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(struct prefix));
  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                 label_ntop(label, 1, &dest->local_label);
  89                 bgp_set_valid_label(&dest->local_label);
  90         }
  91         SET_FLAG(dest->flags, BGP_NODE_LABEL_CHANGED);
  92         bgp_dest_unlock_node(dest);
  93         bgp_process(bgp, dest, afi, safi);
  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 /**
 124  * This is passed as the callback function to bgp_labelpool.c:bgp_lp_get()
 125  * by bgp_reg_dereg_for_label() when a label needs to be obtained from
 126  * label pool.
 127  * Note that it will reject the allocated label if a label index is found,
 128  * because the label index supposes predictable labels
 129  */
 130 int bgp_reg_for_label_callback(mpls_label_t new_label, void *labelid,
 131                                bool allocated)
 132 {
 133         struct bgp_path_info *pi;
 134         struct bgp_dest *dest;
 135
 136         pi = labelid;
 137         /* Is this path still valid? */
 138         if (!bgp_path_info_unlock(pi)) {
 139                 if (BGP_DEBUG(labelpool, LABELPOOL))
 140                         zlog_debug(
 141                                 "%s: bgp_path_info is no longer valid, ignoring",
 142                                 __func__);
 143                 return -1;
 144         }
 145
 146         dest = pi->net;
 147
 148         if (BGP_DEBUG(labelpool, LABELPOOL))
 149                 zlog_debug("%s: FEC %pRN label=%u, allocated=%d", __func__,
 150                            bgp_dest_to_rnode(dest), new_label, allocated);
 151
 152         if (!allocated) {
 153                 /*
 154                  * previously-allocated label is now invalid
 155                  */
 156                 if (pi->attr->label_index == MPLS_INVALID_LABEL_INDEX
 157                     && pi->attr->label != MPLS_LABEL_NONE
 158                     && CHECK_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL)) {
 159                         bgp_unregister_for_label(dest);
 160                         label_ntop(MPLS_LABEL_IMPLICIT_NULL, 1,
 161                                    &dest->local_label);
 162                         bgp_set_valid_label(&dest->local_label);
 163                 }
 164                 return 0;
 165         }
 166
 167         /*
 168          * label index is assigned, this should be handled by SR-related code,
 169          * so retry FEC registration and then reject label allocation for
 170          * it to be released to label pool
 171          */
 172         if (pi->attr->label_index != MPLS_INVALID_LABEL_INDEX) {
 173                 flog_err(
 174                         EC_BGP_LABEL,
 175                         "%s: FEC %pRN Rejecting allocated label %u as Label Index is %u",
 176                         __func__, bgp_dest_to_rnode(dest), new_label,
 177                         pi->attr->label_index);
 178
 179                 bgp_register_for_label(pi->net, pi);
 180
 181                 return -1;
 182         }
 183
 184         if (pi->attr->label != MPLS_INVALID_LABEL) {
 185                 if (new_label == pi->attr->label) {
 186                         /* already have same label, accept but do nothing */
 187                         return 0;
 188                 }
 189                 /* Shouldn't happen: different label allocation */
 190                 flog_err(EC_BGP_LABEL,
 191                          "%s: %pRN had label %u but got new assignment %u",
 192                          __func__, bgp_dest_to_rnode(dest), pi->attr->label,
 193                          new_label);
 194                 /* continue means use new one */
 195         }
 196
 197         label_ntop(new_label, 1, &dest->local_label);
 198         bgp_set_valid_label(&dest->local_label);
 199
 200         /*
 201          * Get back to registering the FEC
 202          */
 203         bgp_register_for_label(pi->net, pi);
 204
 205         return 0;
 206 }
 207
 208 void bgp_reg_dereg_for_label(struct bgp_dest *dest, struct bgp_path_info *pi,
 209                              bool reg)
 210 {
 211         bool with_label_index = false;
 212         struct stream *s;
 213         const struct prefix *p;
 214         mpls_label_t *local_label;
 215         int command;
 216         uint16_t flags = 0;
 217         size_t flags_pos = 0;
 218
 219         p = bgp_dest_get_prefix(dest);
 220         local_label = &(dest->local_label);
 221         /* this prevents the loop when we're called by
 222          * bgp_reg_for_label_callback()
 223          */
 224         bool have_label_to_reg = bgp_is_valid_label(local_label)
 225                         && label_pton(local_label) != MPLS_LABEL_IMPLICIT_NULL;
 226
 227         if (reg) {
 228                 assert(pi);
 229                 /*
 230                  * Determine if we will let zebra should derive label from
 231                  * label index instead of bgpd requesting from label pool
 232                  */
 233                 if (CHECK_FLAG(pi->attr->flag,
 234                             ATTR_FLAG_BIT(BGP_ATTR_PREFIX_SID))
 235                         && pi->attr->label_index != BGP_INVALID_LABEL_INDEX) {
 236                         with_label_index = true;
 237                 } else {
 238                         /*
 239                          * If no label index was provided -- assume any label
 240                          * from label pool will do. This means that label index
 241                          * always takes precedence over auto-assigned labels.
 242                          */
 243                         if (!have_label_to_reg) {
 244                                 if (BGP_DEBUG(labelpool, LABELPOOL))
 245                                         zlog_debug(
 246                                                 "%s: Requesting label from LP for %pFX",
 247                                                 __func__, p);
 248
 249                                 /* bgp_reg_for_label_callback() will call back
 250                                  * __func__ when it gets a label from the pool.
 251                                  * This means we'll never register FECs without
 252                                  * valid labels.
 253                                  */
 254                                 bgp_lp_get(LP_TYPE_BGP_LU, pi,
 255                                     bgp_reg_for_label_callback);
 256                                 return;
 257                         }
 258                 }
 259         }
 260
 261         /* Check socket. */
 262         if (!zclient || zclient->sock < 0)
 263                 return;
 264
 265         /* If the route node has a local_label assigned or the
 266          * path node has an MPLS SR label index allowing zebra to
 267          * derive the label, proceed with registration. */
 268         s = zclient->obuf;
 269         stream_reset(s);
 270         command = (reg) ? ZEBRA_FEC_REGISTER : ZEBRA_FEC_UNREGISTER;
 271         zclient_create_header(s, command, VRF_DEFAULT);
 272         flags_pos = stream_get_endp(s); /* save position of 'flags' */
 273         stream_putw(s, flags);          /* initial flags */
 274         stream_putw(s, PREFIX_FAMILY(p));
 275         stream_put_prefix(s, p);
 276         if (reg) {
 277                 if (have_label_to_reg) {
 278                         flags |= ZEBRA_FEC_REGISTER_LABEL;
 279                         stream_putl(s, label_pton(local_label));
 280                 } else if (with_label_index) {
 281                         flags |= ZEBRA_FEC_REGISTER_LABEL_INDEX;
 282                         stream_putl(s, pi->attr->label_index);
 283                 }
 284                 SET_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 285         } else
 286                 UNSET_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 287
 288         /* Set length and flags */
 289         stream_putw_at(s, 0, stream_get_endp(s));
 290
 291         /*
 292          * We only need to write new flags if this is a register
 293          */
 294         if (reg)
 295                 stream_putw_at(s, flags_pos, flags);
 296
 297         zclient_send_message(zclient);
 298 }
 299
 300 static int bgp_nlri_get_labels(struct peer *peer, uint8_t *pnt, uint8_t plen,
 301                                mpls_label_t *label)
 302 {
 303         uint8_t *data = pnt;
 304         uint8_t *lim = pnt + plen;
 305         uint8_t llen = 0;
 306         uint8_t label_depth = 0;
 307
 308         for (; data < lim; data += BGP_LABEL_BYTES) {
 309                 memcpy(label, data, BGP_LABEL_BYTES);
 310                 llen += BGP_LABEL_BYTES;
 311
 312                 bgp_set_valid_label(label);
 313                 label_depth += 1;
 314
 315                 if (bgp_is_withdraw_label(label) || label_bos(label))
 316                         break;
 317         }
 318
 319         /* If we RX multiple labels we will end up keeping only the last
 320          * one. We do not yet support a label stack greater than 1. */
 321         if (label_depth > 1)
 322                 zlog_info("%s rcvd UPDATE with label stack %d deep", peer->host,
 323                           label_depth);
 324
 325         if (!(bgp_is_withdraw_label(label) || label_bos(label)))
 326                 flog_warn(
 327                         EC_BGP_INVALID_LABEL_STACK,
 328                         "%s rcvd UPDATE with invalid label stack - no bottom of stack",
 329                         peer->host);
 330
 331         return llen;
 332 }
 333
 334 int bgp_nlri_parse_label(struct peer *peer, struct attr *attr,
 335                          struct bgp_nlri *packet)
 336 {
 337         uint8_t *pnt;
 338         uint8_t *lim;
 339         struct prefix p;
 340         int psize = 0;
 341         int prefixlen;
 342         afi_t afi;
 343         safi_t safi;
 344         int addpath_encoded;
 345         uint32_t addpath_id;
 346         mpls_label_t label = MPLS_INVALID_LABEL;
 347         uint8_t llen;
 348
 349         pnt = packet->nlri;
 350         lim = pnt + packet->length;
 351         afi = packet->afi;
 352         safi = packet->safi;
 353         addpath_id = 0;
 354
 355         addpath_encoded =
 356                 (CHECK_FLAG(peer->af_cap[afi][safi], PEER_CAP_ADDPATH_AF_RX_ADV)
 357                  && CHECK_FLAG(peer->af_cap[afi][safi],
 358                                PEER_CAP_ADDPATH_AF_TX_RCV));
 359
 360         for (; pnt < lim; pnt += psize) {
 361                 /* Clear prefix structure. */
 362                 memset(&p, 0, sizeof(struct prefix));
 363
 364                 if (addpath_encoded) {
 365
 366                         /* When packet overflow occurs return immediately. */
 367                         if (pnt + BGP_ADDPATH_ID_LEN > lim)
 368                                 return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 369
 370                         memcpy(&addpath_id, pnt, BGP_ADDPATH_ID_LEN);
 371                         addpath_id = ntohl(addpath_id);
 372                         pnt += BGP_ADDPATH_ID_LEN;
 373                 }
 374
 375                 /* Fetch prefix length. */
 376                 prefixlen = *pnt++;
 377                 p.family = afi2family(packet->afi);
 378                 psize = PSIZE(prefixlen);
 379
 380                 /* sanity check against packet data */
 381                 if ((pnt + psize) > lim) {
 382                         flog_err(
 383                                 EC_BGP_UPDATE_RCV,
 384                                 "%s [Error] Update packet error / L-U (prefix length %d exceeds packet size %u)",
 385                                 peer->host, prefixlen, (uint)(lim - pnt));
 386                         return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 387                 }
 388
 389                 /* Fill in the labels */
 390                 llen = bgp_nlri_get_labels(peer, pnt, psize, &label);
 391                 p.prefixlen = prefixlen - BSIZE(llen);
 392
 393                 /* There needs to be at least one label */
 394                 if (prefixlen < 24) {
 395                         flog_err(EC_BGP_UPDATE_RCV,
 396                                  "%s [Error] Update packet error (wrong label length %d)",
 397                                  peer->host, prefixlen);
 398                         bgp_notify_send(peer, BGP_NOTIFY_UPDATE_ERR,
 399                                         BGP_NOTIFY_UPDATE_INVAL_NETWORK);
 400                         return BGP_NLRI_PARSE_ERROR_LABEL_LENGTH;
 401                 }
 402
 403                 if ((afi == AFI_IP && p.prefixlen > 32)
 404                     || (afi == AFI_IP6 && p.prefixlen > 128))
 405                         return BGP_NLRI_PARSE_ERROR_PREFIX_LENGTH;
 406
 407                 /* Fetch prefix from NLRI packet */
 408                 memcpy(&p.u.prefix, pnt + llen, psize - llen);
 409
 410                 /* Check address. */
 411                 if (afi == AFI_IP && safi == SAFI_LABELED_UNICAST) {
 412                         if (IN_CLASSD(ntohl(p.u.prefix4.s_addr))) {
 413                                 /* From RFC4271 Section 6.3:
 414                                  *
 415                                  * If a prefix in the NLRI field is semantically
 416                                  * incorrect
 417                                  * (e.g., an unexpected multicast IP address),
 418                                  * an error SHOULD
 419                                  * be logged locally, and the prefix SHOULD be
 420                                  * ignored.
 421                                   */
 422                                 flog_err(
 423                                         EC_BGP_UPDATE_RCV,
 424                                         "%s: IPv4 labeled-unicast NLRI is multicast address %pI4, ignoring",
 425                                         peer->host, &p.u.prefix4);
 426                                 continue;
 427                         }
 428                 }
 429
 430                 /* Check address. */
 431                 if (afi == AFI_IP6 && safi == SAFI_LABELED_UNICAST) {
 432                         if (IN6_IS_ADDR_LINKLOCAL(&p.u.prefix6)) {
 433                                 char buf[BUFSIZ];
 434
 435                                 flog_err(
 436                                         EC_BGP_UPDATE_RCV,
 437                                         "%s: IPv6 labeled-unicast NLRI is link-local address %s, ignoring",
 438                                         peer->host,
 439                                         inet_ntop(AF_INET6, &p.u.prefix6, buf,
 440                                                   BUFSIZ));
 441
 442                                 continue;
 443                         }
 444
 445                         if (IN6_IS_ADDR_MULTICAST(&p.u.prefix6)) {
 446                                 char buf[BUFSIZ];
 447
 448                                 flog_err(
 449                                         EC_BGP_UPDATE_RCV,
 450                                         "%s: IPv6 unicast NLRI is multicast address %s, ignoring",
 451                                         peer->host,
 452                                         inet_ntop(AF_INET6, &p.u.prefix6, buf,
 453                                                   BUFSIZ));
 454
 455                                 continue;
 456                         }
 457                 }
 458
 459                 if (attr) {
 460                         bgp_update(peer, &p, addpath_id, attr, packet->afi,
 461                                    SAFI_UNICAST, ZEBRA_ROUTE_BGP,
 462                                    BGP_ROUTE_NORMAL, NULL, &label, 1, 0, NULL);
 463                 } else {
 464                         bgp_withdraw(peer, &p, addpath_id, attr, packet->afi,
 465                                      SAFI_UNICAST, ZEBRA_ROUTE_BGP,
 466                                      BGP_ROUTE_NORMAL, NULL, &label, 1, NULL);
 467                 }
 468         }
 469
 470         /* Packet length consistency check. */
 471         if (pnt != lim) {
 472                 flog_err(
 473                         EC_BGP_UPDATE_RCV,
 474                         "%s [Error] Update packet error / L-U (%td data remaining after parsing)",
 475                         peer->host, lim - pnt);
 476                 return BGP_NLRI_PARSE_ERROR_PACKET_LENGTH;
 477         }
 478
 479         return BGP_NLRI_PARSE_OK;
 480 }