]>
Commit | Line | Data |
---|---|---|
2fe27d5a | 1 | /* |
0bd01224 | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 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" |
6b7c12fd | 34 | #include "socket-util.h" |
cc75061a | 35 | #include "util.h" |
2fe27d5a BP |
36 | #include "vlog.h" |
37 | ||
38 | VLOG_DEFINE_THIS_MODULE(netlink_socket); | |
39 | ||
40 | COVERAGE_DEFINE(netlink_overflow); | |
41 | COVERAGE_DEFINE(netlink_received); | |
fc999dda | 42 | COVERAGE_DEFINE(netlink_recv_jumbo); |
2fe27d5a BP |
43 | COVERAGE_DEFINE(netlink_sent); |
44 | ||
45 | /* Linux header file confusion causes this to be undefined. */ | |
46 | #ifndef SOL_NETLINK | |
47 | #define SOL_NETLINK 270 | |
48 | #endif | |
49 | ||
50 | /* A single (bad) Netlink message can in theory dump out many, many log | |
51 | * messages, so the burst size is set quite high here to avoid missing useful | |
52 | * information. Also, at high logging levels we log *all* Netlink messages. */ | |
53 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600); | |
54 | ||
7d7447df | 55 | static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n); |
2fe27d5a | 56 | static void log_nlmsg(const char *function, int error, |
7041c3a9 | 57 | const void *message, size_t size, int protocol); |
2fe27d5a BP |
58 | \f |
59 | /* Netlink sockets. */ | |
60 | ||
0d121c73 | 61 | struct nl_sock { |
2fe27d5a | 62 | int fd; |
7d7447df | 63 | uint32_t next_seq; |
2fe27d5a | 64 | uint32_t pid; |
7041c3a9 | 65 | int protocol; |
cc75061a | 66 | unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */ |
2fe27d5a BP |
67 | }; |
68 | ||
cc75061a BP |
69 | /* Compile-time limit on iovecs, so that we can allocate a maximum-size array |
70 | * of iovecs on the stack. */ | |
71 | #define MAX_IOVS 128 | |
72 | ||
73 | /* Maximum number of iovecs that may be passed to sendmsg, capped at a | |
74 | * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS. | |
75 | * | |
76 | * Initialized by nl_sock_create(). */ | |
77 | static int max_iovs; | |
78 | ||
a88b4e04 BP |
79 | static int nl_pool_alloc(int protocol, struct nl_sock **sockp); |
80 | static void nl_pool_release(struct nl_sock *); | |
2fe27d5a BP |
81 | |
82 | /* Creates a new netlink socket for the given netlink 'protocol' | |
83 | * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the | |
a88b4e04 | 84 | * new socket if successful, otherwise returns a positive errno value. */ |
2fe27d5a | 85 | int |
cceb11f5 | 86 | nl_sock_create(int protocol, struct nl_sock **sockp) |
2fe27d5a | 87 | { |
0bd01224 | 88 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2fe27d5a BP |
89 | struct nl_sock *sock; |
90 | struct sockaddr_nl local, remote; | |
2c5a6834 | 91 | socklen_t local_size; |
d2b9f5b0 | 92 | int rcvbuf; |
2fe27d5a BP |
93 | int retval = 0; |
94 | ||
0bd01224 | 95 | if (ovsthread_once_start(&once)) { |
cc75061a BP |
96 | int save_errno = errno; |
97 | errno = 0; | |
98 | ||
99 | max_iovs = sysconf(_SC_UIO_MAXIOV); | |
100 | if (max_iovs < _XOPEN_IOV_MAX) { | |
101 | if (max_iovs == -1 && errno) { | |
10a89ef0 | 102 | VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", ovs_strerror(errno)); |
cc75061a BP |
103 | } |
104 | max_iovs = _XOPEN_IOV_MAX; | |
105 | } else if (max_iovs > MAX_IOVS) { | |
106 | max_iovs = MAX_IOVS; | |
107 | } | |
108 | ||
109 | errno = save_errno; | |
0bd01224 | 110 | ovsthread_once_done(&once); |
cc75061a BP |
111 | } |
112 | ||
2fe27d5a | 113 | *sockp = NULL; |
488232b7 | 114 | sock = xmalloc(sizeof *sock); |
2fe27d5a BP |
115 | |
116 | sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol); | |
117 | if (sock->fd < 0) { | |
10a89ef0 | 118 | VLOG_ERR("fcntl: %s", ovs_strerror(errno)); |
2fe27d5a BP |
119 | goto error; |
120 | } | |
7041c3a9 | 121 | sock->protocol = protocol; |
7d7447df | 122 | sock->next_seq = 1; |
2fe27d5a | 123 | |
d2b9f5b0 BP |
124 | rcvbuf = 1024 * 1024; |
125 | if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE, | |
126 | &rcvbuf, sizeof rcvbuf)) { | |
80af5ee5 BP |
127 | /* Only root can use SO_RCVBUFFORCE. Everyone else gets EPERM. |
128 | * Warn only if the failure is therefore unexpected. */ | |
f28b6dd3 | 129 | if (errno != EPERM) { |
80af5ee5 | 130 | VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed " |
10a89ef0 | 131 | "(%s)", rcvbuf, ovs_strerror(errno)); |
80af5ee5 | 132 | } |
d2b9f5b0 BP |
133 | } |
134 | ||
cc75061a BP |
135 | retval = get_socket_rcvbuf(sock->fd); |
136 | if (retval < 0) { | |
137 | retval = -retval; | |
138 | goto error; | |
139 | } | |
140 | sock->rcvbuf = retval; | |
141 | ||
2c5a6834 | 142 | /* Connect to kernel (pid 0) as remote address. */ |
2fe27d5a BP |
143 | memset(&remote, 0, sizeof remote); |
144 | remote.nl_family = AF_NETLINK; | |
145 | remote.nl_pid = 0; | |
146 | if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) { | |
10a89ef0 | 147 | VLOG_ERR("connect(0): %s", ovs_strerror(errno)); |
2c5a6834 BP |
148 | goto error; |
149 | } | |
150 | ||
151 | /* Obtain pid assigned by kernel. */ | |
152 | local_size = sizeof local; | |
153 | if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) { | |
10a89ef0 | 154 | VLOG_ERR("getsockname: %s", ovs_strerror(errno)); |
2c5a6834 BP |
155 | goto error; |
156 | } | |
157 | if (local_size < sizeof local || local.nl_family != AF_NETLINK) { | |
158 | VLOG_ERR("getsockname returned bad Netlink name"); | |
159 | retval = EINVAL; | |
160 | goto error; | |
2fe27d5a | 161 | } |
2c5a6834 | 162 | sock->pid = local.nl_pid; |
2fe27d5a | 163 | |
2fe27d5a BP |
164 | *sockp = sock; |
165 | return 0; | |
166 | ||
2fe27d5a BP |
167 | error: |
168 | if (retval == 0) { | |
169 | retval = errno; | |
170 | if (retval == 0) { | |
171 | retval = EINVAL; | |
172 | } | |
173 | } | |
174 | if (sock->fd >= 0) { | |
175 | close(sock->fd); | |
176 | } | |
177 | free(sock); | |
178 | return retval; | |
179 | } | |
180 | ||
c6eab56d BP |
181 | /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and |
182 | * sets '*sockp' to the new socket if successful, otherwise returns a positive | |
183 | * errno value. */ | |
184 | int | |
185 | nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp) | |
186 | { | |
187 | return nl_sock_create(src->protocol, sockp); | |
188 | } | |
189 | ||
2fe27d5a BP |
190 | /* Destroys netlink socket 'sock'. */ |
191 | void | |
192 | nl_sock_destroy(struct nl_sock *sock) | |
193 | { | |
194 | if (sock) { | |
a88b4e04 BP |
195 | close(sock->fd); |
196 | free(sock); | |
2fe27d5a BP |
197 | } |
198 | } | |
199 | ||
cceb11f5 BP |
200 | /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if |
201 | * successful, otherwise a positive errno value. | |
202 | * | |
a838c4fe BP |
203 | * A socket that is subscribed to a multicast group that receives asynchronous |
204 | * notifications must not be used for Netlink transactions or dumps, because | |
205 | * transactions and dumps can cause notifications to be lost. | |
206 | * | |
cceb11f5 BP |
207 | * Multicast group numbers are always positive. |
208 | * | |
209 | * It is not an error to attempt to join a multicast group to which a socket | |
210 | * already belongs. */ | |
211 | int | |
212 | nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
213 | { | |
214 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, | |
215 | &multicast_group, sizeof multicast_group) < 0) { | |
216 | VLOG_WARN("could not join multicast group %u (%s)", | |
10a89ef0 | 217 | multicast_group, ovs_strerror(errno)); |
cceb11f5 BP |
218 | return errno; |
219 | } | |
220 | return 0; | |
221 | } | |
222 | ||
223 | /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if | |
224 | * successful, otherwise a positive errno value. | |
225 | * | |
226 | * Multicast group numbers are always positive. | |
227 | * | |
228 | * It is not an error to attempt to leave a multicast group to which a socket | |
229 | * does not belong. | |
230 | * | |
231 | * On success, reading from 'sock' will still return any messages that were | |
232 | * received on 'multicast_group' before the group was left. */ | |
233 | int | |
234 | nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
235 | { | |
236 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP, | |
237 | &multicast_group, sizeof multicast_group) < 0) { | |
238 | VLOG_WARN("could not leave multicast group %u (%s)", | |
10a89ef0 | 239 | multicast_group, ovs_strerror(errno)); |
cceb11f5 BP |
240 | return errno; |
241 | } | |
242 | return 0; | |
243 | } | |
244 | ||
c6eab56d | 245 | static int |
ff459dd6 BP |
246 | nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, |
247 | uint32_t nlmsg_seq, bool wait) | |
2fe27d5a BP |
248 | { |
249 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg); | |
250 | int error; | |
251 | ||
252 | nlmsg->nlmsg_len = msg->size; | |
ff459dd6 | 253 | nlmsg->nlmsg_seq = nlmsg_seq; |
2fe27d5a BP |
254 | nlmsg->nlmsg_pid = sock->pid; |
255 | do { | |
256 | int retval; | |
257 | retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT); | |
258 | error = retval < 0 ? errno : 0; | |
259 | } while (error == EINTR); | |
7041c3a9 | 260 | log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol); |
2fe27d5a BP |
261 | if (!error) { |
262 | COVERAGE_INC(netlink_sent); | |
263 | } | |
264 | return error; | |
265 | } | |
266 | ||
c6eab56d | 267 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on |
ff459dd6 BP |
268 | * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid |
269 | * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh | |
270 | * sequence number, before the message is sent. | |
c6eab56d BP |
271 | * |
272 | * Returns 0 if successful, otherwise a positive errno value. If | |
273 | * 'wait' is true, then the send will wait until buffer space is ready; | |
274 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. */ | |
275 | int | |
276 | nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait) | |
ff459dd6 BP |
277 | { |
278 | return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait); | |
279 | } | |
280 | ||
281 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on | |
282 | * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid | |
283 | * will be set to 'sock''s pid, and nlmsg_seq will be initialized to | |
284 | * 'nlmsg_seq', before the message is sent. | |
285 | * | |
286 | * Returns 0 if successful, otherwise a positive errno value. If | |
287 | * 'wait' is true, then the send will wait until buffer space is ready; | |
288 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. | |
289 | * | |
290 | * This function is suitable for sending a reply to a request that was received | |
291 | * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */ | |
292 | int | |
293 | nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg, | |
294 | uint32_t nlmsg_seq, bool wait) | |
c6eab56d | 295 | { |
ff459dd6 | 296 | return nl_sock_send__(sock, msg, nlmsg_seq, wait); |
c6eab56d BP |
297 | } |
298 | ||
c6eab56d | 299 | static int |
72d32ac0 | 300 | nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait) |
2fe27d5a | 301 | { |
72d32ac0 BP |
302 | /* We can't accurately predict the size of the data to be received. The |
303 | * caller is supposed to have allocated enough space in 'buf' to handle the | |
304 | * "typical" case. To handle exceptions, we make available enough space in | |
305 | * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable | |
306 | * figure since that's the maximum length of a Netlink attribute). */ | |
2fe27d5a | 307 | struct nlmsghdr *nlmsghdr; |
72d32ac0 | 308 | uint8_t tail[65536]; |
fc999dda | 309 | struct iovec iov[2]; |
fc999dda BP |
310 | struct msghdr msg; |
311 | ssize_t retval; | |
312 | ||
cb22974d | 313 | ovs_assert(buf->allocated >= sizeof *nlmsghdr); |
72d32ac0 | 314 | ofpbuf_clear(buf); |
2fe27d5a | 315 | |
72d32ac0 BP |
316 | iov[0].iov_base = buf->base; |
317 | iov[0].iov_len = buf->allocated; | |
fc999dda | 318 | iov[1].iov_base = tail; |
72d32ac0 | 319 | iov[1].iov_len = sizeof tail; |
fc999dda BP |
320 | |
321 | memset(&msg, 0, sizeof msg); | |
322 | msg.msg_iov = iov; | |
323 | msg.msg_iovlen = 2; | |
324 | ||
2fe27d5a | 325 | do { |
fc999dda BP |
326 | retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT); |
327 | } while (retval < 0 && errno == EINTR); | |
328 | ||
329 | if (retval < 0) { | |
330 | int error = errno; | |
331 | if (error == ENOBUFS) { | |
332 | /* Socket receive buffer overflow dropped one or more messages that | |
333 | * the kernel tried to send to us. */ | |
334 | COVERAGE_INC(netlink_overflow); | |
335 | } | |
fc999dda | 336 | return error; |
2fe27d5a | 337 | } |
fc999dda | 338 | |
2fe27d5a | 339 | if (msg.msg_flags & MSG_TRUNC) { |
34582733 | 340 | VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)", |
72d32ac0 | 341 | sizeof tail); |
fc999dda | 342 | return E2BIG; |
2fe27d5a | 343 | } |
2fe27d5a | 344 | |
fc999dda BP |
345 | nlmsghdr = buf->data; |
346 | if (retval < sizeof *nlmsghdr | |
2fe27d5a | 347 | || nlmsghdr->nlmsg_len < sizeof *nlmsghdr |
fc999dda | 348 | || nlmsghdr->nlmsg_len > retval) { |
34582733 | 349 | VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE"d bytes < %"PRIuSIZE")", |
72d32ac0 | 350 | retval, sizeof *nlmsghdr); |
2fe27d5a BP |
351 | return EPROTO; |
352 | } | |
353 | ||
72d32ac0 BP |
354 | buf->size = MIN(retval, buf->allocated); |
355 | if (retval > buf->allocated) { | |
356 | COVERAGE_INC(netlink_recv_jumbo); | |
357 | ofpbuf_put(buf, tail, retval - buf->allocated); | |
358 | } | |
359 | ||
7041c3a9 | 360 | log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol); |
2fe27d5a BP |
361 | COVERAGE_INC(netlink_received); |
362 | ||
363 | return 0; | |
364 | } | |
365 | ||
72d32ac0 BP |
366 | /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If |
367 | * 'wait' is true, waits for a message to be ready. Otherwise, fails with | |
368 | * EAGAIN if the 'sock' receive buffer is empty. | |
369 | * | |
370 | * The caller must have initialized 'buf' with an allocation of at least | |
371 | * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough | |
372 | * space for a "typical" message. | |
373 | * | |
374 | * On success, returns 0 and replaces 'buf''s previous content by the received | |
375 | * message. This function expands 'buf''s allocated memory, as necessary, to | |
376 | * hold the actual size of the received message. | |
c6eab56d | 377 | * |
72d32ac0 BP |
378 | * On failure, returns a positive errno value and clears 'buf' to zero length. |
379 | * 'buf' retains its previous memory allocation. | |
380 | * | |
381 | * Regardless of success or failure, this function resets 'buf''s headroom to | |
382 | * 0. */ | |
c6eab56d | 383 | int |
72d32ac0 | 384 | nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait) |
c6eab56d | 385 | { |
72d32ac0 | 386 | return nl_sock_recv__(sock, buf, wait); |
cc75061a BP |
387 | } |
388 | ||
389 | static void | |
390 | nl_sock_record_errors__(struct nl_transaction **transactions, size_t n, | |
391 | int error) | |
392 | { | |
393 | size_t i; | |
394 | ||
395 | for (i = 0; i < n; i++) { | |
72d32ac0 BP |
396 | struct nl_transaction *txn = transactions[i]; |
397 | ||
398 | txn->error = error; | |
399 | if (txn->reply) { | |
400 | ofpbuf_clear(txn->reply); | |
401 | } | |
cc75061a BP |
402 | } |
403 | } | |
404 | ||
405 | static int | |
406 | nl_sock_transact_multiple__(struct nl_sock *sock, | |
407 | struct nl_transaction **transactions, size_t n, | |
408 | size_t *done) | |
409 | { | |
72d32ac0 BP |
410 | uint64_t tmp_reply_stub[1024 / 8]; |
411 | struct nl_transaction tmp_txn; | |
412 | struct ofpbuf tmp_reply; | |
413 | ||
414 | uint32_t base_seq; | |
cc75061a BP |
415 | struct iovec iovs[MAX_IOVS]; |
416 | struct msghdr msg; | |
417 | int error; | |
418 | int i; | |
419 | ||
72d32ac0 | 420 | base_seq = nl_sock_allocate_seq(sock, n); |
cc75061a BP |
421 | *done = 0; |
422 | for (i = 0; i < n; i++) { | |
72d32ac0 BP |
423 | struct nl_transaction *txn = transactions[i]; |
424 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request); | |
cc75061a | 425 | |
72d32ac0 BP |
426 | nlmsg->nlmsg_len = txn->request->size; |
427 | nlmsg->nlmsg_seq = base_seq + i; | |
cc75061a | 428 | nlmsg->nlmsg_pid = sock->pid; |
cc75061a | 429 | |
72d32ac0 BP |
430 | iovs[i].iov_base = txn->request->data; |
431 | iovs[i].iov_len = txn->request->size; | |
cc75061a BP |
432 | } |
433 | ||
434 | memset(&msg, 0, sizeof msg); | |
435 | msg.msg_iov = iovs; | |
436 | msg.msg_iovlen = n; | |
437 | do { | |
438 | error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0; | |
439 | } while (error == EINTR); | |
440 | ||
441 | for (i = 0; i < n; i++) { | |
72d32ac0 | 442 | struct nl_transaction *txn = transactions[i]; |
cc75061a | 443 | |
72d32ac0 | 444 | log_nlmsg(__func__, error, txn->request->data, txn->request->size, |
cc75061a BP |
445 | sock->protocol); |
446 | } | |
447 | if (!error) { | |
448 | COVERAGE_ADD(netlink_sent, n); | |
449 | } | |
450 | ||
451 | if (error) { | |
452 | return error; | |
453 | } | |
454 | ||
72d32ac0 BP |
455 | ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub); |
456 | tmp_txn.request = NULL; | |
457 | tmp_txn.reply = &tmp_reply; | |
458 | tmp_txn.error = 0; | |
cc75061a | 459 | while (n > 0) { |
72d32ac0 BP |
460 | struct nl_transaction *buf_txn, *txn; |
461 | uint32_t seq; | |
462 | ||
463 | /* Find a transaction whose buffer we can use for receiving a reply. | |
464 | * If no such transaction is left, use tmp_txn. */ | |
465 | buf_txn = &tmp_txn; | |
466 | for (i = 0; i < n; i++) { | |
467 | if (transactions[i]->reply) { | |
468 | buf_txn = transactions[i]; | |
469 | break; | |
470 | } | |
471 | } | |
cc75061a | 472 | |
72d32ac0 BP |
473 | /* Receive a reply. */ |
474 | error = nl_sock_recv__(sock, buf_txn->reply, false); | |
475 | if (error) { | |
476 | if (error == EAGAIN) { | |
477 | nl_sock_record_errors__(transactions, n, 0); | |
478 | *done += n; | |
479 | error = 0; | |
480 | } | |
481 | break; | |
cc75061a BP |
482 | } |
483 | ||
72d32ac0 BP |
484 | /* Match the reply up with a transaction. */ |
485 | seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq; | |
486 | if (seq < base_seq || seq >= base_seq + n) { | |
487 | VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq); | |
cc75061a BP |
488 | continue; |
489 | } | |
72d32ac0 BP |
490 | i = seq - base_seq; |
491 | txn = transactions[i]; | |
cc75061a | 492 | |
72d32ac0 BP |
493 | /* Fill in the results for 'txn'. */ |
494 | if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) { | |
495 | if (txn->reply) { | |
496 | ofpbuf_clear(txn->reply); | |
497 | } | |
498 | if (txn->error) { | |
cc75061a | 499 | VLOG_DBG_RL(&rl, "received NAK error=%d (%s)", |
10a89ef0 | 500 | error, ovs_strerror(txn->error)); |
cc75061a | 501 | } |
cc75061a | 502 | } else { |
72d32ac0 BP |
503 | txn->error = 0; |
504 | if (txn->reply && txn != buf_txn) { | |
505 | /* Swap buffers. */ | |
506 | struct ofpbuf *reply = buf_txn->reply; | |
507 | buf_txn->reply = txn->reply; | |
508 | txn->reply = reply; | |
509 | } | |
cc75061a BP |
510 | } |
511 | ||
72d32ac0 BP |
512 | /* Fill in the results for transactions before 'txn'. (We have to do |
513 | * this after the results for 'txn' itself because of the buffer swap | |
514 | * above.) */ | |
515 | nl_sock_record_errors__(transactions, i, 0); | |
516 | ||
517 | /* Advance. */ | |
cc75061a BP |
518 | *done += i + 1; |
519 | transactions += i + 1; | |
520 | n -= i + 1; | |
72d32ac0 | 521 | base_seq += i + 1; |
cc75061a | 522 | } |
72d32ac0 | 523 | ofpbuf_uninit(&tmp_reply); |
cc75061a | 524 | |
72d32ac0 | 525 | return error; |
cc75061a BP |
526 | } |
527 | ||
72d32ac0 BP |
528 | /* Sends the 'request' member of the 'n' transactions in 'transactions' on |
529 | * 'sock', in order, and receives responses to all of them. Fills in the | |
cc75061a | 530 | * 'error' member of each transaction with 0 if it was successful, otherwise |
72d32ac0 BP |
531 | * with a positive errno value. If 'reply' is nonnull, then it will be filled |
532 | * with the reply if the message receives a detailed reply. In other cases, | |
533 | * i.e. where the request failed or had no reply beyond an indication of | |
534 | * success, 'reply' will be cleared if it is nonnull. | |
cc75061a BP |
535 | * |
536 | * The caller is responsible for destroying each request and reply, and the | |
537 | * transactions array itself. | |
538 | * | |
539 | * Before sending each message, this function will finalize nlmsg_len in each | |
72d32ac0 BP |
540 | * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and |
541 | * initialize nlmsg_seq. | |
cc75061a BP |
542 | * |
543 | * Bare Netlink is an unreliable transport protocol. This function layers | |
544 | * reliable delivery and reply semantics on top of bare Netlink. See | |
545 | * nl_sock_transact() for some caveats. | |
546 | */ | |
547 | void | |
548 | nl_sock_transact_multiple(struct nl_sock *sock, | |
549 | struct nl_transaction **transactions, size_t n) | |
550 | { | |
551 | int max_batch_count; | |
552 | int error; | |
553 | ||
554 | if (!n) { | |
555 | return; | |
556 | } | |
557 | ||
cc75061a BP |
558 | /* In theory, every request could have a 64 kB reply. But the default and |
559 | * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to | |
560 | * be a bit below 128 kB, so that would only allow a single message in a | |
561 | * "batch". So we assume that replies average (at most) 4 kB, which allows | |
562 | * a good deal of batching. | |
563 | * | |
564 | * In practice, most of the requests that we batch either have no reply at | |
565 | * all or a brief reply. */ | |
566 | max_batch_count = MAX(sock->rcvbuf / 4096, 1); | |
567 | max_batch_count = MIN(max_batch_count, max_iovs); | |
568 | ||
569 | while (n > 0) { | |
570 | size_t count, bytes; | |
571 | size_t done; | |
572 | ||
573 | /* Batch up to 'max_batch_count' transactions. But cap it at about a | |
574 | * page of requests total because big skbuffs are expensive to | |
575 | * allocate in the kernel. */ | |
576 | #if defined(PAGESIZE) | |
577 | enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) }; | |
578 | #else | |
579 | enum { MAX_BATCH_BYTES = 4096 - 512 }; | |
580 | #endif | |
581 | bytes = transactions[0]->request->size; | |
582 | for (count = 1; count < n && count < max_batch_count; count++) { | |
583 | if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) { | |
584 | break; | |
585 | } | |
586 | bytes += transactions[count]->request->size; | |
587 | } | |
588 | ||
589 | error = nl_sock_transact_multiple__(sock, transactions, count, &done); | |
590 | transactions += done; | |
591 | n -= done; | |
592 | ||
593 | if (error == ENOBUFS) { | |
594 | VLOG_DBG_RL(&rl, "receive buffer overflow, resending request"); | |
595 | } else if (error) { | |
10a89ef0 | 596 | VLOG_ERR_RL(&rl, "transaction error (%s)", ovs_strerror(error)); |
cc75061a BP |
597 | nl_sock_record_errors__(transactions, n, error); |
598 | } | |
599 | } | |
600 | } | |
601 | ||
2fe27d5a BP |
602 | /* Sends 'request' to the kernel via 'sock' and waits for a response. If |
603 | * successful, returns 0. On failure, returns a positive errno value. | |
604 | * | |
605 | * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's | |
606 | * reply, which the caller is responsible for freeing with ofpbuf_delete(), and | |
607 | * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's | |
608 | * reply, if any, is discarded. | |
609 | * | |
7d7447df BP |
610 | * Before the message is sent, nlmsg_len in 'request' will be finalized to |
611 | * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will | |
612 | * be initialized, NLM_F_ACK will be set in nlmsg_flags. | |
2fe27d5a BP |
613 | * |
614 | * The caller is responsible for destroying 'request'. | |
615 | * | |
616 | * Bare Netlink is an unreliable transport protocol. This function layers | |
617 | * reliable delivery and reply semantics on top of bare Netlink. | |
618 | * | |
619 | * In Netlink, sending a request to the kernel is reliable enough, because the | |
620 | * kernel will tell us if the message cannot be queued (and we will in that | |
621 | * case put it on the transmit queue and wait until it can be delivered). | |
622 | * | |
623 | * Receiving the reply is the real problem: if the socket buffer is full when | |
624 | * the kernel tries to send the reply, the reply will be dropped. However, the | |
625 | * kernel sets a flag that a reply has been dropped. The next call to recv | |
626 | * then returns ENOBUFS. We can then re-send the request. | |
627 | * | |
628 | * Caveats: | |
629 | * | |
630 | * 1. Netlink depends on sequence numbers to match up requests and | |
631 | * replies. The sender of a request supplies a sequence number, and | |
632 | * the reply echos back that sequence number. | |
633 | * | |
634 | * This is fine, but (1) some kernel netlink implementations are | |
635 | * broken, in that they fail to echo sequence numbers and (2) this | |
636 | * function will drop packets with non-matching sequence numbers, so | |
637 | * that only a single request can be usefully transacted at a time. | |
638 | * | |
639 | * 2. Resending the request causes it to be re-executed, so the request | |
640 | * needs to be idempotent. | |
641 | */ | |
642 | int | |
cc75061a BP |
643 | nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request, |
644 | struct ofpbuf **replyp) | |
2fe27d5a | 645 | { |
cc75061a BP |
646 | struct nl_transaction *transactionp; |
647 | struct nl_transaction transaction; | |
2fe27d5a | 648 | |
ebc56baa | 649 | transaction.request = CONST_CAST(struct ofpbuf *, request); |
72d32ac0 | 650 | transaction.reply = replyp ? ofpbuf_new(1024) : NULL; |
cc75061a | 651 | transactionp = &transaction; |
72d32ac0 | 652 | |
cc75061a | 653 | nl_sock_transact_multiple(sock, &transactionp, 1); |
72d32ac0 | 654 | |
2fe27d5a | 655 | if (replyp) { |
72d32ac0 BP |
656 | if (transaction.error) { |
657 | ofpbuf_delete(transaction.reply); | |
658 | *replyp = NULL; | |
659 | } else { | |
660 | *replyp = transaction.reply; | |
661 | } | |
2fe27d5a | 662 | } |
72d32ac0 | 663 | |
cc75061a | 664 | return transaction.error; |
2fe27d5a BP |
665 | } |
666 | ||
6b7c12fd BP |
667 | /* Drain all the messages currently in 'sock''s receive queue. */ |
668 | int | |
669 | nl_sock_drain(struct nl_sock *sock) | |
670 | { | |
671 | return drain_rcvbuf(sock->fd); | |
672 | } | |
673 | ||
a88b4e04 BP |
674 | /* Starts a Netlink "dump" operation, by sending 'request' to the kernel on a |
675 | * Netlink socket created with the given 'protocol', and initializes 'dump' to | |
676 | * reflect the state of the operation. | |
2fe27d5a BP |
677 | * |
678 | * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will | |
a88b4e04 BP |
679 | * be set to the Netlink socket's pid, before the message is sent. NLM_F_DUMP |
680 | * and NLM_F_ACK will be set in nlmsg_flags. | |
2fe27d5a | 681 | * |
a88b4e04 | 682 | * The design of this Netlink socket library ensures that the dump is reliable. |
2fe27d5a BP |
683 | * |
684 | * This function provides no status indication. An error status for the entire | |
685 | * dump operation is provided when it is completed by calling nl_dump_done(). | |
686 | * | |
c6eab56d | 687 | * The caller is responsible for destroying 'request'. |
2fe27d5a BP |
688 | */ |
689 | void | |
a88b4e04 | 690 | nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request) |
2fe27d5a | 691 | { |
72d32ac0 | 692 | ofpbuf_init(&dump->buffer, 4096); |
a88b4e04 BP |
693 | dump->status = nl_pool_alloc(protocol, &dump->sock); |
694 | if (dump->status) { | |
695 | return; | |
c6eab56d | 696 | } |
7d7447df BP |
697 | |
698 | nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK; | |
a88b4e04 BP |
699 | dump->status = nl_sock_send__(dump->sock, request, |
700 | nl_sock_allocate_seq(dump->sock, 1), true); | |
7d7447df | 701 | dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq; |
2fe27d5a BP |
702 | } |
703 | ||
704 | /* Helper function for nl_dump_next(). */ | |
705 | static int | |
72d32ac0 | 706 | nl_dump_recv(struct nl_dump *dump) |
2fe27d5a BP |
707 | { |
708 | struct nlmsghdr *nlmsghdr; | |
2fe27d5a BP |
709 | int retval; |
710 | ||
72d32ac0 | 711 | retval = nl_sock_recv__(dump->sock, &dump->buffer, true); |
2fe27d5a BP |
712 | if (retval) { |
713 | return retval == EINTR ? EAGAIN : retval; | |
714 | } | |
2fe27d5a | 715 | |
72d32ac0 | 716 | nlmsghdr = nl_msg_nlmsghdr(&dump->buffer); |
2fe27d5a | 717 | if (dump->seq != nlmsghdr->nlmsg_seq) { |
727ef33f | 718 | VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32, |
2fe27d5a BP |
719 | nlmsghdr->nlmsg_seq, dump->seq); |
720 | return EAGAIN; | |
721 | } | |
722 | ||
72d32ac0 | 723 | if (nl_msg_nlmsgerr(&dump->buffer, &retval)) { |
2fe27d5a | 724 | VLOG_INFO_RL(&rl, "netlink dump request error (%s)", |
10a89ef0 | 725 | ovs_strerror(retval)); |
2fe27d5a BP |
726 | return retval && retval != EAGAIN ? retval : EPROTO; |
727 | } | |
728 | ||
729 | return 0; | |
730 | } | |
731 | ||
732 | /* Attempts to retrieve another reply from 'dump', which must have been | |
733 | * initialized with nl_dump_start(). | |
734 | * | |
735 | * If successful, returns true and points 'reply->data' and 'reply->size' to | |
736 | * the message that was retrieved. The caller must not modify 'reply' (because | |
737 | * it points into the middle of a larger buffer). | |
738 | * | |
739 | * On failure, returns false and sets 'reply->data' to NULL and 'reply->size' | |
740 | * to 0. Failure might indicate an actual error or merely the end of replies. | |
741 | * An error status for the entire dump operation is provided when it is | |
742 | * completed by calling nl_dump_done(). | |
743 | */ | |
744 | bool | |
745 | nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply) | |
746 | { | |
747 | struct nlmsghdr *nlmsghdr; | |
748 | ||
749 | reply->data = NULL; | |
750 | reply->size = 0; | |
751 | if (dump->status) { | |
752 | return false; | |
753 | } | |
754 | ||
72d32ac0 BP |
755 | while (!dump->buffer.size) { |
756 | int retval = nl_dump_recv(dump); | |
2fe27d5a | 757 | if (retval) { |
72d32ac0 | 758 | ofpbuf_clear(&dump->buffer); |
2fe27d5a BP |
759 | if (retval != EAGAIN) { |
760 | dump->status = retval; | |
761 | return false; | |
762 | } | |
763 | } | |
764 | } | |
765 | ||
72d32ac0 | 766 | nlmsghdr = nl_msg_next(&dump->buffer, reply); |
2fe27d5a BP |
767 | if (!nlmsghdr) { |
768 | VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment"); | |
769 | dump->status = EPROTO; | |
770 | return false; | |
771 | } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) { | |
772 | dump->status = EOF; | |
773 | return false; | |
774 | } | |
775 | ||
776 | return true; | |
777 | } | |
778 | ||
779 | /* Completes Netlink dump operation 'dump', which must have been initialized | |
780 | * with nl_dump_start(). Returns 0 if the dump operation was error-free, | |
781 | * otherwise a positive errno value describing the problem. */ | |
782 | int | |
783 | nl_dump_done(struct nl_dump *dump) | |
784 | { | |
785 | /* Drain any remaining messages that the client didn't read. Otherwise the | |
a88b4e04 BP |
786 | * kernel will continue to queue them up and waste buffer space. |
787 | * | |
788 | * XXX We could just destroy and discard the socket in this case. */ | |
2fe27d5a BP |
789 | while (!dump->status) { |
790 | struct ofpbuf reply; | |
791 | if (!nl_dump_next(dump, &reply)) { | |
cb22974d | 792 | ovs_assert(dump->status); |
2fe27d5a BP |
793 | } |
794 | } | |
a88b4e04 | 795 | nl_pool_release(dump->sock); |
72d32ac0 | 796 | ofpbuf_uninit(&dump->buffer); |
2fe27d5a BP |
797 | return dump->status == EOF ? 0 : dump->status; |
798 | } | |
799 | ||
800 | /* Causes poll_block() to wake up when any of the specified 'events' (which is | |
801 | * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */ | |
802 | void | |
803 | nl_sock_wait(const struct nl_sock *sock, short int events) | |
804 | { | |
805 | poll_fd_wait(sock->fd, events); | |
806 | } | |
50802adb | 807 | |
8522ba09 BP |
808 | /* Returns the underlying fd for 'sock', for use in "poll()"-like operations |
809 | * that can't use nl_sock_wait(). | |
810 | * | |
811 | * It's a little tricky to use the returned fd correctly, because nl_sock does | |
812 | * "copy on write" to allow a single nl_sock to be used for notifications, | |
813 | * transactions, and dumps. If 'sock' is used only for notifications and | |
814 | * transactions (and never for dump) then the usage is safe. */ | |
815 | int | |
816 | nl_sock_fd(const struct nl_sock *sock) | |
817 | { | |
818 | return sock->fd; | |
819 | } | |
820 | ||
50802adb JG |
821 | /* Returns the PID associated with this socket. */ |
822 | uint32_t | |
823 | nl_sock_pid(const struct nl_sock *sock) | |
824 | { | |
825 | return sock->pid; | |
826 | } | |
2fe27d5a BP |
827 | \f |
828 | /* Miscellaneous. */ | |
829 | ||
2ad204c8 BP |
830 | struct genl_family { |
831 | struct hmap_node hmap_node; | |
832 | uint16_t id; | |
833 | char *name; | |
834 | }; | |
835 | ||
836 | static struct hmap genl_families = HMAP_INITIALIZER(&genl_families); | |
837 | ||
2fe27d5a BP |
838 | static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = { |
839 | [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16}, | |
213a13ed | 840 | [CTRL_ATTR_MCAST_GROUPS] = {.type = NL_A_NESTED, .optional = true}, |
2fe27d5a BP |
841 | }; |
842 | ||
2ad204c8 BP |
843 | static struct genl_family * |
844 | find_genl_family_by_id(uint16_t id) | |
845 | { | |
846 | struct genl_family *family; | |
847 | ||
848 | HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0), | |
849 | &genl_families) { | |
850 | if (family->id == id) { | |
851 | return family; | |
852 | } | |
853 | } | |
854 | return NULL; | |
855 | } | |
856 | ||
857 | static void | |
858 | define_genl_family(uint16_t id, const char *name) | |
859 | { | |
860 | struct genl_family *family = find_genl_family_by_id(id); | |
861 | ||
862 | if (family) { | |
863 | if (!strcmp(family->name, name)) { | |
864 | return; | |
865 | } | |
866 | free(family->name); | |
867 | } else { | |
868 | family = xmalloc(sizeof *family); | |
869 | family->id = id; | |
870 | hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0)); | |
871 | } | |
872 | family->name = xstrdup(name); | |
873 | } | |
874 | ||
875 | static const char * | |
876 | genl_family_to_name(uint16_t id) | |
877 | { | |
878 | if (id == GENL_ID_CTRL) { | |
879 | return "control"; | |
880 | } else { | |
881 | struct genl_family *family = find_genl_family_by_id(id); | |
882 | return family ? family->name : "unknown"; | |
883 | } | |
884 | } | |
885 | ||
e408762f | 886 | static int |
2a477244 BP |
887 | do_lookup_genl_family(const char *name, struct nlattr **attrs, |
888 | struct ofpbuf **replyp) | |
2fe27d5a BP |
889 | { |
890 | struct nl_sock *sock; | |
891 | struct ofpbuf request, *reply; | |
2a477244 | 892 | int error; |
2fe27d5a | 893 | |
2a477244 BP |
894 | *replyp = NULL; |
895 | error = nl_sock_create(NETLINK_GENERIC, &sock); | |
896 | if (error) { | |
897 | return error; | |
2fe27d5a BP |
898 | } |
899 | ||
900 | ofpbuf_init(&request, 0); | |
901 | nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST, | |
902 | CTRL_CMD_GETFAMILY, 1); | |
903 | nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name); | |
2a477244 | 904 | error = nl_sock_transact(sock, &request, &reply); |
2fe27d5a | 905 | ofpbuf_uninit(&request); |
2a477244 | 906 | if (error) { |
2fe27d5a | 907 | nl_sock_destroy(sock); |
2a477244 | 908 | return error; |
2fe27d5a BP |
909 | } |
910 | ||
911 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN, | |
2a477244 BP |
912 | family_policy, attrs, ARRAY_SIZE(family_policy)) |
913 | || nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]) == 0) { | |
2fe27d5a BP |
914 | nl_sock_destroy(sock); |
915 | ofpbuf_delete(reply); | |
2a477244 | 916 | return EPROTO; |
2fe27d5a BP |
917 | } |
918 | ||
2fe27d5a | 919 | nl_sock_destroy(sock); |
2a477244 BP |
920 | *replyp = reply; |
921 | return 0; | |
2fe27d5a BP |
922 | } |
923 | ||
e408762f EJ |
924 | /* Finds the multicast group called 'group_name' in genl family 'family_name'. |
925 | * When successful, writes its result to 'multicast_group' and returns 0. | |
213a13ed | 926 | * Otherwise, clears 'multicast_group' and returns a positive error code. |
b3dcb73c | 927 | */ |
e408762f EJ |
928 | int |
929 | nl_lookup_genl_mcgroup(const char *family_name, const char *group_name, | |
b3dcb73c | 930 | unsigned int *multicast_group) |
e408762f EJ |
931 | { |
932 | struct nlattr *family_attrs[ARRAY_SIZE(family_policy)]; | |
6d23c6f4 | 933 | const struct nlattr *mc; |
2a477244 | 934 | struct ofpbuf *reply; |
e408762f | 935 | unsigned int left; |
2a477244 | 936 | int error; |
e408762f EJ |
937 | |
938 | *multicast_group = 0; | |
2a477244 BP |
939 | error = do_lookup_genl_family(family_name, family_attrs, &reply); |
940 | if (error) { | |
941 | return error; | |
e408762f EJ |
942 | } |
943 | ||
213a13ed | 944 | if (!family_attrs[CTRL_ATTR_MCAST_GROUPS]) { |
b3dcb73c | 945 | error = EPROTO; |
213a13ed EJ |
946 | goto exit; |
947 | } | |
948 | ||
6d23c6f4 | 949 | NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) { |
e408762f EJ |
950 | static const struct nl_policy mc_policy[] = { |
951 | [CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32}, | |
952 | [CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING}, | |
953 | }; | |
954 | ||
955 | struct nlattr *mc_attrs[ARRAY_SIZE(mc_policy)]; | |
956 | const char *mc_name; | |
957 | ||
958 | if (!nl_parse_nested(mc, mc_policy, mc_attrs, ARRAY_SIZE(mc_policy))) { | |
2a477244 BP |
959 | error = EPROTO; |
960 | goto exit; | |
e408762f EJ |
961 | } |
962 | ||
963 | mc_name = nl_attr_get_string(mc_attrs[CTRL_ATTR_MCAST_GRP_NAME]); | |
964 | if (!strcmp(group_name, mc_name)) { | |
965 | *multicast_group = | |
966 | nl_attr_get_u32(mc_attrs[CTRL_ATTR_MCAST_GRP_ID]); | |
2a477244 BP |
967 | error = 0; |
968 | goto exit; | |
e408762f EJ |
969 | } |
970 | } | |
2a477244 | 971 | error = EPROTO; |
e408762f | 972 | |
2a477244 BP |
973 | exit: |
974 | ofpbuf_delete(reply); | |
975 | return error; | |
e408762f EJ |
976 | } |
977 | ||
2fe27d5a BP |
978 | /* If '*number' is 0, translates the given Generic Netlink family 'name' to a |
979 | * number and stores it in '*number'. If successful, returns 0 and the caller | |
980 | * may use '*number' as the family number. On failure, returns a positive | |
981 | * errno value and '*number' caches the errno value. */ | |
982 | int | |
983 | nl_lookup_genl_family(const char *name, int *number) | |
984 | { | |
985 | if (*number == 0) { | |
2a477244 BP |
986 | struct nlattr *attrs[ARRAY_SIZE(family_policy)]; |
987 | struct ofpbuf *reply; | |
988 | int error; | |
989 | ||
990 | error = do_lookup_genl_family(name, attrs, &reply); | |
991 | if (!error) { | |
992 | *number = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]); | |
993 | define_genl_family(*number, name); | |
994 | } else { | |
995 | *number = -error; | |
996 | } | |
997 | ofpbuf_delete(reply); | |
998 | ||
cb22974d | 999 | ovs_assert(*number != 0); |
2fe27d5a BP |
1000 | } |
1001 | return *number > 0 ? 0 : -*number; | |
1002 | } | |
a88b4e04 BP |
1003 | \f |
1004 | struct nl_pool { | |
1005 | struct nl_sock *socks[16]; | |
1006 | int n; | |
1007 | }; | |
1008 | ||
834d6caf | 1009 | static struct ovs_mutex pool_mutex = OVS_MUTEX_INITIALIZER; |
97be1538 | 1010 | static struct nl_pool pools[MAX_LINKS] OVS_GUARDED_BY(pool_mutex); |
a88b4e04 BP |
1011 | |
1012 | static int | |
1013 | nl_pool_alloc(int protocol, struct nl_sock **sockp) | |
1014 | { | |
0bd01224 | 1015 | struct nl_sock *sock = NULL; |
a88b4e04 BP |
1016 | struct nl_pool *pool; |
1017 | ||
1018 | ovs_assert(protocol >= 0 && protocol < ARRAY_SIZE(pools)); | |
1019 | ||
97be1538 | 1020 | ovs_mutex_lock(&pool_mutex); |
a88b4e04 BP |
1021 | pool = &pools[protocol]; |
1022 | if (pool->n > 0) { | |
0bd01224 BP |
1023 | sock = pool->socks[--pool->n]; |
1024 | } | |
97be1538 | 1025 | ovs_mutex_unlock(&pool_mutex); |
0bd01224 BP |
1026 | |
1027 | if (sock) { | |
1028 | *sockp = sock; | |
a88b4e04 BP |
1029 | return 0; |
1030 | } else { | |
1031 | return nl_sock_create(protocol, sockp); | |
1032 | } | |
1033 | } | |
1034 | ||
1035 | static void | |
1036 | nl_pool_release(struct nl_sock *sock) | |
1037 | { | |
1038 | if (sock) { | |
1039 | struct nl_pool *pool = &pools[sock->protocol]; | |
1040 | ||
97be1538 | 1041 | ovs_mutex_lock(&pool_mutex); |
a88b4e04 BP |
1042 | if (pool->n < ARRAY_SIZE(pool->socks)) { |
1043 | pool->socks[pool->n++] = sock; | |
0bd01224 | 1044 | sock = NULL; |
a88b4e04 | 1045 | } |
97be1538 | 1046 | ovs_mutex_unlock(&pool_mutex); |
0bd01224 BP |
1047 | |
1048 | nl_sock_destroy(sock); | |
a88b4e04 BP |
1049 | } |
1050 | } | |
1051 | ||
1052 | int | |
1053 | nl_transact(int protocol, const struct ofpbuf *request, | |
1054 | struct ofpbuf **replyp) | |
1055 | { | |
1056 | struct nl_sock *sock; | |
1057 | int error; | |
1058 | ||
1059 | error = nl_pool_alloc(protocol, &sock); | |
1060 | if (error) { | |
1061 | *replyp = NULL; | |
1062 | return error; | |
1063 | } | |
1064 | ||
1065 | error = nl_sock_transact(sock, request, replyp); | |
1066 | ||
1067 | nl_pool_release(sock); | |
1068 | return error; | |
1069 | } | |
1070 | ||
1071 | void | |
1072 | nl_transact_multiple(int protocol, | |
1073 | struct nl_transaction **transactions, size_t n) | |
1074 | { | |
1075 | struct nl_sock *sock; | |
1076 | int error; | |
1077 | ||
1078 | error = nl_pool_alloc(protocol, &sock); | |
1079 | if (!error) { | |
1080 | nl_sock_transact_multiple(sock, transactions, n); | |
1081 | nl_pool_release(sock); | |
1082 | } else { | |
1083 | nl_sock_record_errors__(transactions, n, error); | |
1084 | } | |
1085 | } | |
1086 | ||
2fe27d5a | 1087 | \f |
7d7447df BP |
1088 | static uint32_t |
1089 | nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n) | |
1090 | { | |
1091 | uint32_t seq = sock->next_seq; | |
1092 | ||
1093 | sock->next_seq += n; | |
1094 | ||
1095 | /* Make it impossible for the next request for sequence numbers to wrap | |
1096 | * around to 0. Start over with 1 to avoid ever using a sequence number of | |
1097 | * 0, because the kernel uses sequence number 0 for notifications. */ | |
1098 | if (sock->next_seq >= UINT32_MAX / 2) { | |
1099 | sock->next_seq = 1; | |
1100 | } | |
1101 | ||
1102 | return seq; | |
1103 | } | |
1104 | ||
2fe27d5a | 1105 | static void |
2ad204c8 | 1106 | nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds) |
2fe27d5a BP |
1107 | { |
1108 | struct nlmsg_flag { | |
1109 | unsigned int bits; | |
1110 | const char *name; | |
1111 | }; | |
1112 | static const struct nlmsg_flag flags[] = { | |
1113 | { NLM_F_REQUEST, "REQUEST" }, | |
1114 | { NLM_F_MULTI, "MULTI" }, | |
1115 | { NLM_F_ACK, "ACK" }, | |
1116 | { NLM_F_ECHO, "ECHO" }, | |
1117 | { NLM_F_DUMP, "DUMP" }, | |
1118 | { NLM_F_ROOT, "ROOT" }, | |
1119 | { NLM_F_MATCH, "MATCH" }, | |
1120 | { NLM_F_ATOMIC, "ATOMIC" }, | |
1121 | }; | |
1122 | const struct nlmsg_flag *flag; | |
1123 | uint16_t flags_left; | |
1124 | ||
1125 | ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16, | |
1126 | h->nlmsg_len, h->nlmsg_type); | |
1127 | if (h->nlmsg_type == NLMSG_NOOP) { | |
1128 | ds_put_cstr(ds, "(no-op)"); | |
1129 | } else if (h->nlmsg_type == NLMSG_ERROR) { | |
1130 | ds_put_cstr(ds, "(error)"); | |
1131 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1132 | ds_put_cstr(ds, "(done)"); | |
1133 | } else if (h->nlmsg_type == NLMSG_OVERRUN) { | |
1134 | ds_put_cstr(ds, "(overrun)"); | |
1135 | } else if (h->nlmsg_type < NLMSG_MIN_TYPE) { | |
1136 | ds_put_cstr(ds, "(reserved)"); | |
2ad204c8 BP |
1137 | } else if (protocol == NETLINK_GENERIC) { |
1138 | ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type)); | |
2fe27d5a BP |
1139 | } else { |
1140 | ds_put_cstr(ds, "(family-defined)"); | |
1141 | } | |
1142 | ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags); | |
1143 | flags_left = h->nlmsg_flags; | |
1144 | for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) { | |
1145 | if ((flags_left & flag->bits) == flag->bits) { | |
1146 | ds_put_format(ds, "[%s]", flag->name); | |
1147 | flags_left &= ~flag->bits; | |
1148 | } | |
1149 | } | |
1150 | if (flags_left) { | |
1151 | ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left); | |
1152 | } | |
2c5a6834 BP |
1153 | ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32, |
1154 | h->nlmsg_seq, h->nlmsg_pid); | |
2fe27d5a BP |
1155 | } |
1156 | ||
1157 | static char * | |
7041c3a9 | 1158 | nlmsg_to_string(const struct ofpbuf *buffer, int protocol) |
2fe27d5a BP |
1159 | { |
1160 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1161 | const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN); | |
1162 | if (h) { | |
2ad204c8 | 1163 | nlmsghdr_to_string(h, protocol, &ds); |
2fe27d5a BP |
1164 | if (h->nlmsg_type == NLMSG_ERROR) { |
1165 | const struct nlmsgerr *e; | |
1166 | e = ofpbuf_at(buffer, NLMSG_HDRLEN, | |
1167 | NLMSG_ALIGN(sizeof(struct nlmsgerr))); | |
1168 | if (e) { | |
1169 | ds_put_format(&ds, " error(%d", e->error); | |
1170 | if (e->error < 0) { | |
10a89ef0 | 1171 | ds_put_format(&ds, "(%s)", ovs_strerror(-e->error)); |
2fe27d5a BP |
1172 | } |
1173 | ds_put_cstr(&ds, ", in-reply-to("); | |
2ad204c8 | 1174 | nlmsghdr_to_string(&e->msg, protocol, &ds); |
2fe27d5a BP |
1175 | ds_put_cstr(&ds, "))"); |
1176 | } else { | |
1177 | ds_put_cstr(&ds, " error(truncated)"); | |
1178 | } | |
1179 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1180 | int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error); | |
1181 | if (error) { | |
1182 | ds_put_format(&ds, " done(%d", *error); | |
1183 | if (*error < 0) { | |
10a89ef0 | 1184 | ds_put_format(&ds, "(%s)", ovs_strerror(-*error)); |
2fe27d5a BP |
1185 | } |
1186 | ds_put_cstr(&ds, ")"); | |
1187 | } else { | |
1188 | ds_put_cstr(&ds, " done(truncated)"); | |
1189 | } | |
7041c3a9 BP |
1190 | } else if (protocol == NETLINK_GENERIC) { |
1191 | struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer); | |
1192 | if (genl) { | |
1193 | ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")", | |
1194 | genl->cmd, genl->version); | |
1195 | } | |
2fe27d5a BP |
1196 | } |
1197 | } else { | |
1198 | ds_put_cstr(&ds, "nl(truncated)"); | |
1199 | } | |
1200 | return ds.string; | |
1201 | } | |
1202 | ||
1203 | static void | |
1204 | log_nlmsg(const char *function, int error, | |
7041c3a9 | 1205 | const void *message, size_t size, int protocol) |
2fe27d5a BP |
1206 | { |
1207 | struct ofpbuf buffer; | |
1208 | char *nlmsg; | |
1209 | ||
1210 | if (!VLOG_IS_DBG_ENABLED()) { | |
1211 | return; | |
1212 | } | |
1213 | ||
1214 | ofpbuf_use_const(&buffer, message, size); | |
7041c3a9 | 1215 | nlmsg = nlmsg_to_string(&buffer, protocol); |
10a89ef0 | 1216 | VLOG_DBG_RL(&rl, "%s (%s): %s", function, ovs_strerror(error), nlmsg); |
2fe27d5a BP |
1217 | free(nlmsg); |
1218 | } |