]>
Commit | Line | Data |
---|---|---|
a132aa96 EJ |
1 | /* |
2 | * Copyright (c) 2011 Nicira Networks. | |
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 | ||
19 | #include "route-table.h" | |
20 | ||
21 | #include <assert.h> | |
22 | #include <arpa/inet.h> | |
23 | #include <sys/socket.h> | |
24 | #include <linux/rtnetlink.h> | |
25 | #include <net/if.h> | |
26 | ||
27 | #include "hash.h" | |
28 | #include "hmap.h" | |
29 | #include "netlink.h" | |
30 | #include "netlink-socket.h" | |
31 | #include "ofpbuf.h" | |
32 | #include "rtnetlink.h" | |
b46ccdf5 | 33 | #include "rtnetlink-link.h" |
a132aa96 EJ |
34 | #include "vlog.h" |
35 | ||
36 | VLOG_DEFINE_THIS_MODULE(route_table); | |
37 | ||
38 | struct route_data { | |
39 | /* Copied from struct rtmsg. */ | |
40 | unsigned char rtm_dst_len; | |
41 | ||
42 | /* Extracted from Netlink attributes. */ | |
43 | uint32_t rta_dst; /* Destination in host byte order. 0 if missing. */ | |
44 | int rta_oif; /* Output interface index. */ | |
45 | }; | |
46 | ||
47 | /* A digested version of a route message sent down by the kernel to indicate | |
48 | * that a route has changed. */ | |
49 | struct route_table_msg { | |
db2dede4 | 50 | bool relevant; /* Should this message be processed? */ |
a132aa96 EJ |
51 | int nlmsg_type; /* e.g. RTM_NEWROUTE, RTM_DELROUTE. */ |
52 | struct route_data rd; /* Data parsed from this message. */ | |
53 | }; | |
54 | ||
55 | struct route_node { | |
56 | struct hmap_node node; /* Node in route_map. */ | |
57 | struct route_data rd; /* Data associated with this node. */ | |
58 | }; | |
59 | ||
b46ccdf5 EJ |
60 | struct name_node { |
61 | struct hmap_node node; /* Node in name_map. */ | |
62 | uint32_t ifi_index; /* Kernel interface index. */ | |
63 | ||
64 | char ifname[IFNAMSIZ]; /* Interface name. */ | |
65 | }; | |
66 | ||
a132aa96 EJ |
67 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
68 | ||
69 | static unsigned int register_count = 0; | |
70 | static struct rtnetlink *rtn = NULL; | |
71 | static struct route_table_msg rtmsg; | |
b46ccdf5 EJ |
72 | static struct rtnetlink_notifier route_notifier; |
73 | static struct rtnetlink_notifier name_notifier; | |
f0e167f0 EJ |
74 | |
75 | static bool route_table_valid = false; | |
b46ccdf5 | 76 | static bool name_table_valid = false; |
a132aa96 | 77 | static struct hmap route_map; |
b46ccdf5 | 78 | static struct hmap name_map; |
a132aa96 EJ |
79 | |
80 | static int route_table_reset(void); | |
f0e167f0 | 81 | static void route_table_handle_msg(const struct route_table_msg *); |
a132aa96 EJ |
82 | static bool route_table_parse(struct ofpbuf *, struct route_table_msg *); |
83 | static void route_table_change(const struct route_table_msg *, void *); | |
84 | static struct route_node *route_node_lookup(const struct route_data *); | |
85 | static struct route_node *route_node_lookup_by_ip(uint32_t ip); | |
86 | static void route_map_clear(void); | |
87 | static uint32_t hash_route_data(const struct route_data *); | |
88 | ||
b46ccdf5 EJ |
89 | static void name_table_init(void); |
90 | static void name_table_uninit(void); | |
91 | static int name_table_reset(void); | |
92 | static void name_table_change(const struct rtnetlink_link_change *, void *); | |
93 | static void name_map_clear(void); | |
94 | static struct name_node *name_node_lookup(int ifi_index); | |
95 | ||
96 | /* Populates 'name' with the name of the interface traffic destined for 'ip' | |
97 | * is likely to egress out of (see route_table_get_ifindex). | |
98 | * | |
99 | * Returns true if successful, otherwise false. */ | |
100 | bool | |
101 | route_table_get_name(ovs_be32 ip, char name[IFNAMSIZ]) | |
102 | { | |
103 | int ifindex; | |
104 | ||
105 | if (!name_table_valid) { | |
106 | name_table_reset(); | |
107 | } | |
108 | ||
109 | if (route_table_get_ifindex(ip, &ifindex)) { | |
110 | struct name_node *nn; | |
111 | ||
112 | nn = name_node_lookup(ifindex); | |
113 | if (nn) { | |
e868fb3d | 114 | ovs_strlcpy(name, nn->ifname, IFNAMSIZ); |
b46ccdf5 EJ |
115 | return true; |
116 | } | |
117 | } | |
118 | ||
119 | return false; | |
120 | } | |
121 | ||
a132aa96 EJ |
122 | /* Populates 'ifindex' with the interface index traffic destined for 'ip' is |
123 | * likely to egress. There is no hard guarantee that traffic destined for 'ip' | |
124 | * will egress out the specified interface. 'ifindex' may refer to an | |
125 | * interface which is not physical (such as a bridge port). | |
126 | * | |
127 | * Returns true if successful, otherwise false. */ | |
128 | bool | |
129 | route_table_get_ifindex(ovs_be32 ip_, int *ifindex) | |
130 | { | |
131 | struct route_node *rn; | |
132 | uint32_t ip = ntohl(ip_); | |
133 | ||
134 | *ifindex = 0; | |
135 | ||
f0e167f0 EJ |
136 | if (!route_table_valid) { |
137 | route_table_reset(); | |
138 | } | |
139 | ||
a132aa96 EJ |
140 | rn = route_node_lookup_by_ip(ip); |
141 | ||
142 | if (rn) { | |
143 | *ifindex = rn->rd.rta_oif; | |
144 | return true; | |
145 | } | |
146 | ||
147 | /* Choose a default route. */ | |
148 | HMAP_FOR_EACH(rn, node, &route_map) { | |
149 | if (rn->rd.rta_dst == 0 && rn->rd.rtm_dst_len == 0) { | |
150 | *ifindex = rn->rd.rta_oif; | |
151 | return true; | |
152 | } | |
153 | } | |
154 | ||
155 | return false; | |
156 | } | |
157 | ||
158 | /* Users of the route_table module should register themselves with this | |
159 | * function before making any other route_table function calls. */ | |
160 | void | |
161 | route_table_register(void) | |
162 | { | |
163 | if (!register_count) { | |
164 | rtnetlink_parse_func *pf; | |
165 | rtnetlink_notify_func *nf; | |
166 | ||
167 | assert(!rtn); | |
168 | ||
169 | pf = (rtnetlink_parse_func *) route_table_parse; | |
170 | nf = (rtnetlink_notify_func *) route_table_change; | |
171 | ||
172 | rtn = rtnetlink_create(RTNLGRP_IPV4_ROUTE, pf, &rtmsg); | |
b46ccdf5 | 173 | rtnetlink_notifier_register(rtn, &route_notifier, nf, NULL); |
a132aa96 EJ |
174 | |
175 | hmap_init(&route_map); | |
176 | route_table_reset(); | |
b46ccdf5 | 177 | name_table_init(); |
a132aa96 EJ |
178 | } |
179 | ||
180 | register_count++; | |
181 | } | |
182 | ||
183 | /* Users of the route_table module should unregister themselves with this | |
184 | * function when they will no longer be making any more route_table fuction | |
185 | * calls. */ | |
186 | void | |
187 | route_table_unregister(void) | |
188 | { | |
189 | register_count--; | |
190 | ||
191 | if (!register_count) { | |
192 | rtnetlink_destroy(rtn); | |
193 | rtn = NULL; | |
194 | ||
195 | route_map_clear(); | |
196 | hmap_destroy(&route_map); | |
b46ccdf5 | 197 | name_table_uninit(); |
a132aa96 EJ |
198 | } |
199 | } | |
200 | ||
201 | /* Run periodically to update the locally maintained routing table. */ | |
202 | void | |
203 | route_table_run(void) | |
204 | { | |
205 | if (rtn) { | |
b46ccdf5 | 206 | rtnetlink_link_notifier_run(); |
a132aa96 EJ |
207 | rtnetlink_notifier_run(rtn); |
208 | } | |
209 | } | |
210 | ||
211 | /* Causes poll_block() to wake up when route_table updates are required. */ | |
212 | void | |
213 | route_table_wait(void) | |
214 | { | |
215 | if (rtn) { | |
b46ccdf5 | 216 | rtnetlink_link_notifier_wait(); |
a132aa96 EJ |
217 | rtnetlink_notifier_wait(rtn); |
218 | } | |
219 | } | |
220 | ||
221 | static int | |
222 | route_table_reset(void) | |
223 | { | |
224 | int error; | |
225 | struct nl_dump dump; | |
226 | struct rtgenmsg *rtmsg; | |
227 | struct ofpbuf request, reply; | |
228 | static struct nl_sock *rtnl_sock; | |
229 | ||
230 | route_map_clear(); | |
f0e167f0 | 231 | route_table_valid = true; |
a132aa96 | 232 | |
cceb11f5 | 233 | error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); |
a132aa96 EJ |
234 | if (error) { |
235 | VLOG_WARN_RL(&rl, "failed to reset routing table, " | |
236 | "cannot create RTNETLINK_ROUTE socket"); | |
237 | return error; | |
238 | } | |
239 | ||
240 | ofpbuf_init(&request, 0); | |
241 | ||
242 | nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETROUTE, NLM_F_REQUEST); | |
243 | ||
244 | rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg); | |
245 | rtmsg->rtgen_family = AF_INET; | |
246 | ||
247 | nl_dump_start(&dump, rtnl_sock, &request); | |
248 | ||
249 | while (nl_dump_next(&dump, &reply)) { | |
250 | struct route_table_msg msg; | |
251 | ||
252 | if (route_table_parse(&reply, &msg)) { | |
f0e167f0 | 253 | route_table_handle_msg(&msg); |
a132aa96 EJ |
254 | } |
255 | } | |
256 | ||
257 | error = nl_dump_done(&dump); | |
258 | nl_sock_destroy(rtnl_sock); | |
259 | ||
260 | return error; | |
261 | } | |
262 | ||
263 | ||
264 | static bool | |
265 | route_table_parse(struct ofpbuf *buf, struct route_table_msg *change) | |
266 | { | |
267 | bool parsed; | |
268 | ||
269 | static const struct nl_policy policy[] = { | |
270 | [RTA_DST] = { .type = NL_A_U32, .optional = true }, | |
271 | [RTA_OIF] = { .type = NL_A_U32, .optional = false }, | |
272 | }; | |
273 | ||
274 | static struct nlattr *attrs[ARRAY_SIZE(policy)]; | |
275 | ||
276 | parsed = nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct rtmsg), | |
277 | policy, attrs, ARRAY_SIZE(policy)); | |
278 | ||
279 | if (parsed) { | |
280 | const struct rtmsg *rtm; | |
281 | const struct nlmsghdr *nlmsg; | |
282 | ||
283 | nlmsg = buf->data; | |
284 | rtm = (const struct rtmsg *) ((const char *) buf->data + NLMSG_HDRLEN); | |
285 | ||
286 | if (rtm->rtm_family != AF_INET) { | |
287 | VLOG_DBG_RL(&rl, "received non AF_INET rtnetlink route message"); | |
288 | return false; | |
289 | } | |
290 | ||
291 | memset(change, 0, sizeof *change); | |
db2dede4 EJ |
292 | change->relevant = true; |
293 | ||
294 | if (rtm->rtm_scope == RT_SCOPE_NOWHERE) { | |
295 | change->relevant = false; | |
296 | } | |
297 | ||
298 | if (rtm->rtm_type != RTN_UNICAST && | |
299 | rtm->rtm_type != RTN_LOCAL) { | |
300 | change->relevant = false; | |
301 | } | |
a132aa96 EJ |
302 | |
303 | change->nlmsg_type = nlmsg->nlmsg_type; | |
304 | change->rd.rtm_dst_len = rtm->rtm_dst_len; | |
305 | change->rd.rta_oif = nl_attr_get_u32(attrs[RTA_OIF]); | |
306 | ||
307 | if (attrs[RTA_DST]) { | |
308 | change->rd.rta_dst = ntohl(nl_attr_get_be32(attrs[RTA_DST])); | |
309 | } | |
310 | ||
311 | } else { | |
312 | VLOG_DBG_RL(&rl, "received unparseable rtnetlink route message"); | |
313 | } | |
314 | ||
315 | return parsed; | |
316 | } | |
317 | ||
318 | static void | |
f0e167f0 EJ |
319 | route_table_change(const struct route_table_msg *change OVS_UNUSED, |
320 | void *aux OVS_UNUSED) | |
a132aa96 | 321 | { |
f0e167f0 EJ |
322 | route_table_valid = false; |
323 | } | |
324 | ||
325 | static void | |
326 | route_table_handle_msg(const struct route_table_msg *change) | |
327 | { | |
328 | if (change->relevant && change->nlmsg_type == RTM_NEWROUTE && | |
329 | !route_node_lookup(&change->rd)) { | |
a132aa96 EJ |
330 | struct route_node *rn; |
331 | ||
f0e167f0 EJ |
332 | rn = xzalloc(sizeof *rn); |
333 | memcpy(&rn->rd, &change->rd, sizeof change->rd); | |
a132aa96 | 334 | |
f0e167f0 | 335 | hmap_insert(&route_map, &rn->node, hash_route_data(&rn->rd)); |
a132aa96 EJ |
336 | } |
337 | } | |
338 | ||
339 | static struct route_node * | |
340 | route_node_lookup(const struct route_data *rd) | |
341 | { | |
342 | struct route_node *rn; | |
343 | ||
344 | HMAP_FOR_EACH_WITH_HASH(rn, node, hash_route_data(rd), &route_map) { | |
345 | if (!memcmp(&rn->rd, rd, sizeof *rd)) { | |
346 | return rn; | |
347 | } | |
348 | } | |
349 | ||
350 | return NULL; | |
351 | } | |
352 | ||
353 | static struct route_node * | |
354 | route_node_lookup_by_ip(uint32_t ip) | |
355 | { | |
356 | int dst_len; | |
357 | struct route_node *rn, *rn_ret; | |
358 | ||
359 | dst_len = -1; | |
360 | rn_ret = NULL; | |
361 | ||
362 | HMAP_FOR_EACH(rn, node, &route_map) { | |
363 | uint32_t mask = 0xffffffff << (32 - rn->rd.rtm_dst_len); | |
364 | ||
365 | if (rn->rd.rta_dst == 0 && rn->rd.rtm_dst_len == 0) { | |
366 | /* Default route. */ | |
367 | continue; | |
368 | } | |
369 | ||
370 | if (rn->rd.rtm_dst_len > dst_len && | |
371 | (ip & mask) == (rn->rd.rta_dst & mask)) { | |
372 | rn_ret = rn; | |
373 | dst_len = rn->rd.rtm_dst_len; | |
374 | } | |
375 | } | |
376 | ||
377 | return rn_ret; | |
378 | } | |
379 | ||
380 | static void | |
381 | route_map_clear(void) | |
382 | { | |
383 | struct route_node *rn, *rn_next; | |
384 | ||
385 | HMAP_FOR_EACH_SAFE(rn, rn_next, node, &route_map) { | |
386 | hmap_remove(&route_map, &rn->node); | |
387 | free(rn); | |
388 | } | |
389 | } | |
390 | ||
391 | static uint32_t | |
392 | hash_route_data(const struct route_data *rd) | |
393 | { | |
394 | return hash_bytes(rd, sizeof *rd, 0); | |
395 | } | |
b46ccdf5 EJ |
396 | \f |
397 | /* name_table . */ | |
398 | ||
399 | static void | |
400 | name_table_init(void) | |
401 | { | |
402 | hmap_init(&name_map); | |
403 | rtnetlink_link_notifier_register(&name_notifier, name_table_change, NULL); | |
404 | name_table_valid = false; | |
405 | } | |
406 | ||
407 | static void | |
408 | name_table_uninit(void) | |
409 | { | |
410 | rtnetlink_link_notifier_unregister(&name_notifier); | |
411 | name_map_clear(); | |
412 | hmap_destroy(&name_map); | |
413 | } | |
414 | ||
415 | static int | |
416 | name_table_reset(void) | |
417 | { | |
418 | int error; | |
419 | struct nl_dump dump; | |
420 | struct rtgenmsg *rtmsg; | |
421 | struct ofpbuf request, reply; | |
422 | static struct nl_sock *rtnl_sock; | |
423 | ||
424 | name_table_valid = true; | |
425 | name_map_clear(); | |
426 | error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock); | |
427 | if (error) { | |
428 | VLOG_WARN_RL(&rl, "failed to create NETLINK_ROUTE socket"); | |
429 | return error; | |
430 | } | |
431 | ||
432 | ofpbuf_init(&request, 0); | |
433 | nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETLINK, NLM_F_REQUEST); | |
434 | rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg); | |
435 | rtmsg->rtgen_family = AF_INET; | |
436 | ||
437 | nl_dump_start(&dump, rtnl_sock, &request); | |
438 | while (nl_dump_next(&dump, &reply)) { | |
439 | struct rtnetlink_link_change change; | |
440 | ||
441 | if (rtnetlink_link_parse(&reply, &change) | |
442 | && change.nlmsg_type == RTM_NEWLINK | |
443 | && !name_node_lookup(change.ifi_index)) { | |
444 | struct name_node *nn; | |
445 | ||
446 | nn = xzalloc(sizeof *nn); | |
447 | nn->ifi_index = change.ifi_index; | |
e868fb3d | 448 | ovs_strlcpy(nn->ifname, change.ifname, IFNAMSIZ); |
b46ccdf5 EJ |
449 | hmap_insert(&name_map, &nn->node, hash_int(nn->ifi_index, 0)); |
450 | } | |
451 | } | |
452 | nl_sock_destroy(rtnl_sock); | |
453 | return nl_dump_done(&dump); | |
454 | } | |
455 | ||
456 | static void | |
457 | name_table_change(const struct rtnetlink_link_change *change OVS_UNUSED, | |
458 | void *aux OVS_UNUSED) | |
459 | { | |
460 | /* Changes to interface status can cause routing table changes that some | |
461 | * versions of the linux kernel do not advertise for some reason. */ | |
462 | route_table_valid = false; | |
463 | name_table_valid = false; | |
464 | } | |
465 | ||
466 | static struct name_node * | |
467 | name_node_lookup(int ifi_index) | |
468 | { | |
469 | struct name_node *nn; | |
470 | ||
471 | HMAP_FOR_EACH_WITH_HASH(nn, node, hash_int(ifi_index, 0), &name_map) { | |
472 | if (nn->ifi_index == ifi_index) { | |
473 | return nn; | |
474 | } | |
475 | } | |
476 | ||
477 | return NULL; | |
478 | } | |
479 | ||
480 | static void | |
481 | name_map_clear(void) | |
482 | { | |
483 | struct name_node *nn, *nn_next; | |
484 | ||
485 | HMAP_FOR_EACH_SAFE(nn, nn_next, node, &name_map) { | |
486 | hmap_remove(&name_map, &nn->node); | |
487 | free(nn); | |
488 | } | |
489 | } |