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