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