]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
69123704 | 2 | * Copyright (c) 2008, 2009, 2010 Nicira Networks. |
064af421 | 3 | * |
a14bc59f BP |
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: | |
064af421 | 7 | * |
a14bc59f BP |
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. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "netlink.h" | |
19 | #include <assert.h> | |
20 | #include <errno.h> | |
21 | #include <inttypes.h> | |
22 | #include <stdio.h> | |
23 | #include <stdlib.h> | |
24 | #include <string.h> | |
25 | #include <time.h> | |
26 | #include <unistd.h> | |
27 | #include "coverage.h" | |
28 | #include "dynamic-string.h" | |
29 | #include "netlink-protocol.h" | |
30 | #include "ofpbuf.h" | |
31 | #include "poll-loop.h" | |
32 | #include "timeval.h" | |
33 | #include "util.h" | |
064af421 | 34 | #include "vlog.h" |
5136ce49 BP |
35 | |
36 | VLOG_DEFINE_THIS_MODULE(netlink) | |
064af421 BP |
37 | |
38 | /* Linux header file confusion causes this to be undefined. */ | |
39 | #ifndef SOL_NETLINK | |
40 | #define SOL_NETLINK 270 | |
41 | #endif | |
42 | ||
43 | /* A single (bad) Netlink message can in theory dump out many, many log | |
44 | * messages, so the burst size is set quite high here to avoid missing useful | |
45 | * information. Also, at high logging levels we log *all* Netlink messages. */ | |
46 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600); | |
47 | ||
48 | static void log_nlmsg(const char *function, int error, | |
49 | const void *message, size_t size); | |
50 | \f | |
51 | /* Netlink sockets. */ | |
52 | ||
53 | struct nl_sock | |
54 | { | |
55 | int fd; | |
56 | uint32_t pid; | |
57 | }; | |
58 | ||
59 | /* Next nlmsghdr sequence number. | |
d295e8e9 | 60 | * |
064af421 BP |
61 | * This implementation uses sequence numbers that are unique process-wide, to |
62 | * avoid a hypothetical race: send request, close socket, open new socket that | |
63 | * reuses the old socket's PID value, send request on new socket, receive reply | |
64 | * from kernel to old socket but with same PID and sequence number. (This race | |
65 | * could be avoided other ways, e.g. by preventing PIDs from being quickly | |
66 | * reused). */ | |
67 | static uint32_t next_seq; | |
68 | ||
69 | static int alloc_pid(uint32_t *); | |
70 | static void free_pid(uint32_t); | |
71 | ||
72 | /* Creates a new netlink socket for the given netlink 'protocol' | |
73 | * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the | |
74 | * new socket if successful, otherwise returns a positive errno value. | |
75 | * | |
76 | * If 'multicast_group' is nonzero, the new socket subscribes to the specified | |
77 | * netlink multicast group. (A netlink socket may listen to an arbitrary | |
78 | * number of multicast groups, but so far we only need one at a time.) | |
79 | * | |
80 | * Nonzero 'so_sndbuf' or 'so_rcvbuf' override the kernel default send or | |
81 | * receive buffer size, respectively. | |
82 | */ | |
83 | int | |
84 | nl_sock_create(int protocol, int multicast_group, | |
85 | size_t so_sndbuf, size_t so_rcvbuf, struct nl_sock **sockp) | |
86 | { | |
87 | struct nl_sock *sock; | |
88 | struct sockaddr_nl local, remote; | |
89 | int retval = 0; | |
90 | ||
91 | if (next_seq == 0) { | |
92 | /* Pick initial sequence number. */ | |
c73814a3 | 93 | next_seq = getpid() ^ time_wall(); |
064af421 BP |
94 | } |
95 | ||
96 | *sockp = NULL; | |
97 | sock = malloc(sizeof *sock); | |
98 | if (sock == NULL) { | |
99 | return ENOMEM; | |
100 | } | |
101 | ||
102 | sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol); | |
103 | if (sock->fd < 0) { | |
104 | VLOG_ERR("fcntl: %s", strerror(errno)); | |
105 | goto error; | |
106 | } | |
107 | ||
108 | retval = alloc_pid(&sock->pid); | |
109 | if (retval) { | |
110 | goto error; | |
111 | } | |
112 | ||
113 | if (so_sndbuf != 0 | |
114 | && setsockopt(sock->fd, SOL_SOCKET, SO_SNDBUF, | |
115 | &so_sndbuf, sizeof so_sndbuf) < 0) { | |
116 | VLOG_ERR("setsockopt(SO_SNDBUF,%zu): %s", so_sndbuf, strerror(errno)); | |
117 | goto error_free_pid; | |
118 | } | |
119 | ||
120 | if (so_rcvbuf != 0 | |
121 | && setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF, | |
122 | &so_rcvbuf, sizeof so_rcvbuf) < 0) { | |
123 | VLOG_ERR("setsockopt(SO_RCVBUF,%zu): %s", so_rcvbuf, strerror(errno)); | |
124 | goto error_free_pid; | |
125 | } | |
126 | ||
127 | /* Bind local address as our selected pid. */ | |
128 | memset(&local, 0, sizeof local); | |
129 | local.nl_family = AF_NETLINK; | |
130 | local.nl_pid = sock->pid; | |
131 | if (multicast_group > 0 && multicast_group <= 32) { | |
132 | /* This method of joining multicast groups is supported by old kernels, | |
133 | * but it only allows 32 multicast groups per protocol. */ | |
134 | local.nl_groups |= 1ul << (multicast_group - 1); | |
135 | } | |
136 | if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) { | |
137 | VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno)); | |
138 | goto error_free_pid; | |
139 | } | |
140 | ||
141 | /* Bind remote address as the kernel (pid 0). */ | |
142 | memset(&remote, 0, sizeof remote); | |
143 | remote.nl_family = AF_NETLINK; | |
144 | remote.nl_pid = 0; | |
145 | if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) { | |
146 | VLOG_ERR("connect(0): %s", strerror(errno)); | |
147 | goto error_free_pid; | |
148 | } | |
149 | ||
150 | /* Older kernel headers failed to define this macro. We want our programs | |
151 | * to support the newer kernel features even if compiled with older | |
152 | * headers, so define it ourselves in such a case. */ | |
153 | #ifndef NETLINK_ADD_MEMBERSHIP | |
154 | #define NETLINK_ADD_MEMBERSHIP 1 | |
155 | #endif | |
156 | ||
157 | /* This method of joining multicast groups is only supported by newish | |
158 | * kernels, but it allows for an arbitrary number of multicast groups. */ | |
159 | if (multicast_group > 32 | |
160 | && setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, | |
161 | &multicast_group, sizeof multicast_group) < 0) { | |
162 | VLOG_ERR("setsockopt(NETLINK_ADD_MEMBERSHIP,%d): %s", | |
163 | multicast_group, strerror(errno)); | |
164 | goto error_free_pid; | |
165 | } | |
166 | ||
167 | *sockp = sock; | |
168 | return 0; | |
169 | ||
170 | error_free_pid: | |
171 | free_pid(sock->pid); | |
172 | error: | |
173 | if (retval == 0) { | |
174 | retval = errno; | |
175 | if (retval == 0) { | |
176 | retval = EINVAL; | |
177 | } | |
178 | } | |
179 | if (sock->fd >= 0) { | |
180 | close(sock->fd); | |
181 | } | |
182 | free(sock); | |
183 | return retval; | |
184 | } | |
185 | ||
186 | /* Destroys netlink socket 'sock'. */ | |
187 | void | |
d295e8e9 | 188 | nl_sock_destroy(struct nl_sock *sock) |
064af421 BP |
189 | { |
190 | if (sock) { | |
191 | close(sock->fd); | |
192 | free_pid(sock->pid); | |
193 | free(sock); | |
194 | } | |
195 | } | |
196 | ||
197 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on | |
69123704 BP |
198 | * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, and |
199 | * nlmsg_pid will be set to 'sock''s pid, before the message is sent. | |
064af421 BP |
200 | * |
201 | * Returns 0 if successful, otherwise a positive errno value. If | |
202 | * 'wait' is true, then the send will wait until buffer space is ready; | |
203 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. */ | |
204 | int | |
d295e8e9 | 205 | nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait) |
064af421 | 206 | { |
69123704 | 207 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg); |
064af421 BP |
208 | int error; |
209 | ||
69123704 BP |
210 | nlmsg->nlmsg_len = msg->size; |
211 | nlmsg->nlmsg_pid = sock->pid; | |
064af421 BP |
212 | do { |
213 | int retval; | |
214 | retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT); | |
215 | error = retval < 0 ? errno : 0; | |
216 | } while (error == EINTR); | |
217 | log_nlmsg(__func__, error, msg->data, msg->size); | |
218 | if (!error) { | |
219 | COVERAGE_INC(netlink_sent); | |
220 | } | |
221 | return error; | |
222 | } | |
223 | ||
224 | /* Tries to send the 'n_iov' chunks of data in 'iov' to the kernel on 'sock' as | |
225 | * a single Netlink message. (The message must be fully formed and not require | |
69123704 | 226 | * finalization of its nlmsg_len or nlmsg_pid fields.) |
064af421 BP |
227 | * |
228 | * Returns 0 if successful, otherwise a positive errno value. If 'wait' is | |
229 | * true, then the send will wait until buffer space is ready; otherwise, | |
230 | * returns EAGAIN if the 'sock' send buffer is full. */ | |
231 | int | |
232 | nl_sock_sendv(struct nl_sock *sock, const struct iovec iov[], size_t n_iov, | |
d295e8e9 | 233 | bool wait) |
064af421 BP |
234 | { |
235 | struct msghdr msg; | |
236 | int error; | |
237 | ||
238 | COVERAGE_INC(netlink_send); | |
239 | memset(&msg, 0, sizeof msg); | |
240 | msg.msg_iov = (struct iovec *) iov; | |
241 | msg.msg_iovlen = n_iov; | |
242 | do { | |
243 | int retval; | |
244 | retval = sendmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT); | |
245 | error = retval < 0 ? errno : 0; | |
246 | } while (error == EINTR); | |
247 | if (error != EAGAIN) { | |
248 | log_nlmsg(__func__, error, iov[0].iov_base, iov[0].iov_len); | |
249 | if (!error) { | |
250 | COVERAGE_INC(netlink_sent); | |
251 | } | |
252 | } | |
253 | return error; | |
254 | } | |
255 | ||
256 | /* Tries to receive a netlink message from the kernel on 'sock'. If | |
257 | * successful, stores the received message into '*bufp' and returns 0. The | |
258 | * caller is responsible for destroying the message with ofpbuf_delete(). On | |
259 | * failure, returns a positive errno value and stores a null pointer into | |
260 | * '*bufp'. | |
261 | * | |
262 | * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise, | |
263 | * returns EAGAIN if the 'sock' receive buffer is empty. */ | |
264 | int | |
d295e8e9 | 265 | nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait) |
064af421 BP |
266 | { |
267 | uint8_t tmp; | |
268 | ssize_t bufsize = 2048; | |
269 | ssize_t nbytes, nbytes2; | |
270 | struct ofpbuf *buf; | |
271 | struct nlmsghdr *nlmsghdr; | |
272 | struct iovec iov; | |
273 | struct msghdr msg = { | |
274 | .msg_name = NULL, | |
275 | .msg_namelen = 0, | |
276 | .msg_iov = &iov, | |
277 | .msg_iovlen = 1, | |
278 | .msg_control = NULL, | |
279 | .msg_controllen = 0, | |
280 | .msg_flags = 0 | |
281 | }; | |
282 | ||
283 | buf = ofpbuf_new(bufsize); | |
284 | *bufp = NULL; | |
285 | ||
286 | try_again: | |
287 | /* Attempt to read the message. We don't know the size of the data | |
288 | * yet, so we take a guess at 2048. If we're wrong, we keep trying | |
d295e8e9 | 289 | * and doubling the buffer size each time. |
064af421 BP |
290 | */ |
291 | nlmsghdr = ofpbuf_put_uninit(buf, bufsize); | |
292 | iov.iov_base = nlmsghdr; | |
293 | iov.iov_len = bufsize; | |
294 | do { | |
d295e8e9 | 295 | nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK); |
064af421 BP |
296 | } while (nbytes < 0 && errno == EINTR); |
297 | if (nbytes < 0) { | |
298 | ofpbuf_delete(buf); | |
299 | return errno; | |
300 | } | |
301 | if (msg.msg_flags & MSG_TRUNC) { | |
302 | COVERAGE_INC(netlink_recv_retry); | |
303 | bufsize *= 2; | |
304 | ofpbuf_reinit(buf, bufsize); | |
305 | goto try_again; | |
306 | } | |
307 | buf->size = nbytes; | |
308 | ||
309 | /* We successfully read the message, so recv again to clear the queue */ | |
310 | iov.iov_base = &tmp; | |
311 | iov.iov_len = 1; | |
312 | do { | |
313 | nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT); | |
314 | } while (nbytes2 < 0 && errno == EINTR); | |
315 | if (nbytes2 < 0) { | |
316 | if (errno == ENOBUFS) { | |
317 | /* The kernel is notifying us that a message it tried to send to us | |
318 | * was dropped. We have to pass this along to the caller in case | |
319 | * it wants to retry a request. So kill the buffer, which we can | |
320 | * re-read next time. */ | |
321 | COVERAGE_INC(netlink_overflow); | |
322 | ofpbuf_delete(buf); | |
323 | return ENOBUFS; | |
324 | } else { | |
325 | VLOG_ERR_RL(&rl, "failed to remove nlmsg from socket: %s\n", | |
326 | strerror(errno)); | |
327 | } | |
328 | } | |
329 | if (nbytes < sizeof *nlmsghdr | |
330 | || nlmsghdr->nlmsg_len < sizeof *nlmsghdr | |
331 | || nlmsghdr->nlmsg_len > nbytes) { | |
332 | VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)", | |
333 | bufsize, NLMSG_HDRLEN); | |
334 | ofpbuf_delete(buf); | |
335 | return EPROTO; | |
336 | } | |
337 | *bufp = buf; | |
338 | log_nlmsg(__func__, 0, buf->data, buf->size); | |
339 | COVERAGE_INC(netlink_received); | |
340 | return 0; | |
341 | } | |
342 | ||
343 | /* Sends 'request' to the kernel via 'sock' and waits for a response. If | |
8e2093fc BP |
344 | * successful, returns 0. On failure, returns a positive errno value. |
345 | * | |
346 | * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's | |
347 | * reply, which the caller is responsible for freeing with ofpbuf_delete(), and | |
348 | * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's | |
349 | * reply, if any, is discarded. | |
064af421 | 350 | * |
69123704 BP |
351 | * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will |
352 | * be set to 'sock''s pid, before the message is sent. NLM_F_ACK will be set | |
353 | * in nlmsg_flags. | |
354 | * | |
064af421 BP |
355 | * The caller is responsible for destroying 'request'. |
356 | * | |
357 | * Bare Netlink is an unreliable transport protocol. This function layers | |
358 | * reliable delivery and reply semantics on top of bare Netlink. | |
d295e8e9 | 359 | * |
064af421 BP |
360 | * In Netlink, sending a request to the kernel is reliable enough, because the |
361 | * kernel will tell us if the message cannot be queued (and we will in that | |
362 | * case put it on the transmit queue and wait until it can be delivered). | |
d295e8e9 | 363 | * |
064af421 BP |
364 | * Receiving the reply is the real problem: if the socket buffer is full when |
365 | * the kernel tries to send the reply, the reply will be dropped. However, the | |
366 | * kernel sets a flag that a reply has been dropped. The next call to recv | |
367 | * then returns ENOBUFS. We can then re-send the request. | |
368 | * | |
369 | * Caveats: | |
370 | * | |
371 | * 1. Netlink depends on sequence numbers to match up requests and | |
372 | * replies. The sender of a request supplies a sequence number, and | |
373 | * the reply echos back that sequence number. | |
374 | * | |
375 | * This is fine, but (1) some kernel netlink implementations are | |
376 | * broken, in that they fail to echo sequence numbers and (2) this | |
377 | * function will drop packets with non-matching sequence numbers, so | |
378 | * that only a single request can be usefully transacted at a time. | |
379 | * | |
380 | * 2. Resending the request causes it to be re-executed, so the request | |
381 | * needs to be idempotent. | |
382 | */ | |
383 | int | |
384 | nl_sock_transact(struct nl_sock *sock, | |
d295e8e9 | 385 | const struct ofpbuf *request, struct ofpbuf **replyp) |
064af421 BP |
386 | { |
387 | uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq; | |
388 | struct nlmsghdr *nlmsghdr; | |
389 | struct ofpbuf *reply; | |
390 | int retval; | |
391 | ||
8e2093fc BP |
392 | if (replyp) { |
393 | *replyp = NULL; | |
394 | } | |
064af421 BP |
395 | |
396 | /* Ensure that we get a reply even if this message doesn't ordinarily call | |
397 | * for one. */ | |
398 | nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK; | |
d295e8e9 | 399 | |
064af421 BP |
400 | send: |
401 | retval = nl_sock_send(sock, request, true); | |
402 | if (retval) { | |
403 | return retval; | |
404 | } | |
405 | ||
406 | recv: | |
407 | retval = nl_sock_recv(sock, &reply, true); | |
408 | if (retval) { | |
409 | if (retval == ENOBUFS) { | |
410 | COVERAGE_INC(netlink_overflow); | |
411 | VLOG_DBG_RL(&rl, "receive buffer overflow, resending request"); | |
412 | goto send; | |
413 | } else { | |
414 | return retval; | |
415 | } | |
416 | } | |
417 | nlmsghdr = nl_msg_nlmsghdr(reply); | |
418 | if (seq != nlmsghdr->nlmsg_seq) { | |
419 | VLOG_DBG_RL(&rl, "ignoring seq %"PRIu32" != expected %"PRIu32, | |
420 | nl_msg_nlmsghdr(reply)->nlmsg_seq, seq); | |
421 | ofpbuf_delete(reply); | |
422 | goto recv; | |
423 | } | |
a35fbea5 BP |
424 | |
425 | /* If the reply is an error, discard the reply and return the error code. | |
426 | * | |
427 | * Except: if the reply is just an acknowledgement (error code of 0), and | |
428 | * the caller is interested in the reply (replyp != NULL), pass the reply | |
429 | * up to the caller. Otherwise the caller will get a return value of 0 | |
430 | * and null '*replyp', which makes unwary callers likely to segfault. */ | |
431 | if (nl_msg_nlmsgerr(reply, &retval) && (retval || !replyp)) { | |
064af421 BP |
432 | ofpbuf_delete(reply); |
433 | if (retval) { | |
434 | VLOG_DBG_RL(&rl, "received NAK error=%d (%s)", | |
435 | retval, strerror(retval)); | |
436 | } | |
437 | return retval != EAGAIN ? retval : EPROTO; | |
438 | } | |
439 | ||
8e2093fc BP |
440 | if (replyp) { |
441 | *replyp = reply; | |
442 | } else { | |
443 | ofpbuf_delete(reply); | |
444 | } | |
064af421 BP |
445 | return 0; |
446 | } | |
447 | ||
974d6a6d BP |
448 | /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via |
449 | * 'sock', and initializes 'dump' to reflect the state of the operation. | |
450 | * | |
451 | * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will | |
452 | * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and | |
453 | * NLM_F_ACK will be set in nlmsg_flags. | |
454 | * | |
455 | * The properties of Netlink make dump operations reliable as long as all of | |
456 | * the following are true: | |
457 | * | |
458 | * - At most a single dump is in progress at a time on a given nl_sock. | |
459 | * | |
460 | * - The nl_sock is not subscribed to any multicast groups. | |
461 | * | |
462 | * - The nl_sock is not used to send any other messages before the dump | |
463 | * operation is complete. | |
464 | * | |
465 | * This function provides no status indication. An error status for the entire | |
466 | * dump operation is provided when it is completed by calling nl_dump_done(). | |
467 | * | |
468 | * The caller is responsible for destroying 'request'. The caller must not | |
469 | * close 'sock' before it completes the dump operation (by calling | |
470 | * nl_dump_done()). | |
471 | */ | |
472 | void | |
473 | nl_dump_start(struct nl_dump *dump, | |
474 | struct nl_sock *sock, const struct ofpbuf *request) | |
475 | { | |
476 | struct nlmsghdr *nlmsghdr = nl_msg_nlmsghdr(request); | |
477 | nlmsghdr->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK; | |
478 | dump->seq = nlmsghdr->nlmsg_seq; | |
479 | dump->sock = sock; | |
480 | dump->status = nl_sock_send(sock, request, true); | |
481 | dump->buffer = NULL; | |
482 | } | |
483 | ||
484 | /* Helper function for nl_dump_next(). */ | |
485 | static int | |
486 | nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp) | |
487 | { | |
488 | struct nlmsghdr *nlmsghdr; | |
489 | struct ofpbuf *buffer; | |
490 | int retval; | |
491 | ||
492 | retval = nl_sock_recv(dump->sock, bufferp, true); | |
493 | if (retval) { | |
494 | return retval == EINTR ? EAGAIN : retval; | |
495 | } | |
496 | buffer = *bufferp; | |
497 | ||
498 | nlmsghdr = nl_msg_nlmsghdr(buffer); | |
499 | if (dump->seq != nlmsghdr->nlmsg_seq) { | |
500 | VLOG_DBG_RL(&rl, "ignoring seq %"PRIu32" != expected %"PRIu32, | |
501 | nlmsghdr->nlmsg_seq, dump->seq); | |
502 | return EAGAIN; | |
503 | } | |
504 | ||
505 | if (nl_msg_nlmsgerr(buffer, &retval)) { | |
506 | VLOG_INFO_RL(&rl, "netlink dump request error (%s)", | |
507 | strerror(retval)); | |
508 | return retval && retval != EAGAIN ? retval : EPROTO; | |
509 | } | |
510 | ||
511 | return 0; | |
512 | } | |
513 | ||
514 | /* Attempts to retrieve another reply from 'dump', which must have been | |
515 | * initialized with nl_dump_start(). | |
516 | * | |
517 | * If successful, returns true and points 'reply->data' and 'reply->size' to | |
518 | * the message that was retrieved. The caller must not modify 'reply' (because | |
519 | * it points into the middle of a larger buffer). | |
520 | * | |
521 | * On failure, returns false and sets 'reply->data' to NULL and 'reply->size' | |
522 | * to 0. Failure might indicate an actual error or merely the end of replies. | |
523 | * An error status for the entire dump operation is provided when it is | |
524 | * completed by calling nl_dump_done(). | |
525 | */ | |
526 | bool | |
527 | nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply) | |
528 | { | |
529 | struct nlmsghdr *nlmsghdr; | |
530 | ||
531 | reply->data = NULL; | |
532 | reply->size = 0; | |
533 | if (dump->status) { | |
534 | return false; | |
535 | } | |
536 | ||
537 | if (dump->buffer && !dump->buffer->size) { | |
538 | ofpbuf_delete(dump->buffer); | |
539 | dump->buffer = NULL; | |
540 | } | |
541 | while (!dump->buffer) { | |
542 | int retval = nl_dump_recv(dump, &dump->buffer); | |
543 | if (retval) { | |
544 | ofpbuf_delete(dump->buffer); | |
545 | dump->buffer = NULL; | |
546 | if (retval != EAGAIN) { | |
547 | dump->status = retval; | |
548 | return false; | |
549 | } | |
550 | } | |
551 | } | |
552 | ||
553 | nlmsghdr = nl_msg_next(dump->buffer, reply); | |
554 | if (!nlmsghdr) { | |
555 | VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment"); | |
556 | dump->status = EPROTO; | |
557 | return false; | |
558 | } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) { | |
559 | dump->status = EOF; | |
560 | return false; | |
561 | } | |
562 | ||
563 | return true; | |
564 | } | |
565 | ||
566 | /* Completes Netlink dump operation 'dump', which must have been initialized | |
567 | * with nl_dump_start(). Returns 0 if the dump operation was error-free, | |
568 | * otherwise a positive errno value describing the problem. */ | |
569 | int | |
570 | nl_dump_done(struct nl_dump *dump) | |
571 | { | |
572 | /* Drain any remaining messages that the client didn't read. Otherwise the | |
573 | * kernel will continue to queue them up and waste buffer space. */ | |
574 | while (!dump->status) { | |
575 | struct ofpbuf reply; | |
576 | if (!nl_dump_next(dump, &reply)) { | |
577 | assert(dump->status); | |
578 | } | |
579 | } | |
580 | ||
581 | ofpbuf_delete(dump->buffer); | |
582 | return dump->status == EOF ? 0 : dump->status; | |
583 | } | |
584 | ||
064af421 BP |
585 | /* Causes poll_block() to wake up when any of the specified 'events' (which is |
586 | * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */ | |
587 | void | |
588 | nl_sock_wait(const struct nl_sock *sock, short int events) | |
589 | { | |
590 | poll_fd_wait(sock->fd, events); | |
591 | } | |
592 | \f | |
593 | /* Netlink messages. */ | |
594 | ||
595 | /* Returns the nlmsghdr at the head of 'msg'. | |
596 | * | |
597 | * 'msg' must be at least as large as a nlmsghdr. */ | |
598 | struct nlmsghdr * | |
d295e8e9 | 599 | nl_msg_nlmsghdr(const struct ofpbuf *msg) |
064af421 BP |
600 | { |
601 | return ofpbuf_at_assert(msg, 0, NLMSG_HDRLEN); | |
602 | } | |
603 | ||
604 | /* Returns the genlmsghdr just past 'msg''s nlmsghdr. | |
605 | * | |
606 | * Returns a null pointer if 'msg' is not large enough to contain an nlmsghdr | |
607 | * and a genlmsghdr. */ | |
608 | struct genlmsghdr * | |
d295e8e9 | 609 | nl_msg_genlmsghdr(const struct ofpbuf *msg) |
064af421 BP |
610 | { |
611 | return ofpbuf_at(msg, NLMSG_HDRLEN, GENL_HDRLEN); | |
612 | } | |
613 | ||
614 | /* If 'buffer' is a NLMSG_ERROR message, stores 0 in '*errorp' if it is an ACK | |
615 | * message, otherwise a positive errno value, and returns true. If 'buffer' is | |
616 | * not an NLMSG_ERROR message, returns false. | |
617 | * | |
618 | * 'msg' must be at least as large as a nlmsghdr. */ | |
619 | bool | |
d295e8e9 | 620 | nl_msg_nlmsgerr(const struct ofpbuf *msg, int *errorp) |
064af421 BP |
621 | { |
622 | if (nl_msg_nlmsghdr(msg)->nlmsg_type == NLMSG_ERROR) { | |
623 | struct nlmsgerr *err = ofpbuf_at(msg, NLMSG_HDRLEN, sizeof *err); | |
624 | int code = EPROTO; | |
625 | if (!err) { | |
626 | VLOG_ERR_RL(&rl, "received invalid nlmsgerr (%zd bytes < %zd)", | |
627 | msg->size, NLMSG_HDRLEN + sizeof *err); | |
628 | } else if (err->error <= 0 && err->error > INT_MIN) { | |
629 | code = -err->error; | |
630 | } | |
631 | if (errorp) { | |
632 | *errorp = code; | |
633 | } | |
634 | return true; | |
635 | } else { | |
636 | return false; | |
637 | } | |
638 | } | |
639 | ||
640 | /* Ensures that 'b' has room for at least 'size' bytes plus netlink padding at | |
641 | * its tail end, reallocating and copying its data if necessary. */ | |
642 | void | |
d295e8e9 | 643 | nl_msg_reserve(struct ofpbuf *msg, size_t size) |
064af421 BP |
644 | { |
645 | ofpbuf_prealloc_tailroom(msg, NLMSG_ALIGN(size)); | |
646 | } | |
647 | ||
648 | /* Puts a nlmsghdr at the beginning of 'msg', which must be initially empty. | |
69123704 | 649 | * Uses the given 'type' and 'flags'. 'expected_payload' should be |
064af421 BP |
650 | * an estimate of the number of payload bytes to be supplied; if the size of |
651 | * the payload is unknown a value of 0 is acceptable. | |
652 | * | |
653 | * 'type' is ordinarily an enumerated value specific to the Netlink protocol | |
654 | * (e.g. RTM_NEWLINK, for NETLINK_ROUTE protocol). For Generic Netlink, 'type' | |
655 | * is the family number obtained via nl_lookup_genl_family(). | |
656 | * | |
657 | * 'flags' is a bit-mask that indicates what kind of request is being made. It | |
658 | * is often NLM_F_REQUEST indicating that a request is being made, commonly | |
659 | * or'd with NLM_F_ACK to request an acknowledgement. | |
660 | * | |
69123704 BP |
661 | * Sets the new nlmsghdr's nlmsg_pid field to 0 for now. nl_sock_send() will |
662 | * fill it in just before sending the message. | |
663 | * | |
664 | * nl_msg_put_genlmsghdr() is more convenient for composing a Generic Netlink | |
064af421 BP |
665 | * message. */ |
666 | void | |
69123704 | 667 | nl_msg_put_nlmsghdr(struct ofpbuf *msg, |
d295e8e9 | 668 | size_t expected_payload, uint32_t type, uint32_t flags) |
064af421 BP |
669 | { |
670 | struct nlmsghdr *nlmsghdr; | |
671 | ||
672 | assert(msg->size == 0); | |
673 | ||
674 | nl_msg_reserve(msg, NLMSG_HDRLEN + expected_payload); | |
675 | nlmsghdr = nl_msg_put_uninit(msg, NLMSG_HDRLEN); | |
676 | nlmsghdr->nlmsg_len = 0; | |
677 | nlmsghdr->nlmsg_type = type; | |
678 | nlmsghdr->nlmsg_flags = flags; | |
679 | nlmsghdr->nlmsg_seq = ++next_seq; | |
69123704 | 680 | nlmsghdr->nlmsg_pid = 0; |
064af421 BP |
681 | } |
682 | ||
683 | /* Puts a nlmsghdr and genlmsghdr at the beginning of 'msg', which must be | |
69123704 BP |
684 | * initially empty. 'expected_payload' should be an estimate of the number of |
685 | * payload bytes to be supplied; if the size of the payload is unknown a value | |
686 | * of 0 is acceptable. | |
064af421 BP |
687 | * |
688 | * 'family' is the family number obtained via nl_lookup_genl_family(). | |
689 | * | |
690 | * 'flags' is a bit-mask that indicates what kind of request is being made. It | |
691 | * is often NLM_F_REQUEST indicating that a request is being made, commonly | |
692 | * or'd with NLM_F_ACK to request an acknowledgement. | |
693 | * | |
694 | * 'cmd' is an enumerated value specific to the Generic Netlink family | |
695 | * (e.g. CTRL_CMD_NEWFAMILY for the GENL_ID_CTRL family). | |
696 | * | |
697 | * 'version' is a version number specific to the family and command (often 1). | |
698 | * | |
69123704 BP |
699 | * Sets the new nlmsghdr's nlmsg_pid field to 0 for now. nl_sock_send() will |
700 | * fill it in just before sending the message. | |
701 | * | |
702 | * nl_msg_put_nlmsghdr() should be used to compose Netlink messages that are | |
703 | * not Generic Netlink messages. */ | |
064af421 | 704 | void |
69123704 BP |
705 | nl_msg_put_genlmsghdr(struct ofpbuf *msg, size_t expected_payload, |
706 | int family, uint32_t flags, uint8_t cmd, uint8_t version) | |
064af421 BP |
707 | { |
708 | struct genlmsghdr *genlmsghdr; | |
709 | ||
69123704 | 710 | nl_msg_put_nlmsghdr(msg, GENL_HDRLEN + expected_payload, family, flags); |
064af421 BP |
711 | assert(msg->size == NLMSG_HDRLEN); |
712 | genlmsghdr = nl_msg_put_uninit(msg, GENL_HDRLEN); | |
713 | genlmsghdr->cmd = cmd; | |
714 | genlmsghdr->version = version; | |
715 | genlmsghdr->reserved = 0; | |
716 | } | |
717 | ||
718 | /* Appends the 'size' bytes of data in 'p', plus Netlink padding if needed, to | |
719 | * the tail end of 'msg'. Data in 'msg' is reallocated and copied if | |
720 | * necessary. */ | |
721 | void | |
d295e8e9 | 722 | nl_msg_put(struct ofpbuf *msg, const void *data, size_t size) |
064af421 BP |
723 | { |
724 | memcpy(nl_msg_put_uninit(msg, size), data, size); | |
725 | } | |
726 | ||
727 | /* Appends 'size' bytes of data, plus Netlink padding if needed, to the tail | |
728 | * end of 'msg', reallocating and copying its data if necessary. Returns a | |
729 | * pointer to the first byte of the new data, which is left uninitialized. */ | |
730 | void * | |
d295e8e9 | 731 | nl_msg_put_uninit(struct ofpbuf *msg, size_t size) |
064af421 BP |
732 | { |
733 | size_t pad = NLMSG_ALIGN(size) - size; | |
734 | char *p = ofpbuf_put_uninit(msg, size + pad); | |
735 | if (pad) { | |
d295e8e9 | 736 | memset(p + size, 0, pad); |
064af421 BP |
737 | } |
738 | return p; | |
739 | } | |
740 | ||
741 | /* Appends a Netlink attribute of the given 'type' and room for 'size' bytes of | |
742 | * data as its payload, plus Netlink padding if needed, to the tail end of | |
743 | * 'msg', reallocating and copying its data if necessary. Returns a pointer to | |
744 | * the first byte of data in the attribute, which is left uninitialized. */ | |
745 | void * | |
d295e8e9 | 746 | nl_msg_put_unspec_uninit(struct ofpbuf *msg, uint16_t type, size_t size) |
064af421 BP |
747 | { |
748 | size_t total_size = NLA_HDRLEN + size; | |
749 | struct nlattr* nla = nl_msg_put_uninit(msg, total_size); | |
750 | assert(NLA_ALIGN(total_size) <= UINT16_MAX); | |
751 | nla->nla_len = total_size; | |
752 | nla->nla_type = type; | |
753 | return nla + 1; | |
754 | } | |
755 | ||
756 | /* Appends a Netlink attribute of the given 'type' and the 'size' bytes of | |
757 | * 'data' as its payload, to the tail end of 'msg', reallocating and copying | |
758 | * its data if necessary. Returns a pointer to the first byte of data in the | |
759 | * attribute, which is left uninitialized. */ | |
760 | void | |
761 | nl_msg_put_unspec(struct ofpbuf *msg, uint16_t type, | |
d295e8e9 | 762 | const void *data, size_t size) |
064af421 BP |
763 | { |
764 | memcpy(nl_msg_put_unspec_uninit(msg, type, size), data, size); | |
765 | } | |
766 | ||
767 | /* Appends a Netlink attribute of the given 'type' and no payload to 'msg'. | |
768 | * (Some Netlink protocols use the presence or absence of an attribute as a | |
769 | * Boolean flag.) */ | |
770 | void | |
d295e8e9 | 771 | nl_msg_put_flag(struct ofpbuf *msg, uint16_t type) |
064af421 BP |
772 | { |
773 | nl_msg_put_unspec(msg, type, NULL, 0); | |
774 | } | |
775 | ||
776 | /* Appends a Netlink attribute of the given 'type' and the given 8-bit 'value' | |
777 | * to 'msg'. */ | |
778 | void | |
d295e8e9 | 779 | nl_msg_put_u8(struct ofpbuf *msg, uint16_t type, uint8_t value) |
064af421 BP |
780 | { |
781 | nl_msg_put_unspec(msg, type, &value, sizeof value); | |
782 | } | |
783 | ||
784 | /* Appends a Netlink attribute of the given 'type' and the given 16-bit 'value' | |
785 | * to 'msg'. */ | |
786 | void | |
787 | nl_msg_put_u16(struct ofpbuf *msg, uint16_t type, uint16_t value) | |
788 | { | |
789 | nl_msg_put_unspec(msg, type, &value, sizeof value); | |
790 | } | |
791 | ||
792 | /* Appends a Netlink attribute of the given 'type' and the given 32-bit 'value' | |
793 | * to 'msg'. */ | |
794 | void | |
795 | nl_msg_put_u32(struct ofpbuf *msg, uint16_t type, uint32_t value) | |
796 | { | |
797 | nl_msg_put_unspec(msg, type, &value, sizeof value); | |
798 | } | |
799 | ||
800 | /* Appends a Netlink attribute of the given 'type' and the given 64-bit 'value' | |
801 | * to 'msg'. */ | |
802 | void | |
803 | nl_msg_put_u64(struct ofpbuf *msg, uint16_t type, uint64_t value) | |
804 | { | |
805 | nl_msg_put_unspec(msg, type, &value, sizeof value); | |
806 | } | |
807 | ||
808 | /* Appends a Netlink attribute of the given 'type' and the given | |
809 | * null-terminated string 'value' to 'msg'. */ | |
810 | void | |
811 | nl_msg_put_string(struct ofpbuf *msg, uint16_t type, const char *value) | |
812 | { | |
813 | nl_msg_put_unspec(msg, type, value, strlen(value) + 1); | |
814 | } | |
815 | ||
38a99756 BP |
816 | /* Adds the header for nested Netlink attributes to 'msg', with the specified |
817 | * 'type', and returns the header's offset within 'msg'. The caller should add | |
818 | * the content for the nested Netlink attribute to 'msg' (e.g. using the other | |
819 | * nl_msg_*() functions), and then pass the returned offset to | |
820 | * nl_msg_end_nested() to finish up the nested attributes. */ | |
821 | size_t | |
822 | nl_msg_start_nested(struct ofpbuf *msg, uint16_t type) | |
823 | { | |
824 | size_t offset = msg->size; | |
825 | nl_msg_put_unspec(msg, type, NULL, 0); | |
826 | return offset; | |
827 | } | |
828 | ||
829 | /* Finalizes a nested Netlink attribute in 'msg'. 'offset' should be the value | |
830 | * returned by nl_msg_start_nested(). */ | |
831 | void | |
832 | nl_msg_end_nested(struct ofpbuf *msg, size_t offset) | |
833 | { | |
834 | struct nlattr *attr = ofpbuf_at_assert(msg, offset, sizeof *attr); | |
835 | attr->nla_len = msg->size - offset; | |
836 | } | |
837 | ||
838 | /* Appends a nested Netlink attribute of the given 'type', with the 'size' | |
839 | * bytes of content starting at 'data', to 'msg'. */ | |
064af421 BP |
840 | void |
841 | nl_msg_put_nested(struct ofpbuf *msg, | |
38a99756 | 842 | uint16_t type, const void *data, size_t size) |
064af421 | 843 | { |
38a99756 BP |
844 | size_t offset = nl_msg_start_nested(msg, type); |
845 | nl_msg_put(msg, data, size); | |
846 | nl_msg_end_nested(msg, offset); | |
064af421 BP |
847 | } |
848 | ||
974d6a6d BP |
849 | /* If 'buffer' begins with a valid "struct nlmsghdr", pulls the header and its |
850 | * payload off 'buffer', stores header and payload in 'msg->data' and | |
851 | * 'msg->size', and returns a pointer to the header. | |
852 | * | |
853 | * If 'buffer' does not begin with a "struct nlmsghdr" or begins with one that | |
854 | * is invalid, returns NULL without modifying 'buffer'. */ | |
855 | struct nlmsghdr * | |
856 | nl_msg_next(struct ofpbuf *buffer, struct ofpbuf *msg) | |
857 | { | |
858 | if (buffer->size >= sizeof(struct nlmsghdr)) { | |
859 | struct nlmsghdr *nlmsghdr = nl_msg_nlmsghdr(buffer); | |
860 | size_t len = nlmsghdr->nlmsg_len; | |
861 | if (len >= sizeof *nlmsghdr && len <= buffer->size) { | |
862 | msg->data = nlmsghdr; | |
863 | msg->size = len; | |
864 | ofpbuf_pull(buffer, len); | |
865 | return nlmsghdr; | |
866 | } | |
867 | } | |
868 | ||
869 | msg->data = NULL; | |
870 | msg->size = 0; | |
871 | return NULL; | |
872 | } | |
873 | \f | |
874 | /* Attributes. */ | |
875 | ||
064af421 BP |
876 | /* Returns the first byte in the payload of attribute 'nla'. */ |
877 | const void * | |
d295e8e9 | 878 | nl_attr_get(const struct nlattr *nla) |
064af421 BP |
879 | { |
880 | assert(nla->nla_len >= NLA_HDRLEN); | |
881 | return nla + 1; | |
882 | } | |
883 | ||
884 | /* Returns the number of bytes in the payload of attribute 'nla'. */ | |
885 | size_t | |
d295e8e9 | 886 | nl_attr_get_size(const struct nlattr *nla) |
064af421 BP |
887 | { |
888 | assert(nla->nla_len >= NLA_HDRLEN); | |
889 | return nla->nla_len - NLA_HDRLEN; | |
890 | } | |
891 | ||
892 | /* Asserts that 'nla''s payload is at least 'size' bytes long, and returns the | |
893 | * first byte of the payload. */ | |
894 | const void * | |
d295e8e9 | 895 | nl_attr_get_unspec(const struct nlattr *nla, size_t size) |
064af421 BP |
896 | { |
897 | assert(nla->nla_len >= NLA_HDRLEN + size); | |
898 | return nla + 1; | |
899 | } | |
900 | ||
901 | /* Returns true if 'nla' is nonnull. (Some Netlink protocols use the presence | |
902 | * or absence of an attribute as a Boolean flag.) */ | |
903 | bool | |
d295e8e9 | 904 | nl_attr_get_flag(const struct nlattr *nla) |
064af421 BP |
905 | { |
906 | return nla != NULL; | |
907 | } | |
908 | ||
909 | #define NL_ATTR_GET_AS(NLA, TYPE) \ | |
910 | (*(TYPE*) nl_attr_get_unspec(nla, sizeof(TYPE))) | |
911 | ||
912 | /* Returns the 8-bit value in 'nla''s payload. | |
913 | * | |
914 | * Asserts that 'nla''s payload is at least 1 byte long. */ | |
915 | uint8_t | |
d295e8e9 | 916 | nl_attr_get_u8(const struct nlattr *nla) |
064af421 BP |
917 | { |
918 | return NL_ATTR_GET_AS(nla, uint8_t); | |
919 | } | |
920 | ||
921 | /* Returns the 16-bit value in 'nla''s payload. | |
922 | * | |
923 | * Asserts that 'nla''s payload is at least 2 bytes long. */ | |
924 | uint16_t | |
d295e8e9 | 925 | nl_attr_get_u16(const struct nlattr *nla) |
064af421 BP |
926 | { |
927 | return NL_ATTR_GET_AS(nla, uint16_t); | |
928 | } | |
929 | ||
930 | /* Returns the 32-bit value in 'nla''s payload. | |
931 | * | |
932 | * Asserts that 'nla''s payload is at least 4 bytes long. */ | |
933 | uint32_t | |
d295e8e9 | 934 | nl_attr_get_u32(const struct nlattr *nla) |
064af421 BP |
935 | { |
936 | return NL_ATTR_GET_AS(nla, uint32_t); | |
937 | } | |
938 | ||
939 | /* Returns the 64-bit value in 'nla''s payload. | |
940 | * | |
941 | * Asserts that 'nla''s payload is at least 8 bytes long. */ | |
942 | uint64_t | |
d295e8e9 | 943 | nl_attr_get_u64(const struct nlattr *nla) |
064af421 BP |
944 | { |
945 | return NL_ATTR_GET_AS(nla, uint64_t); | |
946 | } | |
947 | ||
948 | /* Returns the null-terminated string value in 'nla''s payload. | |
949 | * | |
950 | * Asserts that 'nla''s payload contains a null-terminated string. */ | |
951 | const char * | |
d295e8e9 | 952 | nl_attr_get_string(const struct nlattr *nla) |
064af421 BP |
953 | { |
954 | assert(nla->nla_len > NLA_HDRLEN); | |
955 | assert(memchr(nl_attr_get(nla), '\0', nla->nla_len - NLA_HDRLEN) != NULL); | |
956 | return nl_attr_get(nla); | |
957 | } | |
958 | ||
25eeae6a BP |
959 | /* Initializes 'nested' to the payload of 'nla'. Doesn't initialize every |
960 | * field in 'nested', but enough to poke around with it in a read-only way. */ | |
961 | void | |
962 | nl_attr_get_nested(const struct nlattr *nla, struct ofpbuf *nested) | |
963 | { | |
964 | nested->data = (void *) nl_attr_get(nla); | |
965 | nested->size = nl_attr_get_size(nla); | |
966 | } | |
967 | ||
064af421 BP |
968 | /* Default minimum and maximum payload sizes for each type of attribute. */ |
969 | static const size_t attr_len_range[][2] = { | |
970 | [0 ... N_NL_ATTR_TYPES - 1] = { 0, SIZE_MAX }, | |
971 | [NL_A_U8] = { 1, 1 }, | |
972 | [NL_A_U16] = { 2, 2 }, | |
973 | [NL_A_U32] = { 4, 4 }, | |
974 | [NL_A_U64] = { 8, 8 }, | |
975 | [NL_A_STRING] = { 1, SIZE_MAX }, | |
976 | [NL_A_FLAG] = { 0, SIZE_MAX }, | |
b4d73e97 | 977 | [NL_A_NESTED] = { 0, SIZE_MAX }, |
064af421 BP |
978 | }; |
979 | ||
980 | /* Parses the 'msg' starting at the given 'nla_offset' as a sequence of Netlink | |
981 | * attributes. 'policy[i]', for 0 <= i < n_attrs, specifies how the attribute | |
982 | * with nla_type == i is parsed; a pointer to attribute i is stored in | |
983 | * attrs[i]. Returns true if successful, false on failure. | |
984 | * | |
985 | * If the Netlink attributes in 'msg' follow a Netlink header and a Generic | |
986 | * Netlink header, then 'nla_offset' should be NLMSG_HDRLEN + GENL_HDRLEN. */ | |
987 | bool | |
988 | nl_policy_parse(const struct ofpbuf *msg, size_t nla_offset, | |
989 | const struct nl_policy policy[], | |
990 | struct nlattr *attrs[], size_t n_attrs) | |
991 | { | |
992 | void *p, *tail; | |
993 | size_t n_required; | |
994 | size_t i; | |
995 | ||
996 | n_required = 0; | |
997 | for (i = 0; i < n_attrs; i++) { | |
998 | attrs[i] = NULL; | |
999 | ||
1000 | assert(policy[i].type < N_NL_ATTR_TYPES); | |
1001 | if (policy[i].type != NL_A_NO_ATTR | |
1002 | && policy[i].type != NL_A_FLAG | |
1003 | && !policy[i].optional) { | |
1004 | n_required++; | |
1005 | } | |
1006 | } | |
1007 | ||
1008 | p = ofpbuf_at(msg, nla_offset, 0); | |
1009 | if (p == NULL) { | |
1010 | VLOG_DBG_RL(&rl, "missing headers in nl_policy_parse"); | |
1011 | return false; | |
1012 | } | |
1013 | tail = ofpbuf_tail(msg); | |
1014 | ||
1015 | while (p < tail) { | |
1016 | size_t offset = (char*)p - (char*)msg->data; | |
1017 | struct nlattr *nla = p; | |
1018 | size_t len, aligned_len; | |
1019 | uint16_t type; | |
1020 | ||
1021 | /* Make sure its claimed length is plausible. */ | |
1022 | if (nla->nla_len < NLA_HDRLEN) { | |
1023 | VLOG_DBG_RL(&rl, "%zu: attr shorter than NLA_HDRLEN (%"PRIu16")", | |
1024 | offset, nla->nla_len); | |
1025 | return false; | |
1026 | } | |
1027 | len = nla->nla_len - NLA_HDRLEN; | |
1028 | aligned_len = NLA_ALIGN(len); | |
1029 | if (aligned_len > (char*)tail - (char*)p) { | |
1030 | VLOG_DBG_RL(&rl, "%zu: attr %"PRIu16" aligned data len (%zu) " | |
1031 | "> bytes left (%tu)", | |
1032 | offset, nla->nla_type, aligned_len, | |
1033 | (char*)tail - (char*)p); | |
1034 | return false; | |
1035 | } | |
1036 | ||
1037 | type = nla->nla_type; | |
1038 | if (type < n_attrs && policy[type].type != NL_A_NO_ATTR) { | |
2a022368 | 1039 | const struct nl_policy *e = &policy[type]; |
064af421 BP |
1040 | size_t min_len, max_len; |
1041 | ||
1042 | /* Validate length and content. */ | |
2a022368 BP |
1043 | min_len = e->min_len ? e->min_len : attr_len_range[e->type][0]; |
1044 | max_len = e->max_len ? e->max_len : attr_len_range[e->type][1]; | |
064af421 BP |
1045 | if (len < min_len || len > max_len) { |
1046 | VLOG_DBG_RL(&rl, "%zu: attr %"PRIu16" length %zu not in " | |
1047 | "allowed range %zu...%zu", | |
1048 | offset, type, len, min_len, max_len); | |
1049 | return false; | |
1050 | } | |
2a022368 | 1051 | if (e->type == NL_A_STRING) { |
064af421 BP |
1052 | if (((char *) nla)[nla->nla_len - 1]) { |
1053 | VLOG_DBG_RL(&rl, "%zu: attr %"PRIu16" lacks null at end", | |
1054 | offset, type); | |
1055 | return false; | |
1056 | } | |
1057 | if (memchr(nla + 1, '\0', len - 1) != NULL) { | |
1058 | VLOG_DBG_RL(&rl, "%zu: attr %"PRIu16" has bad length", | |
1059 | offset, type); | |
1060 | return false; | |
1061 | } | |
1062 | } | |
2a022368 | 1063 | if (!e->optional && attrs[type] == NULL) { |
064af421 BP |
1064 | assert(n_required > 0); |
1065 | --n_required; | |
1066 | } | |
1067 | attrs[type] = nla; | |
1068 | } else { | |
1069 | /* Skip attribute type that we don't care about. */ | |
1070 | } | |
1071 | p = (char*)p + NLA_ALIGN(nla->nla_len); | |
1072 | } | |
1073 | if (n_required) { | |
1074 | VLOG_DBG_RL(&rl, "%zu required attrs missing", n_required); | |
1075 | return false; | |
1076 | } | |
1077 | return true; | |
1078 | } | |
25eeae6a BP |
1079 | |
1080 | /* Parses the Netlink attributes within 'nla'. 'policy[i]', for 0 <= i < | |
1081 | * n_attrs, specifies how the attribute with nla_type == i is parsed; a pointer | |
1082 | * to attribute i is stored in attrs[i]. Returns true if successful, false on | |
1083 | * failure. */ | |
1084 | bool | |
1085 | nl_parse_nested(const struct nlattr *nla, const struct nl_policy policy[], | |
1086 | struct nlattr *attrs[], size_t n_attrs) | |
1087 | { | |
1088 | struct ofpbuf buf; | |
1089 | ||
1090 | nl_attr_get_nested(nla, &buf); | |
1091 | return nl_policy_parse(&buf, 0, policy, attrs, n_attrs); | |
1092 | } | |
064af421 BP |
1093 | \f |
1094 | /* Miscellaneous. */ | |
1095 | ||
d295e8e9 | 1096 | static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = { |
064af421 BP |
1097 | [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16}, |
1098 | }; | |
1099 | ||
d295e8e9 | 1100 | static int do_lookup_genl_family(const char *name) |
064af421 BP |
1101 | { |
1102 | struct nl_sock *sock; | |
1103 | struct ofpbuf request, *reply; | |
1104 | struct nlattr *attrs[ARRAY_SIZE(family_policy)]; | |
1105 | int retval; | |
1106 | ||
1107 | retval = nl_sock_create(NETLINK_GENERIC, 0, 0, 0, &sock); | |
1108 | if (retval) { | |
1109 | return -retval; | |
1110 | } | |
1111 | ||
1112 | ofpbuf_init(&request, 0); | |
69123704 | 1113 | nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST, |
064af421 BP |
1114 | CTRL_CMD_GETFAMILY, 1); |
1115 | nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name); | |
1116 | retval = nl_sock_transact(sock, &request, &reply); | |
1117 | ofpbuf_uninit(&request); | |
1118 | if (retval) { | |
1119 | nl_sock_destroy(sock); | |
1120 | return -retval; | |
1121 | } | |
1122 | ||
1123 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN, | |
1124 | family_policy, attrs, ARRAY_SIZE(family_policy))) { | |
1125 | nl_sock_destroy(sock); | |
1126 | ofpbuf_delete(reply); | |
1127 | return -EPROTO; | |
1128 | } | |
1129 | ||
1130 | retval = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]); | |
1131 | if (retval == 0) { | |
1132 | retval = -EPROTO; | |
1133 | } | |
1134 | nl_sock_destroy(sock); | |
1135 | ofpbuf_delete(reply); | |
1136 | return retval; | |
1137 | } | |
1138 | ||
1139 | /* If '*number' is 0, translates the given Generic Netlink family 'name' to a | |
1140 | * number and stores it in '*number'. If successful, returns 0 and the caller | |
1141 | * may use '*number' as the family number. On failure, returns a positive | |
1142 | * errno value and '*number' caches the errno value. */ | |
1143 | int | |
d295e8e9 | 1144 | nl_lookup_genl_family(const char *name, int *number) |
064af421 BP |
1145 | { |
1146 | if (*number == 0) { | |
1147 | *number = do_lookup_genl_family(name); | |
1148 | assert(*number != 0); | |
1149 | } | |
1150 | return *number > 0 ? 0 : -*number; | |
1151 | } | |
1152 | \f | |
1153 | /* Netlink PID. | |
1154 | * | |
1155 | * Every Netlink socket must be bound to a unique 32-bit PID. By convention, | |
1156 | * programs that have a single Netlink socket use their Unix process ID as PID, | |
1157 | * and programs with multiple Netlink sockets add a unique per-socket | |
1158 | * identifier in the bits above the Unix process ID. | |
1159 | * | |
1160 | * The kernel has Netlink PID 0. | |
1161 | */ | |
1162 | ||
1163 | /* Parameters for how many bits in the PID should come from the Unix process ID | |
1164 | * and how many unique per-socket. */ | |
1165 | #define SOCKET_BITS 10 | |
1166 | #define MAX_SOCKETS (1u << SOCKET_BITS) | |
1167 | ||
1168 | #define PROCESS_BITS (32 - SOCKET_BITS) | |
1169 | #define MAX_PROCESSES (1u << PROCESS_BITS) | |
1170 | #define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1)) | |
1171 | ||
1172 | /* Bit vector of unused socket identifiers. */ | |
1173 | static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)]; | |
1174 | ||
1175 | /* Allocates and returns a new Netlink PID. */ | |
1176 | static int | |
1177 | alloc_pid(uint32_t *pid) | |
1178 | { | |
1179 | int i; | |
1180 | ||
1181 | for (i = 0; i < MAX_SOCKETS; i++) { | |
1182 | if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) { | |
1183 | avail_sockets[i / 32] |= 1u << (i % 32); | |
1184 | *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS); | |
1185 | return 0; | |
1186 | } | |
1187 | } | |
1188 | VLOG_ERR("netlink pid space exhausted"); | |
1189 | return ENOBUFS; | |
1190 | } | |
1191 | ||
1192 | /* Makes the specified 'pid' available for reuse. */ | |
1193 | static void | |
1194 | free_pid(uint32_t pid) | |
1195 | { | |
1196 | int sock = pid >> PROCESS_BITS; | |
1197 | assert(avail_sockets[sock / 32] & (1u << (sock % 32))); | |
1198 | avail_sockets[sock / 32] &= ~(1u << (sock % 32)); | |
1199 | } | |
1200 | \f | |
1201 | static void | |
1202 | nlmsghdr_to_string(const struct nlmsghdr *h, struct ds *ds) | |
1203 | { | |
1204 | struct nlmsg_flag { | |
1205 | unsigned int bits; | |
1206 | const char *name; | |
1207 | }; | |
d295e8e9 | 1208 | static const struct nlmsg_flag flags[] = { |
064af421 BP |
1209 | { NLM_F_REQUEST, "REQUEST" }, |
1210 | { NLM_F_MULTI, "MULTI" }, | |
1211 | { NLM_F_ACK, "ACK" }, | |
1212 | { NLM_F_ECHO, "ECHO" }, | |
1213 | { NLM_F_DUMP, "DUMP" }, | |
1214 | { NLM_F_ROOT, "ROOT" }, | |
1215 | { NLM_F_MATCH, "MATCH" }, | |
1216 | { NLM_F_ATOMIC, "ATOMIC" }, | |
1217 | }; | |
1218 | const struct nlmsg_flag *flag; | |
1219 | uint16_t flags_left; | |
1220 | ||
1221 | ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16, | |
1222 | h->nlmsg_len, h->nlmsg_type); | |
1223 | if (h->nlmsg_type == NLMSG_NOOP) { | |
1224 | ds_put_cstr(ds, "(no-op)"); | |
1225 | } else if (h->nlmsg_type == NLMSG_ERROR) { | |
1226 | ds_put_cstr(ds, "(error)"); | |
1227 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1228 | ds_put_cstr(ds, "(done)"); | |
1229 | } else if (h->nlmsg_type == NLMSG_OVERRUN) { | |
1230 | ds_put_cstr(ds, "(overrun)"); | |
1231 | } else if (h->nlmsg_type < NLMSG_MIN_TYPE) { | |
1232 | ds_put_cstr(ds, "(reserved)"); | |
1233 | } else { | |
1234 | ds_put_cstr(ds, "(family-defined)"); | |
1235 | } | |
1236 | ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags); | |
1237 | flags_left = h->nlmsg_flags; | |
1238 | for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) { | |
1239 | if ((flags_left & flag->bits) == flag->bits) { | |
1240 | ds_put_format(ds, "[%s]", flag->name); | |
1241 | flags_left &= ~flag->bits; | |
1242 | } | |
1243 | } | |
1244 | if (flags_left) { | |
1245 | ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left); | |
1246 | } | |
1247 | ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32"(%d:%d))", | |
1248 | h->nlmsg_seq, h->nlmsg_pid, | |
1249 | (int) (h->nlmsg_pid & PROCESS_MASK), | |
1250 | (int) (h->nlmsg_pid >> PROCESS_BITS)); | |
1251 | } | |
1252 | ||
1253 | static char * | |
1254 | nlmsg_to_string(const struct ofpbuf *buffer) | |
1255 | { | |
1256 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1257 | const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN); | |
1258 | if (h) { | |
1259 | nlmsghdr_to_string(h, &ds); | |
1260 | if (h->nlmsg_type == NLMSG_ERROR) { | |
1261 | const struct nlmsgerr *e; | |
1262 | e = ofpbuf_at(buffer, NLMSG_HDRLEN, | |
1263 | NLMSG_ALIGN(sizeof(struct nlmsgerr))); | |
1264 | if (e) { | |
1265 | ds_put_format(&ds, " error(%d", e->error); | |
1266 | if (e->error < 0) { | |
1267 | ds_put_format(&ds, "(%s)", strerror(-e->error)); | |
1268 | } | |
1269 | ds_put_cstr(&ds, ", in-reply-to("); | |
1270 | nlmsghdr_to_string(&e->msg, &ds); | |
1271 | ds_put_cstr(&ds, "))"); | |
1272 | } else { | |
1273 | ds_put_cstr(&ds, " error(truncated)"); | |
1274 | } | |
1275 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
1276 | int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error); | |
1277 | if (error) { | |
1278 | ds_put_format(&ds, " done(%d", *error); | |
1279 | if (*error < 0) { | |
1280 | ds_put_format(&ds, "(%s)", strerror(-*error)); | |
1281 | } | |
1282 | ds_put_cstr(&ds, ")"); | |
1283 | } else { | |
1284 | ds_put_cstr(&ds, " done(truncated)"); | |
1285 | } | |
1286 | } | |
1287 | } else { | |
1288 | ds_put_cstr(&ds, "nl(truncated)"); | |
1289 | } | |
1290 | return ds.string; | |
1291 | } | |
1292 | ||
1293 | static void | |
1294 | log_nlmsg(const char *function, int error, | |
1295 | const void *message, size_t size) | |
1296 | { | |
1297 | struct ofpbuf buffer; | |
1298 | char *nlmsg; | |
1299 | ||
1300 | if (!VLOG_IS_DBG_ENABLED()) { | |
1301 | return; | |
1302 | } | |
1303 | ||
1304 | buffer.data = (void *) message; | |
1305 | buffer.size = size; | |
1306 | nlmsg = nlmsg_to_string(&buffer); | |
1307 | VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg); | |
1308 | free(nlmsg); | |
1309 | } | |
1310 |