363 | ||
c6eab56d | 364 | static int |
ff459dd6 BP |
365 | nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, |
366 | uint32_t nlmsg_seq, bool wait) | |
2fe27d5a BP |
367 | { |
368 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg); | |
369 | int error; | |
370 | ||
1f317cb5 | 371 | nlmsg->nlmsg_len = ofpbuf_size(msg); |
ff459dd6 | 372 | nlmsg->nlmsg_seq = nlmsg_seq; |
2fe27d5a BP |
373 | nlmsg->nlmsg_pid = sock->pid; |
374 | do { | |
375 | int retval; | |
22326ba6 AS |
376 | #ifdef _WIN32 |
377 | bool result; | |
378 | DWORD last_error = 0; | |
379 | result = WriteFile(sock->handle, ofpbuf_data(msg), ofpbuf_size(msg), | |
380 | &retval, NULL); | |
381 | last_error = GetLastError(); | |
382 | if (last_error != ERROR_SUCCESS && !result) { | |
383 | retval = -1; | |
384 | errno = EAGAIN; | |
385 | } | |
386 | #else | |
1f317cb5 | 387 | retval = send(sock->fd, ofpbuf_data(msg), ofpbuf_size(msg), wait ? 0 : MSG_DONTWAIT); |
22326ba6 | 388 | #endif |
2fe27d5a BP |
389 | error = retval < 0 ? errno : 0; |
390 | } while (error == EINTR); | |
1f317cb5 | 391 | log_nlmsg(__func__, error, ofpbuf_data(msg), ofpbuf_size(msg), sock->protocol); |
2fe27d5a BP |
392 | if (!error) { |
393 | COVERAGE_INC(netlink_sent); | |
394 | } | |
395 | return error; | |
396 | } | |
397 | ||
c6eab56d | 398 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on |
1f317cb5 | 399 | * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid |
ff459dd6 BP |
400 | * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh |
401 | * sequence number, before the message is sent. | |
c6eab56d BP |
402 | * |
403 | * Returns 0 if successful, otherwise a positive errno value. If | |
404 | * 'wait' is true, then the send will wait until buffer space is ready; | |
405 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. */ | |
406 | int | |
407 | nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait) | |
ff459dd6 BP |
408 | { |
409 | return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait); | |
410 | } | |
411 | ||
412 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on | |
1f317cb5 | 413 | * 'sock'. nlmsg_len in 'msg' will be finalized to match ofpbuf_size(msg), nlmsg_pid |
ff459dd6 BP |
414 | * will be set to 'sock''s pid, and nlmsg_seq will be initialized to |
415 | * 'nlmsg_seq', before the message is sent. | |
416 | * | |
417 | * Returns 0 if successful, otherwise a positive errno value. If | |
418 | * 'wait' is true, then the send will wait until buffer space is ready; | |
419 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. | |
420 | * | |
421 | * This function is suitable for sending a reply to a request that was received | |
422 | * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */ | |
423 | int | |
424 | nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg, | |
425 | uint32_t nlmsg_seq, bool wait) | |
c6eab56d | 426 | { |
ff459dd6 | 427 | return nl_sock_send__(sock, msg, nlmsg_seq, wait); |
c6eab56d BP |
428 | } |
429 | ||
c6eab56d | 430 | static int |
72d32ac0 | 431 | nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait) |
2fe27d5a | 432 | { |
72d32ac0 BP |
433 | /* We can't accurately predict the size of the data to be received. The |
434 | * caller is supposed to have allocated enough space in 'buf' to handle the | |
435 | * "typical" case. To handle exceptions, we make available enough space in | |
436 | * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable | |
437 | * figure since that's the maximum length of a Netlink attribute). */ | |
2fe27d5a | 438 | struct nlmsghdr *nlmsghdr; |
22326ba6 AS |
439 | #ifdef _WIN32 |
440 | #define MAX_STACK_LENGTH 81920 | |
441 | uint8_t tail[MAX_STACK_LENGTH]; | |
442 | #else | |
72d32ac0 | 443 | uint8_t tail[65536]; |
22326ba6 | 444 | #endif |
fc999dda | 445 | struct iovec iov[2]; |
fc999dda BP |
446 | struct msghdr msg; |
447 | ssize_t retval; | |
8f20fd98 | 448 | int error; |
fc999dda | 449 | |
cb22974d | 450 | ovs_assert(buf->allocated >= sizeof *nlmsghdr); |
72d32ac0 | 451 | ofpbuf_clear(buf); |
2fe27d5a | 452 | |
1f317cb5 | 453 | iov[0].iov_base = ofpbuf_base(buf); |
72d32ac0 | 454 | iov[0].iov_len = buf->allocated; |
fc999dda | 455 | iov[1].iov_base = tail; |
72d32ac0 | 456 | iov[1].iov_len = sizeof tail; |
fc999dda BP |
457 | |
458 | memset(&msg, 0, sizeof msg); | |
459 | msg.msg_iov = iov; | |
460 | msg.msg_iovlen = 2; | |
461 | ||
8f20fd98 BP |
462 | /* Receive a Netlink message from the kernel. |
463 | * | |
464 | * This works around a kernel bug in which the kernel returns an error code | |
465 | * as if it were the number of bytes read. It doesn't actually modify | |
466 | * anything in the receive buffer in that case, so we can initialize the | |
467 | * Netlink header with an impossible message length and then, upon success, | |
468 | * check whether it changed. */ | |
469 | nlmsghdr = ofpbuf_base(buf); | |
2fe27d5a | 470 | do { |
8f20fd98 | 471 | nlmsghdr->nlmsg_len = UINT32_MAX; |
22326ba6 AS |
472 | #ifdef _WIN32 |
473 | boolean result = false; | |
474 | DWORD last_error = 0; | |
475 | result = ReadFile(sock->handle, tail, MAX_STACK_LENGTH, &retval, NULL); | |
476 | last_error = GetLastError(); | |
477 | if (last_error != ERROR_SUCCESS && !result) { | |
478 | retval = -1; | |
479 | errno = EAGAIN; | |
480 | } else { | |
481 | ofpbuf_put(buf, tail, retval); | |
482 | } | |
483 | #else | |
fc999dda | 484 | retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT); |
22326ba6 | 485 | #endif |
8f20fd98 BP |
486 | error = (retval < 0 ? errno |
487 | : retval == 0 ? ECONNRESET /* not possible? */ | |
488 | : nlmsghdr->nlmsg_len != UINT32_MAX ? 0 | |
7f8e2646 | 489 | : retval); |
8f20fd98 BP |
490 | } while (error == EINTR); |
491 | if (error) { | |
fc999dda BP |
492 | if (error == ENOBUFS) { |
493 | /* Socket receive buffer overflow dropped one or more messages that | |
494 | * the kernel tried to send to us. */ | |
495 | COVERAGE_INC(netlink_overflow); | |
496 | } | |
fc999dda | 497 | return error; |
2fe27d5a | 498 | } |
fc999dda | 499 | |
2fe27d5a | 500 | if (msg.msg_flags & MSG_TRUNC) { |
34582733 | 501 | VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)", |
72d32ac0 | 502 | sizeof tail); |
fc999dda | 503 | return E2BIG; |
2fe27d5a | 504 | } |
2fe27d5a | 505 | |
fc999dda | 506 | if (retval < sizeof *nlmsghdr |
2fe27d5a | 507 | || nlmsghdr->nlmsg_len < sizeof *nlmsghdr |
fc999dda | 508 | || nlmsghdr->nlmsg_len > retval) { |
e5e4b47c | 509 | VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE" bytes < %"PRIuSIZE")", |
72d32ac0 | 510 | retval, sizeof *nlmsghdr); |
2fe27d5a BP |
511 | return EPROTO; |
512 | } | |
22326ba6 | 513 | #ifndef _WIN32 |
1f317cb5 | 514 | ofpbuf_set_size(buf, MIN(retval, buf->allocated)); |
72d32ac0 BP |
515 | if (retval > buf->allocated) { |
516 | COVERAGE_INC(netlink_recv_jumbo); | |
517 | ofpbuf_put(buf, tail, retval - buf->allocated); | |
518 | } | |
22326ba6 | 519 | #endif |
72d32ac0 | 520 | |
1f317cb5 | 521 | log_nlmsg(__func__, 0, ofpbuf_data(buf), ofpbuf_size(buf), sock->protocol); |
2fe27d5a BP |
522 | COVERAGE_INC(netlink_received); |
523 | ||
524 | return 0; | |
525 | } | |
526 | ||
72d32ac0 BP |
527 | /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If |
528 | * 'wait' is true, waits for a message to be ready. Otherwise, fails with | |
529 | * EAGAIN if the 'sock' receive buffer is empty. | |
530 | * | |
531 | * The caller must have initialized 'buf' with an allocation of at least | |
532 | * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough | |
533 | * space for a "typical" message. | |
534 | * | |
535 | * On success, returns 0 and replaces 'buf''s previous content by the received | |
536 | * message. This function expands 'buf''s allocated memory, as necessary, to | |
537 | * hold the actual size of the received message. | |
c6eab56d | 538 | * |
72d32ac0 BP |
539 | * On failure, returns a positive errno value and clears 'buf' to zero length. |
540 | * 'buf' retains its previous memory allocation. | |
541 | * | |
542 | * Regardless of success or failure, this function resets 'buf''s headroom to | |
543 | * 0. */ | |
c6eab56d | 544 | int |
72d32ac0 | 545 | nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait) |
c6eab56d | 546 | { |
72d32ac0 | 547 | return nl_sock_recv__(sock, buf, wait); |
cc75061a BP |
548 | } |
549 | ||
550 | static void | |
551 | nl_sock_record_errors__(struct nl_transaction **transactions, size_t n, | |
552 | int error) | |
553 | { | |
554 | size_t i; | |
555 | ||
556 | for (i = 0; i < n; i++) { | |
72d32ac0 BP |
557 | struct nl_transaction *txn = transactions[i]; |
558 | ||
559 | txn->error = error; | |
560 | if (txn->reply) { | |
561 | ofpbuf_clear(txn->reply); | |
562 | } | |
cc75061a BP |
563 | } |
564 | } | |
565 | ||
566 | static int | |
567 | nl_sock_transact_multiple__(struct nl_sock *sock, | |
568 | struct nl_transaction **transactions, size_t n, | |
569 | size_t *done) | |
570 | { | |
72d32ac0 BP |
571 | uint64_t tmp_reply_stub[1024 / 8]; |
572 | struct nl_transaction tmp_txn; | |
573 | struct ofpbuf tmp_reply; | |
574 | ||
575 | uint32_t base_seq; | |
cc75061a BP |
576 | struct iovec iovs[MAX_IOVS]; |
577 | struct msghdr msg; | |
578 | int error; | |
579 | int i; | |
580 | ||
72d32ac0 | 581 | base_seq = nl_sock_allocate_seq(sock, n); |
cc75061a BP |
582 | *done = 0; |
583 | for (i = 0; i < n; i++) { | |
72d32ac0 BP |
584 | struct nl_transaction *txn = transactions[i]; |
585 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request); | |
cc75061a | 586 | |
1f317cb5 | 587 | nlmsg->nlmsg_len = ofpbuf_size(txn->request); |
72d32ac0 | 588 | nlmsg->nlmsg_seq = base_seq + i; |
cc75061a | 589 | nlmsg->nlmsg_pid = sock->pid; |
cc75061a | 590 | |
1f317cb5 PS |
591 | iovs[i].iov_base = ofpbuf_data(txn->request); |
592 | iovs[i].iov_len = ofpbuf_size(txn->request); | |
cc75061a BP |
593 | } |
594 | ||
595 | memset(&msg, 0, sizeof msg); | |
596 | msg.msg_iov = iovs; | |
597 | msg.msg_iovlen = n; | |
598 | do { | |
22326ba6 AS |
599 | #ifdef _WIN32 |
600 | DWORD last_error = 0; | |
601 | bool result = FALSE; | |
602 | for (i = 0; i < n; i++) { | |
603 | result = WriteFile((HANDLE)sock->handle, iovs[i].iov_base, iovs[i].iov_len, | |
604 | &error, NULL); | |
605 | last_error = GetLastError(); | |
606 | if (last_error != ERROR_SUCCESS && !result) { | |
607 | error = EAGAIN; | |
608 | errno = EAGAIN; | |
609 | } else { | |
610 | error = 0; | |
611 | } | |
612 | } | |
613 | #else | |
cc75061a | 614 | error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0; |
22326ba6 | 615 | #endif |
cc75061a BP |
616 | } while (error == EINTR); |
617 | ||
618 | for (i = 0; i < n; i++) { | |
72d32ac0 | 619 | struct nl_transaction *txn = transactions[i]; |
cc75061a | 620 | |
1f317cb5 | 621 | log_nlmsg(__func__, error, ofpbuf_data(txn->request), ofpbuf_size(txn->request), |
cc75061a BP |
622 | sock->protocol); |
623 | } | |
624 | if (!error) { | |
625 | COVERAGE_ADD(netlink_sent, n); | |
626 | } | |
627 | ||
628 | if (error) { | |
629 | return error; | |
630 | } | |
631 | ||
72d32ac0 BP |
632 | ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub); |
633 | tmp_txn.request = NULL; | |
634 | tmp_txn.reply = &tmp_reply; | |
635 | tmp_txn.error = 0; | |
cc75061a | 636 | while (n > 0) { |
72d32ac0 BP |
637 | struct nl_transaction *buf_txn, *txn; |
638 | uint32_t seq; | |
639 | ||
640 | /* Find a transaction whose buffer we can use for receiving a reply. | |
641 | * If no such transaction is left, use tmp_txn. */ | |
642 | buf_txn = &tmp_txn; | |
643 | for (i = 0; i < n; i++) { | |
644 | if (transactions[i]->reply) { | |
645 | buf_txn = transactions[i]; | |
646 | break; | |
647 | } | |
648 | } | |
cc75061a | 649 | |
72d32ac0 BP |
650 | /* Receive a reply. */ |
651 | error = nl_sock_recv__(sock, buf_txn->reply, false); | |
652 | if (error) { | |
653 | if (error == EAGAIN) { | |
654 | nl_sock_record_errors__(transactions, n, 0); | |
655 | *done += n; | |
656 | error = 0; | |
657 | } | |
658 | break; | |
cc75061a BP |
659 | } |
660 | ||
72d32ac0 BP |
661 | /* Match the reply up with a transaction. */ |
662 | seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq; | |
663 | if (seq < base_seq || seq >= base_seq + n) { | |
664 | VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq); | |
cc75061a BP |
665 | continue; |
666 | } | |
72d32ac0 BP |
667 | i = seq - base_seq; |
668 | txn = transactions[i]; | |
cc75061a | 669 | |
72d32ac0 BP |
670 | /* Fill in the results for 'txn'. */ |
671 | if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) { | |
672 | if (txn->reply) { | |
673 | ofpbuf_clear(txn->reply); | |
674 | } | |
675 | if (txn->error) { | |
cc75061a | 676 | VLOG_DBG_RL(&rl, "received NAK error=%d (%s)", |
10a89ef0 | 677 | error, ovs_strerror(txn->error)); |
cc75061a | 678 | } |
cc75061a | 679 | } else { |
72d32ac0 BP |
680 | txn->error = 0; |
681 | if (txn->reply && txn != buf_txn) { | |
682 | /* Swap buffers. */ | |
683 | struct ofpbuf *reply = buf_txn->reply; | |
684 | buf_txn->reply = txn->reply; | |
685 | txn->reply = reply; | |
686 | } | |
cc75061a BP |
687 | } |
688 | ||
72d32ac0 BP |
689 | /* Fill in the results for transactions before 'txn'. (We have to do |
690 | * this after the results for 'txn' itself because of the buffer swap | |
691 | * above.) */ | |
692 | nl_sock_record_errors__(transactions, i, 0); | |
693 | ||
694 | /* Advance. */ | |
cc75061a BP |
695 | *done += i + 1; |
696 | transactions += i + 1; | |
697 | n -= i + 1; | |
72d32ac0 | 698 | base_seq += i + 1; |
cc75061a | 699 | } |
72d32ac0 | 700 | ofpbuf_uninit(&tmp_reply); |
cc75061a | 701 | |
72d32ac0 | 702 | return error; |
cc75061a BP |
703 | } |
704 | ||
022ad2b9 | 705 | static void |
cc75061a BP |
706 | nl_sock_transact_multiple(struct nl_sock *sock, |
707 | struct nl_transaction **transactions, size_t n) | |
708 | { | |
709 | int max_batch_count; | |
710 | int error; | |
711 | ||
712 | if (!n) { | |
713 | return; | |
714 | } | |
715 | ||
cc75061a BP |
716 | /* In theory, every request could have a 64 kB reply. But the default and |
717 | * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to | |
718 | * be a bit below 128 kB, so that would only allow a single message in a | |
719 | * "batch". So we assume that replies average (at most) 4 kB, which allows | |
720 | * a good deal of batching. | |
721 | * | |
722 | * In practice, most of the requests that we batch either have no reply at | |
723 | * all or a brief reply. */ | |
724 | max_batch_count = MAX(sock->rcvbuf / 4096, 1); | |
725 | max_batch_count = MIN(max_batch_count, max_iovs); | |
726 | ||
727 | while (n > 0) { | |
728 | size_t count, bytes; | |
729 | size_t done; | |
730 | ||
731 | /* Batch up to 'max_batch_count' transactions. But cap it at about a | |
732 | * page of requests total because big skbuffs are expensive to | |
733 | * allocate in the kernel. */ | |
734 | #if defined(PAGESIZE) | |
735 | enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) }; | |
736 | #else | |
737 | enum { MAX_BATCH_BYTES = 4096 - 512 }; | |
738 | #endif | |
1f317cb5 | 739 | bytes = ofpbuf_size(transactions[0]->request); |
cc75061a | 740 | for (count = 1; count < n && count < max_batch_count; count++) { |
1f317cb5 | 741 | if (bytes + ofpbuf_size(transactions[count]->request) > MAX_BATCH_BYTES) { |
cc75061a BP |
742 | break; |
743 | } | |
1f317cb5 | 744 | bytes += ofpbuf_size(transactions[count]->request); |
cc75061a BP |
745 | } |
746 | ||
747 | error = nl_sock_transact_multiple__(sock, transactions, count, &done); | |
748 | transactions += done; | |
749 | n -= done; | |
750 | ||
751 | if (error == ENOBUFS) { | |
752 | VLOG_DBG_RL(&rl, "receive buffer overflow, resending request"); | |
753 | } else if (error) { | |
10a89ef0 | 754 | VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error)); |
cc75061a BP |
755 | nl_sock_record_errors__(transactions, n, error); |
756 | } | |
757 | } | |
758 | } | |
759 | ||
022ad2b9 | 760 | static int |
cc75061a BP |
761 | nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request, |
762 | struct ofpbuf **replyp) | |
2fe27d5a | 763 | { |
cc75061a BP |
764 | struct nl_transaction *transactionp; |
765 | struct nl_transaction transaction; | |
2fe27d5a | 766 | |
ebc56baa | 767 | transaction.request = CONST_CAST(struct ofpbuf *, request); |
72d32ac0 | 768 | transaction.reply = replyp ? ofpbuf_new(1024) : NULL; |
cc75061a | 769 | transactionp = &transaction; |
72d32ac0 | 770 | |
cc75061a | 771 | nl_sock_transact_multiple(sock, &transactionp, 1); |
72d32ac0 | 772 | |
2fe27d5a | 773 | if (replyp) { |
72d32ac0 BP |
774 | if (transaction.error) { |
775 | ofpbuf_delete(transaction.reply); | |
776 | *replyp = NULL; | |
777 | } else { | |
778 | *replyp = transaction.reply; | |
779 | } | |
2fe27d5a | 780 | } |
72d32ac0 | 781 | |
cc75061a | 782 | return transaction.error; |
2fe27d5a BP |
783 | } |
784 | ||
6b7c12fd BP |
785 | /* Drain all the messages currently in 'sock''s receive queue. */ |
786 | int | |
787 | nl_sock_drain(struct nl_sock *sock) | |
788 | { | |
22326ba6 AS |
789 | #ifdef _WIN32 |
790 | return 0; | |
791 | #else | |
6b7c12fd | 792 | return drain_rcvbuf(sock->fd); |
22326ba6 | 793 | #endif |
6b7c12fd BP |
794 | } |
795 | ||
a88b4e04 BP |
796 | /* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a |
797 | * Netlink socket created with the given 'protocol', and initializes 'dump' to | |
798 | * reflect the state of the operation. | |
2fe27d5a | 799 | * |
db1fc210 JS |
800 | * 'request' must contain a Netlink message. Before sending the message, |
801 | * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be | |
802 | * set to the Netlink socket's pid. NLM_F_DUMP and NLM_F_ACK will be set in | |
803 | * nlmsg_flags. | |
2fe27d5a | 804 | * |
a88b4e04 | 805 | * The design of this Netlink socket library ensures that the dump is reliable. |
2fe27d5a | 806 | * |
db1fc210 JS |
807 | * This function provides no status indication. nl_dump_done() provides an |
808 | * error status for the entire dump operation. | |
2fe27d5a | 809 | * |
db1fc210 | 810 | * The caller must eventually destroy 'request'. |
2fe27d5a BP |
811 | */ |
812 | void | |
a88b4e04 | 813 | nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request) |
2fe27d5a | 814 | { |
7d7447df | 815 | nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK; |
93295354 BP |
816 | |
817 | ovs_mutex_init(&dump->mutex); | |
818 | ovs_mutex_lock(&dump->mutex); | |
819 | dump->status = nl_pool_alloc(protocol, &dump->sock); | |
820 | if (!dump->status) { | |
821 | dump->status = nl_sock_send__(dump->sock, request, | |
822 | nl_sock_allocate_seq(dump->sock, 1), | |
823 | true); | |
b2d1c78a | 824 | } |
9c8ad495 | 825 | dump->nl_seq = nl_msg_nlmsghdr(request)->nlmsg_seq; |
93295354 BP |
826 | ovs_mutex_unlock(&dump->mutex); |
827 | } | |
828 | ||
829 | static int | |
830 | nl_dump_refill(struct nl_dump *dump, struct ofpbuf *buffer) | |
831 | OVS_REQUIRES(dump->mutex) | |
832 | { | |
833 | struct nlmsghdr *nlmsghdr; | |
834 | int error; | |
835 | ||
836 | while (!ofpbuf_size(buffer)) { | |
1738803a | 837 | error = nl_sock_recv__(dump->sock, buffer, false); |
93295354 | 838 | if (error) { |
1738803a AW |
839 | /* The kernel never blocks providing the results of a dump, so |
840 | * error == EAGAIN means that we've read the whole thing, and | |
841 | * therefore transform it into EOF. (The kernel always provides | |
842 | * NLMSG_DONE as a sentinel. Some other thread must have received | |
843 | * that already but not yet signaled it in 'status'.) | |
844 | * | |
845 | * Any other error is just an error. */ | |
93295354 BP |
846 | return error == EAGAIN ? EOF : error; |
847 | } | |
848 | ||
849 | nlmsghdr = nl_msg_nlmsghdr(buffer); | |
850 | if (dump->nl_seq != nlmsghdr->nlmsg_seq) { | |
851 | VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32, | |
852 | nlmsghdr->nlmsg_seq, dump->nl_seq); | |
853 | ofpbuf_clear(buffer); | |
854 | } | |
855 | } | |
856 | ||
857 | if (nl_msg_nlmsgerr(buffer, &error) && error) { | |
858 | VLOG_INFO_RL(&rl, "netlink dump request error (%s)", | |
859 | ovs_strerror(error)); | |
860 | ofpbuf_clear(buffer); | |
861 | return error; | |
862 | } | |
863 | ||
864 | return 0; | |
865 | } | |
866 | ||
867 | static int | |
868 | nl_dump_next__(struct ofpbuf *reply, struct ofpbuf *buffer) | |
869 | { | |
870 | struct nlmsghdr *nlmsghdr = nl_msg_next(buffer, reply); | |
871 | if (!nlmsghdr) { | |
872 | VLOG_WARN_RL(&rl, "netlink dump contains message fragment"); | |
873 | return EPROTO; | |
874 | } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) { | |
875 | return EOF; | |
876 | } else { | |
877 | return 0; | |
878 | } | |
2fe27d5a BP |
879 | } |
880 | ||
d57695d7 JS |
881 | /* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must |
882 | * have been initialized with nl_dump_start(), and 'buffer' must have been | |
883 | * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long. | |
2fe27d5a | 884 | * |
19aa20a0 BP |
885 | * If successful, returns true and points 'reply->data' and |
886 | * 'ofpbuf_size(reply)' to the message that was retrieved. The caller must not | |
887 | * modify 'reply' (because it points within 'buffer', which will be used by | |
888 | * future calls to this function). | |
889 | * | |
890 | * On failure, returns false and sets 'reply->data' to NULL and | |
891 | * 'ofpbuf_size(reply)' to 0. Failure might indicate an actual error or merely | |
892 | * the end of replies. An error status for the entire dump operation is | |
893 | * provided when it is completed by calling nl_dump_done(). | |
0672776e JS |
894 | * |
895 | * Multiple threads may call this function, passing the same nl_dump, however | |
896 | * each must provide independent buffers. This function may cache multiple | |
897 | * replies in the buffer, and these will be processed before more replies are | |
898 | * fetched. When this function returns false, other threads may continue to | |
899 | * process replies in their buffers, but they will not fetch more replies. | |
2fe27d5a BP |
900 | */ |
901 | bool | |
d57695d7 | 902 | nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer) |
2fe27d5a | 903 | { |
93295354 | 904 | int retval = 0; |
0672776e | 905 | |
93295354 BP |
906 | /* If the buffer is empty, refill it. |
907 | * | |
908 | * If the buffer is not empty, we don't check the dump's status. | |
909 | * Otherwise, we could end up skipping some of the dump results if thread A | |
910 | * hits EOF while thread B is in the midst of processing a batch. */ | |
911 | if (!ofpbuf_size(buffer)) { | |
0791315e | 912 | ovs_mutex_lock(&dump->mutex); |
93295354 BP |
913 | if (!dump->status) { |
914 | /* Take the mutex here to avoid an in-kernel race. If two threads | |
915 | * try to read from a Netlink dump socket at once, then the socket | |
916 | * error can be set to EINVAL, which will be encountered on the | |
917 | * next recv on that socket, which could be anywhere due to the way | |
918 | * that we pool Netlink sockets. Serializing the recv calls avoids | |
919 | * the issue. */ | |
920 | dump->status = nl_dump_refill(dump, buffer); | |
921 | } | |
922 | retval = dump->status; | |
0791315e | 923 | ovs_mutex_unlock(&dump->mutex); |
93295354 | 924 | } |
0791315e | 925 | |
93295354 BP |
926 | /* Fetch the next message from the buffer. */ |
927 | if (!retval) { | |
928 | retval = nl_dump_next__(reply, buffer); | |
2fe27d5a | 929 | if (retval) { |
93295354 BP |
930 | /* Record 'retval' as the dump status, but don't overwrite an error |
931 | * with EOF. */ | |
932 | ovs_mutex_lock(&dump->mutex); | |
933 | if (dump->status <= 0) { | |
934 | dump->status = retval; | |
2fe27d5a | 935 | } |
93295354 | 936 | ovs_mutex_unlock(&dump->mutex); |
2fe27d5a | 937 | } |
2fe27d5a BP |
938 | } |
939 | ||
93295354 BP |
940 | if (retval) { |
941 | ofpbuf_set_data(reply, NULL); | |
942 | ofpbuf_set_size(reply, 0); | |
0672776e | 943 | } |
93295354 | 944 | return !retval; |
2fe27d5a BP |
945 | } |
946 | ||
947 | /* Completes Netlink dump operation 'dump', which must have been initialized | |
948 | * with nl_dump_start(). Returns 0 if the dump operation was error-free, | |
949 | * otherwise a positive errno value describing the problem. */ | |
950 | int | |
951 | nl_dump_done(struct nl_dump *dump) | |
952 | { | |
0672776e | 953 | int status; |
d57695d7 | 954 | |
93295354 BP |
955 | ovs_mutex_lock(&dump->mutex); |
956 | status = dump->status; | |
957 | ovs_mutex_unlock(&dump->mutex); | |
958 | ||
2fe27d5a | 959 | /* Drain any remaining messages that the client didn't read. Otherwise the |
a88b4e04 BP |
960 | * kernel will continue to queue them up and waste buffer space. |
961 | * | |
962 | * XXX We could just destroy and discard the socket in this case. */ | |
0672776e JS |
963 | if (!status) { |
964 | uint64_t tmp_reply_stub[NL_DUMP_BUFSIZE / 8]; | |
965 | struct ofpbuf reply, buf; | |
966 | ||
967 | ofpbuf_use_stub(&buf, tmp_reply_stub, sizeof tmp_reply_stub); | |
968 | while (nl_dump_next(dump, &reply, &buf)) { | |
969 | /* Nothing to do. */ | |
2fe27d5a | 970 | } |
0672776e | 971 | ofpbuf_uninit(&buf); |
93295354 BP |
972 | |
973 | ovs_mutex_lock(&dump->mutex); | |
974 | status = dump->status; | |
975 | ovs_mutex_unlock(&dump->mutex); | |
976 | ovs_assert(status); | |
2fe27d5a | 977 | } |
93295354 | 978 | |
a88b4e04 | 979 | nl_pool_release(dump->sock); |
0791315e | 980 | ovs_mutex_destroy(&dump->mutex); |
93295354 BP |
981 | |
982 | return status == EOF ? 0 : status; | |
2fe27d5a BP |
983 | } |
984 | ||
985 | /* Causes poll_block() to wake up when any of the specified 'events' (which is | |
986 | * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */ | |
987 | void | |
988 | nl_sock_wait(const struct nl_sock *sock, short int events) | |
989 | { | |
22326ba6 AS |
990 | #ifdef _WIN32 |
991 | poll_fd_wait(sock->handle, events); | |
992 | #else | |
2fe27d5a | 993 | poll_fd_wait(sock->fd, events); |
22326ba6 | 994 | #endif |
2fe27d5a | 995 | } |
50802adb | 996 | |
8522ba09 BP |
997 | /* Returns the underlying fd for 'sock', for use in "poll()"-like operations |
998 | * that can't use nl_sock_wait(). | |
999 | * | |
1000 | * It's a little tricky to use the returned fd correctly, because nl_sock does | |
1001 | * "copy on write" to allow a single nl_sock to be used for notifications, | |
1002 | * transactions, and dumps. If 'sock' is used only for notifications and | |
1003 | * transactions (and never for dump) then the usage is safe. */ | |
1004 | int | |
1005 | nl_sock_fd(const struct nl_sock *sock) | |
1006 | { | |
22326ba6 AS |
1007 | #ifdef _WIN32 |
1008 | return sock->handle; | |
1009 | #else | |
8522ba09 | 1010 | return sock->fd; |
22326ba6 | 1011 | #endif |
8522ba09 BP |
1012 | } |
1013 | ||
50802adb JG |
1014 | /* Returns the PID associated with this socket. */ |
1015 | uint32_t | |
1016 | nl_sock_pid(const struct nl_sock *sock) | |
1017 | { | |
1018 | return sock->pid; | |
1019 | } | |
2fe27d5a BP |
1020 | \f |
1021 | /* Miscellaneous. */ | |
1022 | ||
2ad204c8 BP |
1023 | struct genl_family { |
1024 | struct hmap_node hmap_node; | |
1025 | uint16_t id; | |
1026 | char *name; | |
1027 | }; | |
1028 | ||
1029 | static struct hmap genl_families = HMAP_INITIALIZER(&genl_families); | |
1030 | ||
2fe27d5a BP |
1031 | static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = { |
1032 | [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16}, | |
213a13ed | 1033 | [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true}, |
2fe27d5a BP |
1034 | }; |
1035 | ||
2ad204c8 BP |
1036 | static struct genl_family * |
1037 | find_genl_family_by_id(uint16_t id) | |
1038 | { | |
1039 | struct genl_family *family; | |
1040 | ||
1041 | HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0), | |
1042 | &genl_families) { | |
1043 | if (family->id == id) { | |
1044 | return family; | |
1045 | } | |
1046 | } | |
1047 | return NULL; | |
1048 | } | |
1049 | ||
1050 | static void | |
1051 | define_genl_family(uint16_t id, const char *name) | |
1052 | { | |
1053 | struct genl_family *family = find_genl_family_by_id(id); | |
1054 | ||
1055 | if (family) { | |
1056 | if (!strcmp(family->name, name)) { | |
1057 | return; | |
1058 | } | |
1059 | free(family->name); | |
1060 | } else { | |
1061 | family = xmalloc(sizeof *family); | |
1062 | family->id = id; | |
1063 | hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0)); | |
1064 | } | |
1065 | family->name = xstrdup(name); | |
1066 | } | |
1067 | ||
1068 | static const char * | |
1069 | genl_family_to_name(uint16_t id) | |
1070 | { | |
1071 | if (id == GENL_ID_CTRL) { | |
1072 | return "control"; | |
1073 | } else { | |
1074 | struct genl_family *family = find_genl_family_by_id(id); | |
1075 | return family ? family->name : "unknown"; | |
1076 | } | |
1077 | } | |
1078 | ||
e408762f | 1079 | static int |
2a477244 BP |
1080 | do_lookup_genl_family(const char *name, struct nlattr **attrs, |
1081 | struct ofpbuf **replyp) | |
2fe27d5a BP |
1082 | { |
1083 | struct nl_sock *sock; | |
1084 | struct ofpbuf request, *reply; | |
2a477244 | 1085 | int error; |
2fe27d5a | 1086 | |
2a477244 BP |
1087 | *replyp = NULL; |
1088 | error = nl_sock_create(NETLINK_GENERIC, &sock); | |
1089 | if (error) { | |
1090 | return error; | |
2fe27d5a BP |
1091 | } |
1092 | ||
1093 | ofpbuf_init(&request, 0); | |
1094 | nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST, | |
1095 | CTRL_CMD_GETFAMILY, 1); | |
1096 | nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name); | |
2a477244 | 1097 | error = nl_sock_transact(sock, &request, &reply); |
2fe27d5a | 1098 | ofpbuf_uninit(&request); |
2a477244 | 1099 | if (error) { |
2fe27d5a | 1100 | nl_sock_destroy(sock); |
2a477244 | 1101 | return error; |
2fe27d5a BP |
1102 | } |
1103 | ||
1104 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN, | |
2a477244 BP |
1105 | family_policy, attrs, ARRAY_SIZE(family_policy)) |
1106 | || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) { | |
2fe27d5a BP |
1107 | nl_sock_destroy(sock); |
1108 | ofpbuf_delete(reply); | |
2a477244 | 1109 | return EPROTO; |
2fe27d5a BP |
1110 | } |
1111 | ||
2fe27d5a | 1112 | nl_sock_destroy(sock); |
2a477244 BP |
1113 | *replyp = reply; |
1114 | return 0; | |
2fe27d5a BP |
1115 | } |
1116 | ||
e408762f EJ |
1117 | /* Finds the multicast group called 'group_name' in genl family 'family_name'. |
1118 | * When successful, writes its result to 'multicast_group' and returns 0. | |
213a13ed | 1119 | * Otherwise, clears 'multicast_group' and returns a positive error code. |
b3dcb73c | 1120 | */ |
e408762f EJ |
1121 | int |
1122 | nl_lookup_genl_mcgroup(const char *family_name, const char *group_name, | |
b3dcb73c | 1123 | unsigned int *multicast_group) |
e408762f EJ |
1124 | { |
1125 | struct nlattr *family_attrs[ARRAY_SIZE(family_policy)]; | |
6d23c6f4 | 1126 | const struct nlattr *mc; |
2a477244 | 1127 | struct ofpbuf *reply; |
e408762f | 1128 | unsigned int left; |
2a477244 | 1129 | int error; |
e408762f EJ |
1130 | |
1131 | *multicast_group = 0; | |
2a477244 BP |
1132 | error = do_lookup_genl_family(family_name, family_attrs, &reply); |
1133 | if (error) { | |
1134 | return error; | |
e408762f EJ |
1135 | } |
1136 | ||
213a13ed | 1137 | if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) { |
b3dcb73c | 1138 | error = EPROTO; |
213a13ed EJ |
1139 | goto exit; |
1140 | } | |
1141 | ||
6d23c6f4 | 1142 | NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) { |
e408762f EJ |
1143 | static const struct nl_policy mc_policy[] = { |
1144 | [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32}, | |
1145 | [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING}, | |
1146 | }; | |
1147 | ||
1148 | struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)]; | |
1149 | const char *mc_name; | |
1150 | ||
1151 | if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) { | |
2a477244 BP |
1152 | error = EPROTO; |
1153 | goto exit; | |
e408762f EJ |
1154 | } |
1155 | ||
1156 | mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]); | |
1157 | if (!strcmp(group_name, mc_name)) { | |
1158 | *multicast_group = | |
1159 | nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]); | |
2a477244 BP |
1160 | error = 0; |
1161 | goto exit; | |
e408762f EJ |
1162 | } |
1163 | } | |
2a477244 | 1164 | error = EPROTO; |
e408762f | 1165 | |
2a477244 BP |
1166 | exit: |
1167 | ofpbuf_delete(reply); | |
1168 | return error; | |
e408762f EJ |
1169 | } |
1170 | ||
2fe27d5a BP |
1171 | /* If '*number' is 0, translates the given Generic Netlink family 'name' to a |
1172 | * number and stores it in '*number'. If successful, returns 0 and the caller | |
1173 | * may use '*number' as the family number. On failure, returns a positive | |
1174 | * errno value and '*number' caches the errno value. */ | |
1175 | int | |
1176 | nl_lookup_genl_family(const char *name, int *number) | |
1177 | { | |
1178 | if (*number == 0) { | |
2a477244 BP |
1179 | struct nlattr *attrs[ARRAY_SIZE(family_policy)]; |
1180 | struct ofpbuf *reply; | |
1181 | int error; | |
1182 | ||
1183 | error = do_lookup_genl_family(name, attrs, &reply); | |
1184 | if (!error) { | |
1185 | *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]); | |
1186 | define_genl_family(*number, name); | |
1187 | } else { | |
1188 | *number = -error; | |
1189 | } | |
1190 | ofpbuf_delete(reply); | |
1191 | ||
cb22974d | 1192 | ovs_assert(*number != 0); |
2fe27d5a BP |
1193 | } |
1194 | return *number > 0 ? 0 : -*number; | |
1195 | } | |
a88b4e04 BP |
1196 | \f |
1197 | struct nl_pool { | |
1198 | struct nl_sock *socks[16]; | |
1199 | int n; | |
1200 | }; | |
1201 | ||
834d6caf | 1202 | static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER; |
97be1538 | 1203 | static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex); |
a88b4e04 BP |
1204 | |
1205 | static int | |
1206 | nl_pool_alloc(int protocol, struct nl_sock **sockp) | |
1207 | { | |
0bd01224 | 1208 | struct nl_sock *sock = NULL; |
a88b4e04 BP |
1209 | struct nl_pool *pool; |
1210 | ||
1211 | ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools)); | |
1212 | ||
97be1538 | 1213 | ovs_mutex_lock(&pool_mutex); |
a88b4e04 BP |
1214 | pool = &pools[protocol]; |
1215 | if (pool->n > 0) { | |
0bd01224 BP |
1216 | sock = pool->socks[--pool->n]; |
1217 | } | |
97be1538 | 1218 | ovs_mutex_unlock(&pool_mutex); |
0bd01224 BP |
1219 | |
1220 | if (sock) { | |
1221 | *sockp = sock; | |
a88b4e04 BP |
1222 | return 0; |
1223 | } else { | |
1224 | return nl_sock_create(protocol, sockp); | |
1225 | } | |
1226 | } | |
1227 | ||
1228 | static void | |
1229 | nl_pool_release(struct nl_sock *sock) | |
1230 | { | |
1231 | if (sock) { | |
1232 | struct nl_pool *pool = &pools[sock->protocol]; | |
1233 | ||
97be1538 | 1234 | ovs_mutex_lock(&pool_mutex); |
a88b4e04 BP |
1235 | if (pool->n < ARRAY_SIZE(pool->socks)) { |
1236 | pool->socks[pool->n++] = sock; | |
0bd01224 | 1237 | sock = NULL; |
a88b4e04 | 1238 | } |
97be1538 | 1239 | ovs_mutex_unlock(&pool_mutex); |
0bd01224 BP |
1240 | |
1241 | nl_sock_destroy(sock); | |
a88b4e04 BP |
1242 | } |
1243 | } | |
1244 | ||
022ad2b9 BP |
1245 | /* Sends 'request' to the kernel on a Netlink socket for the given 'protocol' |
1246 | * (e.g. NETLINK_ROUTE or NETLINK_GENERIC) and waits for a response. If | |
1247 | * successful, returns 0. On failure, returns a positive errno value. | |
1248 | * | |
1249 | * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's | |
1250 | * reply, which the caller is responsible for freeing with ofpbuf_delete(), and | |
1251 | * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's | |
1252 | * reply, if any, is discarded. | |
1253 | * | |
1254 | * Before the message is sent, nlmsg_len in 'request' will be finalized to | |
1255 | * match ofpbuf_size(msg), nlmsg_pid will be set to the pid of the socket used | |
1256 | * for sending the request, and nlmsg_seq will be initialized. | |
1257 | * | |
1258 | * The caller is responsible for destroying 'request'. | |
1259 | * | |
1260 | * Bare Netlink is an unreliable transport protocol. This function layers | |
1261 | * reliable delivery and reply semantics on top of bare Netlink. | |
1262 | * | |
1263 | * In Netlink, sending a request to the kernel is reliable enough, because the | |
1264 | * kernel will tell us if the message cannot be queued (and we will in that | |
1265 | * case put it on the transmit queue and wait until it can be delivered). | |
1266 | * | |
1267 | * Receiving the reply is the real problem: if the socket buffer is full when | |
1268 | * the kernel tries to send the reply, the reply will be dropped. However, the | |
1269 | * kernel sets a flag that a reply has been dropped. The next call to recv | |
1270 | * then returns ENOBUFS. We can then re-send the request. | |
1271 | * | |
1272 | * Caveats: | |
1273 | * | |
1274 | * 1. Netlink depends on sequence numbers to match up requests and | |
1275 | * replies. The sender of a request supplies a sequence number, and | |
1276 | * the reply echos back that sequence number. | |
1277 | * | |
1278 | * This is fine, but (1) some kernel netlink implementations are | |
1279 | * broken, in that they fail to echo sequence numbers and (2) this | |
1280 | * function will drop packets with non-matching sequence numbers, so | |
1281 | * that only a single request can be usefully transacted at a time. | |
1282 | * | |
1283 | * 2. Resending the request causes it to be re-executed, so the request | |
1284 | * needs to be idempotent. | |
1285 | */ | |
a88b4e04 BP |
1286 | int |
1287 | nl_transact(int protocol, const struct ofpbuf *request, | |
1288 | struct ofpbuf **replyp) | |
1289 | { | |
1290 | struct nl_sock *sock; | |
1291 | int error; | |
1292 | ||
1293 | error = nl_pool_alloc(protocol, &sock); | |
1294 | if (error) { | |
1295 | *replyp = NULL; | |
1296 | return error; | |
1297 | } | |
1298 | ||
1299 | error = nl_sock_transact(sock, request, replyp); | |
1300 | ||
1301 | nl_pool_release(sock); | |
1302 | return error; | |
1303 | } | |
1304 | ||
022ad2b9 BP |
1305 | /* Sends the 'request' member of the 'n' transactions in 'transactions' on a |
1306 | * Netlink socket for the given 'protocol' (e.g. NETLINK_ROUTE or | |
1307 | * NETLINK_GENERIC), in order, and receives responses to all of them. Fills in | |
1308 | * the 'error' member of each transaction with 0 if it was successful, | |
1309 | * otherwise with a positive errno value. If 'reply' is nonnull, then it will | |
1310 | * be filled with the reply if the message receives a detailed reply. In other | |
1311 | * cases, i.e. where the request failed or had no reply beyond an indication of | |
1312 | * success, 'reply' will be cleared if it is nonnull. | |
1313 | * | |
1314 | * The caller is responsible for destroying each request and reply, and the | |
1315 | * transactions array itself. | |
1316 | * | |
1317 | * Before sending each message, this function will finalize nlmsg_len in each | |
1318 | * 'request' to match the ofpbuf's size, set nlmsg_pid to the pid of the socket | |
1319 | * used for the transaction, and initialize nlmsg_seq. | |
1320 | * | |
1321 | * Bare Netlink is an unreliable transport protocol. This function layers | |
1322 | * reliable delivery and reply semantics on top of bare Netlink. See | |
1323 | * nl_transact() for some caveats. | |
1324 | */ | |
a88b4e04 BP |
1325 | void |
1326 | nl_transact_multiple(int protocol, | |
1327 | struct nl_transaction **transactions, size_t n) | |
1328 | { | |
1329 | struct nl_sock *sock; | |
1330 | int error; | |
1331 | ||
1332 | error = nl_pool_alloc(protocol, &sock); | |
1333 | if (!error) { | |
1334 | nl_sock_transact_multiple(sock, transactions, n); | |
1335 | nl_pool_release(sock); | |
1336 | } else { | |
1337 | nl_sock_record_errors__(transactions, n, error); | |
1338 | } | |
1339 | } | |
1340 | ||
2fe27d5a | 1341 | \f |
7d7447df BP |
1342 | static uint32_t |
1343 | nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n) | |
1344 | { | |
1345 | uint32_t seq = sock->next_seq; | |
1346 | ||
1347 | sock->next_seq += n; | |
1348 | ||
1349 | /* Make it impossible for the next request for sequence numbers to wrap | |
1350 | * around to 0. Start over with 1 to avoid ever using a sequence number of | |
1351 | * 0, because the kernel uses sequence number 0 for notifications. */ | |
1352 | if (sock->next_seq >= UINT32_MAX / 2) { | |
1353 | sock->next_seq = 1; | |
1354 | } | |
1355 | ||
1356 | return seq; | |
1357 | } | |
1358 | ||
2fe27d5a | 1359 | static void |
2ad204c8 | 1360 | nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds) |
2fe27d5a BP |
1361 | { |
1362 | struct nlmsg_flag { | |
1363 | unsigned int bits; | |
1364 | const char *name; | |
1365 | }; | |
1366 | static const struct nlmsg_flag flags[] = { | |
1367 | { NLM_F_REQUEST, "REQUEST" }, | |
1368 | { NLM_F_MULTI, "MULTI" }, | |
1369 | { NLM_F_ACK, "ACK" }, | |
1370 | { NLM_F_ECHO, "ECHO" }, | |
1371 | { NLM_F_DUMP, "DUMP" }, | |
1372 | { NLM_F_ROOT, "ROOT" }, | |
1373 | { NLM_F_MATCH, "MATCH" }, | |
1374 | { NLM_F_ATOMIC, "ATOMIC" }, | |
1375 | }; | |
1376 | const struct nlmsg_flag *flag; | |
1377 | uint16_t flags_left; | |
1378 | ||
1379 | ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16, | |
1380 | h->nlmsg_len, h->nlmsg_type); | |
1381 | if (h->nlmsg_type == NLMSG_NOOP) { | |
1382 | ds_put_cstr(ds, "(no-op)"); | |
1383 | } else if (h->nlmsg_type == NLMSG_ERROR) { | |
1384 | ds_put_cstr(ds, "(error)"); | |
1385 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1386 | ds_put_cstr(ds, "(done)"); | |
1387 | } else if (h->nlmsg_type == NLMSG_OVERRUN) { | |
1388 | ds_put_cstr(ds, "(overrun)"); | |
1389 | } else if (h->nlmsg_type < NLMSG_MIN_TYPE) { | |
1390 | ds_put_cstr(ds, "(reserved)"); | |
2ad204c8 BP |
1391 | } else if (protocol == NETLINK_GENERIC) { |
1392 | ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type)); | |
2fe27d5a BP |
1393 | } else { |
1394 | ds_put_cstr(ds, "(family-defined)"); | |
1395 | } | |
1396 | ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags); | |
1397 | flags_left = h->nlmsg_flags; | |
1398 | for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) { | |
1399 | if ((flags_left & flag->bits) == flag->bits) { | |
1400 | ds_put_format(ds, "[%s]", flag->name); | |
1401 | flags_left &= ~flag->bits; | |
1402 | } | |
1403 | } | |
1404 | if (flags_left) { | |
1405 | ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left); | |
1406 | } | |
2c5a6834 BP |
1407 | ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32, |
1408 | h->nlmsg_seq, h->nlmsg_pid); | |
2fe27d5a BP |
1409 | } |
1410 | ||
1411 | static char * | |
7041c3a9 | 1412 | nlmsg_to_string(const struct ofpbuf *buffer, int protocol) |
2fe27d5a BP |
1413 | { |
1414 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1415 | const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN); | |
1416 | if (h) { | |
2ad204c8 | 1417 | nlmsghdr_to_string(h, protocol, &ds); |
2fe27d5a BP |
1418 | if (h->nlmsg_type == NLMSG_ERROR) { |
1419 | const struct nlmsgerr *e; | |
1420 | e = ofpbuf_at(buffer, NLMSG_HDRLEN, | |
1421 | NLMSG_ALIGN(sizeof(struct nlmsgerr))); | |
1422 | if (e) { | |
1423 | ds_put_format(&ds, " error(%d", e->error); | |
1424 | if (e->error < 0) { | |
10a89ef0 | 1425 | ds_put_format(&ds, "(%s)", ovs_strerror(-e->error)); |
2fe27d5a BP |
1426 | } |
1427 | ds_put_cstr(&ds, ", in-reply-to("); | |
2ad204c8 | 1428 | nlmsghdr_to_string(&e->msg, protocol, &ds); |
2fe27d5a BP |
1429 | ds_put_cstr(&ds, "))"); |
1430 | } else { | |
1431 | ds_put_cstr(&ds, " error(truncated)"); | |
1432 | } | |
1433 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1434 | int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error); | |
1435 | if (error) { | |
1436 | ds_put_format(&ds, " done(%d", *error); | |
1437 | if (*error < 0) { | |
10a89ef0 | 1438 | ds_put_format(&ds, "(%s)", ovs_strerror(-*error)); |
2fe27d5a BP |
1439 | } |
1440 | ds_put_cstr(&ds, ")"); | |
1441 | } else { | |
1442 | ds_put_cstr(&ds, " done(truncated)"); | |
1443 | } | |
7041c3a9 BP |
1444 | } else if (protocol == NETLINK_GENERIC) { |
1445 | struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer); | |
1446 | if (genl) { | |
1447 | ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")", | |
1448 | genl->cmd, genl->version); | |
1449 | } | |
2fe27d5a BP |
1450 | } |
1451 | } else { | |
1452 | ds_put_cstr(&ds, "nl(truncated)"); | |
1453 | } | |
1454 | return ds.string; | |
1455 | } | |
1456 | ||
1457 | static void | |
1458 | log_nlmsg(const char *function, int error, | |
7041c3a9 | 1459 | const void *message, size_t size, int protocol) |
2fe27d5a BP |
1460 | { |
1461 | struct ofpbuf buffer; | |
1462 | char *nlmsg; | |
1463 | ||
1464 | if (!VLOG_IS_DBG_ENABLED()) { | |
1465 | return; | |
1466 | } | |
1467 | ||
1468 | ofpbuf_use_const(&buffer, message, size); | |
7041c3a9 | 1469 | nlmsg = nlmsg_to_string(&buffer, protocol); |
10a89ef0 | 1470 | VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg); |
2fe27d5a BP |
1471 | free(nlmsg); |
1472 | } |