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 #include <stdlib.h>
  14 #include <unistd.h>
  15 #include <string.h>
  16 #include <time.h>
  17 #include <errno.h>
  18 #include <sys/socket.h>
  19 #include <netinet/in.h>
  20 #include <arpa/inet.h>
  21 #include <netdb.h>
  22 #include <fcntl.h>
  23
  24 #include "l2tpns.h"
  25 #include "bgp.h"
  26 #include "util.h"
  27
  28 static void bgp_clear(struct bgp_peer *peer);
  29 static void bgp_set_retry(struct bgp_peer *peer);
  30 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
  31     struct bgp_route_list *new);
  32 static struct bgp_route6_list *bgp_insert_route6(struct bgp_route6_list *head,
  33     struct bgp_route6_list *new);
  34
  35 static void bgp_process_timers(struct bgp_peer *peer);
  36 static void bgp_free_routes(struct bgp_route_list *routes);
  37 static void bgp_free_routes6(struct bgp_route6_list *routes);
  38 static char const *bgp_msg_type_str(uint8_t type);
  39 static int bgp_connect(struct bgp_peer *peer);
  40 static int bgp_handle_connect(struct bgp_peer *peer);
  41 static int bgp_write(struct bgp_peer *peer);
  42 static int bgp_read(struct bgp_peer *peer);
  43 static int bgp_handle_input(struct bgp_peer *peer);
  44 static int bgp_send_open(struct bgp_peer *peer);
  45 static int bgp_send_keepalive(struct bgp_peer *peer);
  46 static int bgp_send_update(struct bgp_peer *peer);
  47 static int bgp_send_update6(struct bgp_peer *peer);
  48 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
  49     uint8_t subcode);
  50
  51 static uint16_t our_as;
  52 static struct bgp_route_list *bgp_routes = 0;
  53 static struct bgp_route6_list *bgp_routes6 = 0;
  54
  55 int bgp_configured = 0;
  56 struct bgp_peer *bgp_peers = 0;
  57
  58 /* prepare peer structure, globals */
  59 int bgp_setup(int as)
  60 {
  61     int i;
  62     struct bgp_peer *peer;
  63
  64     for (i = 0; i < BGP_NUM_PEERS; i++)
  65     {
  66         peer = &bgp_peers[i];
  67         memset(peer, 0, sizeof(*peer));
  68
  69         peer->addr = INADDR_NONE;
  70         peer->sock = -1;
  71         peer->state = peer->next_state = Disabled;
  72
  73         if (!((peer->outbuf = malloc(sizeof(*peer->outbuf)))
  74             && (peer->inbuf = malloc(sizeof(*peer->inbuf)))))
  75         {
  76             LOG(0, 0, 0, "Can't allocate buffers for bgp peer (%s)\n",
  77                 strerror(errno));
  78
  79             return 0;
  80         }
  81
  82         peer->edata.type = FD_TYPE_BGP;
  83         peer->edata.index = i;
  84         peer->events = 0;
  85     }
  86
  87     if (as < 1)
  88         as = 0;
  89
  90     if ((our_as = as))
  91         return 0;
  92
  93     bgp_routes = 0;
  94     bgp_routes6 = 0;
  95     bgp_configured = 0; /* set by bgp_start */
  96
  97     return 1;
  98 }
  99
 100 /* start connection with a peer */
 101 int bgp_start(struct bgp_peer *peer, char *name, int as, int keepalive,
 102     int hold, struct in_addr update_source, int enable)
 103 {
 104     struct hostent *h;
 105     int ibgp;
 106     int i;
 107     struct bgp_path_attr a;
 108     char path_attrs[64];
 109     char *p = path_attrs;
 110     in_addr_t ip;
 111     uint32_t metric = htonl(BGP_METRIC);
 112     uint32_t no_export = htonl(BGP_COMMUNITY_NO_EXPORT);
 113
 114     if (!our_as)
 115         return 0;
 116
 117     if (peer->state != Disabled)
 118         bgp_halt(peer);
 119
 120     snprintf(peer->name, sizeof(peer->name), "%s", name);
 121
 122     if (!(h = gethostbyname(name)) || h->h_addrtype != AF_INET)
 123     {
 124         LOG(0, 0, 0, "Can't get address for BGP peer %s (%s)\n",
 125             name, h ? "no address" : hstrerror(h_errno));
 126
 127         return 0;
 128     }
 129
 130     memcpy(&peer->addr, h->h_addr, sizeof(peer->addr));
 131     peer->source_addr = update_source.s_addr;
 132     peer->as = as > 0 ? as : our_as;
 133     ibgp = peer->as == our_as;
 134
 135     /* set initial timer values */
 136     peer->init_keepalive = keepalive == -1 ? BGP_KEEPALIVE_TIME : keepalive;
 137     peer->init_hold = hold == -1 ? BGP_HOLD_TIME : hold;
 138
 139     if (peer->init_hold < 3)
 140         peer->init_hold = 3;
 141
 142     if (peer->init_keepalive * 3 > peer->init_hold)
 143         peer->init_keepalive = peer->init_hold / 3;
 144
 145     /* clear buffers, go to Idle state */
 146     peer->next_state = Idle;
 147     bgp_clear(peer);
 148
 149     /* set initial routing state */
 150     peer->routing = enable;
 151
 152     /* all our routes use the same attributes, so prepare it in advance */
 153     if (peer->path_attrs)
 154         free(peer->path_attrs);
 155
 156     peer->path_attr_len = 0;
 157
 158     /* ORIGIN */
 159     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 160     a.code = BGP_PATH_ATTR_CODE_ORIGIN;
 161     a.data.s.len = 1;
 162     a.data.s.value[0] = BGP_PATH_ATTR_CODE_ORIGIN_IGP;
 163
 164 #define ADD_ATTRIBUTE()         do { \
 165     i = BGP_PATH_ATTR_SIZE(a);  \
 166     memcpy(p, &a, i);           \
 167     p += i;                     \
 168     peer->path_attr_len += i;   } while (0)
 169
 170     ADD_ATTRIBUTE();
 171
 172     /* AS_PATH */
 173     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 174     a.code = BGP_PATH_ATTR_CODE_AS_PATH;
 175     if (ibgp)
 176     {
 177         /* empty path */
 178         a.data.s.len = 0;
 179     }
 180     else
 181     {
 182         /* just our AS */
 183         struct {
 184             uint8_t type;
 185             uint8_t len;
 186             uint16_t value;
 187         } as_path = {
 188             BGP_PATH_ATTR_CODE_AS_PATH_AS_SEQUENCE,
 189             1,
 190             htons(our_as),
 191         };
 192
 193         a.data.s.len = sizeof(as_path);
 194         memcpy(&a.data.s.value, &as_path, sizeof(as_path));
 195     }
 196
 197     ADD_ATTRIBUTE();
 198
 199     /* MULTI_EXIT_DISC */
 200     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL;
 201     a.code = BGP_PATH_ATTR_CODE_MULTI_EXIT_DISC;
 202     a.data.s.len = sizeof(metric);
 203     memcpy(a.data.s.value, &metric, sizeof(metric));
 204
 205     ADD_ATTRIBUTE();
 206
 207     if (ibgp)
 208     {
 209         uint32_t local_pref = htonl(BGP_LOCAL_PREF);
 210
 211         /* LOCAL_PREF */
 212         a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 213         a.code = BGP_PATH_ATTR_CODE_LOCAL_PREF;
 214         a.data.s.len = sizeof(local_pref);
 215         memcpy(a.data.s.value, &local_pref, sizeof(local_pref));
 216
 217         ADD_ATTRIBUTE();
 218     }
 219
 220     /* COMMUNITIES */
 221     a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL | BGP_PATH_ATTR_FLAG_TRANS;
 222     a.code = BGP_PATH_ATTR_CODE_COMMUNITIES;
 223     a.data.s.len = sizeof(no_export);
 224     memcpy(a.data.s.value, &no_export, sizeof(no_export));
 225
 226     ADD_ATTRIBUTE();
 227
 228     /* remember the len before adding NEXT_HOP */
 229     peer->path_attr_len_without_nexthop = peer->path_attr_len;
 230
 231     /* NEXT_HOP */
 232     a.flags = BGP_PATH_ATTR_FLAG_TRANS;
 233     a.code = BGP_PATH_ATTR_CODE_NEXT_HOP;
 234     if (config->nexthop_address)
 235     {
 236         ip = config->nexthop_address;
 237     }
 238     else
 239     {
 240         ip = my_address; /* we're it */
 241     }
 242     a.data.s.len = sizeof(ip);
 243     memcpy(a.data.s.value, &ip, sizeof(ip));
 244
 245     ADD_ATTRIBUTE();
 246
 247     if (!(peer->path_attrs = malloc(peer->path_attr_len)))
 248     {
 249         LOG(0, 0, 0, "Can't allocate path_attrs for %s (%s)\n",
 250             name, strerror(errno));
 251
 252         return 0;
 253     }
 254
 255     memcpy(peer->path_attrs, path_attrs, peer->path_attr_len);
 256
 257     /* multiprotocol attributes initialization */
 258     if (config->ipv6_prefix.s6_addr[0])
 259     {
 260         struct bgp_attr_mp_reach_nlri_partial mp_reach_nlri_partial;
 261         struct bgp_attr_mp_unreach_nlri_partial mp_unreach_nlri_partial;
 262
 263         a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL;
 264         a.code = BGP_PATH_ATTR_CODE_MP_REACH_NLRI;
 265         a.data.s.len = 0; /* will be set on UPDATE */
 266
 267         mp_reach_nlri_partial.afi = htons(BGP_MP_AFI_IPv6);
 268         mp_reach_nlri_partial.safi = BGP_MP_SAFI_UNICAST;
 269         mp_reach_nlri_partial.reserved = 0;
 270         mp_reach_nlri_partial.next_hop_len = 16;
 271
 272         /* use the defined nexthop6, or our address in ipv6_prefix */
 273         if (config->nexthop6_address.s6_addr[0])
 274             memcpy(&mp_reach_nlri_partial.next_hop,
 275                     &config->nexthop6_address.s6_addr, 16);
 276         else
 277         {
 278             /* our address is ipv6prefix::1 */
 279             memcpy(&mp_reach_nlri_partial.next_hop,
 280                     &config->ipv6_prefix.s6_addr, 16);
 281             mp_reach_nlri_partial.next_hop[15] = 1;
 282         }
 283
 284         memcpy(&a.data.s.value, &mp_reach_nlri_partial,
 285                 sizeof(struct bgp_attr_mp_reach_nlri_partial));
 286         memcpy(&peer->mp_reach_nlri_partial, &a,
 287                 BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE);
 288
 289         a.flags = BGP_PATH_ATTR_FLAG_OPTIONAL | BGP_PATH_ATTR_FLAG_EXTLEN;
 290         a.code = BGP_PATH_ATTR_CODE_MP_UNREACH_NLRI;
 291         a.data.e.len = 0; /* will be set on UPDATE */
 292
 293         mp_unreach_nlri_partial.afi = htons(BGP_MP_AFI_IPv6);
 294         mp_unreach_nlri_partial.safi = BGP_MP_SAFI_UNICAST;
 295
 296         memcpy(&a.data.e.value, &mp_unreach_nlri_partial,
 297                 sizeof(struct bgp_attr_mp_unreach_nlri_partial));
 298         memcpy(&peer->mp_unreach_nlri_partial, &a,
 299                 BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE);
 300     }
 301
 302     peer->mp_handling = HandlingUnknown;
 303
 304     LOG(4, 0, 0, "Initiating BGP connection to %s (routing %s)\n",
 305         name, enable ? "enabled" : "suspended");
 306
 307     /* we have at least one peer configured */
 308     bgp_configured = 1;
 309
 310     /* connect */
 311     return bgp_connect(peer);
 312 }
 313
 314 /* clear counters, timers, routes and buffers; close socket; move to
 315    next_state, which may be Disabled or Idle */
 316 static void bgp_clear(struct bgp_peer *peer)
 317 {
 318     if (peer->sock != -1)
 319     {
 320         close(peer->sock);
 321         peer->sock = -1;
 322     }
 323
 324     peer->keepalive_time = 0;
 325     peer->expire_time = 0;
 326
 327     peer->keepalive = peer->init_keepalive;
 328     peer->hold = peer->init_hold;
 329
 330     bgp_free_routes(peer->routes);
 331     peer->routes = 0;
 332     bgp_free_routes6(peer->routes6);
 333     peer->routes6 = 0;
 334
 335     peer->outbuf->packet.header.len = 0;
 336     peer->outbuf->done = 0;
 337     peer->inbuf->packet.header.len = 0;
 338     peer->inbuf->done = 0;
 339
 340     peer->cli_flag = 0;
 341     peer->events = 0;
 342
 343     if (peer->state != peer->next_state)
 344     {
 345         peer->state = peer->next_state;
 346         peer->state_time = time_now;
 347
 348         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
 349             bgp_state_str(peer->next_state));
 350     }
 351 }
 352
 353 /* initiate a clean shutdown */
 354 void bgp_stop(struct bgp_peer *peer)
 355 {
 356     LOG(4, 0, 0, "Terminating BGP connection to %s\n", peer->name);
 357     bgp_send_notification(peer, BGP_ERR_CEASE, 0);
 358 }
 359
 360 /* drop connection (if any) and set state to Disabled */
 361 void bgp_halt(struct bgp_peer *peer)
 362 {
 363     LOG(4, 0, 0, "Aborting BGP connection to %s\n", peer->name);
 364     peer->next_state = Disabled;
 365     bgp_clear(peer);
 366 }
 367
 368 /* drop connection (if any) and set to Idle for connection retry */
 369 int bgp_restart(struct bgp_peer *peer)
 370 {
 371     peer->next_state = Idle;
 372     bgp_clear(peer);
 373
 374     /* restart now */
 375     peer->retry_time = time_now;
 376     peer->retry_count = 0;
 377
 378     /* connect */
 379     return bgp_connect(peer);
 380 }
 381
 382 static void bgp_set_retry(struct bgp_peer *peer)
 383 {
 384     if (peer->retry_count++ < BGP_MAX_RETRY)
 385     {
 386         peer->retry_time = time_now + (BGP_RETRY_BACKOFF * peer->retry_count);
 387         peer->next_state = Idle;
 388         bgp_clear(peer);
 389     }
 390     else
 391         bgp_halt(peer); /* give up */
 392 }
 393
 394 /* insert route into list; sorted */
 395 static struct bgp_route_list *bgp_insert_route(struct bgp_route_list *head,
 396     struct bgp_route_list *new)
 397 {
 398     struct bgp_route_list *p = head;
 399     struct bgp_route_list *e = 0;
 400
 401     while (p && memcmp(&p->dest, &new->dest, sizeof(p->dest)) < 0)
 402     {
 403         e = p;
 404         p = p->next;
 405     }
 406
 407     if (e)
 408     {
 409         new->next = e->next;
 410         e->next = new;
 411     }
 412     else
 413     {
 414         new->next = head;
 415         head = new;
 416     }
 417
 418     return head;
 419 }
 420
 421 /* insert route6 into list; sorted */
 422 static struct bgp_route6_list *bgp_insert_route6(struct bgp_route6_list *head,
 423     struct bgp_route6_list *new)
 424 {
 425     struct bgp_route6_list *p = head;
 426     struct bgp_route6_list *e = 0;
 427
 428     while (p && memcmp(&p->dest, &new->dest, sizeof(p->dest)) < 0)
 429     {
 430         e = p;
 431         p = p->next;
 432     }
 433
 434     if (e)
 435     {
 436         new->next = e->next;
 437         e->next = new;
 438     }
 439     else
 440     {
 441         new->next = head;
 442         head = new;
 443     }
 444
 445     return head;
 446 }
 447
 448 /* add route to list for peers */
 449 /*
 450  * Note:  this doesn't do route aggregation, nor drop routes if a less
 451  * specific match already exists (partly because I'm lazy, but also so
 452  * that if that route is later deleted we don't have to be concerned
 453  * about adding back the more specific one).
 454  */
 455 int bgp_add_route(in_addr_t ip, int prefixlen)
 456 {
 457     struct bgp_route_list *r = bgp_routes;
 458     struct bgp_route_list add;
 459     int i;
 460
 461     add.dest.prefix = ip;
 462     add.dest.len = prefixlen;
 463     add.next = 0;
 464
 465     /* check for duplicate */
 466     while (r)
 467     {
 468         i = memcmp(&r->dest, &add.dest, sizeof(r->dest));
 469         if (!i)
 470             return 1; /* already covered */
 471
 472         if (i > 0)
 473             break;
 474
 475         r = r->next;
 476     }
 477
 478     /* insert into route list; sorted */
 479     if (!(r = malloc(sizeof(*r))))
 480     {
 481         LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
 482             fmtaddr(add.dest.prefix, 0), add.dest.len, strerror(errno));
 483
 484         return 0;
 485     }
 486
 487     memcpy(r, &add, sizeof(*r));
 488     bgp_routes = bgp_insert_route(bgp_routes, r);
 489
 490     /* flag established peers for update */
 491     for (i = 0; i < BGP_NUM_PEERS; i++)
 492         if (bgp_peers[i].state == Established)
 493             bgp_peers[i].update_routes = 1;
 494
 495     LOG(4, 0, 0, "Registered BGP route %s/%d\n",
 496         fmtaddr(add.dest.prefix, 0), add.dest.len);
 497
 498     return 1;
 499 }
 500
 501 /* add route to list for peers */
 502 /*
 503  * Note: same provisions as above
 504  */
 505 int bgp_add_route6(struct in6_addr ip, int prefixlen)
 506 {
 507     struct bgp_route6_list *r = bgp_routes6;
 508     struct bgp_route6_list add;
 509     int i;
 510     char ipv6addr[INET6_ADDRSTRLEN];
 511
 512     memcpy(&add.dest.prefix, &ip.s6_addr, 16);
 513     add.dest.len = prefixlen;
 514     add.next = 0;
 515
 516     /* check for duplicate */
 517     while (r)
 518     {
 519         i = memcmp(&r->dest, &add.dest, sizeof(r->dest));
 520         if (!i)
 521             return 1; /* already covered */
 522
 523         if (i > 0)
 524             break;
 525
 526         r = r->next;
 527     }
 528
 529     /* insert into route list; sorted */
 530     if (!(r = malloc(sizeof(*r))))
 531     {
 532         LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
 533             inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN), add.dest.len,
 534             strerror(errno));
 535
 536         return 0;
 537     }
 538
 539     memcpy(r, &add, sizeof(*r));
 540     bgp_routes6 = bgp_insert_route6(bgp_routes6, r);
 541
 542     /* flag established peers for update */
 543     for (i = 0; i < BGP_NUM_PEERS; i++)
 544         if (bgp_peers[i].state == Established
 545                 && bgp_peers[i].mp_handling == HandleIPv6Routes)
 546             bgp_peers[i].update_routes6 = 1;
 547
 548     LOG(4, 0, 0, "Registered BGP route %s/%d\n",
 549         inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN), add.dest.len);
 550
 551     return 1;
 552 }
 553
 554 /* remove route from list for peers */
 555 int bgp_del_route(in_addr_t ip, int prefixlen)
 556 {
 557     struct bgp_route_list *r = bgp_routes;
 558     struct bgp_route_list *e = 0;
 559     struct bgp_route_list del;
 560     int i;
 561
 562     del.dest.prefix = ip;
 563     del.dest.len = prefixlen;
 564     del.next = 0;
 565
 566     /* find entry in routes list and remove */
 567     while (r)
 568     {
 569         i = memcmp(&r->dest, &del.dest, sizeof(r->dest));
 570         if (!i)
 571         {
 572             if (e)
 573                 e->next = r->next;
 574             else
 575                 bgp_routes = r->next;
 576
 577             free(r);
 578             break;
 579         }
 580
 581         e = r;
 582
 583         if (i > 0)
 584             r = 0; /* stop */
 585         else
 586             r = r->next;
 587     }
 588
 589     /* not found */
 590     if (!r)
 591         return 1;
 592
 593     /* flag established peers for update */
 594     for (i = 0; i < BGP_NUM_PEERS; i++)
 595         if (bgp_peers[i].state == Established)
 596             bgp_peers[i].update_routes = 1;
 597
 598     LOG(4, 0, 0, "Removed BGP route %s/%d\n",
 599         fmtaddr(del.dest.prefix, 0), del.dest.len);
 600
 601     return 1;
 602 }
 603
 604 /* remove route from list for peers */
 605 int bgp_del_route6(struct in6_addr ip, int prefixlen)
 606 {
 607     struct bgp_route6_list *r = bgp_routes6;
 608     struct bgp_route6_list *e = 0;
 609     struct bgp_route6_list del;
 610     int i;
 611     char ipv6addr[INET6_ADDRSTRLEN];
 612
 613     memcpy(&del.dest.prefix, &ip.s6_addr, 16);
 614     del.dest.len = prefixlen;
 615     del.next = 0;
 616
 617     /* find entry in routes list and remove */
 618     while (r)
 619     {
 620         i = memcmp(&r->dest, &del.dest, sizeof(r->dest));
 621         if (!i)
 622         {
 623             if (e)
 624                 e->next = r->next;
 625             else
 626                 bgp_routes6 = r->next;
 627
 628             free(r);
 629             break;
 630         }
 631
 632         e = r;
 633
 634         if (i > 0)
 635             r = 0; /* stop */
 636         else
 637             r = r->next;
 638     }
 639
 640     /* not found */
 641     if (!r)
 642         return 1;
 643
 644     /* flag established peers for update */
 645     for (i = 0; i < BGP_NUM_PEERS; i++)
 646         if (bgp_peers[i].state == Established
 647                 && bgp_peers[i].mp_handling == HandleIPv6Routes)
 648             bgp_peers[i].update_routes6 = 1;
 649
 650     LOG(4, 0, 0, "Removed BGP route %s/%d\n",
 651         inet_ntop(AF_INET6, &ip, ipv6addr, INET6_ADDRSTRLEN), del.dest.len);
 652
 653     return 1;
 654 }
 655
 656 /* enable or disable routing */
 657 void bgp_enable_routing(int enable)
 658 {
 659     int i;
 660
 661     for (i = 0; i < BGP_NUM_PEERS; i++)
 662     {
 663         bgp_peers[i].routing = enable;
 664
 665         /* flag established peers for update */
 666         if (bgp_peers[i].state == Established)
 667             bgp_peers[i].update_routes = 1;
 668     }
 669
 670     LOG(4, 0, 0, "%s BGP routing\n", enable ? "Enabled" : "Suspended");
 671 }
 672
 673 #ifdef HAVE_EPOLL
 674 # include <sys/epoll.h>
 675 #else
 676 # include "fake_epoll.h"
 677 #endif
 678
 679 /* return a bitmask of the events required to poll this peer's fd */
 680 int bgp_set_poll()
 681 {
 682     int i;
 683
 684     if (!bgp_configured)
 685         return 0;
 686
 687     for (i = 0; i < BGP_NUM_PEERS; i++)
 688     {
 689         struct bgp_peer *peer = &bgp_peers[i];
 690         int events = 0;
 691
 692         if (peer->state == Disabled || peer->state == Idle)
 693             continue;
 694
 695         if (peer->inbuf->done < BGP_MAX_PACKET_SIZE)
 696             events |= EPOLLIN;
 697
 698         if (peer->state == Connect ||           /* connection in progress */
 699             peer->update_routes ||              /* routing updates */
 700             peer->outbuf->packet.header.len)    /* pending output */
 701             events |= EPOLLOUT;
 702
 703         if (peer->events != events)
 704         {
 705             struct epoll_event ev;
 706
 707             ev.events = peer->events = events;
 708             ev.data.ptr = &peer->edata;
 709             epoll_ctl(epollfd, EPOLL_CTL_MOD, peer->sock, &ev);
 710         }
 711     }
 712
 713     return 1;
 714 }
 715
 716 /* process bgp events/timers */
 717 int bgp_process(uint32_t events[])
 718 {
 719     int i;
 720
 721     if (!bgp_configured)
 722         return 0;
 723
 724     for (i = 0; i < BGP_NUM_PEERS; i++)
 725     {
 726         struct bgp_peer *peer = &bgp_peers[i];
 727
 728         if (*peer->name && peer->cli_flag == BGP_CLI_RESTART)
 729         {
 730             bgp_restart(peer);
 731             continue;
 732         }
 733
 734         if (peer->state == Disabled)
 735             continue;
 736
 737         if (peer->cli_flag)
 738         {
 739             switch (peer->cli_flag)
 740             {
 741             case BGP_CLI_SUSPEND:
 742                 if (peer->routing)
 743                 {
 744                     peer->routing = 0;
 745                     if (peer->state == Established)
 746                         peer->update_routes = 1;
 747                 }
 748
 749                 break;
 750
 751             case BGP_CLI_ENABLE:
 752                 if (!peer->routing)
 753                 {
 754                     peer->routing = 1;
 755                     if (peer->state == Established)
 756                         peer->update_routes = 1;
 757                 }
 758
 759                 break;
 760             }
 761
 762             peer->cli_flag = 0;
 763         }
 764
 765         /* handle empty/fill of buffers */
 766         if (events[i] & EPOLLOUT)
 767         {
 768             int r = 1;
 769             if (peer->state == Connect)
 770                 r = bgp_handle_connect(peer);
 771             else if (peer->outbuf->packet.header.len)
 772                 r = bgp_write(peer);
 773
 774             if (!r)
 775                 continue;
 776         }
 777
 778         if (events[i] & (EPOLLIN|EPOLLHUP))
 779         {
 780             if (!bgp_read(peer))
 781                 continue;
 782         }
 783
 784         /* process input buffer contents */
 785         while (peer->inbuf->done >= sizeof(peer->inbuf->packet.header)
 786             && !peer->outbuf->packet.header.len) /* may need to queue a response */
 787         {
 788             if (bgp_handle_input(peer) < 0)
 789                 continue;
 790         }
 791
 792         /* process pending updates */
 793         if (peer->update_routes
 794             && !peer->outbuf->packet.header.len) /* ditto */
 795         {
 796             if (!bgp_send_update(peer))
 797                 continue;
 798         }
 799
 800         /* process pending IPv6 updates */
 801         if (peer->update_routes6
 802             && !peer->outbuf->packet.header.len) /* ditto */
 803         {
 804             if (!bgp_send_update6(peer))
 805                 continue;
 806         }
 807
 808         /* process timers */
 809         bgp_process_timers(peer);
 810     }
 811
 812     return 1;
 813 }
 814
 815 /* process bgp timers only */
 816 void bgp_process_peers_timers()
 817 {
 818     int i;
 819
 820     if (!bgp_configured)
 821         return;
 822
 823     for (i = 0; i < BGP_NUM_PEERS; i++)
 824     {
 825         struct bgp_peer *peer = &bgp_peers[i];
 826
 827         if (peer->state == Disabled)
 828             continue;
 829
 830         bgp_process_timers(peer);
 831     }
 832 }
 833
 834 static void bgp_process_timers(struct bgp_peer *peer)
 835 {
 836     if (peer->state == Established)
 837     {
 838         if (time_now > peer->expire_time)
 839         {
 840             LOG(1, 0, 0, "No message from BGP peer %s in %ds\n",
 841                 peer->name, peer->hold);
 842
 843             bgp_send_notification(peer, BGP_ERR_HOLD_TIMER_EXP, 0);
 844             return;
 845         }
 846
 847         if (time_now > peer->keepalive_time && !peer->outbuf->packet.header.len)
 848             bgp_send_keepalive(peer);
 849     }
 850     else if (peer->state == Idle)
 851     {
 852         if (time_now > peer->retry_time)
 853             bgp_connect(peer);
 854     }
 855     else if (time_now > peer->state_time + BGP_STATE_TIME)
 856     {
 857         LOG(1, 0, 0, "%s timer expired for BGP peer %s\n",
 858             bgp_state_str(peer->state), peer->name);
 859
 860         bgp_restart(peer);
 861     }
 862 }
 863
 864 static void bgp_free_routes(struct bgp_route_list *routes)
 865 {
 866     struct bgp_route_list *tmp;
 867
 868     while ((tmp = routes))
 869     {
 870         routes = tmp->next;
 871         free(tmp);
 872     }
 873 }
 874
 875 static void bgp_free_routes6(struct bgp_route6_list *routes)
 876 {
 877     struct bgp_route6_list *tmp;
 878
 879     while ((tmp = routes))
 880     {
 881         routes = tmp->next;
 882         free(tmp);
 883     }
 884 }
 885
 886 char const *bgp_state_str(enum bgp_state state)
 887 {
 888     switch (state)
 889     {
 890     case Disabled:      return "Disabled";
 891     case Idle:          return "Idle";
 892     case Connect:       return "Connect";
 893     case Active:        return "Active";
 894     case OpenSent:      return "OpenSent";
 895     case OpenConfirm:   return "OpenConfirm";
 896     case Established:   return "Established";
 897     }
 898
 899     return "?";
 900 }
 901
 902 static char const *bgp_msg_type_str(uint8_t type)
 903 {
 904     switch (type)
 905     {
 906     case BGP_MSG_OPEN:          return "OPEN";
 907     case BGP_MSG_UPDATE:        return "UPDATE";
 908     case BGP_MSG_NOTIFICATION:  return "NOTIFICATION";
 909     case BGP_MSG_KEEPALIVE:     return "KEEPALIVE";
 910     }
 911
 912     return "?";
 913 }
 914
 915 /* attempt to connect to peer */
 916 static int bgp_connect(struct bgp_peer *peer)
 917 {
 918     static int bgp_port = 0;
 919     struct sockaddr_in addr;
 920     struct sockaddr_in source_addr;
 921     struct epoll_event ev;
 922
 923     if (!bgp_port)
 924     {
 925         struct servent *serv;
 926         if (!(serv = getservbyname("bgp", "tcp")))
 927         {
 928             LOG(0, 0, 0, "Can't get bgp service (%s)\n", strerror(errno));
 929             return 0;
 930         }
 931
 932         bgp_port = serv->s_port;
 933     }
 934
 935     if ((peer->sock = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
 936     {
 937         LOG(0, 0, 0, "Can't create a socket for BGP peer %s (%s)\n",
 938             peer->name, strerror(errno));
 939
 940         peer->state = peer->next_state = Disabled;
 941         return 0;
 942     }
 943
 944     /* add to poll set */
 945     ev.events = peer->events = EPOLLOUT;
 946     ev.data.ptr = &peer->edata;
 947     epoll_ctl(epollfd, EPOLL_CTL_ADD, peer->sock, &ev);
 948
 949     /* set to non-blocking */
 950     fcntl(peer->sock, F_SETFL, fcntl(peer->sock, F_GETFL, 0) | O_NONBLOCK);
 951
 952     /* set source address */
 953     memset(&source_addr, 0, sizeof(source_addr));
 954     source_addr.sin_family = AF_INET;
 955     source_addr.sin_addr.s_addr = peer->source_addr; /* defaults to INADDR_ANY */
 956     if (bind(peer->sock, (struct sockaddr *) &source_addr, sizeof(source_addr)) < 0)
 957     {
 958         LOG(1, 0, 0, "Can't set source address to %s: %s\n",
 959             inet_ntoa(source_addr.sin_addr), strerror(errno));
 960
 961         bgp_set_retry(peer);
 962         return 0;
 963     }
 964
 965     /* try connect */
 966     memset(&addr, 0, sizeof(addr));
 967     addr.sin_family = AF_INET;
 968     addr.sin_port = bgp_port;
 969     addr.sin_addr.s_addr = peer->addr;
 970
 971     while (connect(peer->sock, (struct sockaddr *) &addr, sizeof(addr)) == -1)
 972     {
 973         if (errno == EINTR) /* SIGALARM handler */
 974             continue;
 975
 976         if (errno != EINPROGRESS)
 977         {
 978             LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n",
 979                 inet_ntoa(addr.sin_addr), strerror(errno));
 980
 981             bgp_set_retry(peer);
 982             return 0;
 983         }
 984
 985         peer->state = Connect;
 986         peer->state_time = time_now;
 987
 988         LOG(4, 0, 0, "BGP peer %s: state Connect\n", peer->name);
 989         return 1;
 990     }
 991
 992     peer->state = Active;
 993     peer->state_time = time_now;
 994     peer->retry_time = peer->retry_count = 0;
 995
 996     LOG(4, 0, 0, "BGP peer %s: state Active\n", inet_ntoa(addr.sin_addr));
 997
 998     return bgp_send_open(peer);
 999 }
1000
1001 /* complete partial connection (state = Connect) */
1002 static int bgp_handle_connect(struct bgp_peer *peer)
1003 {
1004     int err = 0;
1005     socklen_t len = sizeof(int);
1006     getsockopt(peer->sock, SOL_SOCKET, SO_ERROR, &err, &len);
1007     if (err)
1008     {
1009         LOG(1, 0, 0, "Can't connect to BGP peer %s (%s)\n", peer->name,
1010             strerror(err));
1011
1012         bgp_set_retry(peer);
1013         return 0;
1014     }
1015
1016     peer->state = Active;
1017     peer->state_time = time_now;
1018
1019     LOG(4, 0, 0, "BGP peer %s: state Active\n", peer->name);
1020
1021     return bgp_send_open(peer);
1022 }
1023
1024 /* initiate a write */
1025 static int bgp_write(struct bgp_peer *peer)
1026 {
1027     int len = htons(peer->outbuf->packet.header.len);
1028     int r;
1029
1030     while ((r = write(peer->sock, &peer->outbuf->packet + peer->outbuf->done,
1031         len - peer->outbuf->done)) == -1)
1032     {
1033         if (errno == EINTR)
1034             continue;
1035
1036         if (errno == EAGAIN)
1037             return 1;
1038
1039         if (errno == EPIPE)
1040             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
1041         else
1042             LOG(1, 0, 0, "Can't write to BGP peer %s (%s)\n", peer->name,
1043                 strerror(errno));
1044
1045         bgp_set_retry(peer);
1046         return 0;
1047     }
1048
1049     if (r < len)
1050     {
1051         peer->outbuf->done += r;
1052         return 1;
1053     }
1054
1055     LOG(4, 0, 0, "Sent %s to BGP peer %s\n",
1056         bgp_msg_type_str(peer->outbuf->packet.header.type), peer->name);
1057
1058     peer->outbuf->packet.header.len = 0;
1059     peer->outbuf->done = 0;
1060
1061     if (peer->state == Established)
1062         peer->keepalive_time = time_now + peer->keepalive;
1063
1064     if (peer->state != peer->next_state)
1065     {
1066         if (peer->next_state == Disabled || peer->next_state == Idle)
1067         {
1068             bgp_clear(peer);
1069             return 0;
1070         }
1071
1072         peer->state = peer->next_state;
1073         peer->state_time = time_now;
1074
1075         LOG(4, 0, 0, "BGP peer %s: state %s\n", peer->name,
1076             bgp_state_str(peer->state));
1077     }
1078
1079     return 1;
1080 }
1081
1082 /* initiate a read */
1083 static int bgp_read(struct bgp_peer *peer)
1084 {
1085     int r;
1086
1087     while ((r = read(peer->sock, &peer->inbuf->packet + peer->inbuf->done,
1088         BGP_MAX_PACKET_SIZE - peer->inbuf->done)) < 1)
1089     {
1090         if (!r)
1091         {
1092             LOG(1, 0, 0, "Connection to BGP peer %s closed\n", peer->name);
1093         }
1094         else
1095         {
1096             if (errno == EINTR)
1097                 continue;
1098
1099             if (errno == EAGAIN)
1100                 return 1;
1101
1102             LOG(1, 0, 0, "Can't read from BGP peer %s (%s)\n", peer->name,
1103                 strerror(errno));
1104         }
1105
1106         bgp_set_retry(peer);
1107         return 0;
1108     }
1109
1110     peer->inbuf->done += r;
1111     return 1;
1112 }
1113
1114 /* process buffered packets */
1115 static int bgp_handle_input(struct bgp_peer *peer)
1116 {
1117     struct bgp_packet *p = &peer->inbuf->packet;
1118     int len = ntohs(p->header.len);
1119
1120     if (len > BGP_MAX_PACKET_SIZE)
1121     {
1122         LOG(1, 0, 0, "Bad header length from BGP %s\n", peer->name);
1123         bgp_send_notification(peer, BGP_ERR_HEADER, BGP_ERR_HDR_BAD_LEN);
1124         return 0;
1125     }
1126
1127     if (peer->inbuf->done < len)
1128         return 0;
1129
1130     LOG(4, 0, 0, "Received %s from BGP peer %s\n",
1131         bgp_msg_type_str(p->header.type), peer->name);
1132
1133     switch (p->header.type)
1134     {
1135     case BGP_MSG_OPEN:
1136         {
1137             struct bgp_data_open data;
1138             int hold;
1139             int i;
1140             off_t param_offset, capability_offset;
1141             struct bgp_opt_param *param;
1142             uint8_t capabilities_len;
1143             char *capabilities = NULL;
1144             struct bgp_capability *capability;
1145             struct bgp_mp_cap_param *mp_cap;
1146
1147             for (i = 0; i < sizeof(p->header.marker); i++)
1148             {
1149                 if ((unsigned char) p->header.marker[i] != 0xff)
1150                 {
1151                     LOG(1, 0, 0, "Invalid marker from BGP peer %s\n",
1152                         peer->name);
1153
1154                     bgp_send_notification(peer, BGP_ERR_HEADER,
1155                         BGP_ERR_HDR_NOT_SYNC);
1156
1157                     return 0;
1158                 }
1159             }
1160
1161             if (peer->state != OpenSent)
1162             {
1163                 LOG(1, 0, 0, "OPEN from BGP peer %s in %s state\n",
1164                     peer->name, bgp_state_str(peer->state));
1165
1166                 bgp_send_notification(peer, BGP_ERR_FSM, 0);
1167                 return 0;
1168             }
1169
1170             memcpy(&data, p->data, len - sizeof(p->header));
1171
1172             if (data.version != BGP_VERSION)
1173             {
1174                 LOG(1, 0, 0, "Bad version (%d) sent by BGP peer %s\n",
1175                     (int) data.version, peer->name);
1176
1177                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_VERSION);
1178                 return 0;
1179             }
1180
1181             if (ntohs(data.as) != peer->as)
1182             {
1183                 LOG(1, 0, 0, "Bad AS sent by BGP peer %s (got %d, "
1184                     "expected %d)\n", peer->name, (int) htons(data.as),
1185                     (int) peer->as);
1186
1187                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_BAD_AS);
1188                 return 0;
1189             }
1190
1191             if ((hold = ntohs(data.hold_time)) < 3)
1192             {
1193                 LOG(1, 0, 0, "Bad hold time (%d) from BGP peer %s\n",
1194                     hold, peer->name);
1195
1196                 bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_HOLD_TIME);
1197                 return 0;
1198             }
1199
1200             /* pick lowest hold time */
1201             if (hold < peer->hold)
1202                 peer->hold = hold;
1203
1204             /* adjust our keepalive based on negotiated hold value */
1205             if (peer->keepalive * 3 > peer->hold)
1206                 peer->keepalive = peer->hold / 3;
1207
1208             /* check for optional parameters */
1209             /* 2 is for the size of type + len (both uint8_t) */
1210             for (param_offset = 0;
1211                     param_offset < data.opt_len;
1212                     param_offset += 2 + param->len)
1213             {
1214                 param = (struct bgp_opt_param *)((char *)&data.opt_params + param_offset);
1215
1216                 /* sensible check */
1217                 if (data.opt_len - param_offset < 2
1218                         || param->len > data.opt_len - param_offset - 2)
1219                 {
1220                     LOG(1, 0, 0, "Malformed Optional Parameter list from BGP peer %s\n",
1221                         peer->name);
1222
1223                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
1224                     return 0;
1225                 }
1226
1227                 /* we know only one parameter type */
1228                 if (param->type != BGP_PARAM_TYPE_CAPABILITY)
1229                 {
1230                     LOG(1, 0, 0, "Unsupported Optional Parameter type %d from BGP peer %s\n",
1231                         param->type, peer->name);
1232
1233                     bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_OPN_UNSUP_PARAM);
1234                     return 0;
1235                 }
1236
1237                 capabilities_len = param->len;
1238                 capabilities = (char *)&param->value;
1239
1240                 /* look for BGP multiprotocol capability */
1241                 for (capability_offset = 0;
1242                         capability_offset < capabilities_len;
1243                         capability_offset += 2 + capability->len)
1244                 {
1245                     capability = (struct bgp_capability *)(capabilities + capability_offset);
1246
1247                     /* sensible check */
1248                     if (capabilities_len - capability_offset < 2
1249                             || capability->len > capabilities_len - capability_offset - 2)
1250                     {
1251                         LOG(1, 0, 0, "Malformed Capabilities list from BGP peer %s\n",
1252                             peer->name);
1253
1254                         bgp_send_notification(peer, BGP_ERR_OPEN, BGP_ERR_UNSPEC);
1255                         return 0;
1256                     }
1257
1258                     /* we only know one capability code */
1259                     if (capability->code != BGP_CAP_CODE_MP
1260                             && capability->len != sizeof(struct bgp_mp_cap_param))
1261                     {
1262                         LOG(4, 0, 0, "Unsupported Capability code %d from BGP peer %s\n",
1263                             capability->code, peer->name);
1264
1265                         /* we don't terminate, still; we just jump to the next one */
1266                         continue;
1267                     }
1268
1269                     mp_cap = (struct bgp_mp_cap_param *)&capability->value;
1270                     /* the only <AFI, SAFI> tuple we support */
1271                     if (ntohs(mp_cap->afi) != BGP_MP_AFI_IPv6 && mp_cap->safi != BGP_MP_SAFI_UNICAST)
1272                     {
1273                         LOG(4, 0, 0, "Unsupported multiprotocol AFI %d and SAFI %d from BGP peer %s\n",
1274                             mp_cap->afi, mp_cap->safi, peer->name);
1275
1276                         /* we don't terminate, still; we just jump to the next one */
1277                         continue;
1278                     }
1279
1280                     /* yes it can! */
1281                     peer->mp_handling = HandleIPv6Routes;
1282                 }
1283             }
1284
1285             if (peer->mp_handling != HandleIPv6Routes)
1286             {
1287                 peer->mp_handling = DoesntHandleIPv6Routes;
1288                 if (config->ipv6_prefix.s6_addr[0])
1289                     LOG(1, 0, 0, "Warning: BGP peer %s doesn't handle IPv6 prefixes updates\n",
1290                             peer->name);
1291             }
1292
1293             /* next transition requires an exchange of keepalives */
1294             bgp_send_keepalive(peer);
1295         }
1296
1297         break;
1298
1299     case BGP_MSG_KEEPALIVE:
1300         if (peer->state == OpenConfirm)
1301         {
1302             peer->state = peer->next_state = Established;
1303             peer->state_time = time_now;
1304             peer->keepalive_time = time_now + peer->keepalive;
1305             peer->update_routes = 1;
1306             peer->retry_count = 0;
1307             peer->retry_time = 0;
1308
1309             LOG(4, 0, 0, "BGP peer %s: state Established\n", peer->name);
1310         }
1311
1312         break;
1313
1314     case BGP_MSG_NOTIFICATION:
1315         if (len > sizeof(p->header))
1316         {
1317             struct bgp_data_notification *notification =
1318                 (struct bgp_data_notification *) p->data;
1319
1320             if (notification->error_code == BGP_ERR_CEASE)
1321             {
1322                 LOG(4, 0, 0, "BGP peer %s sent CEASE\n", peer->name);
1323                 bgp_set_retry(peer);
1324                 return 0;
1325             }
1326
1327             if (notification->error_code == BGP_ERR_OPEN
1328                     && notification->error_subcode == BGP_ERR_OPN_UNSUP_PARAM)
1329             {
1330                 LOG(4, 0, 0, "BGP peer %s doesn't support BGP Capabilities\n", peer->name);
1331                 peer->mp_handling = DoesntHandleIPv6Routes;
1332                 bgp_set_retry(peer);
1333                 return 0;
1334             }
1335
1336             if (notification->error_code == BGP_ERR_OPEN
1337                     && notification->error_subcode == BGP_ERR_OPN_UNSUP_CAP)
1338             {
1339                 /* the only capability we advertise is this one, so upon receiving
1340                    an "unsupported capability" message, we disable IPv6 routes for
1341                    this peer */
1342                 LOG(4, 0, 0, "BGP peer %s doesn't support IPv6 routes advertisement\n", peer->name);
1343                 peer->mp_handling = DoesntHandleIPv6Routes;
1344                 break;
1345             }
1346
1347             /* FIXME: should handle more notifications */
1348             LOG(4, 0, 0, "BGP peer %s sent unhandled NOTIFICATION %d\n",
1349                 peer->name, (int) notification->error_code);
1350         }
1351
1352         break;
1353     }
1354
1355     /* reset timer */
1356     peer->expire_time = time_now + peer->hold;
1357
1358     /* see if there's another message in the same packet/buffer */
1359     if (peer->inbuf->done > len)
1360     {
1361         peer->inbuf->done -= len;
1362         memmove(p, (char *) p + len, peer->inbuf->done);
1363     }
1364     else
1365     {
1366         peer->inbuf->packet.header.len = 0;
1367         peer->inbuf->done = 0;
1368     }
1369
1370     return peer->inbuf->done;
1371 }
1372
1373 /* send/buffer OPEN message */
1374 static int bgp_send_open(struct bgp_peer *peer)
1375 {
1376     struct bgp_data_open data;
1377     struct bgp_mp_cap_param mp_ipv6 = { htons(BGP_MP_AFI_IPv6), 0, BGP_MP_SAFI_UNICAST };
1378     struct bgp_capability cap_mp_ipv6;
1379     struct bgp_opt_param param_cap_mp_ipv6;
1380     uint16_t len = sizeof(peer->outbuf->packet.header);
1381
1382     memset(peer->outbuf->packet.header.marker, 0xff,
1383         sizeof(peer->outbuf->packet.header.marker));
1384
1385     peer->outbuf->packet.header.type = BGP_MSG_OPEN;
1386
1387     data.version = BGP_VERSION;
1388     data.as = htons(our_as);
1389     data.hold_time = htons(peer->hold);
1390     /* use the source IP we use as identifier, if available */
1391     if (peer->source_addr != INADDR_ANY)
1392         data.identifier = peer->source_addr;
1393     else
1394         data.identifier = my_address;
1395
1396     /* if we know peer doesn't support MP (mp_handling == DoesntHandleIPv6Routes)
1397        then don't add this parameter */
1398     if (config->ipv6_prefix.s6_addr[0]
1399             && (peer->mp_handling == HandlingUnknown
1400                 || peer->mp_handling == HandleIPv6Routes))
1401     {
1402         /* construct the param and capability */
1403         cap_mp_ipv6.code = BGP_CAP_CODE_MP;
1404         cap_mp_ipv6.len = sizeof(mp_ipv6);
1405         memcpy(&cap_mp_ipv6.value, &mp_ipv6, cap_mp_ipv6.len);
1406
1407         param_cap_mp_ipv6.type = BGP_PARAM_TYPE_CAPABILITY;
1408         param_cap_mp_ipv6.len = 2 + sizeof(mp_ipv6);
1409         memcpy(&param_cap_mp_ipv6.value, &cap_mp_ipv6, param_cap_mp_ipv6.len);
1410
1411         data.opt_len = 2 + param_cap_mp_ipv6.len;
1412         memcpy(&data.opt_params, &param_cap_mp_ipv6, data.opt_len);
1413     }
1414     else
1415         data.opt_len = 0;
1416
1417     memcpy(peer->outbuf->packet.data, &data, BGP_DATA_OPEN_SIZE + data.opt_len);
1418     len += BGP_DATA_OPEN_SIZE + data.opt_len;
1419
1420     peer->outbuf->packet.header.len = htons(len);
1421     peer->outbuf->done = 0;
1422     peer->next_state = OpenSent;
1423
1424     return bgp_write(peer);
1425 }
1426
1427 /* send/buffer KEEPALIVE message */
1428 static int bgp_send_keepalive(struct bgp_peer *peer)
1429 {
1430     memset(peer->outbuf->packet.header.marker, 0xff,
1431         sizeof(peer->outbuf->packet.header.marker));
1432
1433     peer->outbuf->packet.header.type = BGP_MSG_KEEPALIVE;
1434     peer->outbuf->packet.header.len =
1435         htons(sizeof(peer->outbuf->packet.header));
1436
1437     peer->outbuf->done = 0;
1438     peer->next_state = (peer->state == OpenSent) ? OpenConfirm : peer->state;
1439
1440     return bgp_write(peer);
1441 }
1442
1443 /* send/buffer UPDATE message */
1444 static int bgp_send_update(struct bgp_peer *peer)
1445 {
1446     uint16_t unf_len = 0;
1447     uint16_t attr_len;
1448     uint16_t len = sizeof(peer->outbuf->packet.header);
1449     struct bgp_route_list *have = peer->routes;
1450     struct bgp_route_list *want = peer->routing ? bgp_routes : 0;
1451     struct bgp_route_list *e = 0;
1452     struct bgp_route_list *add = 0;
1453     int s;
1454
1455     char *data = (char *) &peer->outbuf->packet.data;
1456
1457     /* need leave room for attr_len, bgp_path_attrs and one prefix */
1458     char *max = (char *) &peer->outbuf->packet.data
1459         + sizeof(peer->outbuf->packet.data)
1460         - sizeof(attr_len) - peer->path_attr_len - sizeof(struct bgp_ip_prefix);
1461
1462     /* skip over unf_len */
1463     data += sizeof(unf_len);
1464     len += sizeof(unf_len);
1465
1466     memset(peer->outbuf->packet.header.marker, 0xff,
1467         sizeof(peer->outbuf->packet.header.marker));
1468
1469     peer->outbuf->packet.header.type = BGP_MSG_UPDATE;
1470
1471     peer->update_routes = 0; /* tentatively clear */
1472
1473     /* find differences */
1474     while ((have || want) && data < (max - sizeof(struct bgp_ip_prefix)))
1475     {
1476         if (have)
1477             s = want
1478                 ? memcmp(&have->dest, &want->dest, sizeof(have->dest))
1479                 : -1;
1480         else
1481             s = 1;
1482
1483         if (s < 0) /* found one to delete */
1484         {
1485             struct bgp_route_list *tmp = have;
1486             have = have->next;
1487
1488             s = BGP_IP_PREFIX_SIZE(tmp->dest);
1489             memcpy(data, &tmp->dest, s);
1490             data += s;
1491             unf_len += s;
1492             len += s;
1493
1494             LOG(5, 0, 0, "Withdrawing route %s/%d from BGP peer %s\n",
1495                 fmtaddr(tmp->dest.prefix, 0), tmp->dest.len, peer->name);
1496
1497             free(tmp);
1498
1499             if (e)
1500                 e->next = have;
1501             else
1502                 peer->routes = have;
1503         }
1504         else
1505         {
1506             if (!s) /* same */
1507             {
1508                 e = have; /* stash the last found to relink above */
1509                 have = have->next;
1510                 want = want->next;
1511             }
1512             else if (s > 0) /* addition reqd. */
1513             {
1514                 if (add)
1515                 {
1516                     peer->update_routes = 1; /* only one add per packet */
1517                     if (!have)
1518                         break;
1519                 }
1520                 else
1521                     add = want;
1522
1523                 if (want)
1524                     want = want->next;
1525             }
1526         }
1527     }
1528
1529     if (have || want)
1530         peer->update_routes = 1; /* more to do */
1531
1532     /* anything changed? */
1533     if (!(unf_len || add))
1534         return 1;
1535
1536     /* go back and insert unf_len */
1537     unf_len = htons(unf_len);
1538     memcpy(&peer->outbuf->packet.data, &unf_len, sizeof(unf_len));
1539
1540     if (add)
1541     {
1542         if (!(e = malloc(sizeof(*e))))
1543         {
1544             LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
1545                 fmtaddr(add->dest.prefix, 0), add->dest.len, strerror(errno));
1546
1547             return 0;
1548         }
1549
1550         memcpy(e, add, sizeof(*e));
1551         e->next = 0;
1552         peer->routes = bgp_insert_route(peer->routes, e);
1553
1554         attr_len = htons(peer->path_attr_len);
1555         memcpy(data, &attr_len, sizeof(attr_len));
1556         data += sizeof(attr_len);
1557         len += sizeof(attr_len);
1558
1559         memcpy(data, peer->path_attrs, peer->path_attr_len);
1560         data += peer->path_attr_len;
1561         len += peer->path_attr_len;
1562
1563         s = BGP_IP_PREFIX_SIZE(add->dest);
1564         memcpy(data, &add->dest, s);
1565         data += s;
1566         len += s;
1567
1568         LOG(5, 0, 0, "Advertising route %s/%d to BGP peer %s\n",
1569             fmtaddr(add->dest.prefix, 0), add->dest.len, peer->name);
1570     }
1571     else
1572     {
1573         attr_len = 0;
1574         memcpy(data, &attr_len, sizeof(attr_len));
1575         data += sizeof(attr_len);
1576         len += sizeof(attr_len);
1577     }
1578
1579     peer->outbuf->packet.header.len = htons(len);
1580     peer->outbuf->done = 0;
1581
1582     return bgp_write(peer);
1583 }
1584
1585 /* send/buffer UPDATE message for IPv6 routes */
1586 static int bgp_send_update6(struct bgp_peer *peer)
1587 {
1588     uint16_t attr_len;
1589     uint16_t unreach_len = 0;
1590     char *unreach_len_pos;
1591     uint8_t reach_len;
1592     uint16_t len = sizeof(peer->outbuf->packet.header);
1593     struct bgp_route6_list *have = peer->routes6;
1594     struct bgp_route6_list *want = peer->routing ? bgp_routes6 : 0;
1595     struct bgp_route6_list *e = 0;
1596     struct bgp_route6_list *add = 0;
1597     int s;
1598     char ipv6addr[INET6_ADDRSTRLEN];
1599
1600     char *data = (char *) &peer->outbuf->packet.data;
1601
1602     /* need leave room for attr_len, bgp_path_attrs and one prefix */
1603     char *max = (char *) &peer->outbuf->packet.data
1604         + sizeof(peer->outbuf->packet.data)
1605         - sizeof(attr_len) - peer->path_attr_len_without_nexthop
1606         - BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE - sizeof(struct bgp_ip6_prefix);
1607
1608     memset(peer->outbuf->packet.header.marker, 0xff,
1609         sizeof(peer->outbuf->packet.header.marker));
1610
1611     peer->outbuf->packet.header.type = BGP_MSG_UPDATE;
1612
1613     /* insert non-MP unfeasible routes length */
1614     memcpy(data, &unreach_len, sizeof(unreach_len));
1615     /* skip over it and attr_len too; it will be filled when known */
1616     data += sizeof(unreach_len) + sizeof(attr_len);
1617     len += sizeof(unreach_len) + sizeof(attr_len);
1618
1619     /* copy usual attributes */
1620     memcpy(data, peer->path_attrs, peer->path_attr_len_without_nexthop);
1621     data += peer->path_attr_len_without_nexthop;
1622     attr_len = peer->path_attr_len_without_nexthop;
1623
1624     /* copy MP unreachable NLRI heading */
1625     memcpy(data, peer->mp_unreach_nlri_partial,
1626             BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE);
1627     /* remember where to update this attr len */
1628     unreach_len_pos = data + 2;
1629     data += BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1630     attr_len += BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1631
1632     peer->update_routes6 = 0; /* tentatively clear */
1633
1634     /* find differences */
1635     while ((have || want) && data < (max - sizeof(struct bgp_ip6_prefix)))
1636     {
1637         if (have)
1638             s = want
1639                 ? memcmp(&have->dest, &want->dest, sizeof(have->dest))
1640                 : -1;
1641         else
1642             s = 1;
1643
1644         if (s < 0) /* found one to delete */
1645         {
1646             struct bgp_route6_list *tmp = have;
1647             have = have->next;
1648
1649             s = BGP_IP_PREFIX_SIZE(tmp->dest);
1650             memcpy(data, &tmp->dest, s);
1651             data += s;
1652             unreach_len += s;
1653             attr_len += s;
1654
1655             LOG(5, 0, 0, "Withdrawing route %s/%d from BGP peer %s\n",
1656                 inet_ntop(AF_INET6, &tmp->dest.prefix, ipv6addr, INET6_ADDRSTRLEN),
1657                 tmp->dest.len, peer->name);
1658
1659             free(tmp);
1660
1661             if (e)
1662                 e->next = have;
1663             else
1664                 peer->routes6 = have;
1665         }
1666         else
1667         {
1668             if (!s) /* same */
1669             {
1670                 e = have; /* stash the last found to relink above */
1671                 have = have->next;
1672                 want = want->next;
1673             }
1674             else if (s > 0) /* addition reqd. */
1675             {
1676                 if (add)
1677                 {
1678                     peer->update_routes6 = 1; /* only one add per packet */
1679                     if (!have)
1680                         break;
1681                 }
1682                 else
1683                     add = want;
1684
1685                 if (want)
1686                     want = want->next;
1687             }
1688         }
1689     }
1690
1691     if (have || want)
1692         peer->update_routes6 = 1; /* more to do */
1693
1694     /* anything changed? */
1695     if (!(unreach_len || add))
1696         return 1;
1697
1698     if (unreach_len)
1699     {
1700         /* go back and insert MP unreach_len */
1701         unreach_len += sizeof(struct bgp_attr_mp_unreach_nlri_partial);
1702         unreach_len = htons(unreach_len);
1703         memcpy(unreach_len_pos, &unreach_len, sizeof(unreach_len));
1704     }
1705     else
1706     {
1707         /* we can remove this attribute, then */
1708         data -= BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1709         attr_len -= BGP_PATH_ATTR_MP_UNREACH_NLRI_PARTIAL_SIZE;
1710     }
1711
1712     if (add)
1713     {
1714         if (!(e = malloc(sizeof(*e))))
1715         {
1716             LOG(0, 0, 0, "Can't allocate route for %s/%d (%s)\n",
1717                 inet_ntop(AF_INET6, &add->dest.prefix, ipv6addr, INET6_ADDRSTRLEN),
1718                 add->dest.len, strerror(errno));
1719
1720             return 0;
1721         }
1722
1723         memcpy(e, add, sizeof(*e));
1724         e->next = 0;
1725         peer->routes6 = bgp_insert_route6(peer->routes6, e);
1726
1727         /* copy MP reachable NLRI heading */
1728         memcpy(data, peer->mp_reach_nlri_partial,
1729                 BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE);
1730         /* with proper len */
1731         reach_len = BGP_IP_PREFIX_SIZE(add->dest);
1732         data[2] = sizeof(struct bgp_attr_mp_reach_nlri_partial) + reach_len;
1733         data += BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE;
1734         attr_len += BGP_PATH_ATTR_MP_REACH_NLRI_PARTIAL_SIZE;
1735
1736         memcpy(data, &add->dest, reach_len);
1737         data += reach_len;
1738         attr_len += reach_len;
1739
1740         LOG(5, 0, 0, "Advertising route %s/%d to BGP peer %s\n",
1741             inet_ntop(AF_INET6, &add->dest.prefix, ipv6addr, INET6_ADDRSTRLEN),
1742             add->dest.len, peer->name);
1743     }
1744
1745     /* update len with attributes we added */
1746     len += attr_len;
1747
1748     /* go back and insert attr_len */
1749     attr_len = htons(attr_len);
1750     memcpy((char *)&peer->outbuf->packet.data + 2, &attr_len, sizeof(attr_len));
1751
1752     peer->outbuf->packet.header.len = htons(len);
1753     peer->outbuf->done = 0;
1754
1755     return bgp_write(peer);
1756 }
1757
1758 /* send/buffer NOTIFICATION message */
1759 static int bgp_send_notification(struct bgp_peer *peer, uint8_t code,
1760     uint8_t subcode)
1761 {
1762     struct bgp_data_notification data;
1763     uint16_t len = 0;
1764
1765     data.error_code = code;
1766     len += sizeof(data.error_code);
1767
1768     data.error_subcode = subcode;
1769     len += sizeof(data.error_code);
1770
1771     memset(peer->outbuf->packet.header.marker, 0xff,
1772         sizeof(peer->outbuf->packet.header.marker));
1773
1774     peer->outbuf->packet.header.type = BGP_MSG_NOTIFICATION;
1775     peer->outbuf->packet.header.len =
1776         htons(sizeof(peer->outbuf->packet.header) + len);
1777
1778     memcpy(peer->outbuf->packet.data, &data, len);
1779
1780     peer->outbuf->done = 0;
1781     peer->next_state = code == BGP_ERR_CEASE ? Disabled : Idle;
1782
1783     /* we're dying; ignore any pending input */
1784     peer->inbuf->packet.header.len = 0;
1785     peer->inbuf->done = 0;
1786
1787     return bgp_write(peer);
1788 }