lib/route-table.c

   1 /*
   2  * Copyright (c) 2011, 2012, 2013 Nicira, Inc.
   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 <arpa/inet.h>
  22 #include <sys/socket.h>
  23 #include <linux/rtnetlink.h>
  24 #include <net/if.h>
  25
  26 #include "hash.h"
  27 #include "hmap.h"
  28 #include "netlink.h"
  29 #include "netlink-notifier.h"
  30 #include "netlink-socket.h"
  31 #include "ofpbuf.h"
  32 #include "rtnetlink-link.h"
  33 #include "vlog.h"
  34
  35 VLOG_DEFINE_THIS_MODULE(route_table);
  36
  37 struct route_data {
  38     /* Copied from struct rtmsg. */
  39     unsigned char rtm_dst_len;
  40
  41     /* Extracted from Netlink attributes. */
  42     uint32_t rta_dst; /* Destination in host byte order. 0 if missing. */
  43     int rta_oif;      /* Output interface index. */
  44 };
  45
  46 /* A digested version of a route message sent down by the kernel to indicate
  47  * that a route has changed. */
  48 struct route_table_msg {
  49     bool relevant;        /* Should this message be processed? */
  50     int nlmsg_type;       /* e.g. RTM_NEWROUTE, RTM_DELROUTE. */
  51     struct route_data rd; /* Data parsed from this message. */
  52 };
  53
  54 struct route_node {
  55     struct hmap_node node; /* Node in route_map. */
  56     struct route_data rd;  /* Data associated with this node. */
  57 };
  58
  59 struct name_node {
  60     struct hmap_node node; /* Node in name_map. */
  61     uint32_t ifi_index;    /* Kernel interface index. */
  62
  63     char ifname[IFNAMSIZ]; /* Interface name. */
  64 };
  65
  66 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
  67
  68 static unsigned int register_count = 0;
  69 static struct nln *nln = NULL;
  70 static struct route_table_msg rtmsg;
  71 static struct nln_notifier *route_notifier = NULL;
  72 static struct nln_notifier *name_notifier = NULL;
  73
  74 static bool route_table_valid = false;
  75 static bool name_table_valid = false;
  76 static struct hmap route_map;
  77 static struct hmap name_map;
  78
  79 static int route_table_reset(void);
  80 static void route_table_handle_msg(const struct route_table_msg *);
  81 static bool route_table_parse(struct ofpbuf *, struct route_table_msg *);
  82 static void route_table_change(const struct route_table_msg *, void *);
  83 static struct route_node *route_node_lookup(const struct route_data *);
  84 static struct route_node *route_node_lookup_by_ip(uint32_t ip);
  85 static void route_map_clear(void);
  86 static uint32_t hash_route_data(const struct route_data *);
  87
  88 static void name_table_init(void);
  89 static void name_table_uninit(void);
  90 static int name_table_reset(void);
  91 static void name_table_change(const struct rtnetlink_link_change *, void *);
  92 static void name_map_clear(void);
  93 static struct name_node *name_node_lookup(int ifi_index);
  94
  95 /* Populates 'name' with the name of the interface traffic destined for 'ip'
  96  * is likely to egress out of (see route_table_get_ifindex).
  97  *
  98  * Returns true if successful, otherwise false. */
  99 bool
 100 route_table_get_name(ovs_be32 ip, char name[IFNAMSIZ])
 101 {
 102     int ifindex;
 103
 104     if (!name_table_valid) {
 105         name_table_reset();
 106     }
 107
 108     if (route_table_get_ifindex(ip, &ifindex)) {
 109         struct name_node *nn;
 110
 111         nn = name_node_lookup(ifindex);
 112         if (nn) {
 113             ovs_strlcpy(name, nn->ifname, IFNAMSIZ);
 114             return true;
 115         }
 116     }
 117
 118     return false;
 119 }
 120
 121 /* Populates 'ifindex' with the interface index traffic destined for 'ip' is
 122  * likely to egress.  There is no hard guarantee that traffic destined for 'ip'
 123  * will egress out the specified interface.  'ifindex' may refer to an
 124  * interface which is not physical (such as a bridge port).
 125  *
 126  * Returns true if successful, otherwise false. */
 127 bool
 128 route_table_get_ifindex(ovs_be32 ip_, int *ifindex)
 129 {
 130     struct route_node *rn;
 131     uint32_t ip = ntohl(ip_);
 132
 133     *ifindex = 0;
 134
 135     if (!route_table_valid) {
 136         route_table_reset();
 137     }
 138
 139     rn = route_node_lookup_by_ip(ip);
 140
 141     if (rn) {
 142         *ifindex = rn->rd.rta_oif;
 143         return true;
 144     }
 145
 146     /* Choose a default route. */
 147     HMAP_FOR_EACH(rn, node, &route_map) {
 148         if (rn->rd.rta_dst == 0 && rn->rd.rtm_dst_len == 0) {
 149             *ifindex = rn->rd.rta_oif;
 150             return true;
 151         }
 152     }
 153
 154     return false;
 155 }
 156
 157 /* Users of the route_table module should register themselves with this
 158  * function before making any other route_table function calls. */
 159 void
 160 route_table_register(void)
 161 {
 162     if (!register_count) {
 163         ovs_assert(!nln);
 164         ovs_assert(!route_notifier);
 165
 166         nln = nln_create(NETLINK_ROUTE, RTNLGRP_IPV4_ROUTE,
 167                          (nln_parse_func *) route_table_parse, &rtmsg);
 168
 169         route_notifier =
 170             nln_notifier_create(nln, (nln_notify_func *) route_table_change,
 171                                 NULL);
 172
 173         hmap_init(&route_map);
 174         route_table_reset();
 175         name_table_init();
 176     }
 177
 178     register_count++;
 179 }
 180
 181 /* Users of the route_table module should unregister themselves with this
 182  * function when they will no longer be making any more route_table fuction
 183  * calls. */
 184 void
 185 route_table_unregister(void)
 186 {
 187     register_count--;
 188
 189     if (!register_count) {
 190         nln_notifier_destroy(route_notifier);
 191         route_notifier = NULL;
 192         nln_destroy(nln);
 193         nln = NULL;
 194
 195         route_map_clear();
 196         hmap_destroy(&route_map);
 197         name_table_uninit();
 198     }
 199 }
 200
 201 /* Run periodically to update the locally maintained routing table. */
 202 void
 203 route_table_run(void)
 204 {
 205     if (nln) {
 206         rtnetlink_link_run();
 207         nln_run(nln);
 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 (nln) {
 216         rtnetlink_link_wait();
 217         nln_wait(nln);
 218     }
 219 }
 220
 221 static int
 222 route_table_reset(void)
 223 {
 224     struct nl_dump dump;
 225     struct rtgenmsg *rtmsg;
 226     struct ofpbuf request, reply;
 227
 228     route_map_clear();
 229     route_table_valid = true;
 230
 231     ofpbuf_init(&request, 0);
 232
 233     nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETROUTE, NLM_F_REQUEST);
 234
 235     rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg);
 236     rtmsg->rtgen_family = AF_INET;
 237
 238     nl_dump_start(&dump, NETLINK_ROUTE, &request);
 239     ofpbuf_uninit(&request);
 240
 241     while (nl_dump_next(&dump, &reply)) {
 242         struct route_table_msg msg;
 243
 244         if (route_table_parse(&reply, &msg)) {
 245             route_table_handle_msg(&msg);
 246         }
 247     }
 248
 249     return nl_dump_done(&dump);
 250 }
 251
 252
 253 static bool
 254 route_table_parse(struct ofpbuf *buf, struct route_table_msg *change)
 255 {
 256     bool parsed;
 257
 258     static const struct nl_policy policy[] = {
 259         [RTA_DST] = { .type = NL_A_U32, .optional = true  },
 260         [RTA_OIF] = { .type = NL_A_U32, .optional = false },
 261     };
 262
 263     struct nlattr *attrs[ARRAY_SIZE(policy)];
 264
 265     parsed = nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct rtmsg),
 266                              policy, attrs, ARRAY_SIZE(policy));
 267
 268     if (parsed) {
 269         const struct rtmsg *rtm;
 270         const struct nlmsghdr *nlmsg;
 271
 272         nlmsg = buf->data;
 273         rtm = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *rtm);
 274
 275         if (rtm->rtm_family != AF_INET) {
 276             VLOG_DBG_RL(&rl, "received non AF_INET rtnetlink route message");
 277             return false;
 278         }
 279
 280         memset(change, 0, sizeof *change);
 281         change->relevant = true;
 282
 283         if (rtm->rtm_scope == RT_SCOPE_NOWHERE) {
 284             change->relevant = false;
 285         }
 286
 287         if (rtm->rtm_type != RTN_UNICAST &&
 288             rtm->rtm_type != RTN_LOCAL) {
 289             change->relevant = false;
 290         }
 291
 292         change->nlmsg_type     = nlmsg->nlmsg_type;
 293         change->rd.rtm_dst_len = rtm->rtm_dst_len;
 294         change->rd.rta_oif     = nl_attr_get_u32(attrs[RTA_OIF]);
 295
 296         if (attrs[RTA_DST]) {
 297             change->rd.rta_dst = ntohl(nl_attr_get_be32(attrs[RTA_DST]));
 298         }
 299
 300     } else {
 301         VLOG_DBG_RL(&rl, "received unparseable rtnetlink route message");
 302     }
 303
 304     return parsed;
 305 }
 306
 307 static void
 308 route_table_change(const struct route_table_msg *change OVS_UNUSED,
 309                    void *aux OVS_UNUSED)
 310 {
 311     route_table_valid = false;
 312 }
 313
 314 static void
 315 route_table_handle_msg(const struct route_table_msg *change)
 316 {
 317     if (change->relevant && change->nlmsg_type == RTM_NEWROUTE &&
 318         !route_node_lookup(&change->rd)) {
 319         struct route_node *rn;
 320
 321         rn = xzalloc(sizeof *rn);
 322         memcpy(&rn->rd, &change->rd, sizeof change->rd);
 323
 324         hmap_insert(&route_map, &rn->node, hash_route_data(&rn->rd));
 325     }
 326 }
 327
 328 static struct route_node *
 329 route_node_lookup(const struct route_data *rd)
 330 {
 331     struct route_node *rn;
 332
 333     HMAP_FOR_EACH_WITH_HASH(rn, node, hash_route_data(rd), &route_map) {
 334         if (!memcmp(&rn->rd, rd, sizeof *rd)) {
 335             return rn;
 336         }
 337     }
 338
 339     return NULL;
 340 }
 341
 342 static struct route_node *
 343 route_node_lookup_by_ip(uint32_t ip)
 344 {
 345     int dst_len;
 346     struct route_node *rn, *rn_ret;
 347
 348     dst_len = -1;
 349     rn_ret  = NULL;
 350
 351     HMAP_FOR_EACH(rn, node, &route_map) {
 352         uint32_t mask = 0xffffffff << (32 - rn->rd.rtm_dst_len);
 353
 354         if (rn->rd.rta_dst == 0 && rn->rd.rtm_dst_len == 0) {
 355             /* Default route. */
 356             continue;
 357         }
 358
 359         if (rn->rd.rtm_dst_len > dst_len &&
 360             (ip & mask) == (rn->rd.rta_dst & mask)) {
 361             rn_ret  = rn;
 362             dst_len = rn->rd.rtm_dst_len;
 363         }
 364     }
 365
 366     return rn_ret;
 367 }
 368
 369 static void
 370 route_map_clear(void)
 371 {
 372     struct route_node *rn, *rn_next;
 373
 374     HMAP_FOR_EACH_SAFE(rn, rn_next, node, &route_map) {
 375         hmap_remove(&route_map, &rn->node);
 376         free(rn);
 377     }
 378 }
 379
 380 static uint32_t
 381 hash_route_data(const struct route_data *rd)
 382 {
 383     return hash_bytes(rd, sizeof *rd, 0);
 384 }
 385 \f
 386 /* name_table . */
 387
 388 static void
 389 name_table_init(void)
 390 {
 391     hmap_init(&name_map);
 392     name_notifier = rtnetlink_link_notifier_create(name_table_change, NULL);
 393     name_table_valid = false;
 394 }
 395
 396 static void
 397 name_table_uninit(void)
 398 {
 399     rtnetlink_link_notifier_destroy(name_notifier);
 400     name_notifier = NULL;
 401     name_map_clear();
 402     hmap_destroy(&name_map);
 403 }
 404
 405 static int
 406 name_table_reset(void)
 407 {
 408     struct nl_dump dump;
 409     struct rtgenmsg *rtmsg;
 410     struct ofpbuf request, reply;
 411
 412     name_table_valid = true;
 413     name_map_clear();
 414
 415     ofpbuf_init(&request, 0);
 416     nl_msg_put_nlmsghdr(&request, sizeof *rtmsg, RTM_GETLINK, NLM_F_REQUEST);
 417     rtmsg = ofpbuf_put_zeros(&request, sizeof *rtmsg);
 418     rtmsg->rtgen_family = AF_INET;
 419
 420     nl_dump_start(&dump, NETLINK_ROUTE, &request);
 421     ofpbuf_uninit(&request);
 422
 423     while (nl_dump_next(&dump, &reply)) {
 424         struct rtnetlink_link_change change;
 425
 426         if (rtnetlink_link_parse(&reply, &change)
 427             && change.nlmsg_type == RTM_NEWLINK
 428             && !name_node_lookup(change.ifi_index)) {
 429             struct name_node *nn;
 430
 431             nn = xzalloc(sizeof *nn);
 432             nn->ifi_index = change.ifi_index;
 433             ovs_strlcpy(nn->ifname, change.ifname, IFNAMSIZ);
 434             hmap_insert(&name_map, &nn->node, hash_int(nn->ifi_index, 0));
 435         }
 436     }
 437     return nl_dump_done(&dump);
 438 }
 439
 440 static void
 441 name_table_change(const struct rtnetlink_link_change *change OVS_UNUSED,
 442                   void *aux OVS_UNUSED)
 443 {
 444     /* Changes to interface status can cause routing table changes that some
 445      * versions of the linux kernel do not advertise for some reason. */
 446     route_table_valid = false;
 447     name_table_valid = false;
 448 }
 449
 450 static struct name_node *
 451 name_node_lookup(int ifi_index)
 452 {
 453     struct name_node *nn;
 454
 455     HMAP_FOR_EACH_WITH_HASH(nn, node, hash_int(ifi_index, 0), &name_map) {
 456         if (nn->ifi_index == ifi_index) {
 457             return nn;
 458         }
 459     }
 460
 461     return NULL;
 462 }
 463
 464 static void
 465 name_map_clear(void)
 466 {
 467     struct name_node *nn, *nn_next;
 468
 469     HMAP_FOR_EACH_SAFE(nn, nn_next, node, &name_map) {
 470         hmap_remove(&name_map, &nn->node);
 471         free(nn);
 472     }
 473 }