bgpd/bgp_addpath.c

   1 /*
   2  * Addpath TX ID selection, and related utilities
   3  * Copyright (C) 2018  Amazon.com, Inc. or its affiliates
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License as published by the Free
   7  * Software Foundation; either version 2 of the License, or (at your option)
   8  * any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; see the file COPYING; if not, write to the Free Software
  17  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  18  */
  19
  20 #ifdef HAVE_CONFIG_H
  21 #include "config.h"
  22 #endif
  23
  24 #include "bgp_addpath.h"
  25 #include "bgp_route.h"
  26
  27 static const struct bgp_addpath_strategy_names strat_names[BGP_ADDPATH_MAX] = {
  28         {
  29                 .config_name = "addpath-tx-all-paths",
  30                 .human_name = "All",
  31                 .human_description = "Advertise all paths via addpath",
  32                 .type_json_name = "addpathTxAllPaths",
  33                 .id_json_name = "addpathTxIdAll"
  34         },
  35         {
  36                 .config_name = "addpath-tx-bestpath-per-AS",
  37                 .human_name = "Best-Per-AS",
  38                 .human_description = "Advertise bestpath per AS via addpath",
  39                 .type_json_name = "addpathTxBestpathPerAS",
  40                 .id_json_name = "addpathTxIdBestPerAS"
  41         }
  42 };
  43
  44 static const struct bgp_addpath_strategy_names unknown_names = {
  45         .config_name = "addpath-tx-unknown",
  46         .human_name = "Unknown-Addpath-Strategy",
  47         .human_description = "Unknown Addpath Strategy",
  48         .type_json_name = "addpathTxUnknown",
  49         .id_json_name = "addpathTxIdUnknown"
  50 };
  51
  52 /*
  53  * Returns a structure full of strings associated with an addpath type. Will
  54  * never return null.
  55  */
  56 const struct bgp_addpath_strategy_names *
  57 bgp_addpath_names(enum bgp_addpath_strat strat)
  58 {
  59         if (strat < BGP_ADDPATH_MAX)
  60                 return &(strat_names[strat]);
  61         else
  62                 return &unknown_names;
  63 };
  64
  65 /*
  66  * Returns if any peer is transmitting addpaths for a given afi/safi.
  67  */
  68 bool bgp_addpath_is_addpath_used(struct bgp_addpath_bgp_data *d, afi_t afi,
  69                                  safi_t safi)
  70 {
  71         return d->total_peercount[afi][safi] > 0;
  72 }
  73
  74 /*
  75  * Initialize the BGP instance level data for addpath.
  76  */
  77 void bgp_addpath_init_bgp_data(struct bgp_addpath_bgp_data *d)
  78 {
  79         safi_t safi;
  80         afi_t afi;
  81         int i;
  82
  83         FOREACH_AFI_SAFI (afi, safi) {
  84                 for (i = 0; i < BGP_ADDPATH_MAX; i++) {
  85                         d->id_allocators[afi][safi][i] = NULL;
  86                         d->peercount[afi][safi][i] = 0;
  87                 }
  88                 d->total_peercount[afi][safi] = 0;
  89         }
  90 }
  91
  92 /*
  93  * Free up resources associated with BGP route info structures.
  94  */
  95 void bgp_addpath_free_info_data(struct bgp_addpath_info_data *d,
  96                               struct bgp_addpath_node_data *nd)
  97 {
  98         int i;
  99
 100         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 101                 if (d->addpath_tx_id[i] != IDALLOC_INVALID)
 102                         idalloc_free_to_pool(&nd->free_ids[i],
 103                                              d->addpath_tx_id[i]);
 104         }
 105 }
 106
 107 /*
 108  * Return the addpath ID used to send a particular route, to a particular peer,
 109  * in a particular AFI/SAFI.
 110  */
 111 uint32_t bgp_addpath_id_for_peer(struct peer *peer, afi_t afi, safi_t safi,
 112                                 struct bgp_addpath_info_data *d)
 113 {
 114         if (safi == SAFI_LABELED_UNICAST)
 115                 safi = SAFI_UNICAST;
 116
 117         if (peer->addpath_type[afi][safi] < BGP_ADDPATH_MAX)
 118                 return d->addpath_tx_id[peer->addpath_type[afi][safi]];
 119         else
 120                 return IDALLOC_INVALID;
 121 }
 122
 123 /*
 124  * Returns true if the path has an assigned addpath ID for any of the addpath
 125  * strategies.
 126  */
 127 bool bgp_addpath_info_has_ids(struct bgp_addpath_info_data *d)
 128 {
 129         int i;
 130
 131         for (i = 0; i < BGP_ADDPATH_MAX; i++)
 132                 if (d->addpath_tx_id[i] != 0)
 133                         return true;
 134
 135         return false;
 136 }
 137
 138 /*
 139  * Releases any ID's associated with the BGP prefix.
 140  */
 141 void bgp_addpath_free_node_data(struct bgp_addpath_bgp_data *bd,
 142                               struct bgp_addpath_node_data *nd, afi_t afi,
 143                               safi_t safi)
 144 {
 145         int i;
 146
 147         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 148                 idalloc_drain_pool(bd->id_allocators[afi][safi][i],
 149                                    &(nd->free_ids[i]));
 150         }
 151 }
 152
 153 /*
 154  * Check to see if the addpath strategy requires DMED to be configured to work.
 155  */
 156 bool bgp_addpath_dmed_required(int strategy)
 157 {
 158         return strategy == BGP_ADDPATH_BEST_PER_AS;
 159 }
 160
 161 /*
 162  * Return true if this is a path we should advertise due to a
 163  * configured addpath-tx knob
 164  */
 165 bool bgp_addpath_tx_path(enum bgp_addpath_strat strat, struct bgp_path_info *pi)
 166 {
 167         switch (strat) {
 168         case BGP_ADDPATH_NONE:
 169                 return false;
 170         case BGP_ADDPATH_ALL:
 171                 return true;
 172         case BGP_ADDPATH_BEST_PER_AS:
 173                 if (CHECK_FLAG(pi->flags, BGP_PATH_DMED_SELECTED))
 174                         return true;
 175                 else
 176                         return false;
 177         default:
 178                 return false;
 179         }
 180 }
 181
 182 static void bgp_addpath_flush_type_rn(struct bgp *bgp, afi_t afi, safi_t safi,
 183                                       enum bgp_addpath_strat addpath_type,
 184                                       struct bgp_dest *dest)
 185 {
 186         struct bgp_path_info *pi;
 187
 188         if (safi == SAFI_LABELED_UNICAST)
 189                 safi = SAFI_UNICAST;
 190
 191         idalloc_drain_pool(
 192                 bgp->tx_addpath.id_allocators[afi][safi][addpath_type],
 193                 &(dest->tx_addpath.free_ids[addpath_type]));
 194         for (pi = bgp_dest_get_bgp_path_info(dest); pi; pi = pi->next) {
 195                 if (pi->tx_addpath.addpath_tx_id[addpath_type]
 196                     != IDALLOC_INVALID) {
 197                         idalloc_free(
 198                                 bgp->tx_addpath
 199                                         .id_allocators[afi][safi][addpath_type],
 200                                 pi->tx_addpath.addpath_tx_id[addpath_type]);
 201                         pi->tx_addpath.addpath_tx_id[addpath_type] =
 202                                 IDALLOC_INVALID;
 203                 }
 204         }
 205 }
 206
 207 /*
 208  * Purge all addpath ID's on a BGP instance associated with the addpath
 209  * strategy, and afi/safi combination. This lets us let go of all memory held to
 210  * track ID numbers associated with an addpath type not in use. Since
 211  * post-bestpath ID processing is skipped for types not used, this is the only
 212  * chance to free this data.
 213  */
 214 static void bgp_addpath_flush_type(struct bgp *bgp, afi_t afi, safi_t safi,
 215                                    enum bgp_addpath_strat addpath_type)
 216 {
 217         struct bgp_dest *dest, *ndest;
 218
 219         if (safi == SAFI_LABELED_UNICAST)
 220                 safi = SAFI_UNICAST;
 221
 222         for (dest = bgp_table_top(bgp->rib[afi][safi]); dest;
 223              dest = bgp_route_next(dest)) {
 224                 if (safi == SAFI_MPLS_VPN) {
 225                         struct bgp_table *table;
 226
 227                         table = bgp_dest_get_bgp_table_info(dest);
 228                         if (!table)
 229                                 continue;
 230
 231                         for (ndest = bgp_table_top(table); ndest;
 232                              ndest = bgp_route_next(ndest))
 233                                 bgp_addpath_flush_type_rn(bgp, afi, safi,
 234                                                           addpath_type, ndest);
 235                 } else {
 236                         bgp_addpath_flush_type_rn(bgp, afi, safi, addpath_type,
 237                                                   dest);
 238                 }
 239         }
 240
 241         idalloc_destroy(bgp->tx_addpath.id_allocators[afi][safi][addpath_type]);
 242         bgp->tx_addpath.id_allocators[afi][safi][addpath_type] = NULL;
 243 }
 244
 245 /*
 246  * Allocate an Addpath ID for the given type on a path, if necessary.
 247  */
 248 static void bgp_addpath_populate_path(struct id_alloc *allocator,
 249                                       struct bgp_path_info *path,
 250                                       enum bgp_addpath_strat addpath_type)
 251 {
 252         if (bgp_addpath_tx_path(addpath_type, path)) {
 253                 path->tx_addpath.addpath_tx_id[addpath_type] =
 254                         idalloc_allocate(allocator);
 255         }
 256 }
 257
 258 /*
 259  * Compute addpath ID's on a BGP instance associated with the addpath strategy,
 260  * and afi/safi combination. Since we won't waste the time computing addpath IDs
 261  * for unused strategies, the first time a peer is configured to use a strategy,
 262  * we have to backfill the data.
 263  * In labeled-unicast, addpath allocations SHOULD be done in unicast SAFI.
 264  */
 265 static void bgp_addpath_populate_type(struct bgp *bgp, afi_t afi, safi_t safi,
 266                                     enum bgp_addpath_strat addpath_type)
 267 {
 268         struct bgp_dest *dest, *ndest;
 269         char buf[200];
 270         struct id_alloc *allocator;
 271
 272         if (safi == SAFI_LABELED_UNICAST)
 273                 safi = SAFI_UNICAST;
 274
 275         snprintf(buf, sizeof(buf), "Addpath ID Allocator %s:%d/%d",
 276                  bgp_addpath_names(addpath_type)->config_name, (int)afi,
 277                  (int)safi);
 278         buf[sizeof(buf) - 1] = '\0';
 279         zlog_info("Computing addpath IDs for addpath type %s",
 280                 bgp_addpath_names(addpath_type)->human_name);
 281
 282         bgp->tx_addpath.id_allocators[afi][safi][addpath_type] =
 283                 idalloc_new(buf);
 284
 285         idalloc_reserve(bgp->tx_addpath.id_allocators[afi][safi][addpath_type],
 286                 BGP_ADDPATH_TX_ID_FOR_DEFAULT_ORIGINATE);
 287
 288         allocator = bgp->tx_addpath.id_allocators[afi][safi][addpath_type];
 289
 290         for (dest = bgp_table_top(bgp->rib[afi][safi]); dest;
 291              dest = bgp_route_next(dest)) {
 292                 struct bgp_path_info *bi;
 293
 294                 if (safi == SAFI_MPLS_VPN) {
 295                         struct bgp_table *table;
 296
 297                         table = bgp_dest_get_bgp_table_info(dest);
 298                         if (!table)
 299                                 continue;
 300
 301                         for (ndest = bgp_table_top(table); ndest;
 302                              ndest = bgp_route_next(ndest))
 303                                 for (bi = bgp_dest_get_bgp_path_info(ndest); bi;
 304                                      bi = bi->next)
 305                                         bgp_addpath_populate_path(allocator, bi,
 306                                                                   addpath_type);
 307                 } else {
 308                         for (bi = bgp_dest_get_bgp_path_info(dest); bi;
 309                              bi = bi->next)
 310                                 bgp_addpath_populate_path(allocator, bi,
 311                                                           addpath_type);
 312                 }
 313         }
 314 }
 315
 316 /*
 317  * Handle updates to a peer or group's addpath strategy. If after adjusting
 318  * counts a addpath strategy is in use for the first time, or no longer in use,
 319  * the IDs for that strategy will be populated or flushed.
 320  */
 321 void bgp_addpath_type_changed(struct bgp *bgp)
 322 {
 323         afi_t afi;
 324         safi_t safi;
 325         struct listnode *node, *nnode;
 326         struct peer *peer;
 327         int peer_count[AFI_MAX][SAFI_MAX][BGP_ADDPATH_MAX];
 328         enum bgp_addpath_strat type;
 329
 330         FOREACH_AFI_SAFI(afi, safi) {
 331                 for (type=0; type<BGP_ADDPATH_MAX; type++) {
 332                         peer_count[afi][safi][type] = 0;
 333                 }
 334                 bgp->tx_addpath.total_peercount[afi][safi] = 0;
 335         }
 336
 337         for (ALL_LIST_ELEMENTS(bgp->peer, node, nnode, peer)) {
 338                 FOREACH_AFI_SAFI(afi, safi) {
 339                         type = peer->addpath_type[afi][safi];
 340                         if (type != BGP_ADDPATH_NONE) {
 341                                 peer_count[afi][safi][type] += 1;
 342                                 bgp->tx_addpath.total_peercount[afi][safi] += 1;
 343                         }
 344                 }
 345         }
 346
 347         FOREACH_AFI_SAFI(afi, safi) {
 348                 for (type=0; type<BGP_ADDPATH_MAX; type++) {
 349                         int old = bgp->tx_addpath.peercount[afi][safi][type];
 350                         int new = peer_count[afi][safi][type];
 351
 352                         bgp->tx_addpath.peercount[afi][safi][type] = new;
 353
 354                         if (old == 0 && new != 0) {
 355                                 bgp_addpath_populate_type(bgp, afi, safi,
 356                                         type);
 357                         } else if (old != 0 && new == 0) {
 358                                 bgp_addpath_flush_type(bgp, afi, safi, type);
 359                         }
 360                 }
 361         }
 362 }
 363
 364 /*
 365  * Change the addpath type assigned to a peer, or peer group. In addition to
 366  * adjusting the counts, peer sessions will be reset as needed to make the
 367  * change take effect.
 368  */
 369 void bgp_addpath_set_peer_type(struct peer *peer, afi_t afi, safi_t safi,
 370                               enum bgp_addpath_strat addpath_type)
 371 {
 372         struct bgp *bgp = peer->bgp;
 373         enum bgp_addpath_strat old_type;
 374         struct listnode *node, *nnode;
 375         struct peer *tmp_peer;
 376         struct peer_group *group;
 377
 378         if (safi == SAFI_LABELED_UNICAST)
 379                 safi = SAFI_UNICAST;
 380
 381         old_type = peer->addpath_type[afi][safi];
 382         if (addpath_type == old_type)
 383                 return;
 384
 385         if (addpath_type == BGP_ADDPATH_NONE && peer->group &&
 386             !CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP)) {
 387                 /* A "no" config on a group member inherits group */
 388                 addpath_type = peer->group->conf->addpath_type[afi][safi];
 389         }
 390
 391         peer->addpath_type[afi][safi] = addpath_type;
 392
 393         bgp_addpath_type_changed(bgp);
 394
 395         if (addpath_type != BGP_ADDPATH_NONE) {
 396                 if (bgp_addpath_dmed_required(addpath_type)) {
 397                         if (!CHECK_FLAG(bgp->flags,
 398                                         BGP_FLAG_DETERMINISTIC_MED)) {
 399                                 zlog_warn(
 400                                         "%s: enabling bgp deterministic-med, this is required for addpath-tx-bestpath-per-AS",
 401                                         peer->host);
 402                                 SET_FLAG(bgp->flags,
 403                                          BGP_FLAG_DETERMINISTIC_MED);
 404                                 bgp_recalculate_all_bestpaths(bgp);
 405                         }
 406                 }
 407         }
 408
 409         zlog_info("Resetting peer %s%s due to change in addpath config",
 410                   CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP) ? "group " : "",
 411                   peer->host);
 412
 413         if (CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP)) {
 414                 group = peer->group;
 415
 416                 /* group will be null as peer_group_delete calls peer_delete on
 417                  * group->conf. That peer_delete will eventuallly end up here
 418                  * if the group was configured to tx addpaths.
 419                  */
 420                 if (group != NULL) {
 421                         for (ALL_LIST_ELEMENTS(group->peer, node, nnode,
 422                              tmp_peer)) {
 423                                 if (tmp_peer->addpath_type[afi][safi] ==
 424                                     old_type) {
 425                                         bgp_addpath_set_peer_type(tmp_peer,
 426                                                                  afi,
 427                                                                  safi,
 428                                                                  addpath_type);
 429                                 }
 430                         }
 431                 }
 432         } else {
 433                 peer_change_action(peer, afi, safi, peer_change_reset);
 434         }
 435
 436 }
 437
 438 /*
 439  * Intended to run after bestpath. This function will take TX IDs from paths
 440  * that no longer need them, and give them to paths that do. This prevents
 441  * best-per-as updates from needing to do a separate withdraw and update just to
 442  * swap out which path is sent.
 443  */
 444 void bgp_addpath_update_ids(struct bgp *bgp, struct bgp_dest *bn, afi_t afi,
 445                             safi_t safi)
 446 {
 447         int i;
 448         struct bgp_path_info *pi;
 449         struct id_alloc_pool **pool_ptr;
 450
 451         if (safi == SAFI_LABELED_UNICAST)
 452                 safi = SAFI_UNICAST;
 453
 454         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 455                 struct id_alloc *alloc =
 456                         bgp->tx_addpath.id_allocators[afi][safi][i];
 457                 pool_ptr = &(bn->tx_addpath.free_ids[i]);
 458
 459                 if (bgp->tx_addpath.peercount[afi][safi][i] == 0)
 460                         continue;
 461
 462                 /* Free Unused IDs back to the pool.*/
 463                 for (pi = bgp_dest_get_bgp_path_info(bn); pi; pi = pi->next) {
 464                         if (pi->tx_addpath.addpath_tx_id[i] != IDALLOC_INVALID
 465                             && !bgp_addpath_tx_path(i, pi)) {
 466                                 idalloc_free_to_pool(pool_ptr,
 467                                         pi->tx_addpath.addpath_tx_id[i]);
 468                                 pi->tx_addpath.addpath_tx_id[i] =
 469                                         IDALLOC_INVALID;
 470                         }
 471                 }
 472
 473                 /* Give IDs to paths that need them (pulling from the pool) */
 474                 for (pi = bgp_dest_get_bgp_path_info(bn); pi; pi = pi->next) {
 475                         if (pi->tx_addpath.addpath_tx_id[i] == IDALLOC_INVALID
 476                             && bgp_addpath_tx_path(i, pi)) {
 477                                 pi->tx_addpath.addpath_tx_id[i] =
 478                                         idalloc_allocate_prefer_pool(
 479                                                 alloc, pool_ptr);
 480                         }
 481                 }
 482
 483                 /* Free any IDs left in the pool to the main allocator */
 484                 idalloc_drain_pool(alloc, pool_ptr);
 485         }
 486 }