bgpd/bgp_network.c

   1 /* BGP network related fucntions
   2  * Copyright (C) 1999 Kunihiro Ishiguro
   3  *
   4  * This file is part of GNU Zebra.
   5  *
   6  * GNU Zebra is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; either version 2, or (at your option) any
   9  * later version.
  10  *
  11  * GNU Zebra is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License along
  17  * with this program; see the file COPYING; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include <zebra.h>
  22
  23 #include "thread.h"
  24 #include "sockunion.h"
  25 #include "sockopt.h"
  26 #include "memory.h"
  27 #include "log.h"
  28 #include "if.h"
  29 #include "prefix.h"
  30 #include "command.h"
  31 #include "privs.h"
  32 #include "linklist.h"
  33 #include "network.h"
  34 #include "queue.h"
  35 #include "hash.h"
  36 #include "filter.h"
  37 #include "ns.h"
  38 #include "lib_errors.h"
  39 #include "nexthop.h"
  40
  41 #include "bgpd/bgpd.h"
  42 #include "bgpd/bgp_open.h"
  43 #include "bgpd/bgp_fsm.h"
  44 #include "bgpd/bgp_attr.h"
  45 #include "bgpd/bgp_debug.h"
  46 #include "bgpd/bgp_errors.h"
  47 #include "bgpd/bgp_network.h"
  48 #include "bgpd/bgp_zebra.h"
  49
  50 extern struct zebra_privs_t bgpd_privs;
  51
  52 static char *bgp_get_bound_name(struct peer *peer);
  53
  54 /* BGP listening socket. */
  55 struct bgp_listener {
  56         int fd;
  57         union sockunion su;
  58         struct thread *thread;
  59         struct bgp *bgp;
  60 };
  61
  62 /*
  63  * Set MD5 key for the socket, for the given IPv4 peer address.
  64  * If the password is NULL or zero-length, the option will be disabled.
  65  */
  66 static int bgp_md5_set_socket(int socket, union sockunion *su,
  67                               uint16_t prefixlen, const char *password)
  68 {
  69         int ret = -1;
  70         int en = ENOSYS;
  71 #if HAVE_DECL_TCP_MD5SIG
  72         union sockunion su2;
  73 #endif /* HAVE_TCP_MD5SIG */
  74
  75         assert(socket >= 0);
  76
  77 #if HAVE_DECL_TCP_MD5SIG
  78         /* Ensure there is no extraneous port information. */
  79         memcpy(&su2, su, sizeof(union sockunion));
  80         if (su2.sa.sa_family == AF_INET)
  81                 su2.sin.sin_port = 0;
  82         else
  83                 su2.sin6.sin6_port = 0;
  84
  85         /* For addresses, use the non-extended signature functionality */
  86         if ((su2.sa.sa_family == AF_INET && prefixlen == IPV4_MAX_PREFIXLEN)
  87             || (su2.sa.sa_family == AF_INET6
  88                 && prefixlen == IPV6_MAX_PREFIXLEN))
  89                 ret = sockopt_tcp_signature(socket, &su2, password);
  90         else
  91                 ret = sockopt_tcp_signature_ext(socket, &su2, prefixlen,
  92                                                 password);
  93         en = errno;
  94 #endif /* HAVE_TCP_MD5SIG */
  95
  96         if (ret < 0) {
  97                 char sabuf[SU_ADDRSTRLEN];
  98                 sockunion2str(su, sabuf, sizeof(sabuf));
  99
 100                 switch (ret) {
 101                 case -2:
 102                         flog_warn(
 103                                 EC_BGP_NO_TCP_MD5,
 104                                 "Unable to set TCP MD5 option on socket for peer %s (sock=%d): This platform does not support MD5 auth for prefixes",
 105                                 sabuf, socket);
 106                         break;
 107                 default:
 108                         flog_warn(
 109                                 EC_BGP_NO_TCP_MD5,
 110                                 "Unable to set TCP MD5 option on socket for peer %s (sock=%d): %s",
 111                                 sabuf, socket, safe_strerror(en));
 112                 }
 113         }
 114
 115         return ret;
 116 }
 117
 118 /* Helper for bgp_connect */
 119 static int bgp_md5_set_connect(int socket, union sockunion *su,
 120                                uint16_t prefixlen, const char *password)
 121 {
 122         int ret = -1;
 123
 124 #if HAVE_DECL_TCP_MD5SIG
 125         frr_elevate_privs(&bgpd_privs) {
 126                 ret = bgp_md5_set_socket(socket, su, prefixlen, password);
 127         }
 128 #endif /* HAVE_TCP_MD5SIG */
 129
 130         return ret;
 131 }
 132
 133 static int bgp_md5_set_password(struct peer *peer, const char *password)
 134 {
 135         struct listnode *node;
 136         int ret = 0;
 137         struct bgp_listener *listener;
 138
 139         /*
 140          * Set or unset the password on the listen socket(s). Outbound
 141          * connections are taken care of in bgp_connect() below.
 142          */
 143         frr_elevate_privs(&bgpd_privs)
 144         {
 145                 for (ALL_LIST_ELEMENTS_RO(bm->listen_sockets, node, listener))
 146                         if (listener->su.sa.sa_family
 147                             == peer->su.sa.sa_family) {
 148                                 uint16_t prefixlen =
 149                                         peer->su.sa.sa_family == AF_INET
 150                                                 ? IPV4_MAX_PREFIXLEN
 151                                                 : IPV6_MAX_PREFIXLEN;
 152
 153                                 ret = bgp_md5_set_socket(listener->fd,
 154                                                          &peer->su, prefixlen,
 155                                                          password);
 156                                 break;
 157                         }
 158         }
 159         return ret;
 160 }
 161
 162 int bgp_md5_set_prefix(struct prefix *p, const char *password)
 163 {
 164         int ret = 0;
 165         union sockunion su;
 166         struct listnode *node;
 167         struct bgp_listener *listener;
 168
 169         /* Set or unset the password on the listen socket(s). */
 170         frr_elevate_privs(&bgpd_privs)
 171         {
 172                 for (ALL_LIST_ELEMENTS_RO(bm->listen_sockets, node, listener))
 173                         if (listener->su.sa.sa_family == p->family) {
 174                                 prefix2sockunion(p, &su);
 175                                 ret = bgp_md5_set_socket(listener->fd, &su,
 176                                                          p->prefixlen,
 177                                                          password);
 178                                 break;
 179                         }
 180         }
 181
 182         return ret;
 183 }
 184
 185 int bgp_md5_unset_prefix(struct prefix *p)
 186 {
 187         return bgp_md5_set_prefix(p, NULL);
 188 }
 189
 190 int bgp_md5_set(struct peer *peer)
 191 {
 192         /* Set the password from listen socket. */
 193         return bgp_md5_set_password(peer, peer->password);
 194 }
 195
 196 int bgp_md5_unset(struct peer *peer)
 197 {
 198         /* Unset the password from listen socket. */
 199         return bgp_md5_set_password(peer, NULL);
 200 }
 201
 202 int bgp_set_socket_ttl(struct peer *peer, int bgp_sock)
 203 {
 204         char buf[INET_ADDRSTRLEN];
 205         int ret = 0;
 206
 207         /* In case of peer is EBGP, we should set TTL for this connection.  */
 208         if (!peer->gtsm_hops && (peer_sort(peer) == BGP_PEER_EBGP)) {
 209                 ret = sockopt_ttl(peer->su.sa.sa_family, bgp_sock, peer->ttl);
 210                 if (ret) {
 211                         flog_err(
 212                                 EC_LIB_SOCKET,
 213                                 "%s: Can't set TxTTL on peer (rtrid %s) socket, err = %d",
 214                                 __func__,
 215                                 inet_ntop(AF_INET, &peer->remote_id, buf,
 216                                           sizeof(buf)),
 217                                 errno);
 218                         return ret;
 219                 }
 220         } else if (peer->gtsm_hops) {
 221                 /* On Linux, setting minttl without setting ttl seems to mess
 222                    with the
 223                    outgoing ttl. Therefore setting both.
 224                 */
 225                 ret = sockopt_ttl(peer->su.sa.sa_family, bgp_sock, MAXTTL);
 226                 if (ret) {
 227                         flog_err(
 228                                 EC_LIB_SOCKET,
 229                                 "%s: Can't set TxTTL on peer (rtrid %s) socket, err = %d",
 230                                 __func__,
 231                                 inet_ntop(AF_INET, &peer->remote_id, buf,
 232                                           sizeof(buf)),
 233                                 errno);
 234                         return ret;
 235                 }
 236                 ret = sockopt_minttl(peer->su.sa.sa_family, bgp_sock,
 237                                      MAXTTL + 1 - peer->gtsm_hops);
 238                 if (ret) {
 239                         flog_err(
 240                                 EC_LIB_SOCKET,
 241                                 "%s: Can't set MinTTL on peer (rtrid %s) socket, err = %d",
 242                                 __func__,
 243                                 inet_ntop(AF_INET, &peer->remote_id, buf,
 244                                           sizeof(buf)),
 245                                 errno);
 246                         return ret;
 247                 }
 248         }
 249
 250         return ret;
 251 }
 252
 253 /*
 254  * Obtain the BGP instance that the incoming connection should be processed
 255  * against. This is important because more than one VRF could be using the
 256  * same IP address space. The instance is got by obtaining the device to
 257  * which the incoming connection is bound to. This could either be a VRF
 258  * or it could be an interface, which in turn determines the VRF.
 259  */
 260 static int bgp_get_instance_for_inc_conn(int sock, struct bgp **bgp_inst)
 261 {
 262 #ifndef SO_BINDTODEVICE
 263         /* only Linux has SO_BINDTODEVICE, but we're in Linux-specific code here
 264          * anyway since the assumption is that the interface name returned by
 265          * getsockopt() is useful in identifying the VRF, particularly with
 266          * Linux's
 267          * VRF l3master device.  The whole mechanism is specific to Linux, so...
 268          * when other platforms add VRF support, this will need handling here as
 269          * well.  (or, some restructuring) */
 270         *bgp_inst = bgp_get_default();
 271         return !*bgp_inst;
 272
 273 #else
 274         char name[VRF_NAMSIZ + 1];
 275         socklen_t name_len = VRF_NAMSIZ;
 276         struct bgp *bgp;
 277         int rc;
 278         struct listnode *node, *nnode;
 279
 280         *bgp_inst = NULL;
 281         name[0] = '\0';
 282         rc = getsockopt(sock, SOL_SOCKET, SO_BINDTODEVICE, name, &name_len);
 283         if (rc != 0) {
 284 #if defined(HAVE_CUMULUS)
 285                 flog_err(EC_LIB_SOCKET,
 286                          "[Error] BGP SO_BINDTODEVICE get failed (%s), sock %d",
 287                          safe_strerror(errno), sock);
 288                 return -1;
 289 #endif
 290         }
 291
 292         if (!strlen(name)) {
 293                 *bgp_inst = bgp_get_default();
 294                 return 0; /* default instance. */
 295         }
 296
 297         /* First try match to instance; if that fails, check for interfaces. */
 298         bgp = bgp_lookup_by_name(name);
 299         if (bgp) {
 300                 if (!bgp->vrf_id) // unexpected
 301                         return -1;
 302                 *bgp_inst = bgp;
 303                 return 0;
 304         }
 305
 306         /* TODO - This will be optimized once interfaces move into the NS */
 307         for (ALL_LIST_ELEMENTS(bm->bgp, node, nnode, bgp)) {
 308                 struct interface *ifp;
 309
 310                 if (bgp->inst_type == BGP_INSTANCE_TYPE_VIEW)
 311                         continue;
 312
 313                 ifp = if_lookup_by_name(name, bgp->vrf_id);
 314                 if (ifp) {
 315                         *bgp_inst = bgp;
 316                         return 0;
 317                 }
 318         }
 319
 320         /* We didn't match to either an instance or an interface. */
 321         return -1;
 322 #endif
 323 }
 324
 325 /* Accept bgp connection. */
 326 static int bgp_accept(struct thread *thread)
 327 {
 328         int bgp_sock;
 329         int accept_sock;
 330         union sockunion su;
 331         struct bgp_listener *listener = THREAD_ARG(thread);
 332         struct peer *peer;
 333         struct peer *peer1;
 334         char buf[SU_ADDRSTRLEN];
 335         struct bgp *bgp = NULL;
 336
 337         sockunion_init(&su);
 338
 339         /* Register accept thread. */
 340         accept_sock = THREAD_FD(thread);
 341         if (accept_sock < 0) {
 342                 flog_err_sys(EC_LIB_SOCKET, "accept_sock is nevative value %d",
 343                              accept_sock);
 344                 return -1;
 345         }
 346         listener->thread = NULL;
 347
 348         thread_add_read(bm->master, bgp_accept, listener, accept_sock,
 349                         &listener->thread);
 350
 351         /* Accept client connection. */
 352         bgp_sock = sockunion_accept(accept_sock, &su);
 353         if (bgp_sock < 0) {
 354                 flog_err_sys(EC_LIB_SOCKET,
 355                              "[Error] BGP socket accept failed (%s)",
 356                              safe_strerror(errno));
 357                 return -1;
 358         }
 359         set_nonblocking(bgp_sock);
 360
 361         /* Obtain BGP instance this connection is meant for.
 362          * - if it is a VRF netns sock, then BGP is in listener structure
 363          * - otherwise, the bgp instance need to be demultiplexed
 364          */
 365         if (listener->bgp)
 366                 bgp = listener->bgp;
 367         else if (bgp_get_instance_for_inc_conn(bgp_sock, &bgp)) {
 368                 if (bgp_debug_neighbor_events(NULL))
 369                         zlog_debug(
 370                                 "[Event] Could not get instance for incoming conn from %s",
 371                                 inet_sutop(&su, buf));
 372                 close(bgp_sock);
 373                 return -1;
 374         }
 375
 376         /* Set socket send buffer size */
 377         setsockopt_so_sendbuf(bgp_sock, BGP_SOCKET_SNDBUF_SIZE);
 378
 379         /* Check remote IP address */
 380         peer1 = peer_lookup(bgp, &su);
 381
 382         if (!peer1) {
 383                 peer1 = peer_lookup_dynamic_neighbor(bgp, &su);
 384                 if (peer1) {
 385                         /* Dynamic neighbor has been created, let it proceed */
 386                         peer1->fd = bgp_sock;
 387                         bgp_fsm_change_status(peer1, Active);
 388                         BGP_TIMER_OFF(
 389                                 peer1->t_start); /* created in peer_create() */
 390
 391                         if (peer_active(peer1))
 392                                 BGP_EVENT_ADD(peer1, TCP_connection_open);
 393
 394                         return 0;
 395                 }
 396         }
 397
 398         if (!peer1) {
 399                 if (bgp_debug_neighbor_events(NULL)) {
 400                         zlog_debug(
 401                                 "[Event] %s connection rejected - not configured"
 402                                 " and not valid for dynamic",
 403                                 inet_sutop(&su, buf));
 404                 }
 405                 close(bgp_sock);
 406                 return -1;
 407         }
 408
 409         if (CHECK_FLAG(peer1->flags, PEER_FLAG_SHUTDOWN)) {
 410                 if (bgp_debug_neighbor_events(peer1))
 411                         zlog_debug(
 412                                 "[Event] connection from %s rejected due to admin shutdown",
 413                                 inet_sutop(&su, buf));
 414                 close(bgp_sock);
 415                 return -1;
 416         }
 417
 418         /*
 419          * Do not accept incoming connections in Clearing state. This can result
 420          * in incorect state transitions - e.g., the connection goes back to
 421          * Established and then the Clearing_Completed event is generated. Also,
 422          * block incoming connection in Deleted state.
 423          */
 424         if (peer1->status == Clearing || peer1->status == Deleted) {
 425                 if (bgp_debug_neighbor_events(peer1))
 426                         zlog_debug(
 427                                 "[Event] Closing incoming conn for %s (%p) state %d",
 428                                 peer1->host, peer1, peer1->status);
 429                 close(bgp_sock);
 430                 return -1;
 431         }
 432
 433         /* Check that at least one AF is activated for the peer. */
 434         if (!peer_active(peer1)) {
 435                 if (bgp_debug_neighbor_events(peer1))
 436                         zlog_debug(
 437                                 "%s - incoming conn rejected - no AF activated for peer",
 438                                 peer1->host);
 439                 close(bgp_sock);
 440                 return -1;
 441         }
 442
 443         /* Do not try to reconnect if the peer reached maximum
 444          * prefixes, restart timer is still running or the peer
 445          * is shutdown.
 446          */
 447         if (BGP_PEER_START_SUPPRESSED(peer1)) {
 448                 if (bgp_debug_neighbor_events(peer1))
 449                         zlog_debug(
 450                                 "[Event] Incoming BGP connection rejected from %s "
 451                                 "due to maximum-prefix or shutdown",
 452                                 peer1->host);
 453                 close(bgp_sock);
 454                 return -1;
 455         }
 456
 457         if (bgp_debug_neighbor_events(peer1))
 458                 zlog_debug("[Event] BGP connection from host %s fd %d",
 459                            inet_sutop(&su, buf), bgp_sock);
 460
 461         if (peer1->doppelganger) {
 462                 /* We have an existing connection. Kill the existing one and run
 463                    with this one.
 464                 */
 465                 if (bgp_debug_neighbor_events(peer1))
 466                         zlog_debug(
 467                                 "[Event] New active connection from peer %s, Killing"
 468                                 " previous active connection",
 469                                 peer1->host);
 470                 peer_delete(peer1->doppelganger);
 471         }
 472
 473         if (bgp_set_socket_ttl(peer1, bgp_sock) < 0)
 474                 if (bgp_debug_neighbor_events(peer1))
 475                         zlog_debug(
 476                                 "[Event] Unable to set min/max TTL on peer %s, Continuing",
 477                                 peer1->host);
 478
 479         peer = peer_create(&su, peer1->conf_if, peer1->bgp, peer1->local_as,
 480                            peer1->as, peer1->as_type, 0, 0, NULL);
 481         hash_release(peer->bgp->peerhash, peer);
 482         hash_get(peer->bgp->peerhash, peer, hash_alloc_intern);
 483
 484         peer_xfer_config(peer, peer1);
 485         UNSET_FLAG(peer->flags, PEER_FLAG_CONFIG_NODE);
 486
 487         peer->doppelganger = peer1;
 488         peer1->doppelganger = peer;
 489         peer->fd = bgp_sock;
 490         vrf_bind(peer->bgp->vrf_id, bgp_sock, bgp_get_bound_name(peer));
 491         bgp_fsm_change_status(peer, Active);
 492         BGP_TIMER_OFF(peer->t_start); /* created in peer_create() */
 493
 494         SET_FLAG(peer->sflags, PEER_STATUS_ACCEPT_PEER);
 495
 496         /* Make dummy peer until read Open packet. */
 497         if (peer1->status == Established
 498             && CHECK_FLAG(peer1->sflags, PEER_STATUS_NSF_MODE)) {
 499                 /* If we have an existing established connection with graceful
 500                  * restart
 501                  * capability announced with one or more address families, then
 502                  * drop
 503                  * existing established connection and move state to connect.
 504                  */
 505                 peer1->last_reset = PEER_DOWN_NSF_CLOSE_SESSION;
 506                 SET_FLAG(peer1->sflags, PEER_STATUS_NSF_WAIT);
 507                 bgp_event_update(peer1, TCP_connection_closed);
 508         }
 509
 510         if (peer_active(peer)) {
 511                 BGP_EVENT_ADD(peer, TCP_connection_open);
 512         }
 513
 514         return 0;
 515 }
 516
 517 /* BGP socket bind. */
 518 static char *bgp_get_bound_name(struct peer *peer)
 519 {
 520         char *name = NULL;
 521
 522         if (!peer)
 523                 return NULL;
 524
 525         if ((peer->bgp->vrf_id == VRF_DEFAULT) && !peer->ifname
 526             && !peer->conf_if)
 527                 return NULL;
 528
 529         if (peer->su.sa.sa_family != AF_INET
 530             && peer->su.sa.sa_family != AF_INET6)
 531                 return NULL; // unexpected
 532
 533         /* For IPv6 peering, interface (unnumbered or link-local with interface)
 534          * takes precedence over VRF. For IPv4 peering, explicit interface or
 535          * VRF are the situations to bind.
 536          */
 537         if (peer->su.sa.sa_family == AF_INET6)
 538                 name = (peer->conf_if ? peer->conf_if
 539                                       : (peer->ifname ? peer->ifname
 540                                                       : peer->bgp->name));
 541         else
 542                 name = peer->ifname ? peer->ifname : peer->bgp->name;
 543
 544         return name;
 545 }
 546
 547 static int bgp_update_address(struct interface *ifp, const union sockunion *dst,
 548                               union sockunion *addr)
 549 {
 550         struct prefix *p, *sel, d;
 551         struct connected *connected;
 552         struct listnode *node;
 553         int common;
 554
 555         sockunion2hostprefix(dst, &d);
 556         sel = NULL;
 557         common = -1;
 558
 559         for (ALL_LIST_ELEMENTS_RO(ifp->connected, node, connected)) {
 560                 p = connected->address;
 561                 if (p->family != d.family)
 562                         continue;
 563                 if (prefix_common_bits(p, &d) > common) {
 564                         sel = p;
 565                         common = prefix_common_bits(sel, &d);
 566                 }
 567         }
 568
 569         if (!sel)
 570                 return 1;
 571
 572         prefix2sockunion(sel, addr);
 573         return 0;
 574 }
 575
 576 /* Update source selection.  */
 577 static int bgp_update_source(struct peer *peer)
 578 {
 579         struct interface *ifp;
 580         union sockunion addr;
 581         int ret = 0;
 582
 583         sockunion_init(&addr);
 584
 585         /* Source is specified with interface name.  */
 586         if (peer->update_if) {
 587                 ifp = if_lookup_by_name(peer->update_if, peer->bgp->vrf_id);
 588                 if (!ifp)
 589                         return -1;
 590
 591                 if (bgp_update_address(ifp, &peer->su, &addr))
 592                         return -1;
 593
 594                 ret = sockunion_bind(peer->fd, &addr, 0, &addr);
 595         }
 596
 597         /* Source is specified with IP address.  */
 598         if (peer->update_source)
 599                 ret = sockunion_bind(peer->fd, peer->update_source, 0,
 600                                      peer->update_source);
 601
 602         return ret;
 603 }
 604
 605 #define DATAPLANE_MARK 254      /* main table ID */
 606
 607 /* BGP try to connect to the peer.  */
 608 int bgp_connect(struct peer *peer)
 609 {
 610         assert(!CHECK_FLAG(peer->thread_flags, PEER_THREAD_WRITES_ON));
 611         assert(!CHECK_FLAG(peer->thread_flags, PEER_THREAD_READS_ON));
 612         ifindex_t ifindex = 0;
 613
 614         if (peer->conf_if && BGP_PEER_SU_UNSPEC(peer)) {
 615                 zlog_debug("Peer address not learnt: Returning from connect");
 616                 return 0;
 617         }
 618         frr_elevate_privs(&bgpd_privs) {
 619         /* Make socket for the peer. */
 620                 peer->fd = vrf_sockunion_socket(&peer->su, peer->bgp->vrf_id,
 621                                                 bgp_get_bound_name(peer));
 622         }
 623         if (peer->fd < 0)
 624                 return -1;
 625
 626         set_nonblocking(peer->fd);
 627
 628         /* Set socket send buffer size */
 629         setsockopt_so_sendbuf(peer->fd, BGP_SOCKET_SNDBUF_SIZE);
 630
 631         if (bgp_set_socket_ttl(peer, peer->fd) < 0)
 632                 return -1;
 633
 634         sockopt_reuseaddr(peer->fd);
 635         sockopt_reuseport(peer->fd);
 636         if (sockopt_mark_default(peer->fd, DATAPLANE_MARK, &bgpd_privs) < 0)
 637                 flog_warn(EC_BGP_NO_SOCKOPT_MARK,
 638                           "Unable to set mark on FD for peer %s, err=%s",
 639                           peer->host, safe_strerror(errno));
 640
 641 #ifdef IPTOS_PREC_INTERNETCONTROL
 642         frr_elevate_privs(&bgpd_privs) {
 643                 if (sockunion_family(&peer->su) == AF_INET)
 644                         setsockopt_ipv4_tos(peer->fd,
 645                                             IPTOS_PREC_INTERNETCONTROL);
 646                 else if (sockunion_family(&peer->su) == AF_INET6)
 647                         setsockopt_ipv6_tclass(peer->fd,
 648                                                IPTOS_PREC_INTERNETCONTROL);
 649         }
 650 #endif
 651
 652         if (peer->password) {
 653                 uint16_t prefixlen = peer->su.sa.sa_family == AF_INET
 654                                              ? IPV4_MAX_PREFIXLEN
 655                                              : IPV6_MAX_PREFIXLEN;
 656
 657                 bgp_md5_set_connect(peer->fd, &peer->su, prefixlen,
 658                                     peer->password);
 659         }
 660
 661         /* Update source bind. */
 662         if (bgp_update_source(peer) < 0) {
 663                 return connect_error;
 664         }
 665
 666         if (peer->conf_if || peer->ifname)
 667                 ifindex = ifname2ifindex(peer->conf_if ? peer->conf_if
 668                                                        : peer->ifname,
 669                                          peer->bgp->vrf_id);
 670
 671         if (bgp_debug_neighbor_events(peer))
 672                 zlog_debug("%s [Event] Connect start to %s fd %d", peer->host,
 673                            peer->host, peer->fd);
 674
 675         /* Connect to the remote peer. */
 676         return sockunion_connect(peer->fd, &peer->su, htons(peer->port),
 677                                  ifindex);
 678 }
 679
 680 /* After TCP connection is established.  Get local address and port. */
 681 int bgp_getsockname(struct peer *peer)
 682 {
 683         if (peer->su_local) {
 684                 sockunion_free(peer->su_local);
 685                 peer->su_local = NULL;
 686         }
 687
 688         if (peer->su_remote) {
 689                 sockunion_free(peer->su_remote);
 690                 peer->su_remote = NULL;
 691         }
 692
 693         peer->su_local = sockunion_getsockname(peer->fd);
 694         if (!peer->su_local)
 695                 return -1;
 696         peer->su_remote = sockunion_getpeername(peer->fd);
 697         if (!peer->su_remote)
 698                 return -1;
 699
 700         if (!bgp_zebra_nexthop_set(peer->su_local, peer->su_remote,
 701                                    &peer->nexthop, peer)) {
 702                 flog_err(EC_BGP_NH_UPD,
 703                          "%s: nexthop_set failed, resetting connection - intf %p",
 704                          peer->host, peer->nexthop.ifp);
 705                 return -1;
 706         }
 707         return 0;
 708 }
 709
 710
 711 static int bgp_listener(int sock, struct sockaddr *sa, socklen_t salen,
 712                         struct bgp *bgp)
 713 {
 714         struct bgp_listener *listener;
 715         int ret, en;
 716
 717         sockopt_reuseaddr(sock);
 718         sockopt_reuseport(sock);
 719
 720         frr_elevate_privs(&bgpd_privs) {
 721
 722 #ifdef IPTOS_PREC_INTERNETCONTROL
 723                 if (sa->sa_family == AF_INET)
 724                         setsockopt_ipv4_tos(sock, IPTOS_PREC_INTERNETCONTROL);
 725                 else if (sa->sa_family == AF_INET6)
 726                         setsockopt_ipv6_tclass(sock,
 727                                                IPTOS_PREC_INTERNETCONTROL);
 728 #endif
 729
 730                 sockopt_v6only(sa->sa_family, sock);
 731
 732                 ret = bind(sock, sa, salen);
 733                 en = errno;
 734         }
 735
 736         if (ret < 0) {
 737                 flog_err_sys(EC_LIB_SOCKET, "bind: %s", safe_strerror(en));
 738                 return ret;
 739         }
 740
 741         ret = listen(sock, SOMAXCONN);
 742         if (ret < 0) {
 743                 flog_err_sys(EC_LIB_SOCKET, "listen: %s", safe_strerror(errno));
 744                 return ret;
 745         }
 746
 747         listener = XCALLOC(MTYPE_BGP_LISTENER, sizeof(*listener));
 748         listener->fd = sock;
 749
 750         /* this socket needs a change of ns. record bgp back pointer */
 751         if (bgp->vrf_id != VRF_DEFAULT && vrf_is_backend_netns())
 752                 listener->bgp = bgp;
 753
 754         memcpy(&listener->su, sa, salen);
 755         listener->thread = NULL;
 756         thread_add_read(bm->master, bgp_accept, listener, sock,
 757                         &listener->thread);
 758         listnode_add(bm->listen_sockets, listener);
 759
 760         return 0;
 761 }
 762
 763 /* IPv6 supported version of BGP server socket setup.  */
 764 int bgp_socket(struct bgp *bgp, unsigned short port, const char *address)
 765 {
 766         struct addrinfo *ainfo;
 767         struct addrinfo *ainfo_save;
 768         static const struct addrinfo req = {
 769                 .ai_family = AF_UNSPEC,
 770                 .ai_flags = AI_PASSIVE,
 771                 .ai_socktype = SOCK_STREAM,
 772         };
 773         int ret, count;
 774         char port_str[BUFSIZ];
 775
 776         snprintf(port_str, sizeof(port_str), "%d", port);
 777         port_str[sizeof(port_str) - 1] = '\0';
 778
 779         frr_elevate_privs(&bgpd_privs) {
 780                 ret = vrf_getaddrinfo(address, port_str, &req, &ainfo_save,
 781                                       bgp->vrf_id);
 782         }
 783         if (ret != 0) {
 784                 flog_err_sys(EC_LIB_SOCKET, "getaddrinfo: %s",
 785                              gai_strerror(ret));
 786                 return -1;
 787         }
 788         if (bgp_option_check(BGP_OPT_NO_ZEBRA) &&
 789             bgp->vrf_id != VRF_DEFAULT) {
 790                 freeaddrinfo(ainfo_save);
 791                 return -1;
 792         }
 793         count = 0;
 794         for (ainfo = ainfo_save; ainfo; ainfo = ainfo->ai_next) {
 795                 int sock;
 796
 797                 if (ainfo->ai_family != AF_INET && ainfo->ai_family != AF_INET6)
 798                         continue;
 799
 800                 frr_elevate_privs(&bgpd_privs) {
 801                         sock = vrf_socket(ainfo->ai_family,
 802                                           ainfo->ai_socktype,
 803                                           ainfo->ai_protocol, bgp->vrf_id,
 804                                           (bgp->inst_type
 805                                            == BGP_INSTANCE_TYPE_VRF
 806                                            ? bgp->name : NULL));
 807                 }
 808                 if (sock < 0) {
 809                         flog_err_sys(EC_LIB_SOCKET, "socket: %s",
 810                                      safe_strerror(errno));
 811                         continue;
 812                 }
 813
 814                 /* if we intend to implement ttl-security, this socket needs
 815                  * ttl=255 */
 816                 sockopt_ttl(ainfo->ai_family, sock, MAXTTL);
 817
 818                 ret = bgp_listener(sock, ainfo->ai_addr, ainfo->ai_addrlen,
 819                                    bgp);
 820                 if (ret == 0)
 821                         ++count;
 822                 else
 823                         close(sock);
 824         }
 825         freeaddrinfo(ainfo_save);
 826         if (count == 0 && bgp->inst_type != BGP_INSTANCE_TYPE_VRF) {
 827                 flog_err(
 828                         EC_LIB_SOCKET,
 829                         "%s: no usable addresses please check other programs usage of specified port %d",
 830                         __func__, port);
 831                 flog_err_sys(EC_LIB_SOCKET, "%s: Program cannot continue",
 832                              __func__);
 833                 exit(-1);
 834         }
 835
 836         return 0;
 837 }
 838
 839 /* this function closes vrf socket
 840  * this should be called only for vrf socket with netns backend
 841  */
 842 void bgp_close_vrf_socket(struct bgp *bgp)
 843 {
 844         struct listnode *node, *next;
 845         struct bgp_listener *listener;
 846
 847         if (!bgp)
 848                 return;
 849
 850         if (bm->listen_sockets == NULL)
 851                 return;
 852
 853         for (ALL_LIST_ELEMENTS(bm->listen_sockets, node, next, listener)) {
 854                 if (listener->bgp == bgp) {
 855                         thread_cancel(listener->thread);
 856                         close(listener->fd);
 857                         listnode_delete(bm->listen_sockets, listener);
 858                         XFREE(MTYPE_BGP_LISTENER, listener);
 859                 }
 860         }
 861 }
 862
 863 /* this function closes main socket
 864  */
 865 void bgp_close(void)
 866 {
 867         struct listnode *node, *next;
 868         struct bgp_listener *listener;
 869
 870         if (bm->listen_sockets == NULL)
 871                 return;
 872
 873         for (ALL_LIST_ELEMENTS(bm->listen_sockets, node, next, listener)) {
 874                 if (listener->bgp)
 875                         continue;
 876                 thread_cancel(listener->thread);
 877                 close(listener->fd);
 878                 listnode_delete(bm->listen_sockets, listener);
 879                 XFREE(MTYPE_BGP_LISTENER, listener);
 880         }
 881 }