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_node *rn;
  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         rn = bgp_node_lookup(table, &p);
  79         if (!rn) {
  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(&rn->local_label);
  87         else {
  88                 label_ntop(label, 1, &rn->local_label);
  89                 bgp_set_valid_label(&rn->local_label);
  90         }
  91         SET_FLAG(rn->flags, BGP_NODE_LABEL_CHANGED);
  92         bgp_unlock_node(rn);
  93         bgp_process(bgp, rn, afi, safi);
  94         return 1;
  95 }
  96
  97 mpls_label_t bgp_adv_label(struct bgp_node *rn, 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 (!rn || !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 rn->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_node *rn;
 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         rn = pi->net;
 147
 148         if (BGP_DEBUG(labelpool, LABELPOOL))
 149                 zlog_debug("%s: FEC %pRN label=%u, allocated=%d", __func__, rn,
 150                            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(rn->flags, BGP_NODE_REGISTERED_FOR_LABEL)) {
 159                         bgp_unregister_for_label(rn);
 160                         label_ntop(MPLS_LABEL_IMPLICIT_NULL, 1,
 161                                    &rn->local_label);
 162                         bgp_set_valid_label(&rn->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__, rn, new_label, pi->attr->label_index);
 177
 178                 bgp_register_for_label(pi->net, pi);
 179
 180                 return -1;
 181         }
 182
 183         if (pi->attr->label != MPLS_INVALID_LABEL) {
 184                 if (new_label == pi->attr->label) {
 185                         /* already have same label, accept but do nothing */
 186                         return 0;
 187                 }
 188                 /* Shouldn't happen: different label allocation */
 189                 flog_err(EC_BGP_LABEL,
 190                          "%s: %pRN had label %u but got new assignment %u",
 191                          __func__, rn, pi->attr->label, new_label);
 192                 /* continue means use new one */
 193         }
 194
 195         label_ntop(new_label, 1, &rn->local_label);
 196         bgp_set_valid_label(&rn->local_label);
 197
 198         /*
 199          * Get back to registering the FEC
 200          */
 201         bgp_register_for_label(pi->net, pi);
 202
 203         return 0;
 204 }
 205
 206 void bgp_reg_dereg_for_label(struct bgp_node *rn, struct bgp_path_info *pi,
 207                              bool reg)
 208 {
 209         bool with_label_index = false;
 210         struct stream *s;
 211         const struct prefix *p;
 212         mpls_label_t *local_label;
 213         int command;
 214         uint16_t flags = 0;
 215         size_t flags_pos = 0;
 216         char addr[PREFIX_STRLEN];
 217
 218         p = bgp_node_get_prefix(rn);
 219         local_label = &(rn->local_label);
 220         /* this prevents the loop when we're called by
 221          * bgp_reg_for_label_callback()
 222          */
 223         bool have_label_to_reg = bgp_is_valid_label(local_label)
 224                         && label_pton(local_label) != MPLS_LABEL_IMPLICIT_NULL;
 225
 226         if (reg) {
 227                 assert(pi);
 228                 /*
 229                  * Determine if we will let zebra should derive label from
 230                  * label index instead of bgpd requesting from label pool
 231                  */
 232                 if (CHECK_FLAG(pi->attr->flag,
 233                             ATTR_FLAG_BIT(BGP_ATTR_PREFIX_SID))
 234                         && pi->attr->label_index != BGP_INVALID_LABEL_INDEX) {
 235                         with_label_index = true;
 236                 } else {
 237                         /*
 238                          * If no label index was provided -- assume any label
 239                          * from label pool will do. This means that label index
 240                          * always takes precedence over auto-assigned labels.
 241                          */
 242                         if (!have_label_to_reg) {
 243                                 if (BGP_DEBUG(labelpool, LABELPOOL)) {
 244                                         prefix2str(p, addr, PREFIX_STRLEN);
 245                                         zlog_debug("%s: Requesting label from LP for %s",
 246                                                  __func__, addr);
 247                                 }
 248                                 /* bgp_reg_for_label_callback() will call back
 249                                  * __func__ when it gets a label from the pool.
 250                                  * This means we'll never register FECs without
 251                                  * valid labels.
 252                                  */
 253                                 bgp_lp_get(LP_TYPE_BGP_LU, pi,
 254                                     bgp_reg_for_label_callback);
 255                                 return;
 256                         }
 257                 }
 258         }
 259
 260         /* Check socket. */
 261         if (!zclient || zclient->sock < 0)
 262                 return;
 263
 264         /* If the route node has a local_label assigned or the
 265          * path node has an MPLS SR label index allowing zebra to
 266          * derive the label, proceed with registration. */
 267         s = zclient->obuf;
 268         stream_reset(s);
 269         command = (reg) ? ZEBRA_FEC_REGISTER : ZEBRA_FEC_UNREGISTER;
 270         zclient_create_header(s, command, VRF_DEFAULT);
 271         flags_pos = stream_get_endp(s); /* save position of 'flags' */
 272         stream_putw(s, flags);          /* initial flags */
 273         stream_putw(s, PREFIX_FAMILY(p));
 274         stream_put_prefix(s, p);
 275         if (reg) {
 276                 if (have_label_to_reg) {
 277                         flags |= ZEBRA_FEC_REGISTER_LABEL;
 278                         stream_putl(s, label_pton(local_label));
 279                 } else if (with_label_index) {
 280                         flags |= ZEBRA_FEC_REGISTER_LABEL_INDEX;
 281                         stream_putl(s, pi->attr->label_index);
 282                 }
 283                 SET_FLAG(rn->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 284         } else
 285                 UNSET_FLAG(rn->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 286
 287         /* Set length and flags */
 288         stream_putw_at(s, 0, stream_get_endp(s));
 289
 290         /*
 291          * We only need to write new flags if this is a register
 292          */
 293         if (reg)
 294                 stream_putw_at(s, flags_pos, flags);
 295
 296         zclient_send_message(zclient);
 297 }
 298
 299 static int bgp_nlri_get_labels(struct peer *peer, uint8_t *pnt, uint8_t plen,
 300                                mpls_label_t *label)
 301 {
 302         uint8_t *data = pnt;
 303         uint8_t *lim = pnt + plen;
 304         uint8_t llen = 0;
 305         uint8_t label_depth = 0;
 306
 307         for (; data < lim; data += BGP_LABEL_BYTES) {
 308                 memcpy(label, data, BGP_LABEL_BYTES);
 309                 llen += BGP_LABEL_BYTES;
 310
 311                 bgp_set_valid_label(label);
 312                 label_depth += 1;
 313
 314                 if (bgp_is_withdraw_label(label) || label_bos(label))
 315                         break;
 316         }
 317
 318         /* If we RX multiple labels we will end up keeping only the last
 319          * one. We do not yet support a label stack greater than 1. */
 320         if (label_depth > 1)
 321                 zlog_info("%s rcvd UPDATE with label stack %d deep", peer->host,
 322                           label_depth);
 323
 324         if (!(bgp_is_withdraw_label(label) || label_bos(label)))
 325                 flog_warn(
 326                         EC_BGP_INVALID_LABEL_STACK,
 327                         "%s rcvd UPDATE with invalid label stack - no bottom of stack",
 328                         peer->host);
 329
 330         return llen;
 331 }
 332
 333 int bgp_nlri_parse_label(struct peer *peer, struct attr *attr,
 334                          struct bgp_nlri *packet)
 335 {
 336         uint8_t *pnt;
 337         uint8_t *lim;
 338         struct prefix p;
 339         int psize = 0;
 340         int prefixlen;
 341         afi_t afi;
 342         safi_t safi;
 343         int addpath_encoded;
 344         uint32_t addpath_id;
 345         mpls_label_t label = MPLS_INVALID_LABEL;
 346         uint8_t llen;
 347
 348         pnt = packet->nlri;
 349         lim = pnt + packet->length;
 350         afi = packet->afi;
 351         safi = packet->safi;
 352         addpath_id = 0;
 353
 354         addpath_encoded =
 355                 (CHECK_FLAG(peer->af_cap[afi][safi], PEER_CAP_ADDPATH_AF_RX_ADV)
 356                  && CHECK_FLAG(peer->af_cap[afi][safi],
 357                                PEER_CAP_ADDPATH_AF_TX_RCV));
 358
 359         for (; pnt < lim; pnt += psize) {
 360                 /* Clear prefix structure. */
 361                 memset(&p, 0, sizeof(struct prefix));
 362
 363                 if (addpath_encoded) {
 364
 365                         /* When packet overflow occurs return immediately. */
 366                         if (pnt + BGP_ADDPATH_ID_LEN > lim)
 367                                 return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 368
 369                         memcpy(&addpath_id, pnt, BGP_ADDPATH_ID_LEN);
 370                         addpath_id = ntohl(addpath_id);
 371                         pnt += BGP_ADDPATH_ID_LEN;
 372                 }
 373
 374                 /* Fetch prefix length. */
 375                 prefixlen = *pnt++;
 376                 p.family = afi2family(packet->afi);
 377                 psize = PSIZE(prefixlen);
 378
 379                 /* sanity check against packet data */
 380                 if ((pnt + psize) > lim) {
 381                         flog_err(
 382                                 EC_BGP_UPDATE_RCV,
 383                                 "%s [Error] Update packet error / L-U (prefix length %d exceeds packet size %u)",
 384                                 peer->host, prefixlen, (uint)(lim - pnt));
 385                         return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 386                 }
 387
 388                 /* Fill in the labels */
 389                 llen = bgp_nlri_get_labels(peer, pnt, psize, &label);
 390                 p.prefixlen = prefixlen - BSIZE(llen);
 391
 392                 /* There needs to be at least one label */
 393                 if (prefixlen < 24) {
 394                         flog_err(EC_BGP_UPDATE_RCV,
 395                                  "%s [Error] Update packet error"
 396                                  " (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 %s, ignoring",
 425                                         peer->host, inet_ntoa(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 }