]>
Commit | Line | Data |
---|---|---|
2fe27d5a | 1 | /* |
cceb11f5 | 2 | * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. |
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" | |
19 | #include <assert.h> | |
20 | #include <errno.h> | |
21 | #include <inttypes.h> | |
22 | #include <stdlib.h> | |
23 | #include <sys/types.h> | |
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" | |
32 | #include "poll-loop.h" | |
6b7c12fd | 33 | #include "socket-util.h" |
2fe27d5a BP |
34 | #include "stress.h" |
35 | #include "vlog.h" | |
36 | ||
37 | VLOG_DEFINE_THIS_MODULE(netlink_socket); | |
38 | ||
39 | COVERAGE_DEFINE(netlink_overflow); | |
40 | COVERAGE_DEFINE(netlink_received); | |
41 | COVERAGE_DEFINE(netlink_recv_retry); | |
42 | COVERAGE_DEFINE(netlink_send); | |
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 | ||
55 | static void log_nlmsg(const char *function, int error, | |
7041c3a9 | 56 | const void *message, size_t size, int protocol); |
2fe27d5a BP |
57 | \f |
58 | /* Netlink sockets. */ | |
59 | ||
60 | struct nl_sock | |
61 | { | |
62 | int fd; | |
63 | uint32_t pid; | |
7041c3a9 | 64 | int protocol; |
c6eab56d BP |
65 | bool any_groups; |
66 | struct nl_dump *dump; | |
2fe27d5a BP |
67 | }; |
68 | ||
69 | static int alloc_pid(uint32_t *); | |
70 | static void free_pid(uint32_t); | |
c6eab56d | 71 | static int nl_sock_cow__(struct nl_sock *); |
2fe27d5a BP |
72 | |
73 | /* Creates a new netlink socket for the given netlink 'protocol' | |
74 | * (NETLINK_ROUTE, NETLINK_GENERIC, ...). Returns 0 and sets '*sockp' to the | |
cceb11f5 | 75 | * new socket if successful, otherwise returns a positive errno value. */ |
2fe27d5a | 76 | int |
cceb11f5 | 77 | nl_sock_create(int protocol, struct nl_sock **sockp) |
2fe27d5a BP |
78 | { |
79 | struct nl_sock *sock; | |
80 | struct sockaddr_nl local, remote; | |
81 | int retval = 0; | |
82 | ||
83 | *sockp = NULL; | |
84 | sock = malloc(sizeof *sock); | |
85 | if (sock == NULL) { | |
86 | return ENOMEM; | |
87 | } | |
88 | ||
89 | sock->fd = socket(AF_NETLINK, SOCK_RAW, protocol); | |
90 | if (sock->fd < 0) { | |
91 | VLOG_ERR("fcntl: %s", strerror(errno)); | |
92 | goto error; | |
93 | } | |
7041c3a9 | 94 | sock->protocol = protocol; |
c6eab56d | 95 | sock->any_groups = false; |
2ad204c8 | 96 | sock->dump = NULL; |
2fe27d5a BP |
97 | |
98 | retval = alloc_pid(&sock->pid); | |
99 | if (retval) { | |
100 | goto error; | |
101 | } | |
102 | ||
2fe27d5a BP |
103 | /* Bind local address as our selected pid. */ |
104 | memset(&local, 0, sizeof local); | |
105 | local.nl_family = AF_NETLINK; | |
106 | local.nl_pid = sock->pid; | |
2fe27d5a BP |
107 | if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) { |
108 | VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno)); | |
109 | goto error_free_pid; | |
110 | } | |
111 | ||
112 | /* Bind remote address as the kernel (pid 0). */ | |
113 | memset(&remote, 0, sizeof remote); | |
114 | remote.nl_family = AF_NETLINK; | |
115 | remote.nl_pid = 0; | |
116 | if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) { | |
117 | VLOG_ERR("connect(0): %s", strerror(errno)); | |
118 | goto error_free_pid; | |
119 | } | |
120 | ||
2fe27d5a BP |
121 | *sockp = sock; |
122 | return 0; | |
123 | ||
124 | error_free_pid: | |
125 | free_pid(sock->pid); | |
126 | error: | |
127 | if (retval == 0) { | |
128 | retval = errno; | |
129 | if (retval == 0) { | |
130 | retval = EINVAL; | |
131 | } | |
132 | } | |
133 | if (sock->fd >= 0) { | |
134 | close(sock->fd); | |
135 | } | |
136 | free(sock); | |
137 | return retval; | |
138 | } | |
139 | ||
c6eab56d BP |
140 | /* Creates a new netlink socket for the same protocol as 'src'. Returns 0 and |
141 | * sets '*sockp' to the new socket if successful, otherwise returns a positive | |
142 | * errno value. */ | |
143 | int | |
144 | nl_sock_clone(const struct nl_sock *src, struct nl_sock **sockp) | |
145 | { | |
146 | return nl_sock_create(src->protocol, sockp); | |
147 | } | |
148 | ||
2fe27d5a BP |
149 | /* Destroys netlink socket 'sock'. */ |
150 | void | |
151 | nl_sock_destroy(struct nl_sock *sock) | |
152 | { | |
153 | if (sock) { | |
c6eab56d BP |
154 | if (sock->dump) { |
155 | sock->dump = NULL; | |
156 | } else { | |
157 | close(sock->fd); | |
158 | free_pid(sock->pid); | |
159 | free(sock); | |
160 | } | |
2fe27d5a BP |
161 | } |
162 | } | |
163 | ||
cceb11f5 BP |
164 | /* Tries to add 'sock' as a listener for 'multicast_group'. Returns 0 if |
165 | * successful, otherwise a positive errno value. | |
166 | * | |
167 | * Multicast group numbers are always positive. | |
168 | * | |
169 | * It is not an error to attempt to join a multicast group to which a socket | |
170 | * already belongs. */ | |
171 | int | |
172 | nl_sock_join_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
173 | { | |
c6eab56d BP |
174 | int error = nl_sock_cow__(sock); |
175 | if (error) { | |
176 | return error; | |
177 | } | |
cceb11f5 BP |
178 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, |
179 | &multicast_group, sizeof multicast_group) < 0) { | |
180 | VLOG_WARN("could not join multicast group %u (%s)", | |
181 | multicast_group, strerror(errno)); | |
182 | return errno; | |
183 | } | |
c6eab56d | 184 | sock->any_groups = true; |
cceb11f5 BP |
185 | return 0; |
186 | } | |
187 | ||
188 | /* Tries to make 'sock' stop listening to 'multicast_group'. Returns 0 if | |
189 | * successful, otherwise a positive errno value. | |
190 | * | |
191 | * Multicast group numbers are always positive. | |
192 | * | |
193 | * It is not an error to attempt to leave a multicast group to which a socket | |
194 | * does not belong. | |
195 | * | |
196 | * On success, reading from 'sock' will still return any messages that were | |
197 | * received on 'multicast_group' before the group was left. */ | |
198 | int | |
199 | nl_sock_leave_mcgroup(struct nl_sock *sock, unsigned int multicast_group) | |
200 | { | |
c6eab56d | 201 | assert(!sock->dump); |
cceb11f5 BP |
202 | if (setsockopt(sock->fd, SOL_NETLINK, NETLINK_DROP_MEMBERSHIP, |
203 | &multicast_group, sizeof multicast_group) < 0) { | |
204 | VLOG_WARN("could not leave multicast group %u (%s)", | |
205 | multicast_group, strerror(errno)); | |
206 | return errno; | |
207 | } | |
208 | return 0; | |
209 | } | |
210 | ||
c6eab56d BP |
211 | static int |
212 | nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, bool wait) | |
2fe27d5a BP |
213 | { |
214 | struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg); | |
215 | int error; | |
216 | ||
217 | nlmsg->nlmsg_len = msg->size; | |
218 | nlmsg->nlmsg_pid = sock->pid; | |
219 | do { | |
220 | int retval; | |
221 | retval = send(sock->fd, msg->data, msg->size, wait ? 0 : MSG_DONTWAIT); | |
222 | error = retval < 0 ? errno : 0; | |
223 | } while (error == EINTR); | |
7041c3a9 | 224 | log_nlmsg(__func__, error, msg->data, msg->size, sock->protocol); |
2fe27d5a BP |
225 | if (!error) { |
226 | COVERAGE_INC(netlink_sent); | |
227 | } | |
228 | return error; | |
229 | } | |
230 | ||
c6eab56d BP |
231 | /* Tries to send 'msg', which must contain a Netlink message, to the kernel on |
232 | * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, and | |
233 | * nlmsg_pid will be set to 'sock''s pid, before the message is sent. | |
234 | * | |
235 | * Returns 0 if successful, otherwise a positive errno value. If | |
236 | * 'wait' is true, then the send will wait until buffer space is ready; | |
237 | * otherwise, returns EAGAIN if the 'sock' send buffer is full. */ | |
238 | int | |
239 | nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait) | |
240 | { | |
241 | int error = nl_sock_cow__(sock); | |
242 | if (error) { | |
243 | return error; | |
244 | } | |
245 | return nl_sock_send__(sock, msg, wait); | |
246 | } | |
247 | ||
2fe27d5a BP |
248 | /* This stress option is useful for testing that OVS properly tolerates |
249 | * -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can | |
250 | * occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to | |
251 | * reply to a request. They can also occur if messages arrive on a multicast | |
252 | * channel faster than OVS can process them. */ | |
253 | STRESS_OPTION( | |
254 | netlink_overflow, "simulate netlink socket receive buffer overflow", | |
255 | 5, 1, -1, 100); | |
256 | ||
c6eab56d BP |
257 | static int |
258 | nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait) | |
2fe27d5a BP |
259 | { |
260 | uint8_t tmp; | |
261 | ssize_t bufsize = 2048; | |
262 | ssize_t nbytes, nbytes2; | |
263 | struct ofpbuf *buf; | |
264 | struct nlmsghdr *nlmsghdr; | |
265 | struct iovec iov; | |
266 | struct msghdr msg = { | |
267 | .msg_name = NULL, | |
268 | .msg_namelen = 0, | |
269 | .msg_iov = &iov, | |
270 | .msg_iovlen = 1, | |
271 | .msg_control = NULL, | |
272 | .msg_controllen = 0, | |
273 | .msg_flags = 0 | |
274 | }; | |
275 | ||
276 | buf = ofpbuf_new(bufsize); | |
277 | *bufp = NULL; | |
278 | ||
279 | try_again: | |
280 | /* Attempt to read the message. We don't know the size of the data | |
281 | * yet, so we take a guess at 2048. If we're wrong, we keep trying | |
282 | * and doubling the buffer size each time. | |
283 | */ | |
284 | nlmsghdr = ofpbuf_put_uninit(buf, bufsize); | |
285 | iov.iov_base = nlmsghdr; | |
286 | iov.iov_len = bufsize; | |
287 | do { | |
288 | nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK); | |
289 | } while (nbytes < 0 && errno == EINTR); | |
290 | if (nbytes < 0) { | |
291 | ofpbuf_delete(buf); | |
292 | return errno; | |
293 | } | |
294 | if (msg.msg_flags & MSG_TRUNC) { | |
295 | COVERAGE_INC(netlink_recv_retry); | |
296 | bufsize *= 2; | |
297 | ofpbuf_reinit(buf, bufsize); | |
298 | goto try_again; | |
299 | } | |
300 | buf->size = nbytes; | |
301 | ||
302 | /* We successfully read the message, so recv again to clear the queue */ | |
303 | iov.iov_base = &tmp; | |
304 | iov.iov_len = 1; | |
305 | do { | |
306 | nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT); | |
307 | } while (nbytes2 < 0 && errno == EINTR); | |
308 | if (nbytes2 < 0) { | |
309 | if (errno == ENOBUFS) { | |
310 | /* The kernel is notifying us that a message it tried to send to us | |
311 | * was dropped. We have to pass this along to the caller in case | |
312 | * it wants to retry a request. So kill the buffer, which we can | |
313 | * re-read next time. */ | |
314 | COVERAGE_INC(netlink_overflow); | |
315 | ofpbuf_delete(buf); | |
316 | return ENOBUFS; | |
317 | } else { | |
318 | VLOG_ERR_RL(&rl, "failed to remove nlmsg from socket: %s\n", | |
319 | strerror(errno)); | |
320 | } | |
321 | } | |
322 | if (nbytes < sizeof *nlmsghdr | |
323 | || nlmsghdr->nlmsg_len < sizeof *nlmsghdr | |
324 | || nlmsghdr->nlmsg_len > nbytes) { | |
325 | VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)", | |
326 | bufsize, NLMSG_HDRLEN); | |
327 | ofpbuf_delete(buf); | |
328 | return EPROTO; | |
329 | } | |
330 | ||
331 | if (STRESS(netlink_overflow)) { | |
332 | ofpbuf_delete(buf); | |
333 | return ENOBUFS; | |
334 | } | |
335 | ||
336 | *bufp = buf; | |
7041c3a9 | 337 | log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol); |
2fe27d5a BP |
338 | COVERAGE_INC(netlink_received); |
339 | ||
340 | return 0; | |
341 | } | |
342 | ||
c6eab56d BP |
343 | /* Tries to receive a netlink message from the kernel on 'sock'. If |
344 | * successful, stores the received message into '*bufp' and returns 0. The | |
345 | * caller is responsible for destroying the message with ofpbuf_delete(). On | |
346 | * failure, returns a positive errno value and stores a null pointer into | |
347 | * '*bufp'. | |
348 | * | |
349 | * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise, | |
350 | * returns EAGAIN if the 'sock' receive buffer is empty. */ | |
351 | int | |
352 | nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait) | |
353 | { | |
354 | int error = nl_sock_cow__(sock); | |
355 | if (error) { | |
356 | return error; | |
357 | } | |
358 | return nl_sock_recv__(sock, bufp, wait); | |
359 | } | |
360 | ||
2fe27d5a BP |
361 | /* Sends 'request' to the kernel via 'sock' and waits for a response. If |
362 | * successful, returns 0. On failure, returns a positive errno value. | |
363 | * | |
364 | * If 'replyp' is nonnull, then on success '*replyp' is set to the kernel's | |
365 | * reply, which the caller is responsible for freeing with ofpbuf_delete(), and | |
366 | * on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's | |
367 | * reply, if any, is discarded. | |
368 | * | |
369 | * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will | |
370 | * be set to 'sock''s pid, before the message is sent. NLM_F_ACK will be set | |
371 | * in nlmsg_flags. | |
372 | * | |
373 | * The caller is responsible for destroying 'request'. | |
374 | * | |
375 | * Bare Netlink is an unreliable transport protocol. This function layers | |
376 | * reliable delivery and reply semantics on top of bare Netlink. | |
377 | * | |
378 | * In Netlink, sending a request to the kernel is reliable enough, because the | |
379 | * kernel will tell us if the message cannot be queued (and we will in that | |
380 | * case put it on the transmit queue and wait until it can be delivered). | |
381 | * | |
382 | * Receiving the reply is the real problem: if the socket buffer is full when | |
383 | * the kernel tries to send the reply, the reply will be dropped. However, the | |
384 | * kernel sets a flag that a reply has been dropped. The next call to recv | |
385 | * then returns ENOBUFS. We can then re-send the request. | |
386 | * | |
387 | * Caveats: | |
388 | * | |
389 | * 1. Netlink depends on sequence numbers to match up requests and | |
390 | * replies. The sender of a request supplies a sequence number, and | |
391 | * the reply echos back that sequence number. | |
392 | * | |
393 | * This is fine, but (1) some kernel netlink implementations are | |
394 | * broken, in that they fail to echo sequence numbers and (2) this | |
395 | * function will drop packets with non-matching sequence numbers, so | |
396 | * that only a single request can be usefully transacted at a time. | |
397 | * | |
398 | * 2. Resending the request causes it to be re-executed, so the request | |
399 | * needs to be idempotent. | |
400 | */ | |
401 | int | |
402 | nl_sock_transact(struct nl_sock *sock, | |
403 | const struct ofpbuf *request, struct ofpbuf **replyp) | |
404 | { | |
405 | uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq; | |
406 | struct nlmsghdr *nlmsghdr; | |
407 | struct ofpbuf *reply; | |
408 | int retval; | |
409 | ||
410 | if (replyp) { | |
411 | *replyp = NULL; | |
412 | } | |
413 | ||
414 | /* Ensure that we get a reply even if this message doesn't ordinarily call | |
415 | * for one. */ | |
416 | nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK; | |
417 | ||
418 | send: | |
419 | retval = nl_sock_send(sock, request, true); | |
420 | if (retval) { | |
421 | return retval; | |
422 | } | |
423 | ||
424 | recv: | |
425 | retval = nl_sock_recv(sock, &reply, true); | |
426 | if (retval) { | |
427 | if (retval == ENOBUFS) { | |
428 | COVERAGE_INC(netlink_overflow); | |
429 | VLOG_DBG_RL(&rl, "receive buffer overflow, resending request"); | |
430 | goto send; | |
431 | } else { | |
432 | return retval; | |
433 | } | |
434 | } | |
435 | nlmsghdr = nl_msg_nlmsghdr(reply); | |
436 | if (seq != nlmsghdr->nlmsg_seq) { | |
727ef33f | 437 | VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32, |
2fe27d5a BP |
438 | nl_msg_nlmsghdr(reply)->nlmsg_seq, seq); |
439 | ofpbuf_delete(reply); | |
440 | goto recv; | |
441 | } | |
442 | ||
443 | /* If the reply is an error, discard the reply and return the error code. | |
444 | * | |
445 | * Except: if the reply is just an acknowledgement (error code of 0), and | |
446 | * the caller is interested in the reply (replyp != NULL), pass the reply | |
447 | * up to the caller. Otherwise the caller will get a return value of 0 | |
448 | * and null '*replyp', which makes unwary callers likely to segfault. */ | |
449 | if (nl_msg_nlmsgerr(reply, &retval) && (retval || !replyp)) { | |
450 | ofpbuf_delete(reply); | |
451 | if (retval) { | |
452 | VLOG_DBG_RL(&rl, "received NAK error=%d (%s)", | |
453 | retval, strerror(retval)); | |
454 | } | |
455 | return retval != EAGAIN ? retval : EPROTO; | |
456 | } | |
457 | ||
458 | if (replyp) { | |
459 | *replyp = reply; | |
460 | } else { | |
461 | ofpbuf_delete(reply); | |
462 | } | |
463 | return 0; | |
464 | } | |
465 | ||
6b7c12fd BP |
466 | /* Drain all the messages currently in 'sock''s receive queue. */ |
467 | int | |
468 | nl_sock_drain(struct nl_sock *sock) | |
469 | { | |
c6eab56d BP |
470 | int error = nl_sock_cow__(sock); |
471 | if (error) { | |
472 | return error; | |
473 | } | |
6b7c12fd BP |
474 | return drain_rcvbuf(sock->fd); |
475 | } | |
476 | ||
c6eab56d BP |
477 | /* The client is attempting some operation on 'sock'. If 'sock' has an ongoing |
478 | * dump operation, then replace 'sock''s fd with a new socket and hand 'sock''s | |
479 | * old fd over to the dump. */ | |
480 | static int | |
481 | nl_sock_cow__(struct nl_sock *sock) | |
482 | { | |
483 | struct nl_sock *copy; | |
484 | uint32_t tmp_pid; | |
485 | int tmp_fd; | |
486 | int error; | |
487 | ||
488 | if (!sock->dump) { | |
489 | return 0; | |
490 | } | |
491 | ||
492 | error = nl_sock_clone(sock, ©); | |
493 | if (error) { | |
494 | return error; | |
495 | } | |
496 | ||
497 | tmp_fd = sock->fd; | |
498 | sock->fd = copy->fd; | |
499 | copy->fd = tmp_fd; | |
500 | ||
501 | tmp_pid = sock->pid; | |
502 | sock->pid = copy->pid; | |
503 | copy->pid = tmp_pid; | |
504 | ||
505 | sock->dump->sock = copy; | |
506 | sock->dump = NULL; | |
507 | ||
508 | return 0; | |
509 | } | |
510 | ||
2fe27d5a BP |
511 | /* Starts a Netlink "dump" operation, by sending 'request' to the kernel via |
512 | * 'sock', and initializes 'dump' to reflect the state of the operation. | |
513 | * | |
514 | * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will | |
515 | * be set to 'sock''s pid, before the message is sent. NLM_F_DUMP and | |
516 | * NLM_F_ACK will be set in nlmsg_flags. | |
517 | * | |
c6eab56d BP |
518 | * This Netlink socket library is designed to ensure that the dump is reliable |
519 | * and that it will not interfere with other operations on 'sock', including | |
520 | * destroying or sending and receiving messages on 'sock'. One corner case is | |
521 | * not handled: | |
2fe27d5a | 522 | * |
c6eab56d BP |
523 | * - If 'sock' has been used to send a request (e.g. with nl_sock_send()) |
524 | * whose response has not yet been received (e.g. with nl_sock_recv()). | |
525 | * This is unusual: usually nl_sock_transact() is used to send a message | |
526 | * and receive its reply all in one go. | |
2fe27d5a BP |
527 | * |
528 | * This function provides no status indication. An error status for the entire | |
529 | * dump operation is provided when it is completed by calling nl_dump_done(). | |
530 | * | |
c6eab56d BP |
531 | * The caller is responsible for destroying 'request'. |
532 | * | |
533 | * The new 'dump' is independent of 'sock'. 'sock' and 'dump' may be destroyed | |
534 | * in either order. | |
2fe27d5a BP |
535 | */ |
536 | void | |
537 | nl_dump_start(struct nl_dump *dump, | |
538 | struct nl_sock *sock, const struct ofpbuf *request) | |
539 | { | |
540 | struct nlmsghdr *nlmsghdr = nl_msg_nlmsghdr(request); | |
541 | nlmsghdr->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK; | |
542 | dump->seq = nlmsghdr->nlmsg_seq; | |
2fe27d5a | 543 | dump->buffer = NULL; |
c6eab56d BP |
544 | if (sock->any_groups || sock->dump) { |
545 | /* 'sock' might belong to some multicast group, or it already has an | |
546 | * onoging dump. Clone the socket to avoid possibly intermixing | |
547 | * multicast messages or previous dump results with our results. */ | |
548 | dump->status = nl_sock_clone(sock, &dump->sock); | |
549 | if (dump->status) { | |
550 | return; | |
551 | } | |
552 | } else { | |
553 | sock->dump = dump; | |
554 | dump->sock = sock; | |
555 | dump->status = 0; | |
556 | } | |
557 | dump->status = nl_sock_send__(sock, request, true); | |
2fe27d5a BP |
558 | } |
559 | ||
560 | /* Helper function for nl_dump_next(). */ | |
561 | static int | |
562 | nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp) | |
563 | { | |
564 | struct nlmsghdr *nlmsghdr; | |
565 | struct ofpbuf *buffer; | |
566 | int retval; | |
567 | ||
c6eab56d | 568 | retval = nl_sock_recv__(dump->sock, bufferp, true); |
2fe27d5a BP |
569 | if (retval) { |
570 | return retval == EINTR ? EAGAIN : retval; | |
571 | } | |
572 | buffer = *bufferp; | |
573 | ||
574 | nlmsghdr = nl_msg_nlmsghdr(buffer); | |
575 | if (dump->seq != nlmsghdr->nlmsg_seq) { | |
727ef33f | 576 | VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32, |
2fe27d5a BP |
577 | nlmsghdr->nlmsg_seq, dump->seq); |
578 | return EAGAIN; | |
579 | } | |
580 | ||
581 | if (nl_msg_nlmsgerr(buffer, &retval)) { | |
582 | VLOG_INFO_RL(&rl, "netlink dump request error (%s)", | |
583 | strerror(retval)); | |
584 | return retval && retval != EAGAIN ? retval : EPROTO; | |
585 | } | |
586 | ||
587 | return 0; | |
588 | } | |
589 | ||
590 | /* Attempts to retrieve another reply from 'dump', which must have been | |
591 | * initialized with nl_dump_start(). | |
592 | * | |
593 | * If successful, returns true and points 'reply->data' and 'reply->size' to | |
594 | * the message that was retrieved. The caller must not modify 'reply' (because | |
595 | * it points into the middle of a larger buffer). | |
596 | * | |
597 | * On failure, returns false and sets 'reply->data' to NULL and 'reply->size' | |
598 | * to 0. Failure might indicate an actual error or merely the end of replies. | |
599 | * An error status for the entire dump operation is provided when it is | |
600 | * completed by calling nl_dump_done(). | |
601 | */ | |
602 | bool | |
603 | nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply) | |
604 | { | |
605 | struct nlmsghdr *nlmsghdr; | |
606 | ||
607 | reply->data = NULL; | |
608 | reply->size = 0; | |
609 | if (dump->status) { | |
610 | return false; | |
611 | } | |
612 | ||
613 | if (dump->buffer && !dump->buffer->size) { | |
614 | ofpbuf_delete(dump->buffer); | |
615 | dump->buffer = NULL; | |
616 | } | |
617 | while (!dump->buffer) { | |
618 | int retval = nl_dump_recv(dump, &dump->buffer); | |
619 | if (retval) { | |
620 | ofpbuf_delete(dump->buffer); | |
621 | dump->buffer = NULL; | |
622 | if (retval != EAGAIN) { | |
623 | dump->status = retval; | |
624 | return false; | |
625 | } | |
626 | } | |
627 | } | |
628 | ||
629 | nlmsghdr = nl_msg_next(dump->buffer, reply); | |
630 | if (!nlmsghdr) { | |
631 | VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment"); | |
632 | dump->status = EPROTO; | |
633 | return false; | |
634 | } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) { | |
635 | dump->status = EOF; | |
636 | return false; | |
637 | } | |
638 | ||
639 | return true; | |
640 | } | |
641 | ||
642 | /* Completes Netlink dump operation 'dump', which must have been initialized | |
643 | * with nl_dump_start(). Returns 0 if the dump operation was error-free, | |
644 | * otherwise a positive errno value describing the problem. */ | |
645 | int | |
646 | nl_dump_done(struct nl_dump *dump) | |
647 | { | |
648 | /* Drain any remaining messages that the client didn't read. Otherwise the | |
649 | * kernel will continue to queue them up and waste buffer space. */ | |
650 | while (!dump->status) { | |
651 | struct ofpbuf reply; | |
652 | if (!nl_dump_next(dump, &reply)) { | |
653 | assert(dump->status); | |
654 | } | |
655 | } | |
656 | ||
c6eab56d BP |
657 | if (dump->sock) { |
658 | if (dump->sock->dump) { | |
659 | dump->sock->dump = NULL; | |
660 | } else { | |
661 | nl_sock_destroy(dump->sock); | |
662 | } | |
663 | } | |
2fe27d5a BP |
664 | ofpbuf_delete(dump->buffer); |
665 | return dump->status == EOF ? 0 : dump->status; | |
666 | } | |
667 | ||
668 | /* Causes poll_block() to wake up when any of the specified 'events' (which is | |
669 | * a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */ | |
670 | void | |
671 | nl_sock_wait(const struct nl_sock *sock, short int events) | |
672 | { | |
673 | poll_fd_wait(sock->fd, events); | |
674 | } | |
675 | \f | |
676 | /* Miscellaneous. */ | |
677 | ||
2ad204c8 BP |
678 | struct genl_family { |
679 | struct hmap_node hmap_node; | |
680 | uint16_t id; | |
681 | char *name; | |
682 | }; | |
683 | ||
684 | static struct hmap genl_families = HMAP_INITIALIZER(&genl_families); | |
685 | ||
2fe27d5a BP |
686 | static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = { |
687 | [CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16}, | |
688 | }; | |
689 | ||
2ad204c8 BP |
690 | static struct genl_family * |
691 | find_genl_family_by_id(uint16_t id) | |
692 | { | |
693 | struct genl_family *family; | |
694 | ||
695 | HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0), | |
696 | &genl_families) { | |
697 | if (family->id == id) { | |
698 | return family; | |
699 | } | |
700 | } | |
701 | return NULL; | |
702 | } | |
703 | ||
704 | static void | |
705 | define_genl_family(uint16_t id, const char *name) | |
706 | { | |
707 | struct genl_family *family = find_genl_family_by_id(id); | |
708 | ||
709 | if (family) { | |
710 | if (!strcmp(family->name, name)) { | |
711 | return; | |
712 | } | |
713 | free(family->name); | |
714 | } else { | |
715 | family = xmalloc(sizeof *family); | |
716 | family->id = id; | |
717 | hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0)); | |
718 | } | |
719 | family->name = xstrdup(name); | |
720 | } | |
721 | ||
722 | static const char * | |
723 | genl_family_to_name(uint16_t id) | |
724 | { | |
725 | if (id == GENL_ID_CTRL) { | |
726 | return "control"; | |
727 | } else { | |
728 | struct genl_family *family = find_genl_family_by_id(id); | |
729 | return family ? family->name : "unknown"; | |
730 | } | |
731 | } | |
732 | ||
2fe27d5a BP |
733 | static int do_lookup_genl_family(const char *name) |
734 | { | |
735 | struct nl_sock *sock; | |
736 | struct ofpbuf request, *reply; | |
737 | struct nlattr *attrs[ARRAY_SIZE(family_policy)]; | |
738 | int retval; | |
739 | ||
cceb11f5 | 740 | retval = nl_sock_create(NETLINK_GENERIC, &sock); |
2fe27d5a BP |
741 | if (retval) { |
742 | return -retval; | |
743 | } | |
744 | ||
745 | ofpbuf_init(&request, 0); | |
746 | nl_msg_put_genlmsghdr(&request, 0, GENL_ID_CTRL, NLM_F_REQUEST, | |
747 | CTRL_CMD_GETFAMILY, 1); | |
748 | nl_msg_put_string(&request, CTRL_ATTR_FAMILY_NAME, name); | |
749 | retval = nl_sock_transact(sock, &request, &reply); | |
750 | ofpbuf_uninit(&request); | |
751 | if (retval) { | |
752 | nl_sock_destroy(sock); | |
753 | return -retval; | |
754 | } | |
755 | ||
756 | if (!nl_policy_parse(reply, NLMSG_HDRLEN + GENL_HDRLEN, | |
757 | family_policy, attrs, ARRAY_SIZE(family_policy))) { | |
758 | nl_sock_destroy(sock); | |
759 | ofpbuf_delete(reply); | |
760 | return -EPROTO; | |
761 | } | |
762 | ||
763 | retval = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]); | |
764 | if (retval == 0) { | |
765 | retval = -EPROTO; | |
2ad204c8 BP |
766 | } else { |
767 | define_genl_family(retval, name); | |
2fe27d5a BP |
768 | } |
769 | nl_sock_destroy(sock); | |
770 | ofpbuf_delete(reply); | |
2ad204c8 | 771 | |
2fe27d5a BP |
772 | return retval; |
773 | } | |
774 | ||
775 | /* If '*number' is 0, translates the given Generic Netlink family 'name' to a | |
776 | * number and stores it in '*number'. If successful, returns 0 and the caller | |
777 | * may use '*number' as the family number. On failure, returns a positive | |
778 | * errno value and '*number' caches the errno value. */ | |
779 | int | |
780 | nl_lookup_genl_family(const char *name, int *number) | |
781 | { | |
782 | if (*number == 0) { | |
783 | *number = do_lookup_genl_family(name); | |
784 | assert(*number != 0); | |
785 | } | |
786 | return *number > 0 ? 0 : -*number; | |
787 | } | |
788 | \f | |
789 | /* Netlink PID. | |
790 | * | |
791 | * Every Netlink socket must be bound to a unique 32-bit PID. By convention, | |
792 | * programs that have a single Netlink socket use their Unix process ID as PID, | |
793 | * and programs with multiple Netlink sockets add a unique per-socket | |
794 | * identifier in the bits above the Unix process ID. | |
795 | * | |
796 | * The kernel has Netlink PID 0. | |
797 | */ | |
798 | ||
799 | /* Parameters for how many bits in the PID should come from the Unix process ID | |
800 | * and how many unique per-socket. */ | |
801 | #define SOCKET_BITS 10 | |
802 | #define MAX_SOCKETS (1u << SOCKET_BITS) | |
803 | ||
804 | #define PROCESS_BITS (32 - SOCKET_BITS) | |
805 | #define MAX_PROCESSES (1u << PROCESS_BITS) | |
806 | #define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1)) | |
807 | ||
808 | /* Bit vector of unused socket identifiers. */ | |
809 | static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)]; | |
810 | ||
811 | /* Allocates and returns a new Netlink PID. */ | |
812 | static int | |
813 | alloc_pid(uint32_t *pid) | |
814 | { | |
815 | int i; | |
816 | ||
817 | for (i = 0; i < MAX_SOCKETS; i++) { | |
818 | if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) { | |
819 | avail_sockets[i / 32] |= 1u << (i % 32); | |
820 | *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS); | |
821 | return 0; | |
822 | } | |
823 | } | |
824 | VLOG_ERR("netlink pid space exhausted"); | |
825 | return ENOBUFS; | |
826 | } | |
827 | ||
828 | /* Makes the specified 'pid' available for reuse. */ | |
829 | static void | |
830 | free_pid(uint32_t pid) | |
831 | { | |
832 | int sock = pid >> PROCESS_BITS; | |
833 | assert(avail_sockets[sock / 32] & (1u << (sock % 32))); | |
834 | avail_sockets[sock / 32] &= ~(1u << (sock % 32)); | |
835 | } | |
836 | \f | |
837 | static void | |
2ad204c8 | 838 | nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds) |
2fe27d5a BP |
839 | { |
840 | struct nlmsg_flag { | |
841 | unsigned int bits; | |
842 | const char *name; | |
843 | }; | |
844 | static const struct nlmsg_flag flags[] = { | |
845 | { NLM_F_REQUEST, "REQUEST" }, | |
846 | { NLM_F_MULTI, "MULTI" }, | |
847 | { NLM_F_ACK, "ACK" }, | |
848 | { NLM_F_ECHO, "ECHO" }, | |
849 | { NLM_F_DUMP, "DUMP" }, | |
850 | { NLM_F_ROOT, "ROOT" }, | |
851 | { NLM_F_MATCH, "MATCH" }, | |
852 | { NLM_F_ATOMIC, "ATOMIC" }, | |
853 | }; | |
854 | const struct nlmsg_flag *flag; | |
855 | uint16_t flags_left; | |
856 | ||
857 | ds_put_format(ds, "nl(len:%"PRIu32", type=%"PRIu16, | |
858 | h->nlmsg_len, h->nlmsg_type); | |
859 | if (h->nlmsg_type == NLMSG_NOOP) { | |
860 | ds_put_cstr(ds, "(no-op)"); | |
861 | } else if (h->nlmsg_type == NLMSG_ERROR) { | |
862 | ds_put_cstr(ds, "(error)"); | |
863 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
864 | ds_put_cstr(ds, "(done)"); | |
865 | } else if (h->nlmsg_type == NLMSG_OVERRUN) { | |
866 | ds_put_cstr(ds, "(overrun)"); | |
867 | } else if (h->nlmsg_type < NLMSG_MIN_TYPE) { | |
868 | ds_put_cstr(ds, "(reserved)"); | |
2ad204c8 BP |
869 | } else if (protocol == NETLINK_GENERIC) { |
870 | ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type)); | |
2fe27d5a BP |
871 | } else { |
872 | ds_put_cstr(ds, "(family-defined)"); | |
873 | } | |
874 | ds_put_format(ds, ", flags=%"PRIx16, h->nlmsg_flags); | |
875 | flags_left = h->nlmsg_flags; | |
876 | for (flag = flags; flag < &flags[ARRAY_SIZE(flags)]; flag++) { | |
877 | if ((flags_left & flag->bits) == flag->bits) { | |
878 | ds_put_format(ds, "[%s]", flag->name); | |
879 | flags_left &= ~flag->bits; | |
880 | } | |
881 | } | |
882 | if (flags_left) { | |
883 | ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left); | |
884 | } | |
885 | ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32"(%d:%d))", | |
886 | h->nlmsg_seq, h->nlmsg_pid, | |
887 | (int) (h->nlmsg_pid & PROCESS_MASK), | |
888 | (int) (h->nlmsg_pid >> PROCESS_BITS)); | |
889 | } | |
890 | ||
891 | static char * | |
7041c3a9 | 892 | nlmsg_to_string(const struct ofpbuf *buffer, int protocol) |
2fe27d5a BP |
893 | { |
894 | struct ds ds = DS_EMPTY_INITIALIZER; | |
895 | const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN); | |
896 | if (h) { | |
2ad204c8 | 897 | nlmsghdr_to_string(h, protocol, &ds); |
2fe27d5a BP |
898 | if (h->nlmsg_type == NLMSG_ERROR) { |
899 | const struct nlmsgerr *e; | |
900 | e = ofpbuf_at(buffer, NLMSG_HDRLEN, | |
901 | NLMSG_ALIGN(sizeof(struct nlmsgerr))); | |
902 | if (e) { | |
903 | ds_put_format(&ds, " error(%d", e->error); | |
904 | if (e->error < 0) { | |
905 | ds_put_format(&ds, "(%s)", strerror(-e->error)); | |
906 | } | |
907 | ds_put_cstr(&ds, ", in-reply-to("); | |
2ad204c8 | 908 | nlmsghdr_to_string(&e->msg, protocol, &ds); |
2fe27d5a BP |
909 | ds_put_cstr(&ds, "))"); |
910 | } else { | |
911 | ds_put_cstr(&ds, " error(truncated)"); | |
912 | } | |
913 | } else if (h->nlmsg_type == NLMSG_DONE) { | |
914 | int *error = ofpbuf_at(buffer, NLMSG_HDRLEN, sizeof *error); | |
915 | if (error) { | |
916 | ds_put_format(&ds, " done(%d", *error); | |
917 | if (*error < 0) { | |
918 | ds_put_format(&ds, "(%s)", strerror(-*error)); | |
919 | } | |
920 | ds_put_cstr(&ds, ")"); | |
921 | } else { | |
922 | ds_put_cstr(&ds, " done(truncated)"); | |
923 | } | |
7041c3a9 BP |
924 | } else if (protocol == NETLINK_GENERIC) { |
925 | struct genlmsghdr *genl = nl_msg_genlmsghdr(buffer); | |
926 | if (genl) { | |
927 | ds_put_format(&ds, ",genl(cmd=%"PRIu8",version=%"PRIu8")", | |
928 | genl->cmd, genl->version); | |
929 | } | |
2fe27d5a BP |
930 | } |
931 | } else { | |
932 | ds_put_cstr(&ds, "nl(truncated)"); | |
933 | } | |
934 | return ds.string; | |
935 | } | |
936 | ||
937 | static void | |
938 | log_nlmsg(const char *function, int error, | |
7041c3a9 | 939 | const void *message, size_t size, int protocol) |
2fe27d5a BP |
940 | { |
941 | struct ofpbuf buffer; | |
942 | char *nlmsg; | |
943 | ||
944 | if (!VLOG_IS_DBG_ENABLED()) { | |
945 | return; | |
946 | } | |
947 | ||
948 | ofpbuf_use_const(&buffer, message, size); | |
7041c3a9 | 949 | nlmsg = nlmsg_to_string(&buffer, protocol); |
2fe27d5a BP |
950 | VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg); |
951 | free(nlmsg); | |
952 | } | |
953 | ||
954 |