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 (peer->addpath_type[afi][safi] < BGP_ADDPATH_MAX)
 115                 return d->addpath_tx_id[peer->addpath_type[afi][safi]];
 116         else
 117                 return IDALLOC_INVALID;
 118 }
 119
 120 /*
 121  * Returns true if the path has an assigned addpath ID for any of the addpath
 122  * strategies.
 123  */
 124 bool bgp_addpath_info_has_ids(struct bgp_addpath_info_data *d)
 125 {
 126         int i;
 127
 128         for (i = 0; i < BGP_ADDPATH_MAX; i++)
 129                 if (d->addpath_tx_id[i] != 0)
 130                         return true;
 131
 132         return false;
 133 }
 134
 135 /*
 136  * Releases any ID's associated with the BGP prefix.
 137  */
 138 void bgp_addpath_free_node_data(struct bgp_addpath_bgp_data *bd,
 139                               struct bgp_addpath_node_data *nd, afi_t afi,
 140                               safi_t safi)
 141 {
 142         int i;
 143
 144         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 145                 idalloc_drain_pool(bd->id_allocators[afi][safi][i],
 146                                    &(nd->free_ids[i]));
 147         }
 148 }
 149
 150 /*
 151  * Check to see if the addpath strategy requires DMED to be configured to work.
 152  */
 153 bool bgp_addpath_dmed_required(int strategy)
 154 {
 155         return strategy == BGP_ADDPATH_BEST_PER_AS;
 156 }
 157
 158 /*
 159  * Return true if this is a path we should advertise due to a
 160  * configured addpath-tx knob
 161  */
 162 bool bgp_addpath_tx_path(enum bgp_addpath_strat strat, struct bgp_path_info *pi)
 163 {
 164         switch (strat) {
 165         case BGP_ADDPATH_NONE:
 166                 return false;
 167         case BGP_ADDPATH_ALL:
 168                 return true;
 169         case BGP_ADDPATH_BEST_PER_AS:
 170                 if (CHECK_FLAG(pi->flags, BGP_PATH_DMED_SELECTED))
 171                         return true;
 172                 else
 173                         return false;
 174         default:
 175                 return false;
 176         }
 177 }
 178
 179 static void bgp_addpath_flush_type_rn(struct bgp *bgp, afi_t afi, safi_t safi,
 180                                       enum bgp_addpath_strat addpath_type,
 181                                       struct bgp_dest *dest)
 182 {
 183         struct bgp_path_info *pi;
 184
 185         idalloc_drain_pool(
 186                 bgp->tx_addpath.id_allocators[afi][safi][addpath_type],
 187                 &(dest->tx_addpath.free_ids[addpath_type]));
 188         for (pi = bgp_dest_get_bgp_path_info(dest); pi; pi = pi->next) {
 189                 if (pi->tx_addpath.addpath_tx_id[addpath_type]
 190                     != IDALLOC_INVALID) {
 191                         idalloc_free(
 192                                 bgp->tx_addpath
 193                                         .id_allocators[afi][safi][addpath_type],
 194                                 pi->tx_addpath.addpath_tx_id[addpath_type]);
 195                         pi->tx_addpath.addpath_tx_id[addpath_type] =
 196                                 IDALLOC_INVALID;
 197                 }
 198         }
 199 }
 200
 201 /*
 202  * Purge all addpath ID's on a BGP instance associated with the addpath
 203  * strategy, and afi/safi combination. This lets us let go of all memory held to
 204  * track ID numbers associated with an addpath type not in use. Since
 205  * post-bestpath ID processing is skipped for types not used, this is the only
 206  * chance to free this data.
 207  */
 208 static void bgp_addpath_flush_type(struct bgp *bgp, afi_t afi, safi_t safi,
 209                                    enum bgp_addpath_strat addpath_type)
 210 {
 211         struct bgp_dest *dest, *ndest;
 212
 213         for (dest = bgp_table_top(bgp->rib[afi][safi]); dest;
 214              dest = bgp_route_next(dest)) {
 215                 if (safi == SAFI_MPLS_VPN) {
 216                         struct bgp_table *table;
 217
 218                         table = bgp_dest_get_bgp_table_info(dest);
 219                         if (!table)
 220                                 continue;
 221
 222                         for (ndest = bgp_table_top(table); ndest;
 223                              ndest = bgp_route_next(ndest))
 224                                 bgp_addpath_flush_type_rn(bgp, afi, safi,
 225                                                           addpath_type, ndest);
 226                 } else {
 227                         bgp_addpath_flush_type_rn(bgp, afi, safi, addpath_type,
 228                                                   dest);
 229                 }
 230         }
 231
 232         idalloc_destroy(bgp->tx_addpath.id_allocators[afi][safi][addpath_type]);
 233         bgp->tx_addpath.id_allocators[afi][safi][addpath_type] = NULL;
 234 }
 235
 236 /*
 237  * Allocate an Addpath ID for the given type on a path, if necessary.
 238  */
 239 static void bgp_addpath_populate_path(struct id_alloc *allocator,
 240                                       struct bgp_path_info *path,
 241                                       enum bgp_addpath_strat addpath_type)
 242 {
 243         if (bgp_addpath_tx_path(addpath_type, path)) {
 244                 path->tx_addpath.addpath_tx_id[addpath_type] =
 245                         idalloc_allocate(allocator);
 246         }
 247 }
 248
 249 /*
 250  * Compute addpath ID's on a BGP instance associated with the addpath strategy,
 251  * and afi/safi combination. Since we won't waste the time computing addpath IDs
 252  * for unused strategies, the first time a peer is configured to use a strategy,
 253  * we have to backfill the data.
 254  */
 255 static void bgp_addpath_populate_type(struct bgp *bgp, afi_t afi, safi_t safi,
 256                                     enum bgp_addpath_strat addpath_type)
 257 {
 258         struct bgp_dest *dest, *ndest;
 259         char buf[200];
 260         struct id_alloc *allocator;
 261
 262         snprintf(buf, sizeof(buf), "Addpath ID Allocator %s:%d/%d",
 263                  bgp_addpath_names(addpath_type)->config_name, (int)afi,
 264                  (int)safi);
 265         buf[sizeof(buf) - 1] = '\0';
 266         zlog_info("Computing addpath IDs for addpath type %s",
 267                 bgp_addpath_names(addpath_type)->human_name);
 268
 269         bgp->tx_addpath.id_allocators[afi][safi][addpath_type] =
 270                 idalloc_new(buf);
 271
 272         idalloc_reserve(bgp->tx_addpath.id_allocators[afi][safi][addpath_type],
 273                 BGP_ADDPATH_TX_ID_FOR_DEFAULT_ORIGINATE);
 274
 275         allocator = bgp->tx_addpath.id_allocators[afi][safi][addpath_type];
 276
 277         for (dest = bgp_table_top(bgp->rib[afi][safi]); dest;
 278              dest = bgp_route_next(dest)) {
 279                 struct bgp_path_info *bi;
 280
 281                 if (safi == SAFI_MPLS_VPN) {
 282                         struct bgp_table *table;
 283
 284                         table = bgp_dest_get_bgp_table_info(dest);
 285                         if (!table)
 286                                 continue;
 287
 288                         for (ndest = bgp_table_top(table); ndest;
 289                              ndest = bgp_route_next(ndest))
 290                                 for (bi = bgp_dest_get_bgp_path_info(ndest); bi;
 291                                      bi = bi->next)
 292                                         bgp_addpath_populate_path(allocator, bi,
 293                                                                   addpath_type);
 294                 } else {
 295                         for (bi = bgp_dest_get_bgp_path_info(dest); bi;
 296                              bi = bi->next)
 297                                 bgp_addpath_populate_path(allocator, bi,
 298                                                           addpath_type);
 299                 }
 300         }
 301 }
 302
 303 /*
 304  * Handle updates to a peer or group's addpath strategy. If after adjusting
 305  * counts a addpath strategy is in use for the first time, or no longer in use,
 306  * the IDs for that strategy will be populated or flushed.
 307  */
 308 void bgp_addpath_type_changed(struct bgp *bgp)
 309 {
 310         afi_t afi;
 311         safi_t safi;
 312         struct listnode *node, *nnode;
 313         struct peer *peer;
 314         int peer_count[AFI_MAX][SAFI_MAX][BGP_ADDPATH_MAX];
 315         enum bgp_addpath_strat type;
 316
 317         FOREACH_AFI_SAFI(afi, safi) {
 318                 for (type=0; type<BGP_ADDPATH_MAX; type++) {
 319                         peer_count[afi][safi][type] = 0;
 320                 }
 321                 bgp->tx_addpath.total_peercount[afi][safi] = 0;
 322         }
 323
 324         for (ALL_LIST_ELEMENTS(bgp->peer, node, nnode, peer)) {
 325                 FOREACH_AFI_SAFI(afi, safi) {
 326                         type = peer->addpath_type[afi][safi];
 327                         if (type != BGP_ADDPATH_NONE) {
 328                                 peer_count[afi][safi][type] += 1;
 329                                 bgp->tx_addpath.total_peercount[afi][safi] += 1;
 330                         }
 331                 }
 332         }
 333
 334         FOREACH_AFI_SAFI(afi, safi) {
 335                 for (type=0; type<BGP_ADDPATH_MAX; type++) {
 336                         int old = bgp->tx_addpath.peercount[afi][safi][type];
 337                         int new = peer_count[afi][safi][type];
 338
 339                         bgp->tx_addpath.peercount[afi][safi][type] = new;
 340
 341                         if (old == 0 && new != 0) {
 342                                 bgp_addpath_populate_type(bgp, afi, safi,
 343                                         type);
 344                         } else if (old != 0 && new == 0) {
 345                                 bgp_addpath_flush_type(bgp, afi, safi, type);
 346                         }
 347                 }
 348         }
 349 }
 350
 351 /*
 352  * Change the addpath type assigned to a peer, or peer group. In addition to
 353  * adjusting the counts, peer sessions will be reset as needed to make the
 354  * change take effect.
 355  */
 356 void bgp_addpath_set_peer_type(struct peer *peer, afi_t afi, safi_t safi,
 357                               enum bgp_addpath_strat addpath_type)
 358 {
 359         struct bgp *bgp = peer->bgp;
 360         enum bgp_addpath_strat old_type = peer->addpath_type[afi][safi];
 361         struct listnode *node, *nnode;
 362         struct peer *tmp_peer;
 363         struct peer_group *group;
 364
 365         if (addpath_type == old_type)
 366                 return;
 367
 368         if (addpath_type == BGP_ADDPATH_NONE && peer->group &&
 369             !CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP)) {
 370                 /* A "no" config on a group member inherits group */
 371                 addpath_type = peer->group->conf->addpath_type[afi][safi];
 372         }
 373
 374         peer->addpath_type[afi][safi] = addpath_type;
 375
 376         bgp_addpath_type_changed(bgp);
 377
 378         if (addpath_type != BGP_ADDPATH_NONE) {
 379                 if (bgp_addpath_dmed_required(addpath_type)) {
 380                         if (!CHECK_FLAG(bgp->flags,
 381                                         BGP_FLAG_DETERMINISTIC_MED)) {
 382                                 zlog_warn(
 383                                         "%s: enabling bgp deterministic-med, this is required for addpath-tx-bestpath-per-AS",
 384                                         peer->host);
 385                                 SET_FLAG(bgp->flags,
 386                                          BGP_FLAG_DETERMINISTIC_MED);
 387                                 bgp_recalculate_all_bestpaths(bgp);
 388                         }
 389                 }
 390         }
 391
 392         zlog_info("Resetting peer %s%s due to change in addpath config",
 393                   CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP) ? "group " : "",
 394                   peer->host);
 395
 396         if (CHECK_FLAG(peer->sflags, PEER_STATUS_GROUP)) {
 397                 group = peer->group;
 398
 399                 /* group will be null as peer_group_delete calls peer_delete on
 400                  * group->conf. That peer_delete will eventuallly end up here
 401                  * if the group was configured to tx addpaths.
 402                  */
 403                 if (group != NULL) {
 404                         for (ALL_LIST_ELEMENTS(group->peer, node, nnode,
 405                              tmp_peer)) {
 406                                 if (tmp_peer->addpath_type[afi][safi] ==
 407                                     old_type) {
 408                                         bgp_addpath_set_peer_type(tmp_peer,
 409                                                                  afi,
 410                                                                  safi,
 411                                                                  addpath_type);
 412                                 }
 413                         }
 414                 }
 415         } else {
 416                 peer_change_action(peer, afi, safi, peer_change_reset);
 417         }
 418
 419 }
 420
 421 /*
 422  * Intended to run after bestpath. This function will take TX IDs from paths
 423  * that no longer need them, and give them to paths that do. This prevents
 424  * best-per-as updates from needing to do a separate withdraw and update just to
 425  * swap out which path is sent.
 426  */
 427 void bgp_addpath_update_ids(struct bgp *bgp, struct bgp_dest *bn, afi_t afi,
 428                             safi_t safi)
 429 {
 430         int i;
 431         struct bgp_path_info *pi;
 432         struct id_alloc_pool **pool_ptr;
 433
 434         for (i = 0; i < BGP_ADDPATH_MAX; i++) {
 435                 struct id_alloc *alloc =
 436                         bgp->tx_addpath.id_allocators[afi][safi][i];
 437                 pool_ptr = &(bn->tx_addpath.free_ids[i]);
 438
 439                 if (bgp->tx_addpath.peercount[afi][safi][i] == 0)
 440                         continue;
 441
 442                 /* Free Unused IDs back to the pool.*/
 443                 for (pi = bgp_dest_get_bgp_path_info(bn); pi; pi = pi->next) {
 444                         if (pi->tx_addpath.addpath_tx_id[i] != IDALLOC_INVALID
 445                             && !bgp_addpath_tx_path(i, pi)) {
 446                                 idalloc_free_to_pool(pool_ptr,
 447                                         pi->tx_addpath.addpath_tx_id[i]);
 448                                 pi->tx_addpath.addpath_tx_id[i] =
 449                                         IDALLOC_INVALID;
 450                         }
 451                 }
 452
 453                 /* Give IDs to paths that need them (pulling from the pool) */
 454                 for (pi = bgp_dest_get_bgp_path_info(bn); pi; pi = pi->next) {
 455                         if (pi->tx_addpath.addpath_tx_id[i] == IDALLOC_INVALID
 456                             && bgp_addpath_tx_path(i, pi)) {
 457                                 pi->tx_addpath.addpath_tx_id[i] =
 458                                         idalloc_allocate_prefer_pool(
 459                                                 alloc, pool_ptr);
 460                         }
 461                 }
 462
 463                 /* Free any IDs left in the pool to the main allocator */
 464                 idalloc_drain_pool(alloc, pool_ptr);
 465         }
 466 }