2 * Copyright (c) 2011, 2012 Nicira, Inc.
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:
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
19 #include "route-table.h"
22 #include <arpa/inet.h>
23 #include <sys/socket.h>
24 #include <linux/rtnetlink.h>
30 #include "netlink-notifier.h"
31 #include "netlink-socket.h"
33 #include "rtnetlink-link.h"
36 VLOG_DEFINE_THIS_MODULE(route_table);
39 /* Copied from struct rtmsg. */
40 unsigned char rtm_dst_len;
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. */
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 {
50 bool relevant; /* Should this message be processed? */
51 int nlmsg_type; /* e.g. RTM_NEWROUTE, RTM_DELROUTE. */
52 struct route_data rd; /* Data parsed from this message. */
56 struct hmap_node node; /* Node in route_map. */
57 struct route_data rd; /* Data associated with this node. */
61 struct hmap_node node; /* Node in name_map. */
62 uint32_t ifi_index; /* Kernel interface index. */
64 char ifname[IFNAMSIZ]; /* Interface name. */
67 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
69 static unsigned int register_count = 0;
70 static struct nln *nln = NULL;
71 static struct route_table_msg rtmsg;
72 static struct nln_notifier *route_notifier = NULL;
73 static struct nln_notifier *name_notifier = NULL;
75 static bool route_table_valid = false;
76 static bool name_table_valid = false;
77 static struct hmap route_map;
78 static struct hmap name_map;
80 static int route_table_reset(void);
81 static void route_table_handle_msg(const struct route_table_msg *);
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 *);
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);
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).
99 * Returns true if successful, otherwise false. */
101 route_table_get_name(ovs_be32 ip, char name[IFNAMSIZ])
105 if (!name_table_valid) {
109 if (route_table_get_ifindex(ip, &ifindex)) {
110 struct name_node *nn;
112 nn = name_node_lookup(ifindex);
114 ovs_strlcpy(name, nn->ifname, IFNAMSIZ);
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).
127 * Returns true if successful, otherwise false. */
129 route_table_get_ifindex(ovs_be32 ip_, int *ifindex)
131 struct route_node *rn;
132 uint32_t ip = ntohl(ip_);
136 if (!route_table_valid) {
140 rn = route_node_lookup_by_ip(ip);
143 *ifindex = rn->rd.rta_oif;
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;
158 /* Users of the route_table module should register themselves with this
159 * function before making any other route_table function calls. */
161 route_table_register(void)
163 if (!register_count) {
165 assert(!route_notifier);
167 nln = nln_create(NETLINK_ROUTE, RTNLGRP_IPV4_ROUTE,
168 (nln_parse_func *) route_table_parse, &rtmsg);
171 nln_notifier_create(nln, (nln_notify_func *) route_table_change,
174 hmap_init(&route_map);
182 /* Users of the route_table module should unregister themselves with this
183 * function when they will no longer be making any more route_table fuction
186 route_table_unregister(void)
190 if (!register_count) {
191 nln_notifier_destroy(route_notifier);
192 route_notifier = NULL;
197 hmap_destroy(&route_map);
202 /* Run periodically to update the locally maintained routing table. */
204 route_table_run(void)
207 rtnetlink_link_run();
212 /* Causes poll_block() to wake up when route_table updates are required. */
214 route_table_wait(void)
217 rtnetlink_link_wait();
223 route_table_reset(void)
227 struct rtgenmsg *rtmsg;
228 struct ofpbuf request, reply;
229 static struct nl_sock *rtnl_sock;
232 route_table_valid = true;
234 error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
236 VLOG_WARN_RL(&rl, "failed to reset routing table, "
237 "cannot create RTNETLINK_ROUTE socket");
241 ofpbuf_init(&request, 0);
243 nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETROUTE, NLM_F_REQUEST);
245 rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg);
246 rtmsg->rtgen_family = AF_INET;
248 nl_dump_start(&dump, rtnl_sock, &request);
249 ofpbuf_uninit(&request);
251 while (nl_dump_next(&dump, &reply)) {
252 struct route_table_msg msg;
254 if (route_table_parse(&reply, &msg)) {
255 route_table_handle_msg(&msg);
259 error = nl_dump_done(&dump);
260 nl_sock_destroy(rtnl_sock);
267 route_table_parse(struct ofpbuf *buf, struct route_table_msg *change)
271 static const struct nl_policy policy[] = {
272 [RTA_DST] = { .type = NL_A_U32, .optional = true },
273 [RTA_OIF] = { .type = NL_A_U32, .optional = false },
276 static struct nlattr *attrs[ARRAY_SIZE(policy)];
278 parsed = nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct rtmsg),
279 policy, attrs, ARRAY_SIZE(policy));
282 const struct rtmsg *rtm;
283 const struct nlmsghdr *nlmsg;
286 rtm = (const struct rtmsg *) ((const char *) buf->data + NLMSG_HDRLEN);
288 if (rtm->rtm_family != AF_INET) {
289 VLOG_DBG_RL(&rl, "received non AF_INET rtnetlink route message");
293 memset(change, 0, sizeof *change);
294 change->relevant = true;
296 if (rtm->rtm_scope == RT_SCOPE_NOWHERE) {
297 change->relevant = false;
300 if (rtm->rtm_type != RTN_UNICAST &&
301 rtm->rtm_type != RTN_LOCAL) {
302 change->relevant = false;
305 change->nlmsg_type = nlmsg->nlmsg_type;
306 change->rd.rtm_dst_len = rtm->rtm_dst_len;
307 change->rd.rta_oif = nl_attr_get_u32(attrs[RTA_OIF]);
309 if (attrs[RTA_DST]) {
310 change->rd.rta_dst = ntohl(nl_attr_get_be32(attrs[RTA_DST]));
314 VLOG_DBG_RL(&rl, "received unparseable rtnetlink route message");
321 route_table_change(const struct route_table_msg *change OVS_UNUSED,
322 void *aux OVS_UNUSED)
324 route_table_valid = false;
328 route_table_handle_msg(const struct route_table_msg *change)
330 if (change->relevant && change->nlmsg_type == RTM_NEWROUTE &&
331 !route_node_lookup(&change->rd)) {
332 struct route_node *rn;
334 rn = xzalloc(sizeof *rn);
335 memcpy(&rn->rd, &change->rd, sizeof change->rd);
337 hmap_insert(&route_map, &rn->node, hash_route_data(&rn->rd));
341 static struct route_node *
342 route_node_lookup(const struct route_data *rd)
344 struct route_node *rn;
346 HMAP_FOR_EACH_WITH_HASH(rn, node, hash_route_data(rd), &route_map) {
347 if (!memcmp(&rn->rd, rd, sizeof *rd)) {
355 static struct route_node *
356 route_node_lookup_by_ip(uint32_t ip)
359 struct route_node *rn, *rn_ret;
364 HMAP_FOR_EACH(rn, node, &route_map) {
365 uint32_t mask = 0xffffffff << (32 - rn->rd.rtm_dst_len);
367 if (rn->rd.rta_dst == 0 && rn->rd.rtm_dst_len == 0) {
372 if (rn->rd.rtm_dst_len > dst_len &&
373 (ip & mask) == (rn->rd.rta_dst & mask)) {
375 dst_len = rn->rd.rtm_dst_len;
383 route_map_clear(void)
385 struct route_node *rn, *rn_next;
387 HMAP_FOR_EACH_SAFE(rn, rn_next, node, &route_map) {
388 hmap_remove(&route_map, &rn->node);
394 hash_route_data(const struct route_data *rd)
396 return hash_bytes(rd, sizeof *rd, 0);
402 name_table_init(void)
404 hmap_init(&name_map);
405 name_notifier = rtnetlink_link_notifier_create(name_table_change, NULL);
406 name_table_valid = false;
410 name_table_uninit(void)
412 rtnetlink_link_notifier_destroy(name_notifier);
413 name_notifier = NULL;
415 hmap_destroy(&name_map);
419 name_table_reset(void)
423 struct rtgenmsg *rtmsg;
424 struct ofpbuf request, reply;
425 static struct nl_sock *rtnl_sock;
427 name_table_valid = true;
429 error = nl_sock_create(NETLINK_ROUTE, &rtnl_sock);
431 VLOG_WARN_RL(&rl, "failed to create NETLINK_ROUTE socket");
435 ofpbuf_init(&request, 0);
436 nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETLINK, NLM_F_REQUEST);
437 rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg);
438 rtmsg->rtgen_family = AF_INET;
440 nl_dump_start(&dump, rtnl_sock, &request);
441 ofpbuf_uninit(&request);
443 while (nl_dump_next(&dump, &reply)) {
444 struct rtnetlink_link_change change;
446 if (rtnetlink_link_parse(&reply, &change)
447 && change.nlmsg_type == RTM_NEWLINK
448 && !name_node_lookup(change.ifi_index)) {
449 struct name_node *nn;
451 nn = xzalloc(sizeof *nn);
452 nn->ifi_index = change.ifi_index;
453 ovs_strlcpy(nn->ifname, change.ifname, IFNAMSIZ);
454 hmap_insert(&name_map, &nn->node, hash_int(nn->ifi_index, 0));
457 nl_sock_destroy(rtnl_sock);
458 return nl_dump_done(&dump);
462 name_table_change(const struct rtnetlink_link_change *change OVS_UNUSED,
463 void *aux OVS_UNUSED)
465 /* Changes to interface status can cause routing table changes that some
466 * versions of the linux kernel do not advertise for some reason. */
467 route_table_valid = false;
468 name_table_valid = false;
471 static struct name_node *
472 name_node_lookup(int ifi_index)
474 struct name_node *nn;
476 HMAP_FOR_EACH_WITH_HASH(nn, node, hash_int(ifi_index, 0), &name_map) {
477 if (nn->ifi_index == ifi_index) {
488 struct name_node *nn, *nn_next;
490 HMAP_FOR_EACH_SAFE(nn, nn_next, node, &name_map) {
491 hmap_remove(&name_map, &nn->node);