pimd/pim_nht.c

   1 /*
   2  * PIM for Quagga
   3  * Copyright (C) 2017 Cumulus Networks, Inc.
   4  * Chirag Shah
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program 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 #include <zebra.h>
  21 #include "network.h"
  22 #include "zclient.h"
  23 #include "stream.h"
  24 #include "nexthop.h"
  25 #include "if.h"
  26 #include "hash.h"
  27 #include "jhash.h"
  28
  29 #include "lib/printfrr.h"
  30
  31 #include "pimd.h"
  32 #include "pimd/pim_nht.h"
  33 #include "pim_instance.h"
  34 #include "log.h"
  35 #include "pim_time.h"
  36 #include "pim_oil.h"
  37 #include "pim_ifchannel.h"
  38 #include "pim_mroute.h"
  39 #include "pim_zebra.h"
  40 #include "pim_upstream.h"
  41 #include "pim_join.h"
  42 #include "pim_jp_agg.h"
  43 #include "pim_zebra.h"
  44 #include "pim_zlookup.h"
  45 #include "pim_rp.h"
  46 #include "pim_addr.h"
  47
  48 /**
  49  * pim_sendmsg_zebra_rnh -- Format and send a nexthop register/Unregister
  50  *   command to Zebra.
  51  */
  52 void pim_sendmsg_zebra_rnh(struct pim_instance *pim, struct zclient *zclient,
  53                            struct pim_nexthop_cache *pnc, int command)
  54 {
  55         struct prefix p;
  56         int ret;
  57
  58         pim_addr_to_prefix(&p, pnc->rpf.rpf_addr);
  59         ret = zclient_send_rnh(zclient, command, &p, SAFI_UNICAST, false, false,
  60                                pim->vrf->vrf_id);
  61         if (ret == ZCLIENT_SEND_FAILURE)
  62                 zlog_warn("sendmsg_nexthop: zclient_send_message() failed");
  63
  64         if (PIM_DEBUG_PIM_NHT)
  65                 zlog_debug(
  66                         "%s: NHT %sregistered addr %pFX(%s) with Zebra ret:%d ",
  67                         __func__,
  68                         (command == ZEBRA_NEXTHOP_REGISTER) ? " " : "de", &p,
  69                         pim->vrf->name, ret);
  70
  71         return;
  72 }
  73
  74 struct pim_nexthop_cache *pim_nexthop_cache_find(struct pim_instance *pim,
  75                                                  struct pim_rpf *rpf)
  76 {
  77         struct pim_nexthop_cache *pnc = NULL;
  78         struct pim_nexthop_cache lookup;
  79
  80         lookup.rpf.rpf_addr = rpf->rpf_addr;
  81         pnc = hash_lookup(pim->rpf_hash, &lookup);
  82
  83         return pnc;
  84 }
  85
  86 static struct pim_nexthop_cache *pim_nexthop_cache_add(struct pim_instance *pim,
  87                                                        struct pim_rpf *rpf_addr)
  88 {
  89         struct pim_nexthop_cache *pnc;
  90         char hash_name[64];
  91
  92         pnc = XCALLOC(MTYPE_PIM_NEXTHOP_CACHE,
  93                       sizeof(struct pim_nexthop_cache));
  94         pnc->rpf.rpf_addr = rpf_addr->rpf_addr;
  95
  96         pnc = hash_get(pim->rpf_hash, pnc, hash_alloc_intern);
  97
  98         pnc->rp_list = list_new();
  99         pnc->rp_list->cmp = pim_rp_list_cmp;
 100
 101         snprintfrr(hash_name, sizeof(hash_name), "PNC %pPA(%s) Upstream Hash",
 102                    &pnc->rpf.rpf_addr, pim->vrf->name);
 103         pnc->upstream_hash = hash_create_size(8192, pim_upstream_hash_key,
 104                                               pim_upstream_equal, hash_name);
 105
 106         return pnc;
 107 }
 108
 109 static struct pim_nexthop_cache *pim_nht_get(struct pim_instance *pim,
 110                                              pim_addr addr)
 111 {
 112         struct pim_nexthop_cache *pnc = NULL;
 113         struct pim_rpf rpf;
 114         struct zclient *zclient = NULL;
 115
 116         zclient = pim_zebra_zclient_get();
 117         memset(&rpf, 0, sizeof(rpf));
 118         rpf.rpf_addr = addr;
 119
 120         pnc = pim_nexthop_cache_find(pim, &rpf);
 121         if (!pnc) {
 122                 pnc = pim_nexthop_cache_add(pim, &rpf);
 123                 pim_sendmsg_zebra_rnh(pim, zclient, pnc,
 124                                       ZEBRA_NEXTHOP_REGISTER);
 125                 if (PIM_DEBUG_PIM_NHT_DETAIL)
 126                         zlog_debug(
 127                                 "%s: NHT cache and zebra notification added for %pPA(%s)",
 128                                 __func__, &addr, pim->vrf->name);
 129         }
 130
 131         return pnc;
 132 }
 133
 134 /* TBD: this does several distinct things and should probably be split up.
 135  * (checking state vs. returning pnc vs. adding upstream vs. adding rp)
 136  */
 137 int pim_find_or_track_nexthop(struct pim_instance *pim, pim_addr addr,
 138                               struct pim_upstream *up, struct rp_info *rp,
 139                               struct pim_nexthop_cache *out_pnc)
 140 {
 141         struct pim_nexthop_cache *pnc;
 142         struct listnode *ch_node = NULL;
 143
 144         pnc = pim_nht_get(pim, addr);
 145
 146         assertf(up || rp, "addr=%pPA", &addr);
 147
 148         if (rp != NULL) {
 149                 ch_node = listnode_lookup(pnc->rp_list, rp);
 150                 if (ch_node == NULL)
 151                         listnode_add_sort(pnc->rp_list, rp);
 152         }
 153
 154         if (up != NULL)
 155                 (void)hash_get(pnc->upstream_hash, up, hash_alloc_intern);
 156
 157         if (CHECK_FLAG(pnc->flags, PIM_NEXTHOP_VALID)) {
 158                 if (out_pnc)
 159                         memcpy(out_pnc, pnc, sizeof(struct pim_nexthop_cache));
 160                 return 1;
 161         }
 162
 163         return 0;
 164 }
 165
 166 #if PIM_IPV == 4
 167 void pim_nht_bsr_add(struct pim_instance *pim, struct in_addr addr)
 168 {
 169         struct pim_nexthop_cache *pnc;
 170
 171         pnc = pim_nht_get(pim, addr);
 172
 173         pnc->bsr_count++;
 174 }
 175 #endif /* PIM_IPV == 4 */
 176
 177 static void pim_nht_drop_maybe(struct pim_instance *pim,
 178                                struct pim_nexthop_cache *pnc)
 179 {
 180         if (PIM_DEBUG_PIM_NHT)
 181                 zlog_debug(
 182                         "%s: NHT %pPA(%s) rp_list count:%d upstream count:%ld BSR count:%u",
 183                         __func__, &pnc->rpf.rpf_addr, pim->vrf->name,
 184                         pnc->rp_list->count, pnc->upstream_hash->count,
 185                         pnc->bsr_count);
 186
 187         if (pnc->rp_list->count == 0 && pnc->upstream_hash->count == 0
 188             && pnc->bsr_count == 0) {
 189                 struct zclient *zclient = pim_zebra_zclient_get();
 190
 191                 pim_sendmsg_zebra_rnh(pim, zclient, pnc,
 192                                       ZEBRA_NEXTHOP_UNREGISTER);
 193
 194                 list_delete(&pnc->rp_list);
 195                 hash_free(pnc->upstream_hash);
 196
 197                 hash_release(pim->rpf_hash, pnc);
 198                 if (pnc->nexthop)
 199                         nexthops_free(pnc->nexthop);
 200                 XFREE(MTYPE_PIM_NEXTHOP_CACHE, pnc);
 201         }
 202 }
 203
 204 void pim_delete_tracked_nexthop(struct pim_instance *pim, pim_addr addr,
 205                                 struct pim_upstream *up, struct rp_info *rp)
 206 {
 207         struct pim_nexthop_cache *pnc = NULL;
 208         struct pim_nexthop_cache lookup;
 209         struct pim_upstream *upstream = NULL;
 210
 211         /* Remove from RPF hash if it is the last entry */
 212         lookup.rpf.rpf_addr = addr;
 213         pnc = hash_lookup(pim->rpf_hash, &lookup);
 214         if (!pnc) {
 215                 zlog_warn("attempting to delete nonexistent NHT entry %pPA",
 216                           &addr);
 217                 return;
 218         }
 219
 220         if (rp) {
 221                 /* Release the (*, G)upstream from pnc->upstream_hash,
 222                  * whose Group belongs to the RP getting deleted
 223                  */
 224                 frr_each (rb_pim_upstream, &pim->upstream_head, upstream) {
 225                         struct prefix grp;
 226                         struct rp_info *trp_info;
 227
 228                         if (!pim_addr_is_any(upstream->sg.src))
 229                                 continue;
 230
 231                         pim_addr_to_prefix(&grp, upstream->sg.grp);
 232                         trp_info = pim_rp_find_match_group(pim, &grp);
 233                         if (trp_info == rp)
 234                                 hash_release(pnc->upstream_hash, upstream);
 235                 }
 236                 listnode_delete(pnc->rp_list, rp);
 237         }
 238
 239         if (up)
 240                 hash_release(pnc->upstream_hash, up);
 241
 242         pim_nht_drop_maybe(pim, pnc);
 243 }
 244
 245 #if PIM_IPV == 4
 246 void pim_nht_bsr_del(struct pim_instance *pim, struct in_addr addr)
 247 {
 248         struct pim_nexthop_cache *pnc = NULL;
 249         struct pim_nexthop_cache lookup;
 250
 251         /*
 252          * Nothing to do here if the address to unregister
 253          * is 0.0.0.0 as that the BSR has not been registered
 254          * for tracking yet.
 255          */
 256         if (addr.s_addr == INADDR_ANY)
 257                 return;
 258
 259         lookup.rpf.rpf_addr = addr;
 260
 261         pnc = hash_lookup(pim->rpf_hash, &lookup);
 262
 263         if (!pnc) {
 264                 zlog_warn("attempting to delete nonexistent NHT BSR entry %pI4",
 265                           &addr);
 266                 return;
 267         }
 268
 269         assertf(pnc->bsr_count > 0, "addr=%pI4", &addr);
 270         pnc->bsr_count--;
 271
 272         pim_nht_drop_maybe(pim, pnc);
 273 }
 274
 275 bool pim_nht_bsr_rpf_check(struct pim_instance *pim, struct in_addr bsr_addr,
 276                            struct interface *src_ifp, pim_addr src_ip)
 277 {
 278         struct pim_nexthop_cache *pnc = NULL;
 279         struct pim_nexthop_cache lookup;
 280         struct pim_neighbor *nbr = NULL;
 281         struct nexthop *nh;
 282         struct interface *ifp;
 283
 284         lookup.rpf.rpf_addr = bsr_addr;
 285
 286         pnc = hash_lookup(pim->rpf_hash, &lookup);
 287         if (!pnc || !CHECK_FLAG(pnc->flags, PIM_NEXTHOP_ANSWER_RECEIVED)) {
 288                 /* BSM from a new freshly registered BSR - do a synchronous
 289                  * zebra query since otherwise we'd drop the first packet,
 290                  * leading to additional delay in picking up BSM data
 291                  */
 292
 293                 /* FIXME: this should really be moved into a generic NHT
 294                  * function that does "add and get immediate result" or maybe
 295                  * "check cache or get immediate result." But until that can
 296                  * be worked in, here's a copy of the code below :(
 297                  */
 298                 struct pim_zlookup_nexthop nexthop_tab[router->multipath];
 299                 ifindex_t i;
 300                 struct interface *ifp = NULL;
 301                 int num_ifindex;
 302
 303                 memset(nexthop_tab, 0, sizeof(nexthop_tab));
 304                 num_ifindex = zclient_lookup_nexthop(
 305                         pim, nexthop_tab, router->multipath, bsr_addr,
 306                         PIM_NEXTHOP_LOOKUP_MAX);
 307
 308                 if (num_ifindex <= 0)
 309                         return false;
 310
 311                 for (i = 0; i < num_ifindex; i++) {
 312                         struct pim_zlookup_nexthop *znh = &nexthop_tab[i];
 313
 314                         /* pim_zlookup_nexthop has no ->type */
 315
 316                         /* 1:1 match code below with znh instead of nh */
 317                         ifp = if_lookup_by_index(znh->ifindex,
 318                                                  pim->vrf->vrf_id);
 319
 320                         if (!ifp || !ifp->info)
 321                                 continue;
 322
 323                         if (if_is_loopback(ifp) && if_is_loopback(src_ifp))
 324                                 return true;
 325
 326                         nbr = pim_neighbor_find(ifp, znh->nexthop_addr);
 327                         if (!nbr)
 328                                 continue;
 329
 330                         return znh->ifindex == src_ifp->ifindex &&
 331                                (!pim_addr_cmp(znh->nexthop_addr, src_ip));
 332                 }
 333                 return false;
 334         }
 335
 336         if (!CHECK_FLAG(pnc->flags, PIM_NEXTHOP_VALID))
 337                 return false;
 338
 339         /* if we accept BSMs from more than one ECMP nexthop, this will cause
 340          * BSM message "multiplication" for each ECMP hop.  i.e. if you have
 341          * 4-way ECMP and 4 hops you end up with 256 copies of each BSM
 342          * message.
 343          *
 344          * so...  only accept the first (IPv4) valid nexthop as source.
 345          */
 346
 347         for (nh = pnc->nexthop; nh; nh = nh->next) {
 348                 pim_addr nhaddr;
 349
 350                 switch (nh->type) {
 351 #if PIM_IPV == 4
 352                 case NEXTHOP_TYPE_IPV4:
 353                         if (nh->ifindex == IFINDEX_INTERNAL)
 354                                 continue;
 355
 356                         /* fallthru */
 357                 case NEXTHOP_TYPE_IPV4_IFINDEX:
 358                         nhaddr = nh->gate.ipv4;
 359                         break;
 360 #else
 361                 case NEXTHOP_TYPE_IPV6:
 362                         if (nh->ifindex == IFINDEX_INTERNAL)
 363                                 continue;
 364
 365                         /* fallthru */
 366                 case NEXTHOP_TYPE_IPV6_IFINDEX:
 367                         nhaddr = nh->gate.ipv6;
 368                         break;
 369 #endif
 370                 case NEXTHOP_TYPE_IFINDEX:
 371                         nhaddr = bsr_addr;
 372                         break;
 373
 374                 default:
 375                         continue;
 376                 }
 377
 378                 ifp = if_lookup_by_index(nh->ifindex, pim->vrf->vrf_id);
 379                 if (!ifp || !ifp->info)
 380                         continue;
 381
 382                 if (if_is_loopback(ifp) && if_is_loopback(src_ifp))
 383                         return true;
 384
 385                 /* MRIB (IGP) may be pointing at a router where PIM is down */
 386                 nbr = pim_neighbor_find(ifp, nhaddr);
 387                 if (!nbr)
 388                         continue;
 389
 390                 return nh->ifindex == src_ifp->ifindex &&
 391                        (!pim_addr_cmp(nhaddr, src_ip));
 392         }
 393         return false;
 394 }
 395 #endif /* PIM_IPV == 4 */
 396
 397 void pim_rp_nexthop_del(struct rp_info *rp_info)
 398 {
 399         rp_info->rp.source_nexthop.interface = NULL;
 400         rp_info->rp.source_nexthop.mrib_nexthop_addr = PIMADDR_ANY;
 401         rp_info->rp.source_nexthop.mrib_metric_preference =
 402                 router->infinite_assert_metric.metric_preference;
 403         rp_info->rp.source_nexthop.mrib_route_metric =
 404                 router->infinite_assert_metric.route_metric;
 405 }
 406
 407 /* Update RP nexthop info based on Nexthop update received from Zebra.*/
 408 static void pim_update_rp_nh(struct pim_instance *pim,
 409                              struct pim_nexthop_cache *pnc)
 410 {
 411         struct listnode *node = NULL;
 412         struct rp_info *rp_info = NULL;
 413
 414         /*Traverse RP list and update each RP Nexthop info */
 415         for (ALL_LIST_ELEMENTS_RO(pnc->rp_list, node, rp_info)) {
 416                 if (pim_rpf_addr_is_inaddr_any(&rp_info->rp))
 417                         continue;
 418
 419                 // Compute PIM RPF using cached nexthop
 420                 if (!pim_ecmp_nexthop_lookup(pim, &rp_info->rp.source_nexthop,
 421                                              rp_info->rp.rpf_addr,
 422                                              &rp_info->group, 1))
 423                         pim_rp_nexthop_del(rp_info);
 424         }
 425 }
 426
 427 /* Update Upstream nexthop info based on Nexthop update received from Zebra.*/
 428 static int pim_update_upstream_nh_helper(struct hash_bucket *bucket, void *arg)
 429 {
 430         struct pim_instance *pim = (struct pim_instance *)arg;
 431         struct pim_upstream *up = (struct pim_upstream *)bucket->data;
 432
 433         enum pim_rpf_result rpf_result;
 434         struct pim_rpf old;
 435
 436         old.source_nexthop.interface = up->rpf.source_nexthop.interface;
 437         rpf_result = pim_rpf_update(pim, up, &old, __func__);
 438
 439         /* update kernel multicast forwarding cache (MFC); if the
 440          * RPF nbr is now unreachable the MFC has already been updated
 441          * by pim_rpf_clear
 442          */
 443         if (rpf_result != PIM_RPF_FAILURE)
 444                 pim_upstream_mroute_iif_update(up->channel_oil, __func__);
 445
 446         if (rpf_result == PIM_RPF_CHANGED ||
 447                 (rpf_result == PIM_RPF_FAILURE && old.source_nexthop.interface))
 448                 pim_zebra_upstream_rpf_changed(pim, up, &old);
 449
 450
 451         if (PIM_DEBUG_PIM_NHT) {
 452                 zlog_debug(
 453                         "%s: NHT upstream %s(%s) old ifp %s new ifp %s",
 454                         __func__, up->sg_str, pim->vrf->name,
 455                         old.source_nexthop.interface ? old.source_nexthop
 456                                                                .interface->name
 457                                                      : "Unknown",
 458                         up->rpf.source_nexthop.interface ? up->rpf.source_nexthop
 459                                                                    .interface->name
 460                                                          : "Unknown");
 461         }
 462
 463         return HASHWALK_CONTINUE;
 464 }
 465
 466 static int pim_update_upstream_nh(struct pim_instance *pim,
 467                                   struct pim_nexthop_cache *pnc)
 468 {
 469         hash_walk(pnc->upstream_hash, pim_update_upstream_nh_helper, pim);
 470
 471         pim_zebra_update_all_interfaces(pim);
 472
 473         return 0;
 474 }
 475
 476 uint32_t pim_compute_ecmp_hash(struct prefix *src, struct prefix *grp)
 477 {
 478         uint32_t hash_val;
 479
 480         if (!src)
 481                 return 0;
 482
 483         hash_val = prefix_hash_key(src);
 484         if (grp)
 485                 hash_val ^= prefix_hash_key(grp);
 486         return hash_val;
 487 }
 488
 489 static int pim_ecmp_nexthop_search(struct pim_instance *pim,
 490                                    struct pim_nexthop_cache *pnc,
 491                                    struct pim_nexthop *nexthop, pim_addr src,
 492                                    struct prefix *grp, int neighbor_needed)
 493 {
 494         struct pim_neighbor *nbrs[router->multipath], *nbr = NULL;
 495         struct interface *ifps[router->multipath];
 496         struct nexthop *nh_node = NULL;
 497         ifindex_t first_ifindex;
 498         struct interface *ifp = NULL;
 499         uint32_t hash_val = 0, mod_val = 0;
 500         uint8_t nh_iter = 0, found = 0;
 501         uint32_t i, num_nbrs = 0;
 502         pim_addr nh_addr = nexthop->mrib_nexthop_addr;
 503         pim_addr grp_addr = pim_addr_from_prefix(grp);
 504
 505         if (!pnc || !pnc->nexthop_num || !nexthop)
 506                 return 0;
 507
 508         memset(&nbrs, 0, sizeof(nbrs));
 509         memset(&ifps, 0, sizeof(ifps));
 510
 511
 512         // Current Nexthop is VALID, check to stay on the current path.
 513         if (nexthop->interface && nexthop->interface->info &&
 514             (!pim_addr_is_any(nh_addr))) {
 515                 /* User configured knob to explicitly switch
 516                    to new path is disabled or current path
 517                    metric is less than nexthop update.
 518                  */
 519
 520                 if (pim->ecmp_rebalance_enable == 0) {
 521                         uint8_t curr_route_valid = 0;
 522                         // Check if current nexthop is present in new updated
 523                         // Nexthop list.
 524                         // If the current nexthop is not valid, candidate to
 525                         // choose new Nexthop.
 526                         for (nh_node = pnc->nexthop; nh_node;
 527                              nh_node = nh_node->next) {
 528                                 curr_route_valid = (nexthop->interface->ifindex
 529                                                     == nh_node->ifindex);
 530                                 if (curr_route_valid)
 531                                         break;
 532                         }
 533
 534                         if (curr_route_valid &&
 535                             !pim_if_connected_to_source(nexthop->interface,
 536                                                         src)) {
 537                                 nbr = pim_neighbor_find(
 538                                         nexthop->interface,
 539                                         nexthop->mrib_nexthop_addr);
 540                                 if (!nbr
 541                                     && !if_is_loopback(nexthop->interface)) {
 542                                         if (PIM_DEBUG_PIM_NHT)
 543                                                 zlog_debug(
 544                                                         "%s: current nexthop does not have nbr ",
 545                                                         __func__);
 546                                 } else {
 547                                         /* update metric even if the upstream
 548                                          * neighbor stays unchanged
 549                                          */
 550                                         nexthop->mrib_metric_preference =
 551                                                 pnc->distance;
 552                                         nexthop->mrib_route_metric =
 553                                                 pnc->metric;
 554                                         if (PIM_DEBUG_PIM_NHT)
 555                                                 zlog_debug(
 556                                                         "%s: (%pPA,%pPA)(%s) current nexthop %s is valid, skipping new path selection",
 557                                                         __func__, &src,
 558                                                         &grp_addr,
 559                                                         pim->vrf->name,
 560                                                         nexthop->interface->name);
 561                                         return 1;
 562                                 }
 563                         }
 564                 }
 565         }
 566
 567         /*
 568          * Look up all interfaces and neighbors,
 569          * store for later usage
 570          */
 571         for (nh_node = pnc->nexthop, i = 0; nh_node;
 572              nh_node = nh_node->next, i++) {
 573                 ifps[i] =
 574                         if_lookup_by_index(nh_node->ifindex, pim->vrf->vrf_id);
 575                 if (ifps[i]) {
 576 #if PIM_IPV == 4
 577                         pim_addr nhaddr = nh_node->gate.ipv4;
 578 #else
 579                         pim_addr nhaddr = nh_node->gate.ipv6;
 580 #endif
 581                         nbrs[i] = pim_neighbor_find(ifps[i], nhaddr);
 582                         if (nbrs[i] || pim_if_connected_to_source(ifps[i], src))
 583                                 num_nbrs++;
 584                 }
 585         }
 586         if (pim->ecmp_enable) {
 587                 struct prefix src_pfx;
 588                 uint32_t consider = pnc->nexthop_num;
 589
 590                 if (neighbor_needed && num_nbrs < consider)
 591                         consider = num_nbrs;
 592
 593                 if (consider == 0)
 594                         return 0;
 595
 596                 // PIM ECMP flag is enable then choose ECMP path.
 597                 pim_addr_to_prefix(&src_pfx, src);
 598                 hash_val = pim_compute_ecmp_hash(&src_pfx, grp);
 599                 mod_val = hash_val % consider;
 600         }
 601
 602         for (nh_node = pnc->nexthop; nh_node && (found == 0);
 603              nh_node = nh_node->next) {
 604                 first_ifindex = nh_node->ifindex;
 605                 ifp = ifps[nh_iter];
 606                 if (!ifp) {
 607                         if (PIM_DEBUG_PIM_NHT)
 608                                 zlog_debug(
 609                                         "%s %s: could not find interface for ifindex %d (address %pPA(%s))",
 610                                         __FILE__, __func__, first_ifindex, &src,
 611                                         pim->vrf->name);
 612                         if (nh_iter == mod_val)
 613                                 mod_val++; // Select nexthpath
 614                         nh_iter++;
 615                         continue;
 616                 }
 617                 if (!ifp->info) {
 618                         if (PIM_DEBUG_PIM_NHT)
 619                                 zlog_debug(
 620                                         "%s: multicast not enabled on input interface %s(%s) (ifindex=%d, RPF for source %pPA)",
 621                                         __func__, ifp->name, pim->vrf->name,
 622                                         first_ifindex, &src);
 623                         if (nh_iter == mod_val)
 624                                 mod_val++; // Select nexthpath
 625                         nh_iter++;
 626                         continue;
 627                 }
 628
 629                 if (neighbor_needed && !pim_if_connected_to_source(ifp, src)) {
 630                         nbr = nbrs[nh_iter];
 631                         if (!nbr && !if_is_loopback(ifp)) {
 632                                 if (PIM_DEBUG_PIM_NHT)
 633                                         zlog_debug(
 634                                                 "%s: pim nbr not found on input interface %s(%s)",
 635                                                 __func__, ifp->name,
 636                                                 pim->vrf->name);
 637                                 if (nh_iter == mod_val)
 638                                         mod_val++; // Select nexthpath
 639                                 nh_iter++;
 640                                 continue;
 641                         }
 642                 }
 643
 644                 if (nh_iter == mod_val) {
 645                         nexthop->interface = ifp;
 646 #if PIM_IPV == 4
 647                         nexthop->mrib_nexthop_addr = nh_node->gate.ipv4;
 648 #else
 649                         nexthop->mrib_nexthop_addr = nh_node->gate.ipv6;
 650 #endif
 651                         nexthop->mrib_metric_preference = pnc->distance;
 652                         nexthop->mrib_route_metric = pnc->metric;
 653                         nexthop->last_lookup = src;
 654                         nexthop->last_lookup_time = pim_time_monotonic_usec();
 655                         nexthop->nbr = nbr;
 656                         found = 1;
 657                         if (PIM_DEBUG_PIM_NHT)
 658                                 zlog_debug(
 659                                         "%s: (%pPA,%pPA)(%s) selected nhop interface %s addr %pPAs mod_val %u iter %d ecmp %d",
 660                                         __func__, &src, &grp_addr,
 661                                         pim->vrf->name, ifp->name, &nh_addr,
 662                                         mod_val, nh_iter, pim->ecmp_enable);
 663                 }
 664                 nh_iter++;
 665         }
 666
 667         if (found)
 668                 return 1;
 669         else
 670                 return 0;
 671 }
 672
 673 /* This API is used to parse Registered address nexthop update coming from Zebra
 674  */
 675 int pim_parse_nexthop_update(ZAPI_CALLBACK_ARGS)
 676 {
 677         struct nexthop *nexthop;
 678         struct nexthop *nhlist_head = NULL;
 679         struct nexthop *nhlist_tail = NULL;
 680         int i;
 681         struct pim_rpf rpf;
 682         struct pim_nexthop_cache *pnc = NULL;
 683         struct interface *ifp = NULL;
 684         struct vrf *vrf = vrf_lookup_by_id(vrf_id);
 685         struct pim_instance *pim;
 686         struct zapi_route nhr;
 687         struct prefix match;
 688
 689         if (!vrf)
 690                 return 0;
 691         pim = vrf->info;
 692
 693         if (!zapi_nexthop_update_decode(zclient->ibuf, &match, &nhr)) {
 694                 zlog_err("%s: Decode of nexthop update from zebra failed",
 695                          __func__);
 696                 return 0;
 697         }
 698
 699         if (cmd == ZEBRA_NEXTHOP_UPDATE) {
 700                 rpf.rpf_addr = pim_addr_from_prefix(&match);
 701                 pnc = pim_nexthop_cache_find(pim, &rpf);
 702                 if (!pnc) {
 703                         if (PIM_DEBUG_PIM_NHT)
 704                                 zlog_debug(
 705                                         "%s: Skipping NHT update, addr %pPA is not in local cached DB.",
 706                                         __func__, &rpf.rpf_addr);
 707                         return 0;
 708                 }
 709         } else {
 710                 /*
 711                  * We do not currently handle ZEBRA_IMPORT_CHECK_UPDATE
 712                  */
 713                 return 0;
 714         }
 715
 716         pnc->last_update = pim_time_monotonic_usec();
 717
 718         if (nhr.nexthop_num) {
 719                 pnc->nexthop_num = 0; // Only increment for pim enabled rpf.
 720
 721                 for (i = 0; i < nhr.nexthop_num; i++) {
 722                         nexthop = nexthop_from_zapi_nexthop(&nhr.nexthops[i]);
 723                         switch (nexthop->type) {
 724                         case NEXTHOP_TYPE_IFINDEX:
 725                                 /*
 726                                  * Connected route (i.e. no nexthop), use
 727                                  * RPF address from nexthop cache (i.e.
 728                                  * destination) as PIM nexthop.
 729                                  */
 730 #if PIM_IPV == 4
 731                                 nexthop->type = NEXTHOP_TYPE_IPV4_IFINDEX;
 732                                 nexthop->gate.ipv4 = pnc->rpf.rpf_addr;
 733 #else
 734                                 nexthop->type = NEXTHOP_TYPE_IPV6_IFINDEX;
 735                                 nexthop->gate.ipv6 = pnc->rpf.rpf_addr;
 736 #endif
 737                                 break;
 738 #if PIM_IPV == 4
 739                         /* RFC5549 IPv4-over-IPv6 nexthop handling:
 740                          * if we get an IPv6 nexthop in IPv4 PIM, hunt down a
 741                          * PIM neighbor and use that instead.
 742                          */
 743                         case NEXTHOP_TYPE_IPV6_IFINDEX: {
 744                                 struct interface *ifp1 = NULL;
 745                                 struct pim_neighbor *nbr = NULL;
 746
 747                                 ifp1 = if_lookup_by_index(nexthop->ifindex,
 748                                                           pim->vrf->vrf_id);
 749
 750                                 if (!ifp1)
 751                                         nbr = NULL;
 752                                 else
 753                                         /* FIXME: should really use nbr's
 754                                          * secondary address list here
 755                                          */
 756                                         nbr = pim_neighbor_find_if(ifp1);
 757
 758                                 /* Overwrite with Nbr address as NH addr */
 759                                 if (nbr)
 760                                         nexthop->gate.ipv4 = nbr->source_addr;
 761                                 else
 762                                         // Mark nexthop address to 0 until PIM
 763                                         // Nbr is resolved.
 764                                         nexthop->gate.ipv4 = PIMADDR_ANY;
 765
 766                                 break;
 767                         }
 768 #else
 769                         case NEXTHOP_TYPE_IPV6_IFINDEX:
 770 #endif
 771                         case NEXTHOP_TYPE_IPV6:
 772                         case NEXTHOP_TYPE_IPV4:
 773                         case NEXTHOP_TYPE_IPV4_IFINDEX:
 774                         case NEXTHOP_TYPE_BLACKHOLE:
 775                                 /* nothing to do for the other nexthop types */
 776                                 break;
 777                         }
 778
 779                         ifp = if_lookup_by_index(nexthop->ifindex,
 780                                                  pim->vrf->vrf_id);
 781                         if (!ifp) {
 782                                 if (PIM_DEBUG_PIM_NHT) {
 783                                         char buf[NEXTHOP_STRLEN];
 784                                         zlog_debug(
 785                                                 "%s: could not find interface for ifindex %d(%s) (addr %s)",
 786                                                 __func__, nexthop->ifindex,
 787                                                 pim->vrf->name,
 788                                                 nexthop2str(nexthop, buf,
 789                                                             sizeof(buf)));
 790                                 }
 791                                 nexthop_free(nexthop);
 792                                 continue;
 793                         }
 794
 795                         if (PIM_DEBUG_PIM_NHT) {
 796 #if PIM_IPV == 4
 797                                 pim_addr nhaddr = nexthop->gate.ipv4;
 798 #else
 799                                 pim_addr nhaddr = nexthop->gate.ipv6;
 800 #endif
 801                                 zlog_debug(
 802                                         "%s: NHT addr %pFX(%s) %d-nhop via %pPA(%s) type %d distance:%u metric:%u ",
 803                                         __func__, &match, pim->vrf->name, i + 1,
 804                                         &nhaddr, ifp->name, nexthop->type,
 805                                         nhr.distance, nhr.metric);
 806                         }
 807
 808                         if (!ifp->info) {
 809                                 /*
 810                                  * Though Multicast is not enabled on this
 811                                  * Interface store it in database otheriwse we
 812                                  * may miss this update and this will not cause
 813                                  * any issue, because while choosing the path we
 814                                  * are ommitting the Interfaces which are not
 815                                  * multicast enabled
 816                                  */
 817                                 if (PIM_DEBUG_PIM_NHT) {
 818                                         char buf[NEXTHOP_STRLEN];
 819
 820                                         zlog_debug(
 821                                                 "%s: multicast not enabled on input interface %s(%s) (ifindex=%d, addr %s)",
 822                                                 __func__, ifp->name,
 823                                                 pim->vrf->name,
 824                                                 nexthop->ifindex,
 825                                                 nexthop2str(nexthop, buf,
 826                                                             sizeof(buf)));
 827                                 }
 828                         }
 829
 830                         if (nhlist_tail) {
 831                                 nhlist_tail->next = nexthop;
 832                                 nhlist_tail = nexthop;
 833                         } else {
 834                                 nhlist_tail = nexthop;
 835                                 nhlist_head = nexthop;
 836                         }
 837                         // Only keep track of nexthops which are PIM enabled.
 838                         pnc->nexthop_num++;
 839                 }
 840                 /* Reset existing pnc->nexthop before assigning new list */
 841                 nexthops_free(pnc->nexthop);
 842                 pnc->nexthop = nhlist_head;
 843                 if (pnc->nexthop_num) {
 844                         pnc->flags |= PIM_NEXTHOP_VALID;
 845                         pnc->distance = nhr.distance;
 846                         pnc->metric = nhr.metric;
 847                 }
 848         } else {
 849                 pnc->flags &= ~PIM_NEXTHOP_VALID;
 850                 pnc->nexthop_num = nhr.nexthop_num;
 851                 nexthops_free(pnc->nexthop);
 852                 pnc->nexthop = NULL;
 853         }
 854         SET_FLAG(pnc->flags, PIM_NEXTHOP_ANSWER_RECEIVED);
 855
 856         if (PIM_DEBUG_PIM_NHT)
 857                 zlog_debug(
 858                         "%s: NHT Update for %pFX(%s) num_nh %d num_pim_nh %d vrf:%u up %ld rp %d",
 859                         __func__, &match, pim->vrf->name, nhr.nexthop_num,
 860                         pnc->nexthop_num, vrf_id, pnc->upstream_hash->count,
 861                         listcount(pnc->rp_list));
 862
 863         pim_rpf_set_refresh_time(pim);
 864
 865         if (listcount(pnc->rp_list))
 866                 pim_update_rp_nh(pim, pnc);
 867         if (pnc->upstream_hash->count)
 868                 pim_update_upstream_nh(pim, pnc);
 869
 870         return 0;
 871 }
 872
 873 int pim_ecmp_nexthop_lookup(struct pim_instance *pim,
 874                             struct pim_nexthop *nexthop, pim_addr src,
 875                             struct prefix *grp, int neighbor_needed)
 876 {
 877         struct pim_nexthop_cache *pnc;
 878         struct pim_zlookup_nexthop nexthop_tab[router->multipath];
 879         struct pim_neighbor *nbrs[router->multipath], *nbr = NULL;
 880         struct pim_rpf rpf;
 881         int num_ifindex;
 882         struct interface *ifps[router->multipath], *ifp;
 883         int first_ifindex;
 884         int found = 0;
 885         uint8_t i = 0;
 886         uint32_t hash_val = 0, mod_val = 0;
 887         uint32_t num_nbrs = 0;
 888
 889         if (PIM_DEBUG_PIM_NHT_DETAIL)
 890                 zlog_debug("%s: Looking up: %pPA(%s), last lookup time: %lld",
 891                            __func__, &src, pim->vrf->name,
 892                            nexthop->last_lookup_time);
 893
 894         rpf.rpf_addr = src;
 895
 896         pnc = pim_nexthop_cache_find(pim, &rpf);
 897         if (pnc) {
 898                 if (CHECK_FLAG(pnc->flags, PIM_NEXTHOP_ANSWER_RECEIVED))
 899                     return pim_ecmp_nexthop_search(pim, pnc, nexthop, src, grp,
 900                                                    neighbor_needed);
 901         }
 902
 903         memset(nexthop_tab, 0,
 904                sizeof(struct pim_zlookup_nexthop) * router->multipath);
 905         num_ifindex =
 906                 zclient_lookup_nexthop(pim, nexthop_tab, router->multipath, src,
 907                                        PIM_NEXTHOP_LOOKUP_MAX);
 908         if (num_ifindex < 1) {
 909                 if (PIM_DEBUG_PIM_NHT)
 910                         zlog_warn(
 911                                 "%s: could not find nexthop ifindex for address %pPA(%s)",
 912                                 __func__, &src, pim->vrf->name);
 913                 return 0;
 914         }
 915
 916         memset(&nbrs, 0, sizeof(nbrs));
 917         memset(&ifps, 0, sizeof(ifps));
 918
 919         /*
 920          * Look up all interfaces and neighbors,
 921          * store for later usage
 922          */
 923         for (i = 0; i < num_ifindex; i++) {
 924                 ifps[i] = if_lookup_by_index(nexthop_tab[i].ifindex,
 925                                              pim->vrf->vrf_id);
 926                 if (ifps[i]) {
 927                         nbrs[i] = pim_neighbor_find(
 928                                 ifps[i], nexthop_tab[i].nexthop_addr);
 929                         if (nbrs[i] || pim_if_connected_to_source(ifps[i], src))
 930                                 num_nbrs++;
 931                 }
 932         }
 933
 934         // If PIM ECMP enable then choose ECMP path.
 935         if (pim->ecmp_enable) {
 936                 struct prefix src_pfx;
 937                 uint32_t consider = num_ifindex;
 938
 939                 if (neighbor_needed && num_nbrs < consider)
 940                         consider = num_nbrs;
 941
 942                 if (consider == 0)
 943                         return 0;
 944
 945                 pim_addr_to_prefix(&src_pfx, src);
 946                 hash_val = pim_compute_ecmp_hash(&src_pfx, grp);
 947                 mod_val = hash_val % consider;
 948                 if (PIM_DEBUG_PIM_NHT_DETAIL)
 949                         zlog_debug("%s: hash_val %u mod_val %u", __func__,
 950                                    hash_val, mod_val);
 951         }
 952
 953         i = 0;
 954         while (!found && (i < num_ifindex)) {
 955                 first_ifindex = nexthop_tab[i].ifindex;
 956
 957                 ifp = ifps[i];
 958                 if (!ifp) {
 959                         if (PIM_DEBUG_PIM_NHT)
 960                                 zlog_debug(
 961                                         "%s %s: could not find interface for ifindex %d (address %pPA(%s))",
 962                                         __FILE__, __func__, first_ifindex, &src,
 963                                         pim->vrf->name);
 964                         if (i == mod_val)
 965                                 mod_val++;
 966                         i++;
 967                         continue;
 968                 }
 969
 970                 if (!ifp->info) {
 971                         if (PIM_DEBUG_PIM_NHT)
 972                                 zlog_debug(
 973                                         "%s: multicast not enabled on input interface %s(%s) (ifindex=%d, RPF for source %pPA)",
 974                                         __func__, ifp->name, pim->vrf->name,
 975                                         first_ifindex, &src);
 976                         if (i == mod_val)
 977                                 mod_val++;
 978                         i++;
 979                         continue;
 980                 }
 981                 if (neighbor_needed && !pim_if_connected_to_source(ifp, src)) {
 982                         nbr = nbrs[i];
 983                         if (PIM_DEBUG_PIM_NHT_DETAIL)
 984                                 zlog_debug("ifp name: %s(%s), pim nbr: %p",
 985                                            ifp->name, pim->vrf->name, nbr);
 986                         if (!nbr && !if_is_loopback(ifp)) {
 987                                 if (i == mod_val)
 988                                         mod_val++;
 989                                 if (PIM_DEBUG_PIM_NHT)
 990                                         zlog_debug(
 991                                                 "%s: NBR (%pPA) not found on input interface %s(%s) (RPF for source %pPA)",
 992                                                 __func__,
 993                                                 &nexthop_tab[i].nexthop_addr,
 994                                                 ifp->name, pim->vrf->name,
 995                                                 &src);
 996                                 i++;
 997                                 continue;
 998                         }
 999                 }
1000
1001                 if (i == mod_val) {
1002                         if (PIM_DEBUG_PIM_NHT)
1003                                 zlog_debug(
1004                                         "%s: found nhop %pPA for addr %pPA interface %s(%s) metric %d dist %d",
1005                                         __func__, &nexthop_tab[i].nexthop_addr,
1006                                         &src, ifp->name, pim->vrf->name,
1007                                         nexthop_tab[i].route_metric,
1008                                         nexthop_tab[i].protocol_distance);
1009                         /* update nexthop data */
1010                         nexthop->interface = ifp;
1011                         nexthop->mrib_nexthop_addr =
1012                                 nexthop_tab[i].nexthop_addr;
1013                         nexthop->mrib_metric_preference =
1014                                 nexthop_tab[i].protocol_distance;
1015                         nexthop->mrib_route_metric =
1016                                 nexthop_tab[i].route_metric;
1017                         nexthop->last_lookup = src;
1018                         nexthop->last_lookup_time = pim_time_monotonic_usec();
1019                         nexthop->nbr = nbr;
1020                         found = 1;
1021                 }
1022                 i++;
1023         }
1024
1025         if (found)
1026                 return 1;
1027         else
1028                 return 0;
1029 }
1030
1031 int pim_ecmp_fib_lookup_if_vif_index(struct pim_instance *pim, pim_addr src,
1032                                      struct prefix *grp)
1033 {
1034         struct pim_nexthop nhop;
1035         int vif_index;
1036         ifindex_t ifindex;
1037
1038         memset(&nhop, 0, sizeof(nhop));
1039         if (!pim_ecmp_nexthop_lookup(pim, &nhop, src, grp, 1)) {
1040                 if (PIM_DEBUG_PIM_NHT)
1041                         zlog_debug(
1042                                 "%s: could not find nexthop ifindex for address %pPA(%s)",
1043                                 __func__, &src, pim->vrf->name);
1044                 return -1;
1045         }
1046
1047         ifindex = nhop.interface->ifindex;
1048         if (PIM_DEBUG_PIM_NHT)
1049                 zlog_debug(
1050                         "%s: found nexthop ifindex=%d (interface %s(%s)) for address %pPA",
1051                         __func__, ifindex,
1052                         ifindex2ifname(ifindex, pim->vrf->vrf_id),
1053                         pim->vrf->name, &src);
1054
1055         vif_index = pim_if_find_vifindex_by_ifindex(pim, ifindex);
1056
1057         if (vif_index < 0) {
1058                 if (PIM_DEBUG_PIM_NHT) {
1059                         zlog_debug(
1060                                 "%s: low vif_index=%d(%s) < 1 nexthop for address %pPA",
1061                                 __func__, vif_index, pim->vrf->name, &src);
1062                 }
1063                 return -2;
1064         }
1065
1066         return vif_index;
1067 }