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_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         bool have_label_to_reg =
 133                 bgp_is_valid_label(local_label)
 134                 && label_pton(local_label) != MPLS_LABEL_IMPLICIT_NULL;
 135
 136         p = bgp_dest_get_prefix(dest);
 137
 138         /* Check socket. */
 139         if (!zclient || zclient->sock < 0)
 140                 return;
 141
 142         if (BGP_DEBUG(labelpool, LABELPOOL))
 143                 zlog_debug("%s: FEC %sregister %pRN label_index=%u label=%u",
 144                            __func__, reg ? "" : "un", bgp_dest_to_rnode(dest),
 145                            label_index, label_pton(local_label));
 146         /* If the route node has a local_label assigned or the
 147          * path node has an MPLS SR label index allowing zebra to
 148          * derive the label, proceed with registration. */
 149         s = zclient->obuf;
 150         stream_reset(s);
 151         command = (reg) ? ZEBRA_FEC_REGISTER : ZEBRA_FEC_UNREGISTER;
 152         zclient_create_header(s, command, VRF_DEFAULT);
 153         flags_pos = stream_get_endp(s); /* save position of 'flags' */
 154         stream_putw(s, flags);          /* initial flags */
 155         stream_putw(s, PREFIX_FAMILY(p));
 156         stream_put_prefix(s, p);
 157         if (reg) {
 158                 /* label index takes precedence over auto-assigned label. */
 159                 if (label_index != 0) {
 160                         flags |= ZEBRA_FEC_REGISTER_LABEL_INDEX;
 161                         stream_putl(s, label_index);
 162                 } else if (have_label_to_reg) {
 163                         flags |= ZEBRA_FEC_REGISTER_LABEL;
 164                         stream_putl(s, label_pton(local_label));
 165                 }
 166                 SET_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 167         } else
 168                 UNSET_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL);
 169
 170         /* Set length and flags */
 171         stream_putw_at(s, 0, stream_get_endp(s));
 172
 173         /*
 174          * We only need to write new flags if this is a register
 175          */
 176         if (reg)
 177                 stream_putw_at(s, flags_pos, flags);
 178
 179         zclient_send_message(zclient);
 180 }
 181
 182 /**
 183  * This is passed as the callback function to bgp_labelpool.c:bgp_lp_get()
 184  * by bgp_reg_dereg_for_label() when a label needs to be obtained from
 185  * label pool.
 186  * Note that it will reject the allocated label if a label index is found,
 187  * because the label index supposes predictable labels
 188  */
 189 int bgp_reg_for_label_callback(mpls_label_t new_label, void *labelid,
 190                                bool allocated)
 191 {
 192         struct bgp_dest *dest;
 193
 194         dest = labelid;
 195
 196         /*
 197          * if the route had been removed or the request has gone then reject
 198          * the allocated label. The requesting code will have done what is
 199          * required to allocate the correct label
 200          */
 201         if (!CHECK_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED)) {
 202                 bgp_dest_unlock_node(dest);
 203                 return -1;
 204         }
 205
 206         bgp_dest_unlock_node(dest);
 207
 208         if (BGP_DEBUG(labelpool, LABELPOOL))
 209                 zlog_debug("%s: FEC %pRN label=%u, allocated=%d", __func__,
 210                            bgp_dest_to_rnode(dest), new_label, allocated);
 211
 212         if (!allocated) {
 213                 /*
 214                  * previously-allocated label is now invalid, set to implicit
 215                  * null until new label arrives
 216                  */
 217                 if (CHECK_FLAG(dest->flags, BGP_NODE_REGISTERED_FOR_LABEL)) {
 218                         UNSET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 219                         label_ntop(MPLS_LABEL_IMPLICIT_NULL, 1,
 220                                    &dest->local_label);
 221                         bgp_set_valid_label(&dest->local_label);
 222                 }
 223         }
 224
 225         label_ntop(new_label, 1, &dest->local_label);
 226         bgp_set_valid_label(&dest->local_label);
 227
 228         /*
 229          * Get back to registering the FEC
 230          */
 231         bgp_send_fec_register_label_msg(dest, true, 0);
 232
 233         return 0;
 234 }
 235
 236 void bgp_reg_dereg_for_label(struct bgp_dest *dest, struct bgp_path_info *pi,
 237                              bool reg)
 238 {
 239         bool with_label_index = false;
 240         const struct prefix *p;
 241         bool have_label_to_reg =
 242                 bgp_is_valid_label(&dest->local_label)
 243                 && label_pton(&dest->local_label) != MPLS_LABEL_IMPLICIT_NULL;
 244
 245         p = bgp_dest_get_prefix(dest);
 246
 247         if (reg) {
 248                 assert(pi);
 249                 /*
 250                  * Determine if we will let zebra should derive label from
 251                  * label index instead of bgpd requesting from label pool
 252                  */
 253                 if (CHECK_FLAG(pi->attr->flag,
 254                             ATTR_FLAG_BIT(BGP_ATTR_PREFIX_SID))
 255                         && pi->attr->label_index != BGP_INVALID_LABEL_INDEX) {
 256                         with_label_index = true;
 257                         UNSET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 258                 } else {
 259                         /*
 260                          * If no label has been registered -- assume any label
 261                          * from label pool will do. This means that label index
 262                          * always takes precedence over auto-assigned labels.
 263                          */
 264                         if (!have_label_to_reg) {
 265                                 SET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 266                                 if (BGP_DEBUG(labelpool, LABELPOOL))
 267                                         zlog_debug(
 268                                                 "%s: Requesting label from LP for %pFX",
 269                                                 __func__, p);
 270                                 /* bgp_reg_for_label_callback() will deal with
 271                                  * fec registration when it gets a label from
 272                                  * the pool. This means we'll never register
 273                                  * FECs withoutvalid labels.
 274                                  */
 275                                 bgp_lp_get(LP_TYPE_BGP_LU, dest,
 276                                            bgp_reg_for_label_callback);
 277                                 return;
 278                         }
 279                 }
 280         } else {
 281                 UNSET_FLAG(dest->flags, BGP_NODE_LABEL_REQUESTED);
 282                 bgp_lp_release(LP_TYPE_BGP_LU, dest,
 283                                label_pton(&dest->local_label));
 284         }
 285
 286         bgp_send_fec_register_label_msg(
 287                 dest, reg, with_label_index ? pi->attr->label_index : 0);
 288 }
 289
 290 static int bgp_nlri_get_labels(struct peer *peer, uint8_t *pnt, uint8_t plen,
 291                                mpls_label_t *label)
 292 {
 293         uint8_t *data = pnt;
 294         uint8_t *lim = pnt + plen;
 295         uint8_t llen = 0;
 296         uint8_t label_depth = 0;
 297
 298         for (; data < lim; data += BGP_LABEL_BYTES) {
 299                 memcpy(label, data, BGP_LABEL_BYTES);
 300                 llen += BGP_LABEL_BYTES;
 301
 302                 bgp_set_valid_label(label);
 303                 label_depth += 1;
 304
 305                 if (bgp_is_withdraw_label(label) || label_bos(label))
 306                         break;
 307         }
 308
 309         /* If we RX multiple labels we will end up keeping only the last
 310          * one. We do not yet support a label stack greater than 1. */
 311         if (label_depth > 1)
 312                 zlog_info("%s rcvd UPDATE with label stack %d deep", peer->host,
 313                           label_depth);
 314
 315         if (!(bgp_is_withdraw_label(label) || label_bos(label)))
 316                 flog_warn(
 317                         EC_BGP_INVALID_LABEL_STACK,
 318                         "%s rcvd UPDATE with invalid label stack - no bottom of stack",
 319                         peer->host);
 320
 321         return llen;
 322 }
 323
 324 int bgp_nlri_parse_label(struct peer *peer, struct attr *attr,
 325                          struct bgp_nlri *packet)
 326 {
 327         uint8_t *pnt;
 328         uint8_t *lim;
 329         struct prefix p;
 330         int psize = 0;
 331         int prefixlen;
 332         afi_t afi;
 333         safi_t safi;
 334         int addpath_encoded;
 335         uint32_t addpath_id;
 336         mpls_label_t label = MPLS_INVALID_LABEL;
 337         uint8_t llen;
 338
 339         pnt = packet->nlri;
 340         lim = pnt + packet->length;
 341         afi = packet->afi;
 342         safi = packet->safi;
 343         addpath_id = 0;
 344
 345         addpath_encoded =
 346                 (CHECK_FLAG(peer->af_cap[afi][safi], PEER_CAP_ADDPATH_AF_RX_ADV)
 347                  && CHECK_FLAG(peer->af_cap[afi][safi],
 348                                PEER_CAP_ADDPATH_AF_TX_RCV));
 349
 350         for (; pnt < lim; pnt += psize) {
 351                 /* Clear prefix structure. */
 352                 memset(&p, 0, sizeof(struct prefix));
 353
 354                 if (addpath_encoded) {
 355
 356                         /* When packet overflow occurs return immediately. */
 357                         if (pnt + BGP_ADDPATH_ID_LEN > lim)
 358                                 return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 359
 360                         memcpy(&addpath_id, pnt, BGP_ADDPATH_ID_LEN);
 361                         addpath_id = ntohl(addpath_id);
 362                         pnt += BGP_ADDPATH_ID_LEN;
 363                 }
 364
 365                 /* Fetch prefix length. */
 366                 prefixlen = *pnt++;
 367                 p.family = afi2family(packet->afi);
 368                 psize = PSIZE(prefixlen);
 369
 370                 /* sanity check against packet data */
 371                 if ((pnt + psize) > lim) {
 372                         flog_err(
 373                                 EC_BGP_UPDATE_RCV,
 374                                 "%s [Error] Update packet error / L-U (prefix length %d exceeds packet size %u)",
 375                                 peer->host, prefixlen, (uint)(lim - pnt));
 376                         return BGP_NLRI_PARSE_ERROR_PACKET_OVERFLOW;
 377                 }
 378
 379                 /* Fill in the labels */
 380                 llen = bgp_nlri_get_labels(peer, pnt, psize, &label);
 381                 p.prefixlen = prefixlen - BSIZE(llen);
 382
 383                 /* There needs to be at least one label */
 384                 if (prefixlen < 24) {
 385                         flog_err(EC_BGP_UPDATE_RCV,
 386                                  "%s [Error] Update packet error (wrong label length %d)",
 387                                  peer->host, prefixlen);
 388                         bgp_notify_send(peer, BGP_NOTIFY_UPDATE_ERR,
 389                                         BGP_NOTIFY_UPDATE_INVAL_NETWORK);
 390                         return BGP_NLRI_PARSE_ERROR_LABEL_LENGTH;
 391                 }
 392
 393                 if ((afi == AFI_IP && p.prefixlen > 32)
 394                     || (afi == AFI_IP6 && p.prefixlen > 128))
 395                         return BGP_NLRI_PARSE_ERROR_PREFIX_LENGTH;
 396
 397                 /* Fetch prefix from NLRI packet */
 398                 memcpy(&p.u.prefix, pnt + llen, psize - llen);
 399
 400                 /* Check address. */
 401                 if (afi == AFI_IP && safi == SAFI_LABELED_UNICAST) {
 402                         if (IN_CLASSD(ntohl(p.u.prefix4.s_addr))) {
 403                                 /* From RFC4271 Section 6.3:
 404                                  *
 405                                  * If a prefix in the NLRI field is semantically
 406                                  * incorrect
 407                                  * (e.g., an unexpected multicast IP address),
 408                                  * an error SHOULD
 409                                  * be logged locally, and the prefix SHOULD be
 410                                  * ignored.
 411                                   */
 412                                 flog_err(
 413                                         EC_BGP_UPDATE_RCV,
 414                                         "%s: IPv4 labeled-unicast NLRI is multicast address %pI4, ignoring",
 415                                         peer->host, &p.u.prefix4);
 416                                 continue;
 417                         }
 418                 }
 419
 420                 /* Check address. */
 421                 if (afi == AFI_IP6 && safi == SAFI_LABELED_UNICAST) {
 422                         if (IN6_IS_ADDR_LINKLOCAL(&p.u.prefix6)) {
 423                                 flog_err(
 424                                         EC_BGP_UPDATE_RCV,
 425                                         "%s: IPv6 labeled-unicast NLRI is link-local address %pI6, ignoring",
 426                                         peer->host, &p.u.prefix6);
 427
 428                                 continue;
 429                         }
 430
 431                         if (IN6_IS_ADDR_MULTICAST(&p.u.prefix6)) {
 432                                 flog_err(
 433                                         EC_BGP_UPDATE_RCV,
 434                                         "%s: IPv6 unicast NLRI is multicast address %pI6, ignoring",
 435                                         peer->host, &p.u.prefix6);
 436
 437                                 continue;
 438                         }
 439                 }
 440
 441                 if (attr) {
 442                         bgp_update(peer, &p, addpath_id, attr, packet->afi,
 443                                    SAFI_UNICAST, ZEBRA_ROUTE_BGP,
 444                                    BGP_ROUTE_NORMAL, NULL, &label, 1, 0, NULL);
 445                 } else {
 446                         bgp_withdraw(peer, &p, addpath_id, attr, packet->afi,
 447                                      SAFI_UNICAST, ZEBRA_ROUTE_BGP,
 448                                      BGP_ROUTE_NORMAL, NULL, &label, 1, NULL);
 449                 }
 450         }
 451
 452         /* Packet length consistency check. */
 453         if (pnt != lim) {
 454                 flog_err(
 455                         EC_BGP_UPDATE_RCV,
 456                         "%s [Error] Update packet error / L-U (%td data remaining after parsing)",
 457                         peer->host, lim - pnt);
 458                 return BGP_NLRI_PARSE_ERROR_PACKET_LENGTH;
 459         }
 460
 461         return BGP_NLRI_PARSE_OK;
 462 }