bgpd/bgp_label.c

   1 /* BGP carrying label information
   2  * Copyright (C) 2013 Cumulus Networks, Inc.
   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 "command.h"
  24 #include "thread.h"
  25 #include "prefix.h"
  26 #include "zclient.h"
  27 #include "stream.h"
  28 #include "network.h"
  29 #include "log.h"
  30 #include "memory.h"
  31 #include "nexthop.h"
  32 #include "mpls.h"
  33
  34 #include "bgpd/bgpd.h"
  35 #include "bgpd/bgp_table.h"
  36 #include "bgpd/bgp_route.h"
  37 #include "bgpd/bgp_attr.h"
  38 #include "bgpd/bgp_label.h"
  39 #include "bgpd/bgp_packet.h"
  40 #include "bgpd/bgp_debug.h"
  41 #include "bgpd/bgp_errors.h"
  42
  43 extern struct zclient *zclient;
  44
  45 int bgp_parse_fec_update(void)
  46 {
  47         struct stream *s;
  48         struct bgp_dest *dest;
  49         struct bgp *bgp;
  50         struct bgp_table *table;
  51         struct prefix p;
  52         uint32_t label;
  53         afi_t afi;
  54         safi_t safi;
  55
  56         s = zclient->ibuf;
  57
  58         memset(&p, 0, sizeof(p));
  59         p.family = stream_getw(s);
  60         p.prefixlen = stream_getc(s);
  61         stream_get(p.u.val, s, PSIZE(p.prefixlen));
  62         label = stream_getl(s);
  63
  64         /* hack for the bgp instance & SAFI = have to send/receive it */
  65         afi = family2afi(p.family);
  66         safi = SAFI_UNICAST;
  67         bgp = bgp_get_default();
  68         if (!bgp) {
  69                 zlog_debug("no default bgp instance");
  70                 return -1;
  71         }
  72
  73         table = bgp->rib[afi][safi];
  74         if (!table) {
  75                 zlog_debug("no %u unicast table", p.family);
  76                 return -1;
  77         }
  78         dest = bgp_node_lookup(table, &p);
  79         if (!dest) {
  80                 zlog_debug("no node for the prefix");
  81                 return -1;
  82         }
  83
  84         /* treat it as implicit withdraw - the label is invalid */
  85         if (label == MPLS_INVALID_LABEL)
  86                 bgp_unset_valid_label(&dest->local_label);
  87         else {
  88                 dest->local_label = mpls_lse_encode(label, 0, 0, 1);
  89                 bgp_set_valid_label(&dest->local_label);
  90         }
  91         SET_FLAG(dest->flags, BGP_NODE_LABEL_CHANGED);
  92         bgp_process(bgp, dest, afi, safi);
  93         bgp_dest_unlock_node(dest);
  94         return 1;
  95 }
  96
  97 mpls_label_t bgp_adv_label(struct bgp_dest *dest, struct bgp_path_info *pi,
  98                            struct peer *to, afi_t afi, safi_t safi)
  99 {
 100         struct peer *from;
 101         mpls_label_t remote_label;
 102         int reflect;
 103
 104         if (!dest || !pi || !to)
 105                 return MPLS_INVALID_LABEL;
 106
 107         remote_label = pi->extra ? pi->extra->label[0] : MPLS_INVALID_LABEL;
 108         from = pi->peer;
 109         reflect =
 110                 ((from->sort == BGP_PEER_IBGP) && (to->sort == BGP_PEER_IBGP));
 111
 112         if (reflect
 113             && !CHECK_FLAG(to->af_flags[afi][safi],
 114                            PEER_FLAG_FORCE_NEXTHOP_SELF))
 115                 return remote_label;
 116
 117         if (CHECK_FLAG(to->af_flags[afi][safi], PEER_FLAG_NEXTHOP_UNCHANGED))
 118                 return remote_label;
 119
 120         return dest->local_label;
 121 }
 122
 123 static void bgp_send_fec_register_label_msg(struct bgp_dest *dest, bool reg,
 124                                             uint32_t label_index)
 125 {
 126         struct stream *s;
 127         int command;
 128         const struct prefix *p;
 129         uint16_t flags = 0;
 130         size_t flags_pos = 0;
 131         mpls_label_t *local_label = &(dest->local_label);
 132         uint32_t ttl = 0;
 133         uint32_t bos = 0;
 134         uint32_t exp = 0;
 135         mpls_label_t label = MPLS_INVALID_LABEL;
 136         bool have_label_to_reg;
 137
 138         mpls_lse_decode(*local_label, &label, &ttl, &exp, &bos);
 139
 140         have_label_to_reg = bgp_is_valid_label(local_label) &&
 141                             label != MPLS_LABEL_IMPLICIT_NULL;
 142
 143         p = bgp_dest_get_prefix(dest);
 144
 145         /* Check socket. */
 146         if (!zclient || zclient->sock < 0)
 147                 return;
 148
 149         if (BGP_DEBUG(labelpool, LABELPOOL))
 150                 zlog_debug("%s: FEC %sregister %pRN label_index=%u label=%u",
 151                            __func__, reg ? "" : "un", bgp_dest_to_rnode(dest),
 152                            label_index, label);
 153         /* If the route node has a local_label assigned or the
 154          * path node has an MPLS SR label index allowing zebra to
 155          * derive the label, proceed with registration. */
 156         s = zclient->obuf;
 157         stream_reset(s);
 158         command = (reg) ? ZEBRA_FEC_REGISTER : ZEBRA_FEC_UNREGISTER;
 159         zclient_create_header(s, command, VRF_DEFAULT);
 160         flags_pos = stream_get_endp(s); /* save position of 'flags' */
 161         stream_putw(s, flags);          /* initial flags */
 162         stream_putw(s, PREFIX_FAMILY(p));
 163         stream_put_prefix(s, p);
 164         if (reg) {
 165                 /* label index takes precedence over auto-assigned label. */
 166                 if (label_index != 0) {
 167                         flags |= ZEBRA_FEC_REGISTER_LABEL_INDEX;
 168                         stream_putl(s, label_index);
 169                 } else if (have_label_to_reg) {
 170                         flags |= ZEBRA_FEC_REGISTER_LABEL;
 171                         stream_putl(s, label);
 172                 }
 173                 SET_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 174         } else
 175                 UNSET_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 176
 177         /* Set length and flags */
 178         stream_putw_at(s, 0, stream_get_endp(s));
 179
 180         /*
 181          * We only need to write new flags if this is a register
 182          */
 183         if (reg)
 184                 stream_putw_at(s, flags_pos, flags);
 185
 186         zclient_send_message(zclient);
 187 }
 188
 189 /**
 190  * This is passed as the callback function to bgp_labelpool.c:bgp_lp_get()
 191  * by bgp_reg_dereg_for_label() when a label needs to be obtained from
 192  * label pool.
 193  * Note that it will reject the allocated label if a label index is found,
 194  * because the label index supposes predictable labels
 195  */
 196 int bgp_reg_for_label_callback(mpls_label_t new_label, void *labelid,
 197                                bool allocated)
 198 {
 199         struct bgp_dest *dest;
 200
 201         dest = labelid;
 202
 203         /*
 204          * if the route had been removed or the request has gone then reject
 205          * the allocated label. The requesting code will have done what is
 206          * required to allocate the correct label
 207          */
 208         if (!CHECK_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED)) {
 209                 bgp_dest_unlock_node(dest);
 210                 return -1;
 211         }
 212
 213         bgp_dest_unlock_node(dest);
 214
 215         if (BGP_DEBUG(labelpool, LABELPOOL))
 216                 zlog_debug("%s: FEC %pRN label=%u, allocated=%d", __func__,
 217                            bgp_dest_to_rnode(dest), new_label, allocated);
 218
 219         if (!allocated) {
 220                 /*
 221                  * previously-allocated label is now invalid, set to implicit
 222                  * null until new label arrives
 223                  */
 224                 if (CHECK_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL)) {
 225                         UNSET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 226                         dest->local_label = mpls_lse_encode(
 227                                 MPLS_LABEL_IMPLICIT_NULL, 0, 0, 1);
 228                         bgp_set_valid_label(&dest->local_label);
 229                 }
 230         }
 231
 232         dest->local_label = mpls_lse_encode(new_label, 0, 0, 1);
 233         bgp_set_valid_label(&dest->local_label);
 234
 235         /*
 236          * Get back to registering the FEC
 237          */
 238         bgp_send_fec_register_label_msg(dest, true, 0);
 239
 240         return 0;
 241 }
 242
 243 void bgp_reg_dereg_for_label(struct bgp_dest *dest, struct bgp_path_info *pi,
 244                              bool reg)
 245 {
 246         bool with_label_index = false;
 247         const struct prefix *p;
 248         bool have_label_to_reg;
 249         uint32_t ttl = 0;
 250         uint32_t bos = 0;
 251         uint32_t exp = 0;
 252         mpls_label_t label = MPLS_INVALID_LABEL;
 253
 254         mpls_lse_decode(dest->local_label, &label, &ttl, &exp, &bos);
 255
 256         have_label_to_reg = bgp_is_valid_label(&dest->local_label) &&
 257                             label != MPLS_LABEL_IMPLICIT_NULL;
 258
 259         p = bgp_dest_get_prefix(dest);
 260
 261         if (BGP_DEBUG(labelpool, LABELPOOL))
 262                 zlog_debug("%s: %pFX: %s ", __func__, p,
 263                            (reg ? "reg" : "dereg"));
 264
 265         if (reg) {
 266                 assert(pi);
 267                 /*
 268                  * Determine if we will let zebra should derive label from
 269                  * label index instead of bgpd requesting from label pool
 270                  */
 271                 if (CHECK_FLAG(pi->attr->flag,
 272                             ATTR_FLAG_BIT(BGP_ATTR_PREFIX_SID))
 273                         && pi->attr->label_index != BGP_INVALID_LABEL_INDEX) {
 274                         with_label_index = true;
 275                         UNSET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 276                 } else {
 277                         /*
 278                          * If no label has been registered -- assume any label
 279                          * from label pool will do. This means that label index
 280                          * always takes precedence over auto-assigned labels.
 281                          */
 282                         if (!have_label_to_reg) {
 283                                 SET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 284                                 if (BGP_DEBUG(labelpool, LABELPOOL))
 285                                         zlog_debug(
 286                                                 "%s: Requesting label from LP for %pFX",
 287                                                 __func__, p);
 288                                 /* bgp_reg_for_label_callback() will deal with
 289                                  * fec registration when it gets a label from
 290                                  * the pool. This means we'll never register
 291                                  * FECs withoutvalid labels.
 292                                  */
 293                                 bgp_lp_get(LP_TYPE_BGP_LU, dest,
 294                                            bgp_reg_for_label_callback);
 295                                 return;
 296                         }
 297                 }
 298         } else {
 299                 UNSET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 300                 bgp_lp_release(LP_TYPE_BGP_LU, dest, label);
 301         }
 302
 303         bgp_send_fec_register_label_msg(
 304                 dest, reg, with_label_index ? pi->attr->label_index : 0);
 305 }
 306
 307 static int bgp_nlri_get_labels(struct peer *peer, uint8_t *pnt, uint8_t plen,
 308                                mpls_label_t *label)
 309 {
 310         uint8_t *data = pnt;
 311         uint8_t *lim = pnt + plen;
 312         uint8_t llen = 0;
 313         uint8_t label_depth = 0;
 314
 315         for (; data < lim; data += BGP_LABEL_BYTES) {
 316                 memcpy(label, data, BGP_LABEL_BYTES);
 317                 llen += BGP_LABEL_BYTES;
 318
 319                 bgp_set_valid_label(label);
 320                 label_depth += 1;
 321
 322                 if (bgp_is_withdraw_label(label) || label_bos(label))
 323                         break;
 324         }
 325
 326         /* If we RX multiple labels we will end up keeping only the last
 327          * one. We do not yet support a label stack greater than 1. */
 328         if (label_depth > 1)
 329                 zlog_info("%pBP rcvd UPDATE with label stack %d deep", peer,
 330                           label_depth);
 331
 332         if (!(bgp_is_withdraw_label(label) || label_bos(label)))
 333                 flog_warn(
 334                         EC_BGP_INVALID_LABEL_STACK,
 335                         "%pBP rcvd UPDATE with invalid label stack - no bottom of stack",
 336                         peer);
 337
 338         return llen;
 339 }
 340
 341 int bgp_nlri_parse_label(struct peer *peer, struct attr *attr,
 342                          struct bgp_nlri *packet)
 343 {
 344         uint8_t *pnt;
 345         uint8_t *lim;
 346         struct prefix p;
 347         int psize = 0;
 348         int prefixlen;
 349         afi_t afi;
 350         safi_t safi;
 351         bool addpath_capable;
 352         uint32_t addpath_id;
 353         mpls_label_t label = MPLS_INVALID_LABEL;
 354         uint8_t llen;
 355
 356         pnt = packet->nlri;
 357         lim = pnt + packet->length;
 358         afi = packet->afi;
 359         safi = packet->safi;
 360         addpath_id = 0;
 361
 362         addpath_capable = bgp_addpath_encode_rx(peer, afi, safi);
 363
 364         for (; pnt < lim; pnt += psize) {
 365                 /* Clear prefix structure. */
 366                 memset(&p, 0, sizeof(p));
 367
 368                 if (addpath_capable) {
 369
 370                         /* When packet overflow occurs return immediately. */
 371                         if (pnt + BGP_ADDPATH_ID_LEN > lim)
 372                                 return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 373
 374                         memcpy(&addpath_id, pnt, BGP_ADDPATH_ID_LEN);
 375                         addpath_id = ntohl(addpath_id);
 376                         pnt += BGP_ADDPATH_ID_LEN;
 377                 }
 378
 379                 /* Fetch prefix length. */
 380                 prefixlen = *pnt++;
 381                 p.family = afi2family(packet->afi);
 382                 psize = PSIZE(prefixlen);
 383
 384                 /* sanity check against packet data */
 385                 if ((pnt + psize) > lim) {
 386                         flog_err(
 387                                 EC_BGP_UPDATE_RCV,
 388                                 "%s [Error] Update packet error / L-U (prefix length %d exceeds packet size %u)",
 389                                 peer->host, prefixlen, (uint)(lim - pnt));
 390                         return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 391                 }
 392
 393                 /* Fill in the labels */
 394                 llen = bgp_nlri_get_labels(peer, pnt, psize, &label);
 395                 p.prefixlen = prefixlen - BSIZE(llen);
 396
 397                 /* There needs to be at least one label */
 398                 if (prefixlen < 24) {
 399                         flog_err(EC_BGP_UPDATE_RCV,
 400                                  "%s [Error] Update packet error (wrong label length %d)",
 401                                  peer->host, prefixlen);
 402                         bgp_notify_send(peer, BGP_NOTIFY_UPDATE_ERR,
 403                                         BGP_NOTIFY_UPDATE_INVAL_NETWORK);
 404                         return BGP_NLRI_PARSE_ERROR_LABEL_LENGTH;
 405                 }
 406
 407                 if ((afi == AFI_IP && p.prefixlen > IPV4_MAX_BITLEN)
 408                     || (afi == AFI_IP6 && p.prefixlen > IPV6_MAX_BITLEN))
 409                         return BGP_NLRI_PARSE_ERROR_PREFIX_LENGTH;
 410
 411                 /* Fetch prefix from NLRI packet */
 412                 memcpy(&p.u.prefix, pnt + llen, psize - llen);
 413
 414                 /* Check address. */
 415                 if (afi == AFI_IP && safi == SAFI_LABELED_UNICAST) {
 416                         if (IN_CLASSD(ntohl(p.u.prefix4.s_addr))) {
 417                                 /* From RFC4271 Section 6.3:
 418                                  *
 419                                  * If a prefix in the NLRI field is semantically
 420                                  * incorrect
 421                                  * (e.g., an unexpected multicast IP address),
 422                                  * an error SHOULD
 423                                  * be logged locally, and the prefix SHOULD be
 424                                  * ignored.
 425                                   */
 426                                 flog_err(
 427                                         EC_BGP_UPDATE_RCV,
 428                                         "%s: IPv4 labeled-unicast NLRI is multicast address %pI4, ignoring",
 429                                         peer->host, &p.u.prefix4);
 430                                 continue;
 431                         }
 432                 }
 433
 434                 /* Check address. */
 435                 if (afi == AFI_IP6 && safi == SAFI_LABELED_UNICAST) {
 436                         if (IN6_IS_ADDR_LINKLOCAL(&p.u.prefix6)) {
 437                                 flog_err(
 438                                         EC_BGP_UPDATE_RCV,
 439                                         "%s: IPv6 labeled-unicast NLRI is link-local address %pI6, ignoring",
 440                                         peer->host, &p.u.prefix6);
 441
 442                                 continue;
 443                         }
 444
 445                         if (IN6_IS_ADDR_MULTICAST(&p.u.prefix6)) {
 446                                 flog_err(
 447                                         EC_BGP_UPDATE_RCV,
 448                                         "%s: IPv6 unicast NLRI is multicast address %pI6, ignoring",
 449                                         peer->host, &p.u.prefix6);
 450
 451                                 continue;
 452                         }
 453                 }
 454
 455                 if (attr) {
 456                         bgp_update(peer, &p, addpath_id, attr, packet->afi,
 457                                    safi, ZEBRA_ROUTE_BGP, BGP_ROUTE_NORMAL,
 458                                    NULL, &label, 1, 0, NULL);
 459                 } else {
 460                         bgp_withdraw(peer, &p, addpath_id, attr, packet->afi,
 461                                      SAFI_UNICAST, ZEBRA_ROUTE_BGP,
 462                                      BGP_ROUTE_NORMAL, NULL, &label, 1, NULL);
 463                 }
 464         }
 465
 466         /* Packet length consistency check. */
 467         if (pnt != lim) {
 468                 flog_err(
 469                         EC_BGP_UPDATE_RCV,
 470                         "%s [Error] Update packet error / L-U (%td data remaining after parsing)",
 471                         peer->host, lim - pnt);
 472                 return BGP_NLRI_PARSE_ERROR_PACKET_LENGTH;
 473         }
 474
 475         return BGP_NLRI_PARSE_OK;
 476 }