bgp.c

   1 /*
   2  * BGPv4
   3  * Used to advertise routes for upstream (l2tp port, rather than gratiutious
   4  * arp) and downstream--allowing routers to load-balance both.
   5  *
   6  * Implementation limitations:
   7  * - We never listen for incoming connections (session always initiated by us).
   8  * - Any routes advertised by the peer are accepted, but ignored.
   9  * - No password support; neither RFC1771 (which no-one seems to do anyway)
  10  *   nor RFC2385 (which requires a kernel patch on 2.4 kernels).
  11  */
  12
  13 char const *cvs_id_bgp = "$Id: bgp.c,v 1.12 2005/09/02 23:39:36 bodea Exp $";
  14
  15 #include <stdlib.h>
  16 #include <unistd.h>
  17 #include <string.h>
  18 #include <time.h>
  19 #include <errno.h>
  20 #include <sys/socket.h>
  21 #include <netinet/in.h>
  22 #include <arpa/inet.h>
  23 #include <netdb.h>
  24 #include <fcntl.h>
  25
  26 #include "l2tpns.h"
  27 #include "bgp.h"
  28 #include "util.h"
  29
  30 static void bgp_clear(struct bgp_peer *peer);
  31 static void bgp_set_retry(struct bgp_peer *peer);
  32 static void bgp_cidr(in_addr_t ip, in_addr_t mask, struct bgp_ip_prefix *pfx);
  33 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
  34     struct bgp_route_list *new);
  35
  36 static void bgp_free_routes(struct bgp_route_list *routes);
  37 static char const *bgp_msg_type_str(uint8_t type);
  38 static int bgp_connect(struct bgp_peer *peer);
  39 static int bgp_handle_connect(struct bgp_peer *peer);
  40 static int bgp_write(struct bgp_peer *peer);
  41 static int bgp_read(struct bgp_peer *peer);
  42 static int bgp_handle_input(struct bgp_peer *peer);
  43 static int bgp_send_open(struct bgp_peer *peer);
  44 static int bgp_send_keepalive(struct bgp_peer *peer);
  45 static int bgp_send_update(struct bgp_peer *peer);
  46 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
  47     uint8_t subcode);
  48 static int bgp_send_notification_full(struct bgp_peer *peer, uint8_t code,
  49     uint8_t subcode, char *notification_data, uint16_t data_len);
  50
  51 static uint16_t our_as;
  52 static struct bgp_route_list *bgp_routes = 0;
  53
  54 int bgp_configured = 0;
  55 struct bgp_peer *bgp_peers = 0;
  56
  57 /* prepare peer structure, globals */
  58 int bgp_setup(int as)
  59 {
  60     int i;
  61     struct bgp_peer *peer;
  62
  63     for (i = 0; i < BGP_NUM_PEERS; i++)
  64     {
  65         peer = &bgp_peers[i];
  66         memset(peer, 0, sizeof(*peer));
  67
  68         peer->addr = INADDR_NONE;
  69         peer->sock = -1;
  70         peer->state = peer->next_state = Disabled;
  71
  72         if (!((peer->outbuf = malloc(sizeof(*peer->outbuf)))
  73             && (peer->inbuf = malloc(sizeof(*peer->inbuf)))))
  74         {
  75             LOG(0, 0, 0, "Can't allocate buffers for bgp peer (%s)\n",
  76                 strerror(errno));
  77
  78             return 0;
  79         }
  80
  81         peer->edata.type = FD_TYPE_BGP;
  82         peer->edata.index = i;
  83         peer->events = 0;
  84     }
  85
  86     if (as < 1)
  87         as = 0;
  88
  89     if ((our_as = as))
  90         return 0;
  91
  92     bgp_routes = 0;
  93     bgp_configured = 0; /* set by bgp_start */
  94
  95     return 1;
  96 }
  97
  98 /* start connection with a peer */
  99 int bgp_start(struct bgp_peer *peer, char *name, int as, int keepalive,
 100     int hold, int enable)
 101 {
 102     struct hostent *h;
 103     int ibgp;
 104     int i;
 105     struct bgp_path_attr a;
 106     char path_attrs[64];
 107     char *p = path_attrs;
 108     in_addr_t ip;
 109     uint32_t metric = htonl(BGP_METRIC);
 110     uint32_t no_export = htonl(BGP_COMMUNITY_NO_EXPORT);
 111
 112     if (!our_as)
 113         return 0;
 114
 115     if (peer->state != Disabled)
 116         bgp_halt(peer);
 117
 118     snprintf(peer->name, sizeof(peer->name), "%s", name);
 119
 120     if (!(h = gethostbyname(name)) || h->h_addrtype != AF_INET)
 121     {
 122         LOG(0, 0, 0, "Can't get address for BGP peer %s (%s)\n",
 123             name, h ? "no address" : hstrerror(h_errno));
 124
 125         return 0;
 126     }
 127
 128     memcpy(&peer->addr, h->h_addr, sizeof(peer->addr));
 129     peer->as = as > 0 ? as : our_as;
 130     ibgp = peer->as == our_as;
 131
 132     /* set initial timer values */
 133     peer->init_keepalive = keepalive == -1 ? BGP_KEEPALIVE_TIME : keepalive;
 134     peer->init_hold = hold == -1 ? BGP_HOLD_TIME : hold;
 135
 136     if (peer->init_hold < 3)
 137         peer->init_hold = 3;
 138
 139     if (peer->init_keepalive * 3 > peer->init_hold)
 140         peer->init_keepalive = peer->init_hold / 3;
 141
 142     /* clear buffers, go to Idle state */
 143     peer->next_state = Idle;
 144     bgp_clear(peer);
 145
 146     /* set initial routing state */
 147     peer->routing = enable;
 148
 149     /* all our routes use the same attributes, so prepare it in advance */
 150     if (peer->path_attrs)
 151         free(peer->path_attrs);
 152
 153     peer->path_attr_len = 0;
 154
 155     /* ORIGIN */
 156     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 157     a.code = BGP_PATH_ATTR_CODE_ORIGIN;
 158     a.data.s.len = 1;
 159     a.data.s.value[0] = BGP_PATH_ATTR_CODE_ORIGIN_IGP;
 160
 161 #define ADD_ATTRIBUTE()         do { \
 162     i = BGP_PATH_ATTR_SIZE(a);  \
 163     memcpy(p, &a, i);           \
 164     p += i;                     \
 165     peer->path_attr_len += i;   } while (0)
 166
 167     ADD_ATTRIBUTE();
 168
 169     /* AS_PATH */
 170     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 171     a.code = BGP_PATH_ATTR_CODE_AS_PATH;
 172     if (ibgp)
 173     {
 174         /* empty path */
 175         a.data.s.len = 0;
 176     }
 177     else
 178     {
 179         /* just our AS */
 180         struct {
 181             uint8_t type;
 182             uint8_t len;
 183             uint16_t value;
 184         } as_path = {
 185             BGP_PATH_ATTR_CODE_AS_PATH_AS_SEQUENCE,
 186             1,
 187             htons(our_as),
 188         };
 189
 190         a.data.s.len = sizeof(as_path);
 191         memcpy(&a.data.s.value, &as_path, sizeof(as_path));
 192     }
 193
 194     ADD_ATTRIBUTE();
 195
 196     /* MULTI_EXIT_DISC */
 197     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL;
 198     a.code = BGP_PATH_ATTR_CODE_MULTI_EXIT_DISC;
 199     a.data.s.len = sizeof(metric);
 200     memcpy(a.data.s.value, &metric, sizeof(metric));
 201
 202     ADD_ATTRIBUTE();
 203
 204     if (ibgp)
 205     {
 206         uint32_t local_pref = htonl(BGP_LOCAL_PREF);
 207
 208         /* LOCAL_PREF */
 209         a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 210         a.code = BGP_PATH_ATTR_CODE_LOCAL_PREF;
 211         a.data.s.len = sizeof(local_pref);
 212         memcpy(a.data.s.value, &local_pref, sizeof(local_pref));
 213
 214         ADD_ATTRIBUTE();
 215     }
 216
 217     /* COMMUNITIES */
 218     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL | BGP_PATH_ATTR_FLAG_TRANS;
 219     a.code = BGP_PATH_ATTR_CODE_COMMUNITIES;
 220     a.data.s.len = sizeof(no_export);
 221     memcpy(a.data.s.value, &no_export, sizeof(no_export));
 222
 223     ADD_ATTRIBUTE();
 224
 225     /* remember the len before adding NEXT_HOP */
 226     peer->path_attr_len_without_nexthop = peer->path_attr_len;
 227
 228     /* NEXT_HOP */
 229     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 230     a.code = BGP_PATH_ATTR_CODE_NEXT_HOP;
 231     ip = my_address; /* we're it */
 232     a.data.s.len = sizeof(ip);
 233     memcpy(a.data.s.value, &ip, sizeof(ip));
 234
 235     ADD_ATTRIBUTE();
 236
 237     if (!(peer->path_attrs = malloc(peer->path_attr_len)))
 238     {
 239         LOG(0, 0, 0, "Can't allocate path_attrs for %s (%s)\n",
 240             name, strerror(errno));
 241
 242         return 0;
 243     }
 244
 245     memcpy(peer->path_attrs, path_attrs, peer->path_attr_len);
 246
 247     LOG(4, 0, 0, "Initiating BGP connection to %s (routing %s)\n",
 248         name, enable ? "enabled" : "suspended");
 249
 250     /* we have at least one peer configured */
 251     bgp_configured = 1;
 252
 253     /* connect */
 254     return bgp_connect(peer);
 255 }
 256
 257 /* clear counters, timers, routes and buffers; close socket; move to
 258    next_state, which may be Disabled or Idle */
 259 static void bgp_clear(struct bgp_peer *peer)
 260 {
 261     if (peer->sock != -1)
 262     {
 263         close(peer->sock);
 264         peer->sock = -1;
 265     }
 266
 267     peer->keepalive_time = 0;
 268     peer->expire_time = 0;
 269
 270     peer->keepalive = peer->init_keepalive;
 271     peer->hold = peer->init_hold;
 272
 273     bgp_free_routes(peer->routes);
 274     peer->routes = 0;
 275
 276     peer->outbuf->packet.header.len = 0;
 277     peer->outbuf->done = 0;
 278     peer->inbuf->packet.header.len = 0;
 279     peer->inbuf->done = 0;
 280
 281     peer->cli_flag = 0;
 282     peer->events = 0;
 283
 284     if (peer->state != peer->next_state)
 285     {
 286         peer->state = peer->next_state;
 287         peer->state_time = time_now;
 288
 289         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
 290             bgp_state_str(peer->next_state));
 291     }
 292 }
 293
 294 /* initiate a clean shutdown */
 295 void bgp_stop(struct bgp_peer *peer)
 296 {
 297     LOG(4, 0, 0, "Terminating BGP connection to %s\n", peer->name);
 298     bgp_send_notification(peer, BGP_ERR_CEASE, 0);
 299 }
 300
 301 /* drop connection (if any) and set state to Disabled */
 302 void bgp_halt(struct bgp_peer *peer)
 303 {
 304     LOG(4, 0, 0, "Aborting BGP connection to %s\n", peer->name);
 305     peer->next_state = Disabled;
 306     bgp_clear(peer);
 307 }
 308
 309 /* drop connection (if any) and set to Idle for connection retry */
 310 int bgp_restart(struct bgp_peer *peer)
 311 {
 312     peer->next_state = Idle;
 313     bgp_clear(peer);
 314
 315     /* restart now */
 316     peer->retry_time = time_now;
 317     peer->retry_count = 0;
 318
 319     /* connect */
 320     return bgp_connect(peer);
 321 }
 322
 323 static void bgp_set_retry(struct bgp_peer *peer)
 324 {
 325     if (peer->retry_count++ < BGP_MAX_RETRY)
 326     {
 327         peer->retry_time = time_now + (BGP_RETRY_BACKOFF * peer->retry_count);
 328         peer->next_state = Idle;
 329         bgp_clear(peer);
 330     }
 331     else
 332         bgp_halt(peer); /* give up */
 333 }
 334
 335 /* convert ip/mask to CIDR notation */
 336 static void bgp_cidr(in_addr_t ip, in_addr_t mask, struct bgp_ip_prefix *pfx)
 337 {
 338     int i;
 339     uint32_t b;
 340
 341     /* convert to prefix notation */
 342     pfx->len = 32;
 343     pfx->prefix = ip;
 344
 345     if (!mask) /* bogus */
 346         mask = 0xffffffff;
 347
 348     for (i = 0; i < 32 && ((b = ntohl(1 << i)), !(mask & b)); i++)
 349     {
 350         pfx->len--;
 351         pfx->prefix &= ~b;
 352     }
 353 }
 354
 355 /* insert route into list; sorted */
 356 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
 357     struct bgp_route_list *new)
 358 {
 359     struct bgp_route_list *p = head;
 360     struct bgp_route_list *e = 0;
 361
 362     while (p && memcmp(&p->dest, &new->dest, sizeof(p->dest)) < 0)
 363     {
 364         e = p;
 365         p = p->next;
 366     }
 367
 368     if (e)
 369     {
 370         new->next = e->next;
 371         e->next = new;
 372     }
 373     else
 374     {
 375         new->next = head;
 376         head = new;
 377     }
 378
 379     return head;
 380 }
 381
 382 /* add route to list for peers */
 383 /*
 384  * Note:  this doesn't do route aggregation, nor drop routes if a less
 385  * specific match already exists (partly because I'm lazy, but also so
 386  * that if that route is later deleted we don't have to be concerned
 387  * about adding back the more specific one).
 388  */
 389 int bgp_add_route(in_addr_t ip, in_addr_t mask)
 390 {
 391     struct bgp_route_list *r = bgp_routes;
 392     struct bgp_route_list add;
 393     int i;
 394
 395     bgp_cidr(ip, mask, &add.dest);
 396     add.next = 0;
 397
 398     /* check for duplicate */
 399     while (r)
 400     {
 401         i = memcmp(&r->dest, &add.dest, sizeof(r->dest));
 402         if (!i)
 403             return 1; /* already covered */
 404
 405         if (i > 0)
 406             break;
 407
 408         r = r->next;
 409     }
 410
 411     /* insert into route list; sorted */
 412     if (!(r = malloc(sizeof(*r))))
 413     {
 414         LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
 415             fmtaddr(add.dest.prefix, 0), add.dest.len, strerror(errno));
 416
 417         return 0;
 418     }
 419
 420     memcpy(r, &add, sizeof(*r));
 421     bgp_routes = bgp_insert_route(bgp_routes, r);
 422
 423     /* flag established peers for update */
 424     for (i = 0; i < BGP_NUM_PEERS; i++)
 425         if (bgp_peers[i].state == Established)
 426             bgp_peers[i].update_routes = 1;
 427
 428     LOG(4, 0, 0, "Registered BGP route %s/%d\n",
 429         fmtaddr(add.dest.prefix, 0), add.dest.len);
 430
 431     return 1;
 432 }
 433
 434 /* remove route from list for peers */
 435 int bgp_del_route(in_addr_t ip, in_addr_t mask)
 436 {
 437     struct bgp_route_list *r = bgp_routes;
 438     struct bgp_route_list *e = 0;
 439     struct bgp_route_list del;
 440     int i;
 441
 442     bgp_cidr(ip, mask, &del.dest);
 443     del.next = 0;
 444
 445     /* find entry in routes list and remove */
 446     while (r)
 447     {
 448         i = memcmp(&r->dest, &del.dest, sizeof(r->dest));
 449         if (!i)
 450         {
 451             if (e)
 452                 e->next = r->next;
 453             else
 454                 bgp_routes = r->next;
 455
 456             free(r);
 457             break;
 458         }
 459
 460         e = r;
 461
 462         if (i > 0)
 463             r = 0; /* stop */
 464         else
 465             r = r->next;
 466     }
 467
 468     /* not found */
 469     if (!r)
 470         return 1;
 471
 472     /* flag established peers for update */
 473     for (i = 0; i < BGP_NUM_PEERS; i++)
 474         if (bgp_peers[i].state == Established)
 475             bgp_peers[i].update_routes = 1;
 476
 477     LOG(4, 0, 0, "Removed BGP route %s/%d\n",
 478         fmtaddr(del.dest.prefix, 0), del.dest.len);
 479
 480     return 1;
 481 }
 482
 483 /* enable or disable routing */
 484 void bgp_enable_routing(int enable)
 485 {
 486     int i;
 487
 488     for (i = 0; i < BGP_NUM_PEERS; i++)
 489     {
 490         bgp_peers[i].routing = enable;
 491
 492         /* flag established peers for update */
 493         if (bgp_peers[i].state == Established)
 494             bgp_peers[i].update_routes = 1;
 495     }
 496
 497     LOG(4, 0, 0, "%s BGP routing\n", enable ? "Enabled" : "Suspended");
 498 }
 499
 500 #ifdef HAVE_EPOLL
 501 # include <sys/epoll.h>
 502 #else
 503 # include "fake_epoll.h"
 504 #endif
 505
 506 /* return a bitmask of the events required to poll this peer's fd */
 507 int bgp_set_poll()
 508 {
 509     int i;
 510
 511     if (!bgp_configured)
 512         return 0;
 513
 514     for (i = 0; i < BGP_NUM_PEERS; i++)
 515     {
 516         struct bgp_peer *peer = &bgp_peers[i];
 517         int events = 0;
 518
 519         if (peer->state == Disabled || peer->state == Idle)
 520             continue;
 521
 522         if (peer->inbuf->done < BGP_MAX_PACKET_SIZE)
 523             events |= EPOLLIN;
 524
 525         if (peer->state == Connect ||           /* connection in progress */
 526             peer->update_routes ||              /* routing updates */
 527             peer->outbuf->packet.header.len)    /* pending output */
 528             events |= EPOLLOUT;
 529
 530         if (peer->events != events)
 531         {
 532             struct epoll_event ev;
 533
 534             ev.events = peer->events = events;
 535             ev.data.ptr = &peer->edata;
 536             epoll_ctl(epollfd, EPOLL_CTL_MOD, peer->sock, &ev);
 537         }
 538     }
 539
 540     return 1;
 541 }
 542
 543 /* process bgp events/timers */
 544 int bgp_process(uint32_t events[])
 545 {
 546     int i;
 547
 548     if (!bgp_configured)
 549         return 0;
 550
 551     for (i = 0; i < BGP_NUM_PEERS; i++)
 552     {
 553         struct bgp_peer *peer = &bgp_peers[i];
 554
 555         if (*peer->name && peer->cli_flag == BGP_CLI_RESTART)
 556         {
 557             bgp_restart(peer);
 558             continue;
 559         }
 560
 561         if (peer->state == Disabled)
 562             continue;
 563
 564         if (peer->cli_flag)
 565         {
 566             switch (peer->cli_flag)
 567             {
 568             case BGP_CLI_SUSPEND:
 569                 if (peer->routing)
 570                 {
 571                     peer->routing = 0;
 572                     if (peer->state == Established)
 573                         peer->update_routes = 1;
 574                 }
 575
 576                 break;
 577
 578             case BGP_CLI_ENABLE:
 579                 if (!peer->routing)
 580                 {
 581                     peer->routing = 1;
 582                     if (peer->state == Established)
 583                         peer->update_routes = 1;
 584                 }
 585
 586                 break;
 587             }
 588
 589             peer->cli_flag = 0;
 590         }
 591
 592         /* handle empty/fill of buffers */
 593         if (events[i] & EPOLLOUT)
 594         {
 595             int r = 1;
 596             if (peer->state == Connect)
 597                 r = bgp_handle_connect(peer);
 598             else if (peer->outbuf->packet.header.len)
 599                 r = bgp_write(peer);
 600
 601             if (!r)
 602                 continue;
 603         }
 604
 605         if (events[i] & (EPOLLIN|EPOLLHUP))
 606         {
 607             if (!bgp_read(peer))
 608                 continue;
 609         }
 610
 611         /* process input buffer contents */
 612         while (peer->inbuf->done >= sizeof(peer->inbuf->packet.header)
 613             && !peer->outbuf->packet.header.len) /* may need to queue a response */
 614         {
 615             if (bgp_handle_input(peer) < 0)
 616                 continue;
 617         }
 618
 619         /* process pending updates */
 620         if (peer->update_routes
 621             && !peer->outbuf->packet.header.len) /* ditto */
 622         {
 623             if (!bgp_send_update(peer))
 624                 continue;
 625         }
 626
 627         /* process timers */
 628         if (peer->state == Established)
 629         {
 630             if (time_now > peer->expire_time)
 631             {
 632                 LOG(1, 0, 0, "No message from BGP peer %s in %ds\n",
 633                     peer->name, peer->hold);
 634
 635                 bgp_send_notification(peer, BGP_ERR_HOLD_TIMER_EXP, 0);
 636                 continue;
 637             }
 638
 639             if (time_now > peer->keepalive_time && !peer->outbuf->packet.header.len)
 640                 bgp_send_keepalive(peer);
 641         }
 642         else if (peer->state == Idle)
 643         {
 644             if (time_now > peer->retry_time)
 645                 bgp_connect(peer);
 646         }
 647         else if (time_now > peer->state_time + BGP_STATE_TIME)
 648         {
 649             LOG(1, 0, 0, "%s timer expired for BGP peer %s\n",
 650                 bgp_state_str(peer->state), peer->name);
 651
 652             bgp_restart(peer);
 653         }
 654     }
 655
 656     return 1;
 657 }
 658
 659 static void bgp_free_routes(struct bgp_route_list *routes)
 660 {
 661     struct bgp_route_list *tmp;
 662
 663     while ((tmp = routes))
 664     {
 665         routes = tmp->next;
 666         free(tmp);
 667     }
 668 }
 669
 670 char const *bgp_state_str(enum bgp_state state)
 671 {
 672     switch (state)
 673     {
 674     case Disabled:      return "Disabled";
 675     case Idle:          return "Idle";
 676     case Connect:       return "Connect";
 677     case Active:        return "Active";
 678     case OpenSent:      return "OpenSent";
 679     case OpenConfirm:   return "OpenConfirm";
 680     case Established:   return "Established";
 681     }
 682
 683     return "?";
 684 }
 685
 686 static char const *bgp_msg_type_str(uint8_t type)
 687 {
 688     switch (type)
 689     {
 690     case BGP_MSG_OPEN:          return "OPEN";
 691     case BGP_MSG_UPDATE:        return "UPDATE";
 692     case BGP_MSG_NOTIFICATION:  return "NOTIFICATION";
 693     case BGP_MSG_KEEPALIVE:     return "KEEPALIVE";
 694     }
 695
 696     return "?";
 697 }
 698
 699 /* attempt to connect to peer */
 700 static int bgp_connect(struct bgp_peer *peer)
 701 {
 702     static int bgp_port = 0;
 703     struct sockaddr_in addr;
 704     struct epoll_event ev;
 705
 706     if (!bgp_port)
 707     {
 708         struct servent *serv;
 709         if (!(serv = getservbyname("bgp", "tcp")))
 710         {
 711             LOG(0, 0, 0, "Can't get bgp service (%s)\n", strerror(errno));
 712             return 0;
 713         }
 714
 715         bgp_port = serv->s_port;
 716     }
 717
 718     if ((peer->sock = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
 719     {
 720         LOG(0, 0, 0, "Can't create a socket for BGP peer %s (%s)\n",
 721             peer->name, strerror(errno));
 722
 723         peer->state = peer->next_state = Disabled;
 724         return 0;
 725     }
 726
 727     /* add to poll set */
 728     ev.events = peer->events = EPOLLOUT;
 729     ev.data.ptr = &peer->edata;
 730     epoll_ctl(epollfd, EPOLL_CTL_ADD, peer->sock, &ev);
 731
 732     /* set to non-blocking */
 733     fcntl(peer->sock, F_SETFL, fcntl(peer->sock, F_GETFL, 0) | O_NONBLOCK);
 734
 735     /* try connect */
 736     memset(&addr, 0, sizeof(addr));
 737     addr.sin_family = AF_INET;
 738     addr.sin_port = bgp_port;
 739     addr.sin_addr.s_addr = peer->addr;
 740
 741     while (connect(peer->sock, (struct sockaddr *) &addr, sizeof(addr)) == -1)
 742     {
 743         if (errno == EINTR) /* SIGALARM handler */
 744             continue;
 745
 746         if (errno != EINPROGRESS)
 747         {
 748             LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n",
 749                 inet_ntoa(addr.sin_addr), strerror(errno));
 750
 751             bgp_set_retry(peer);
 752             return 0;
 753         }
 754
 755         peer->state = Connect;
 756         peer->state_time = time_now;
 757
 758         LOG(4, 0, 0, "BGP peer %s: state Connect\n", peer->name);
 759         return 1;
 760     }
 761
 762     peer->state = Active;
 763     peer->state_time = time_now;
 764     peer->retry_time = peer->retry_count = 0;
 765
 766     LOG(4, 0, 0, "BGP peer %s: state Active\n", inet_ntoa(addr.sin_addr));
 767
 768     peer->handle_ipv6_routes = 0;
 769
 770     return bgp_send_open(peer);
 771 }
 772
 773 /* complete partial connection (state = Connect) */
 774 static int bgp_handle_connect(struct bgp_peer *peer)
 775 {
 776     int err = 0;
 777     socklen_t len = sizeof(int);
 778     getsockopt(peer->sock, SOL_SOCKET, SO_ERROR, &err, &len);
 779     if (err)
 780     {
 781         LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n", peer->name,
 782             strerror(err));
 783
 784         bgp_set_retry(peer);
 785         return 0;
 786     }
 787
 788     peer->state = Active;
 789     peer->state_time = time_now;
 790
 791     LOG(4, 0, 0, "BGP peer %s: state Active\n", peer->name);
 792
 793     peer->handle_ipv6_routes = 0;
 794
 795     return bgp_send_open(peer);
 796 }
 797
 798 /* initiate a write */
 799 static int bgp_write(struct bgp_peer *peer)
 800 {
 801     int len = htons(peer->outbuf->packet.header.len);
 802     int r;
 803
 804     while ((r = write(peer->sock, &peer->outbuf->packet + peer->outbuf->done,
 805         len - peer->outbuf->done)) == -1)
 806     {
 807         if (errno == EINTR)
 808             continue;
 809
 810         if (errno == EAGAIN)
 811             return 1;
 812
 813         if (errno == EPIPE)
 814             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
 815         else
 816             LOG(1, 0, 0, "Can't write to BGP peer %s (%s)\n", peer->name,
 817                 strerror(errno));
 818
 819         bgp_set_retry(peer);
 820         return 0;
 821     }
 822
 823     if (r < len)
 824     {
 825         peer->outbuf->done += r;
 826         return 1;
 827     }
 828
 829     LOG(4, 0, 0, "Sent %s to BGP peer %s\n",
 830         bgp_msg_type_str(peer->outbuf->packet.header.type), peer->name);
 831
 832     peer->outbuf->packet.header.len = 0;
 833     peer->outbuf->done = 0;
 834
 835     if (peer->state == Established)
 836         peer->keepalive_time = time_now + peer->keepalive;
 837
 838     if (peer->state != peer->next_state)
 839     {
 840         if (peer->next_state == Disabled || peer->next_state == Idle)
 841         {
 842             bgp_clear(peer);
 843             return 0;
 844         }
 845
 846         peer->state = peer->next_state;
 847         peer->state_time = time_now;
 848
 849         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
 850             bgp_state_str(peer->state));
 851     }
 852
 853     return 1;
 854 }
 855
 856 /* initiate a read */
 857 static int bgp_read(struct bgp_peer *peer)
 858 {
 859     int r;
 860
 861     while ((r = read(peer->sock, &peer->inbuf->packet + peer->inbuf->done,
 862         BGP_MAX_PACKET_SIZE - peer->inbuf->done)) < 1)
 863     {
 864         if (!r)
 865         {
 866             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
 867         }
 868         else
 869         {
 870             if (errno == EINTR)
 871                 continue;
 872
 873             if (errno == EAGAIN)
 874                 return 1;
 875
 876             LOG(1, 0, 0, "Can't read from BGP peer %s (%s)\n", peer->name,
 877                 strerror(errno));
 878         }
 879
 880         bgp_set_retry(peer);
 881         return 0;
 882     }
 883
 884     peer->inbuf->done += r;
 885     return 1;
 886 }
 887
 888 /* process buffered packets */
 889 static int bgp_handle_input(struct bgp_peer *peer)
 890 {
 891     struct bgp_packet *p = &peer->inbuf->packet;
 892     int len = ntohs(p->header.len);
 893
 894     if (len > BGP_MAX_PACKET_SIZE)
 895     {
 896         LOG(1, 0, 0, "Bad header length from BGP %s\n", peer->name);
 897         bgp_send_notification(peer, BGP_ERR_HEADER, BGP_ERR_HDR_BAD_LEN);
 898         return 0;
 899     }
 900
 901     if (peer->inbuf->done < len)
 902         return 0;
 903
 904     LOG(4, 0, 0, "Received %s from BGP peer %s\n",
 905         bgp_msg_type_str(p->header.type), peer->name);
 906
 907     switch (p->header.type)
 908     {
 909     case BGP_MSG_OPEN:
 910         {
 911             struct bgp_data_open data;
 912             int hold;
 913             int i;
 914             off_t param_offset, capability_offset;
 915             struct bgp_opt_param *param;
 916             uint8_t capabilities_len;
 917             char *capabilities = NULL;
 918             struct bgp_capability *capability;
 919             struct bgp_mp_cap_param *mp_cap;
 920
 921             for (i = 0; i < sizeof(p->header.marker); i++)
 922             {
 923                 if ((unsigned char) p->header.marker[i] != 0xff)
 924                 {
 925                     LOG(1, 0, 0, "Invalid marker from BGP peer %s\n",
 926                         peer->name);
 927
 928                     bgp_send_notification(peer, BGP_ERR_HEADER,
 929                         BGP_ERR_HDR_NOT_SYNC);
 930
 931                     return 0;
 932                 }
 933             }
 934
 935             if (peer->state != OpenSent)
 936             {
 937                 LOG(1, 0, 0, "OPEN from BGP peer %s in %s state\n",
 938                     peer->name, bgp_state_str(peer->state));
 939
 940                 bgp_send_notification(peer, BGP_ERR_FSM, 0);
 941                 return 0;
 942             }
 943
 944             memcpy(&data, p->data, len - sizeof(p->header));
 945
 946             if (data.version != BGP_VERSION)
 947             {
 948                 LOG(1, 0, 0, "Bad version (%d) sent by BGP peer %s\n",
 949                     (int) data.version, peer->name);
 950
 951                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_VERSION);
 952                 return 0;
 953             }
 954
 955             if (ntohs(data.as) != peer->as)
 956             {
 957                 LOG(1, 0, 0, "Bad AS sent by BGP peer %s (got %d, "
 958                     "expected %d)\n", peer->name, (int) htons(data.as),
 959                     (int) peer->as);
 960
 961                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_BAD_AS);
 962                 return 0;
 963             }
 964
 965             if ((hold = ntohs(data.hold_time)) < 3)
 966             {
 967                 LOG(1, 0, 0, "Bad hold time (%d) from BGP peer %s\n",
 968                     hold, peer->name);
 969
 970                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_HOLD_TIME);
 971                 return 0;
 972             }
 973
 974             /* pick lowest hold time */
 975             if (hold < peer->hold)
 976                 peer->hold = hold;
 977
 978             /* adjust our keepalive based on negotiated hold value */
 979             if (peer->keepalive * 3 > peer->hold)
 980                 peer->keepalive = peer->hold / 3;
 981
 982             /* check for optional parameters */
 983             /* 2 is for the size of type + len (both uint8_t) */
 984             for (param_offset = 0;
 985                     param_offset < data.opt_len;
 986                     param_offset += 2 + param->len)
 987             {
 988                 param = (struct bgp_opt_param *)(&data.opt_params + param_offset);
 989
 990                 /* sensible check */
 991                 if (data.opt_len - param_offset < 2
 992                         || param->len > data.opt_len - param_offset - 2)
 993                 {
 994                     LOG(1, 0, 0, "Malformed Optional Parameter list from BGP peer %s\n",
 995                         peer->name);
 996
 997                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
 998                     return 0;
 999                 }
1000
1001                 /* we know only one parameter type */
1002                 if (param->type != BGP_PARAM_TYPE_CAPABILITY)
1003                 {
1004                     LOG(1, 0, 0, "Unsupported Optional Parameter type %d from BGP peer %s\n",
1005                         param->type, peer->name);
1006
1007                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_PARAM);
1008                     return 0;
1009                 }
1010
1011                 capabilities_len = param->len;
1012                 capabilities = (char *)&param->value;
1013             }
1014
1015             /* look for BGP multiprotocol capability */
1016             if (capabilities)
1017             {
1018                 for (capability_offset = 0;
1019                         capability_offset < capabilities_len;
1020                         capability_offset += 2 + capability->len)
1021                 {
1022                     capability = (struct bgp_capability *)(capabilities + capability_offset);
1023
1024                     /* sensible check */
1025                     if (capabilities_len - capability_offset < 2
1026                             || capability->len > capabilities_len - capability_offset - 2)
1027                     {
1028                         LOG(1, 0, 0, "Malformed Capabilities list from BGP peer %s\n",
1029                             peer->name);
1030
1031                         bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
1032                         return 0;
1033                     }
1034
1035                     /* we only know one capability code */
1036                     if (capability->code != BGP_CAP_CODE_MP
1037                             && capability->len != sizeof(struct bgp_mp_cap_param))
1038                     {
1039                         LOG(4, 0, 0, "Unsupported Capability code %d from BGP peer %s\n",
1040                             capability->code, peer->name);
1041
1042                         bgp_send_notification_full(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_CAP,
1043                                 (char *)capability, 2 + capability->len);
1044                         /* we don't terminate, still; we just jump to the next one */
1045                         continue;
1046                     }
1047
1048                     mp_cap = (struct bgp_mp_cap_param *)&capability->value;
1049                     /* the only <AFI, SAFI> tuple we support */
1050                     if (ntohs(mp_cap->afi) != AF_INET6 && mp_cap->safi != BGP_MP_SAFI_UNICAST)
1051                     {
1052                         LOG(4, 0, 0, "Unsupported multiprotocol AFI %d and SAFI %d from BGP peer %s\n",
1053                             mp_cap->afi, mp_cap->safi, peer->name);
1054
1055                         bgp_send_notification_full(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_CAP,
1056                                 (char *)capability, 2 + capability->len);
1057                         /* we don't terminate, still; we just jump to the next one */
1058                         continue;
1059                     }
1060
1061                     peer->handle_ipv6_routes = 1;
1062                 }
1063             }
1064
1065             /* next transition requires an exchange of keepalives */
1066             bgp_send_keepalive(peer);
1067         }
1068
1069         break;
1070
1071     case BGP_MSG_KEEPALIVE:
1072         if (peer->state == OpenConfirm)
1073         {
1074             peer->state = peer->next_state = Established;
1075             peer->state_time = time_now;
1076             peer->keepalive_time = time_now + peer->keepalive;
1077             peer->update_routes = 1;
1078             peer->retry_count = 0;
1079             peer->retry_time = 0;
1080
1081             LOG(4, 0, 0, "BGP peer %s: state Established\n", peer->name);
1082         }
1083
1084         break;
1085
1086     case BGP_MSG_NOTIFICATION:
1087         if (len > sizeof(p->header))
1088         {
1089             struct bgp_data_notification *notification =
1090                 (struct bgp_data_notification *) p->data;
1091
1092             if (notification->error_code == BGP_ERR_CEASE)
1093             {
1094                 LOG(4, 0, 0, "BGP peer %s sent CEASE\n", peer->name);
1095                 bgp_restart(peer);
1096                 return 0;
1097             }
1098
1099             if (notification->error_code == BGP_ERR_OPEN
1100                     && notification->error_subcode == BGP_ERR_OPN_UNSUP_CAP)
1101             {
1102                 /* the only capability we advertise is this one, so upon receiving
1103                    an "unsupported capability" message, we disable IPv6 routes for
1104                    this peer */
1105                 LOG(4, 0, 0, "BGP peer %s doesn't support IPv6 routes advertisement\n", peer->name);
1106                 peer->handle_ipv6_routes = 0;
1107                 break;
1108             }
1109
1110             /* FIXME: should handle more notifications */
1111             LOG(4, 0, 0, "BGP peer %s sent unhandled NOTIFICATION %d\n",
1112                 peer->name, (int) notification->error_code);
1113         }
1114
1115         break;
1116     }
1117
1118     /* reset timer */
1119     peer->expire_time = time_now + peer->hold;
1120
1121     /* see if there's another message in the same packet/buffer */
1122     if (peer->inbuf->done > len)
1123     {
1124         peer->inbuf->done -= len;
1125         memmove(p, (char *) p + len, peer->inbuf->done);
1126     }
1127     else
1128     {
1129         peer->inbuf->packet.header.len = 0;
1130         peer->inbuf->done = 0;
1131     }
1132
1133     return peer->inbuf->done;
1134 }
1135
1136 /* send/buffer OPEN message */
1137 static int bgp_send_open(struct bgp_peer *peer)
1138 {
1139     struct bgp_data_open data;
1140     struct bgp_mp_cap_param mp_ipv6 = { htons(AF_INET6), 0, BGP_MP_SAFI_UNICAST };
1141     struct bgp_capability cap_mp_ipv6;
1142     struct bgp_opt_param param_cap_mp_ipv6;
1143     uint16_t len = sizeof(peer->outbuf->packet.header);
1144
1145     memset(peer->outbuf->packet.header.marker, 0xff,
1146         sizeof(peer->outbuf->packet.header.marker));
1147
1148     peer->outbuf->packet.header.type = BGP_MSG_OPEN;
1149
1150     data.version = BGP_VERSION;
1151     data.as = htons(our_as);
1152     data.hold_time = htons(peer->hold);
1153     data.identifier = my_address;
1154
1155     /* construct the param and capability */
1156     cap_mp_ipv6.code = BGP_CAP_CODE_MP;
1157     cap_mp_ipv6.len = sizeof(mp_ipv6);
1158     memcpy(&cap_mp_ipv6.value, &mp_ipv6, cap_mp_ipv6.len);
1159
1160     param_cap_mp_ipv6.type = BGP_PARAM_TYPE_CAPABILITY;
1161     param_cap_mp_ipv6.len = 2 + sizeof(mp_ipv6);
1162     memcpy(&param_cap_mp_ipv6.value, &cap_mp_ipv6, param_cap_mp_ipv6.len);
1163
1164     data.opt_len = 2 + param_cap_mp_ipv6.len;
1165     memcpy(&data.opt_params, &param_cap_mp_ipv6, data.opt_len);
1166
1167     memcpy(peer->outbuf->packet.data, &data, BGP_DATA_OPEN_SIZE);
1168     len += BGP_DATA_OPEN_SIZE;
1169
1170     peer->outbuf->packet.header.len = htons(len);
1171     peer->outbuf->done = 0;
1172     peer->next_state = OpenSent;
1173
1174     return bgp_write(peer);
1175 }
1176
1177 /* send/buffer KEEPALIVE message */
1178 static int bgp_send_keepalive(struct bgp_peer *peer)
1179 {
1180     memset(peer->outbuf->packet.header.marker, 0xff,
1181         sizeof(peer->outbuf->packet.header.marker));
1182
1183     peer->outbuf->packet.header.type = BGP_MSG_KEEPALIVE;
1184     peer->outbuf->packet.header.len =
1185         htons(sizeof(peer->outbuf->packet.header));
1186
1187     peer->outbuf->done = 0;
1188     peer->next_state = (peer->state == OpenSent) ? OpenConfirm : peer->state;
1189
1190     return bgp_write(peer);
1191 }
1192
1193 /* send/buffer UPDATE message */
1194 static int bgp_send_update(struct bgp_peer *peer)
1195 {
1196     uint16_t unf_len = 0;
1197     uint16_t attr_len;
1198     uint16_t len = sizeof(peer->outbuf->packet.header);
1199     struct bgp_route_list *have = peer->routes;
1200     struct bgp_route_list *want = peer->routing ? bgp_routes : 0;
1201     struct bgp_route_list *e = 0;
1202     struct bgp_route_list *add = 0;
1203     int s;
1204
1205     char *data = (char *) &peer->outbuf->packet.data;
1206
1207     /* need leave room for attr_len, bgp_path_attrs and one prefix */
1208     char *max = (char *) &peer->outbuf->packet.data
1209         + sizeof(peer->outbuf->packet.data)
1210         - sizeof(attr_len) - peer->path_attr_len - sizeof(struct bgp_ip_prefix);
1211
1212     /* skip over unf_len */
1213     data += sizeof(unf_len);
1214     len += sizeof(unf_len);
1215
1216     memset(peer->outbuf->packet.header.marker, 0xff,
1217         sizeof(peer->outbuf->packet.header.marker));
1218
1219     peer->outbuf->packet.header.type = BGP_MSG_UPDATE;
1220
1221     peer->update_routes = 0; /* tentatively clear */
1222
1223     /* find differences */
1224     while ((have || want) && data < (max - sizeof(struct bgp_ip_prefix)))
1225     {
1226         if (have)
1227             s = want
1228                 ? memcmp(&have->dest, &want->dest, sizeof(have->dest))
1229                 : -1;
1230         else
1231             s = 1;
1232
1233         if (s < 0) /* found one to delete */
1234         {
1235             struct bgp_route_list *tmp = have;
1236             have = have->next;
1237
1238             s = BGP_IP_PREFIX_SIZE(tmp->dest);
1239             memcpy(data, &tmp->dest, s);
1240             data += s;
1241             unf_len += s;
1242             len += s;
1243
1244             LOG(5, 0, 0, "Withdrawing route %s/%d from BGP peer %s\n",
1245                 fmtaddr(tmp->dest.prefix, 0), tmp->dest.len, peer->name);
1246
1247             free(tmp);
1248
1249             if (e)
1250                 e->next = have;
1251             else
1252                 peer->routes = have;
1253         }
1254         else
1255         {
1256             if (!s) /* same */
1257             {
1258                 e = have; /* stash the last found to relink above */
1259                 have = have->next;
1260                 want = want->next;
1261             }
1262             else if (s > 0) /* addition reqd. */
1263             {
1264                 if (add)
1265                 {
1266                     peer->update_routes = 1; /* only one add per packet */
1267                     if (!have)
1268                         break;
1269                 }
1270                 else
1271                     add = want;
1272
1273                 if (want)
1274                     want = want->next;
1275             }
1276         }
1277     }
1278
1279     if (have || want)
1280         peer->update_routes = 1; /* more to do */
1281
1282     /* anything changed? */
1283     if (!(unf_len || add))
1284         return 1;
1285
1286     /* go back and insert unf_len */
1287     unf_len = htons(unf_len);
1288     memcpy(&peer->outbuf->packet.data, &unf_len, sizeof(unf_len));
1289
1290     if (add)
1291     {
1292         if (!(e = malloc(sizeof(*e))))
1293         {
1294             LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
1295                 fmtaddr(add->dest.prefix, 0), add->dest.len, strerror(errno));
1296
1297             return 0;
1298         }
1299
1300         memcpy(e, add, sizeof(*e));
1301         e->next = 0;
1302         peer->routes = bgp_insert_route(peer->routes, e);
1303
1304         attr_len = htons(peer->path_attr_len);
1305         memcpy(data, &attr_len, sizeof(attr_len));
1306         data += sizeof(attr_len);
1307         len += sizeof(attr_len);
1308
1309         memcpy(data, peer->path_attrs, peer->path_attr_len);
1310         data += peer->path_attr_len;
1311         len += peer->path_attr_len;
1312
1313         s = BGP_IP_PREFIX_SIZE(add->dest);
1314         memcpy(data, &add->dest, s);
1315         data += s;
1316         len += s;
1317
1318         LOG(5, 0, 0, "Advertising route %s/%d to BGP peer %s\n",
1319             fmtaddr(add->dest.prefix, 0), add->dest.len, peer->name);
1320     }
1321     else
1322     {
1323         attr_len = 0;
1324         memcpy(data, &attr_len, sizeof(attr_len));
1325         data += sizeof(attr_len);
1326         len += sizeof(attr_len);
1327     }
1328
1329     peer->outbuf->packet.header.len = htons(len);
1330     peer->outbuf->done = 0;
1331
1332     return bgp_write(peer);
1333 }
1334
1335 /* send/buffer NOTIFICATION message */
1336 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
1337     uint8_t subcode)
1338 {
1339     return bgp_send_notification_full(peer, code, subcode, NULL, 0);
1340 }
1341
1342 static int bgp_send_notification_full(struct bgp_peer *peer, uint8_t code,
1343     uint8_t subcode, char *notification_data, uint16_t data_len)
1344 {
1345     struct bgp_data_notification data;
1346     uint16_t len = 0;
1347
1348     data.error_code = code;
1349     len += sizeof(data.error_code);
1350
1351     data.error_subcode = subcode;
1352     len += sizeof(data.error_code);
1353
1354     memcpy(data.data, notification_data, data_len);
1355     len += data_len;
1356
1357     memset(peer->outbuf->packet.header.marker, 0xff,
1358         sizeof(peer->outbuf->packet.header.marker));
1359
1360     peer->outbuf->packet.header.type = BGP_MSG_NOTIFICATION;
1361     peer->outbuf->packet.header.len =
1362         htons(sizeof(peer->outbuf->packet.header) + len);
1363
1364     memcpy(peer->outbuf->packet.data, &data, len);
1365
1366     peer->outbuf->done = 0;
1367     peer->next_state = code == BGP_ERR_CEASE ? Disabled : Idle;
1368
1369     /* we're dying; ignore any pending input */
1370     peer->inbuf->packet.header.len = 0;
1371     peer->inbuf->done = 0;
1372
1373     return bgp_write(peer);
1374 }