drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c

   1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
   2 /* Copyright (c) 2016-2018 Mellanox Technologies. All rights reserved */
   3
   4 #include <linux/kernel.h>
   5 #include <linux/types.h>
   6 #include <linux/rhashtable.h>
   7 #include <linux/bitops.h>
   8 #include <linux/in6.h>
   9 #include <linux/notifier.h>
  10 #include <linux/inetdevice.h>
  11 #include <linux/netdevice.h>
  12 #include <linux/if_bridge.h>
  13 #include <linux/socket.h>
  14 #include <linux/route.h>
  15 #include <linux/gcd.h>
  16 #include <linux/if_macvlan.h>
  17 #include <linux/refcount.h>
  18 #include <linux/jhash.h>
  19 #include <linux/net_namespace.h>
  20 #include <linux/mutex.h>
  21 #include <net/netevent.h>
  22 #include <net/neighbour.h>
  23 #include <net/arp.h>
  24 #include <net/ip_fib.h>
  25 #include <net/ip6_fib.h>
  26 #include <net/nexthop.h>
  27 #include <net/fib_rules.h>
  28 #include <net/ip_tunnels.h>
  29 #include <net/l3mdev.h>
  30 #include <net/addrconf.h>
  31 #include <net/ndisc.h>
  32 #include <net/ipv6.h>
  33 #include <net/fib_notifier.h>
  34 #include <net/switchdev.h>
  35
  36 #include "spectrum.h"
  37 #include "core.h"
  38 #include "reg.h"
  39 #include "spectrum_cnt.h"
  40 #include "spectrum_dpipe.h"
  41 #include "spectrum_ipip.h"
  42 #include "spectrum_mr.h"
  43 #include "spectrum_mr_tcam.h"
  44 #include "spectrum_router.h"
  45 #include "spectrum_span.h"
  46
  47 struct mlxsw_sp_fib;
  48 struct mlxsw_sp_vr;
  49 struct mlxsw_sp_lpm_tree;
  50 struct mlxsw_sp_rif_ops;
  51
  52 struct mlxsw_sp_rif {
  53         struct list_head nexthop_list;
  54         struct list_head neigh_list;
  55         struct net_device *dev; /* NULL for underlay RIF */
  56         struct mlxsw_sp_fid *fid;
  57         unsigned char addr[ETH_ALEN];
  58         int mtu;
  59         u16 rif_index;
  60         u16 vr_id;
  61         const struct mlxsw_sp_rif_ops *ops;
  62         struct mlxsw_sp *mlxsw_sp;
  63
  64         unsigned int counter_ingress;
  65         bool counter_ingress_valid;
  66         unsigned int counter_egress;
  67         bool counter_egress_valid;
  68 };
  69
  70 struct mlxsw_sp_rif_params {
  71         struct net_device *dev;
  72         union {
  73                 u16 system_port;
  74                 u16 lag_id;
  75         };
  76         u16 vid;
  77         bool lag;
  78 };
  79
  80 struct mlxsw_sp_rif_subport {
  81         struct mlxsw_sp_rif common;
  82         refcount_t ref_count;
  83         union {
  84                 u16 system_port;
  85                 u16 lag_id;
  86         };
  87         u16 vid;
  88         bool lag;
  89 };
  90
  91 struct mlxsw_sp_rif_ipip_lb {
  92         struct mlxsw_sp_rif common;
  93         struct mlxsw_sp_rif_ipip_lb_config lb_config;
  94         u16 ul_vr_id; /* Reserved for Spectrum-2. */
  95         u16 ul_rif_id; /* Reserved for Spectrum. */
  96 };
  97
  98 struct mlxsw_sp_rif_params_ipip_lb {
  99         struct mlxsw_sp_rif_params common;
 100         struct mlxsw_sp_rif_ipip_lb_config lb_config;
 101 };
 102
 103 struct mlxsw_sp_rif_ops {
 104         enum mlxsw_sp_rif_type type;
 105         size_t rif_size;
 106
 107         void (*setup)(struct mlxsw_sp_rif *rif,
 108                       const struct mlxsw_sp_rif_params *params);
 109         int (*configure)(struct mlxsw_sp_rif *rif);
 110         void (*deconfigure)(struct mlxsw_sp_rif *rif);
 111         struct mlxsw_sp_fid * (*fid_get)(struct mlxsw_sp_rif *rif,
 112                                          struct netlink_ext_ack *extack);
 113         void (*fdb_del)(struct mlxsw_sp_rif *rif, const char *mac);
 114 };
 115
 116 static struct mlxsw_sp_rif *
 117 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
 118                          const struct net_device *dev);
 119 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif);
 120 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree);
 121 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
 122                                   struct mlxsw_sp_lpm_tree *lpm_tree);
 123 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
 124                                      const struct mlxsw_sp_fib *fib,
 125                                      u8 tree_id);
 126 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
 127                                        const struct mlxsw_sp_fib *fib);
 128
 129 static unsigned int *
 130 mlxsw_sp_rif_p_counter_get(struct mlxsw_sp_rif *rif,
 131                            enum mlxsw_sp_rif_counter_dir dir)
 132 {
 133         switch (dir) {
 134         case MLXSW_SP_RIF_COUNTER_EGRESS:
 135                 return &rif->counter_egress;
 136         case MLXSW_SP_RIF_COUNTER_INGRESS:
 137                 return &rif->counter_ingress;
 138         }
 139         return NULL;
 140 }
 141
 142 static bool
 143 mlxsw_sp_rif_counter_valid_get(struct mlxsw_sp_rif *rif,
 144                                enum mlxsw_sp_rif_counter_dir dir)
 145 {
 146         switch (dir) {
 147         case MLXSW_SP_RIF_COUNTER_EGRESS:
 148                 return rif->counter_egress_valid;
 149         case MLXSW_SP_RIF_COUNTER_INGRESS:
 150                 return rif->counter_ingress_valid;
 151         }
 152         return false;
 153 }
 154
 155 static void
 156 mlxsw_sp_rif_counter_valid_set(struct mlxsw_sp_rif *rif,
 157                                enum mlxsw_sp_rif_counter_dir dir,
 158                                bool valid)
 159 {
 160         switch (dir) {
 161         case MLXSW_SP_RIF_COUNTER_EGRESS:
 162                 rif->counter_egress_valid = valid;
 163                 break;
 164         case MLXSW_SP_RIF_COUNTER_INGRESS:
 165                 rif->counter_ingress_valid = valid;
 166                 break;
 167         }
 168 }
 169
 170 static int mlxsw_sp_rif_counter_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
 171                                      unsigned int counter_index, bool enable,
 172                                      enum mlxsw_sp_rif_counter_dir dir)
 173 {
 174         char ritr_pl[MLXSW_REG_RITR_LEN];
 175         bool is_egress = false;
 176         int err;
 177
 178         if (dir == MLXSW_SP_RIF_COUNTER_EGRESS)
 179                 is_egress = true;
 180         mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
 181         err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
 182         if (err)
 183                 return err;
 184
 185         mlxsw_reg_ritr_counter_pack(ritr_pl, counter_index, enable,
 186                                     is_egress);
 187         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
 188 }
 189
 190 int mlxsw_sp_rif_counter_value_get(struct mlxsw_sp *mlxsw_sp,
 191                                    struct mlxsw_sp_rif *rif,
 192                                    enum mlxsw_sp_rif_counter_dir dir, u64 *cnt)
 193 {
 194         char ricnt_pl[MLXSW_REG_RICNT_LEN];
 195         unsigned int *p_counter_index;
 196         bool valid;
 197         int err;
 198
 199         valid = mlxsw_sp_rif_counter_valid_get(rif, dir);
 200         if (!valid)
 201                 return -EINVAL;
 202
 203         p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
 204         if (!p_counter_index)
 205                 return -EINVAL;
 206         mlxsw_reg_ricnt_pack(ricnt_pl, *p_counter_index,
 207                              MLXSW_REG_RICNT_OPCODE_NOP);
 208         err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
 209         if (err)
 210                 return err;
 211         *cnt = mlxsw_reg_ricnt_good_unicast_packets_get(ricnt_pl);
 212         return 0;
 213 }
 214
 215 static int mlxsw_sp_rif_counter_clear(struct mlxsw_sp *mlxsw_sp,
 216                                       unsigned int counter_index)
 217 {
 218         char ricnt_pl[MLXSW_REG_RICNT_LEN];
 219
 220         mlxsw_reg_ricnt_pack(ricnt_pl, counter_index,
 221                              MLXSW_REG_RICNT_OPCODE_CLEAR);
 222         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ricnt), ricnt_pl);
 223 }
 224
 225 int mlxsw_sp_rif_counter_alloc(struct mlxsw_sp *mlxsw_sp,
 226                                struct mlxsw_sp_rif *rif,
 227                                enum mlxsw_sp_rif_counter_dir dir)
 228 {
 229         unsigned int *p_counter_index;
 230         int err;
 231
 232         p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
 233         if (!p_counter_index)
 234                 return -EINVAL;
 235         err = mlxsw_sp_counter_alloc(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
 236                                      p_counter_index);
 237         if (err)
 238                 return err;
 239
 240         err = mlxsw_sp_rif_counter_clear(mlxsw_sp, *p_counter_index);
 241         if (err)
 242                 goto err_counter_clear;
 243
 244         err = mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
 245                                         *p_counter_index, true, dir);
 246         if (err)
 247                 goto err_counter_edit;
 248         mlxsw_sp_rif_counter_valid_set(rif, dir, true);
 249         return 0;
 250
 251 err_counter_edit:
 252 err_counter_clear:
 253         mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
 254                               *p_counter_index);
 255         return err;
 256 }
 257
 258 void mlxsw_sp_rif_counter_free(struct mlxsw_sp *mlxsw_sp,
 259                                struct mlxsw_sp_rif *rif,
 260                                enum mlxsw_sp_rif_counter_dir dir)
 261 {
 262         unsigned int *p_counter_index;
 263
 264         if (!mlxsw_sp_rif_counter_valid_get(rif, dir))
 265                 return;
 266
 267         p_counter_index = mlxsw_sp_rif_p_counter_get(rif, dir);
 268         if (WARN_ON(!p_counter_index))
 269                 return;
 270         mlxsw_sp_rif_counter_edit(mlxsw_sp, rif->rif_index,
 271                                   *p_counter_index, false, dir);
 272         mlxsw_sp_counter_free(mlxsw_sp, MLXSW_SP_COUNTER_SUB_POOL_RIF,
 273                               *p_counter_index);
 274         mlxsw_sp_rif_counter_valid_set(rif, dir, false);
 275 }
 276
 277 static void mlxsw_sp_rif_counters_alloc(struct mlxsw_sp_rif *rif)
 278 {
 279         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
 280         struct devlink *devlink;
 281
 282         devlink = priv_to_devlink(mlxsw_sp->core);
 283         if (!devlink_dpipe_table_counter_enabled(devlink,
 284                                                  MLXSW_SP_DPIPE_TABLE_NAME_ERIF))
 285                 return;
 286         mlxsw_sp_rif_counter_alloc(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
 287 }
 288
 289 static void mlxsw_sp_rif_counters_free(struct mlxsw_sp_rif *rif)
 290 {
 291         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
 292
 293         mlxsw_sp_rif_counter_free(mlxsw_sp, rif, MLXSW_SP_RIF_COUNTER_EGRESS);
 294 }
 295
 296 #define MLXSW_SP_PREFIX_COUNT (sizeof(struct in6_addr) * BITS_PER_BYTE + 1)
 297
 298 struct mlxsw_sp_prefix_usage {
 299         DECLARE_BITMAP(b, MLXSW_SP_PREFIX_COUNT);
 300 };
 301
 302 #define mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) \
 303         for_each_set_bit(prefix, (prefix_usage)->b, MLXSW_SP_PREFIX_COUNT)
 304
 305 static bool
 306 mlxsw_sp_prefix_usage_eq(struct mlxsw_sp_prefix_usage *prefix_usage1,
 307                          struct mlxsw_sp_prefix_usage *prefix_usage2)
 308 {
 309         return !memcmp(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
 310 }
 311
 312 static void
 313 mlxsw_sp_prefix_usage_cpy(struct mlxsw_sp_prefix_usage *prefix_usage1,
 314                           struct mlxsw_sp_prefix_usage *prefix_usage2)
 315 {
 316         memcpy(prefix_usage1, prefix_usage2, sizeof(*prefix_usage1));
 317 }
 318
 319 static void
 320 mlxsw_sp_prefix_usage_set(struct mlxsw_sp_prefix_usage *prefix_usage,
 321                           unsigned char prefix_len)
 322 {
 323         set_bit(prefix_len, prefix_usage->b);
 324 }
 325
 326 static void
 327 mlxsw_sp_prefix_usage_clear(struct mlxsw_sp_prefix_usage *prefix_usage,
 328                             unsigned char prefix_len)
 329 {
 330         clear_bit(prefix_len, prefix_usage->b);
 331 }
 332
 333 struct mlxsw_sp_fib_key {
 334         unsigned char addr[sizeof(struct in6_addr)];
 335         unsigned char prefix_len;
 336 };
 337
 338 enum mlxsw_sp_fib_entry_type {
 339         MLXSW_SP_FIB_ENTRY_TYPE_REMOTE,
 340         MLXSW_SP_FIB_ENTRY_TYPE_LOCAL,
 341         MLXSW_SP_FIB_ENTRY_TYPE_TRAP,
 342         MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE,
 343         MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE,
 344
 345         /* This is a special case of local delivery, where a packet should be
 346          * decapsulated on reception. Note that there is no corresponding ENCAP,
 347          * because that's a type of next hop, not of FIB entry. (There can be
 348          * several next hops in a REMOTE entry, and some of them may be
 349          * encapsulating entries.)
 350          */
 351         MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP,
 352         MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP,
 353 };
 354
 355 struct mlxsw_sp_nexthop_group_info;
 356 struct mlxsw_sp_nexthop_group;
 357 struct mlxsw_sp_fib_entry;
 358
 359 struct mlxsw_sp_fib_node {
 360         struct mlxsw_sp_fib_entry *fib_entry;
 361         struct list_head list;
 362         struct rhash_head ht_node;
 363         struct mlxsw_sp_fib *fib;
 364         struct mlxsw_sp_fib_key key;
 365 };
 366
 367 struct mlxsw_sp_fib_entry_decap {
 368         struct mlxsw_sp_ipip_entry *ipip_entry;
 369         u32 tunnel_index;
 370 };
 371
 372 static struct mlxsw_sp_fib_entry_priv *
 373 mlxsw_sp_fib_entry_priv_create(const struct mlxsw_sp_router_ll_ops *ll_ops)
 374 {
 375         struct mlxsw_sp_fib_entry_priv *priv;
 376
 377         if (!ll_ops->fib_entry_priv_size)
 378                 /* No need to have priv */
 379                 return NULL;
 380
 381         priv = kzalloc(sizeof(*priv) + ll_ops->fib_entry_priv_size, GFP_KERNEL);
 382         if (!priv)
 383                 return ERR_PTR(-ENOMEM);
 384         refcount_set(&priv->refcnt, 1);
 385         return priv;
 386 }
 387
 388 static void
 389 mlxsw_sp_fib_entry_priv_destroy(struct mlxsw_sp_fib_entry_priv *priv)
 390 {
 391         kfree(priv);
 392 }
 393
 394 static void mlxsw_sp_fib_entry_priv_hold(struct mlxsw_sp_fib_entry_priv *priv)
 395 {
 396         refcount_inc(&priv->refcnt);
 397 }
 398
 399 static void mlxsw_sp_fib_entry_priv_put(struct mlxsw_sp_fib_entry_priv *priv)
 400 {
 401         if (!priv || !refcount_dec_and_test(&priv->refcnt))
 402                 return;
 403         mlxsw_sp_fib_entry_priv_destroy(priv);
 404 }
 405
 406 static void mlxsw_sp_fib_entry_op_ctx_priv_hold(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
 407                                                 struct mlxsw_sp_fib_entry_priv *priv)
 408 {
 409         if (!priv)
 410                 return;
 411         mlxsw_sp_fib_entry_priv_hold(priv);
 412         list_add(&priv->list, &op_ctx->fib_entry_priv_list);
 413 }
 414
 415 static void mlxsw_sp_fib_entry_op_ctx_priv_put_all(struct mlxsw_sp_fib_entry_op_ctx *op_ctx)
 416 {
 417         struct mlxsw_sp_fib_entry_priv *priv, *tmp;
 418
 419         list_for_each_entry_safe(priv, tmp, &op_ctx->fib_entry_priv_list, list)
 420                 mlxsw_sp_fib_entry_priv_put(priv);
 421         INIT_LIST_HEAD(&op_ctx->fib_entry_priv_list);
 422 }
 423
 424 struct mlxsw_sp_fib_entry {
 425         struct mlxsw_sp_fib_node *fib_node;
 426         enum mlxsw_sp_fib_entry_type type;
 427         struct list_head nexthop_group_node;
 428         struct mlxsw_sp_nexthop_group *nh_group;
 429         struct mlxsw_sp_fib_entry_decap decap; /* Valid for decap entries. */
 430         struct mlxsw_sp_fib_entry_priv *priv;
 431 };
 432
 433 struct mlxsw_sp_fib4_entry {
 434         struct mlxsw_sp_fib_entry common;
 435         struct fib_info *fi;
 436         u32 tb_id;
 437         u8 tos;
 438         u8 type;
 439 };
 440
 441 struct mlxsw_sp_fib6_entry {
 442         struct mlxsw_sp_fib_entry common;
 443         struct list_head rt6_list;
 444         unsigned int nrt6;
 445 };
 446
 447 struct mlxsw_sp_rt6 {
 448         struct list_head list;
 449         struct fib6_info *rt;
 450 };
 451
 452 struct mlxsw_sp_lpm_tree {
 453         u8 id; /* tree ID */
 454         unsigned int ref_count;
 455         enum mlxsw_sp_l3proto proto;
 456         unsigned long prefix_ref_count[MLXSW_SP_PREFIX_COUNT];
 457         struct mlxsw_sp_prefix_usage prefix_usage;
 458 };
 459
 460 struct mlxsw_sp_fib {
 461         struct rhashtable ht;
 462         struct list_head node_list;
 463         struct mlxsw_sp_vr *vr;
 464         struct mlxsw_sp_lpm_tree *lpm_tree;
 465         enum mlxsw_sp_l3proto proto;
 466         const struct mlxsw_sp_router_ll_ops *ll_ops;
 467 };
 468
 469 struct mlxsw_sp_vr {
 470         u16 id; /* virtual router ID */
 471         u32 tb_id; /* kernel fib table id */
 472         unsigned int rif_count;
 473         struct mlxsw_sp_fib *fib4;
 474         struct mlxsw_sp_fib *fib6;
 475         struct mlxsw_sp_mr_table *mr_table[MLXSW_SP_L3_PROTO_MAX];
 476         struct mlxsw_sp_rif *ul_rif;
 477         refcount_t ul_rif_refcnt;
 478 };
 479
 480 static int mlxsw_sp_router_ll_basic_ralta_write(struct mlxsw_sp *mlxsw_sp, char *xralta_pl)
 481 {
 482         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralta),
 483                                xralta_pl + MLXSW_REG_XRALTA_RALTA_OFFSET);
 484 }
 485
 486 static int mlxsw_sp_router_ll_basic_ralst_write(struct mlxsw_sp *mlxsw_sp, char *xralst_pl)
 487 {
 488         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralst),
 489                                xralst_pl + MLXSW_REG_XRALST_RALST_OFFSET);
 490 }
 491
 492 static int mlxsw_sp_router_ll_basic_raltb_write(struct mlxsw_sp *mlxsw_sp, char *xraltb_pl)
 493 {
 494         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raltb),
 495                                xraltb_pl + MLXSW_REG_XRALTB_RALTB_OFFSET);
 496 }
 497
 498 static const struct rhashtable_params mlxsw_sp_fib_ht_params;
 499
 500 static struct mlxsw_sp_fib *mlxsw_sp_fib_create(struct mlxsw_sp *mlxsw_sp,
 501                                                 struct mlxsw_sp_vr *vr,
 502                                                 enum mlxsw_sp_l3proto proto)
 503 {
 504         const struct mlxsw_sp_router_ll_ops *ll_ops = mlxsw_sp->router->proto_ll_ops[proto];
 505         struct mlxsw_sp_lpm_tree *lpm_tree;
 506         struct mlxsw_sp_fib *fib;
 507         int err;
 508
 509         lpm_tree = mlxsw_sp->router->lpm.proto_trees[proto];
 510         fib = kzalloc(sizeof(*fib), GFP_KERNEL);
 511         if (!fib)
 512                 return ERR_PTR(-ENOMEM);
 513         err = rhashtable_init(&fib->ht, &mlxsw_sp_fib_ht_params);
 514         if (err)
 515                 goto err_rhashtable_init;
 516         INIT_LIST_HEAD(&fib->node_list);
 517         fib->proto = proto;
 518         fib->vr = vr;
 519         fib->lpm_tree = lpm_tree;
 520         fib->ll_ops = ll_ops;
 521         mlxsw_sp_lpm_tree_hold(lpm_tree);
 522         err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, lpm_tree->id);
 523         if (err)
 524                 goto err_lpm_tree_bind;
 525         return fib;
 526
 527 err_lpm_tree_bind:
 528         mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
 529 err_rhashtable_init:
 530         kfree(fib);
 531         return ERR_PTR(err);
 532 }
 533
 534 static void mlxsw_sp_fib_destroy(struct mlxsw_sp *mlxsw_sp,
 535                                  struct mlxsw_sp_fib *fib)
 536 {
 537         mlxsw_sp_vr_lpm_tree_unbind(mlxsw_sp, fib);
 538         mlxsw_sp_lpm_tree_put(mlxsw_sp, fib->lpm_tree);
 539         WARN_ON(!list_empty(&fib->node_list));
 540         rhashtable_destroy(&fib->ht);
 541         kfree(fib);
 542 }
 543
 544 static struct mlxsw_sp_lpm_tree *
 545 mlxsw_sp_lpm_tree_find_unused(struct mlxsw_sp *mlxsw_sp)
 546 {
 547         static struct mlxsw_sp_lpm_tree *lpm_tree;
 548         int i;
 549
 550         for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
 551                 lpm_tree = &mlxsw_sp->router->lpm.trees[i];
 552                 if (lpm_tree->ref_count == 0)
 553                         return lpm_tree;
 554         }
 555         return NULL;
 556 }
 557
 558 static int mlxsw_sp_lpm_tree_alloc(struct mlxsw_sp *mlxsw_sp,
 559                                    const struct mlxsw_sp_router_ll_ops *ll_ops,
 560                                    struct mlxsw_sp_lpm_tree *lpm_tree)
 561 {
 562         char xralta_pl[MLXSW_REG_XRALTA_LEN];
 563
 564         mlxsw_reg_xralta_pack(xralta_pl, true,
 565                               (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
 566                               lpm_tree->id);
 567         return ll_ops->ralta_write(mlxsw_sp, xralta_pl);
 568 }
 569
 570 static void mlxsw_sp_lpm_tree_free(struct mlxsw_sp *mlxsw_sp,
 571                                    const struct mlxsw_sp_router_ll_ops *ll_ops,
 572                                    struct mlxsw_sp_lpm_tree *lpm_tree)
 573 {
 574         char xralta_pl[MLXSW_REG_XRALTA_LEN];
 575
 576         mlxsw_reg_xralta_pack(xralta_pl, false,
 577                               (enum mlxsw_reg_ralxx_protocol) lpm_tree->proto,
 578                               lpm_tree->id);
 579         ll_ops->ralta_write(mlxsw_sp, xralta_pl);
 580 }
 581
 582 static int
 583 mlxsw_sp_lpm_tree_left_struct_set(struct mlxsw_sp *mlxsw_sp,
 584                                   const struct mlxsw_sp_router_ll_ops *ll_ops,
 585                                   struct mlxsw_sp_prefix_usage *prefix_usage,
 586                                   struct mlxsw_sp_lpm_tree *lpm_tree)
 587 {
 588         char xralst_pl[MLXSW_REG_XRALST_LEN];
 589         u8 root_bin = 0;
 590         u8 prefix;
 591         u8 last_prefix = MLXSW_REG_RALST_BIN_NO_CHILD;
 592
 593         mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage)
 594                 root_bin = prefix;
 595
 596         mlxsw_reg_xralst_pack(xralst_pl, root_bin, lpm_tree->id);
 597         mlxsw_sp_prefix_usage_for_each(prefix, prefix_usage) {
 598                 if (prefix == 0)
 599                         continue;
 600                 mlxsw_reg_xralst_bin_pack(xralst_pl, prefix, last_prefix,
 601                                           MLXSW_REG_RALST_BIN_NO_CHILD);
 602                 last_prefix = prefix;
 603         }
 604         return ll_ops->ralst_write(mlxsw_sp, xralst_pl);
 605 }
 606
 607 static struct mlxsw_sp_lpm_tree *
 608 mlxsw_sp_lpm_tree_create(struct mlxsw_sp *mlxsw_sp,
 609                          const struct mlxsw_sp_router_ll_ops *ll_ops,
 610                          struct mlxsw_sp_prefix_usage *prefix_usage,
 611                          enum mlxsw_sp_l3proto proto)
 612 {
 613         struct mlxsw_sp_lpm_tree *lpm_tree;
 614         int err;
 615
 616         lpm_tree = mlxsw_sp_lpm_tree_find_unused(mlxsw_sp);
 617         if (!lpm_tree)
 618                 return ERR_PTR(-EBUSY);
 619         lpm_tree->proto = proto;
 620         err = mlxsw_sp_lpm_tree_alloc(mlxsw_sp, ll_ops, lpm_tree);
 621         if (err)
 622                 return ERR_PTR(err);
 623
 624         err = mlxsw_sp_lpm_tree_left_struct_set(mlxsw_sp, ll_ops, prefix_usage, lpm_tree);
 625         if (err)
 626                 goto err_left_struct_set;
 627         memcpy(&lpm_tree->prefix_usage, prefix_usage,
 628                sizeof(lpm_tree->prefix_usage));
 629         memset(&lpm_tree->prefix_ref_count, 0,
 630                sizeof(lpm_tree->prefix_ref_count));
 631         lpm_tree->ref_count = 1;
 632         return lpm_tree;
 633
 634 err_left_struct_set:
 635         mlxsw_sp_lpm_tree_free(mlxsw_sp, ll_ops, lpm_tree);
 636         return ERR_PTR(err);
 637 }
 638
 639 static void mlxsw_sp_lpm_tree_destroy(struct mlxsw_sp *mlxsw_sp,
 640                                       const struct mlxsw_sp_router_ll_ops *ll_ops,
 641                                       struct mlxsw_sp_lpm_tree *lpm_tree)
 642 {
 643         mlxsw_sp_lpm_tree_free(mlxsw_sp, ll_ops, lpm_tree);
 644 }
 645
 646 static struct mlxsw_sp_lpm_tree *
 647 mlxsw_sp_lpm_tree_get(struct mlxsw_sp *mlxsw_sp,
 648                       struct mlxsw_sp_prefix_usage *prefix_usage,
 649                       enum mlxsw_sp_l3proto proto)
 650 {
 651         const struct mlxsw_sp_router_ll_ops *ll_ops = mlxsw_sp->router->proto_ll_ops[proto];
 652         struct mlxsw_sp_lpm_tree *lpm_tree;
 653         int i;
 654
 655         for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
 656                 lpm_tree = &mlxsw_sp->router->lpm.trees[i];
 657                 if (lpm_tree->ref_count != 0 &&
 658                     lpm_tree->proto == proto &&
 659                     mlxsw_sp_prefix_usage_eq(&lpm_tree->prefix_usage,
 660                                              prefix_usage)) {
 661                         mlxsw_sp_lpm_tree_hold(lpm_tree);
 662                         return lpm_tree;
 663                 }
 664         }
 665         return mlxsw_sp_lpm_tree_create(mlxsw_sp, ll_ops, prefix_usage, proto);
 666 }
 667
 668 static void mlxsw_sp_lpm_tree_hold(struct mlxsw_sp_lpm_tree *lpm_tree)
 669 {
 670         lpm_tree->ref_count++;
 671 }
 672
 673 static void mlxsw_sp_lpm_tree_put(struct mlxsw_sp *mlxsw_sp,
 674                                   struct mlxsw_sp_lpm_tree *lpm_tree)
 675 {
 676         const struct mlxsw_sp_router_ll_ops *ll_ops =
 677                                 mlxsw_sp->router->proto_ll_ops[lpm_tree->proto];
 678
 679         if (--lpm_tree->ref_count == 0)
 680                 mlxsw_sp_lpm_tree_destroy(mlxsw_sp, ll_ops, lpm_tree);
 681 }
 682
 683 #define MLXSW_SP_LPM_TREE_MIN 1 /* tree 0 is reserved */
 684
 685 static int mlxsw_sp_lpm_init(struct mlxsw_sp *mlxsw_sp)
 686 {
 687         struct mlxsw_sp_prefix_usage req_prefix_usage = {{ 0 } };
 688         struct mlxsw_sp_lpm_tree *lpm_tree;
 689         u64 max_trees;
 690         int err, i;
 691
 692         if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_LPM_TREES))
 693                 return -EIO;
 694
 695         max_trees = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_LPM_TREES);
 696         mlxsw_sp->router->lpm.tree_count = max_trees - MLXSW_SP_LPM_TREE_MIN;
 697         mlxsw_sp->router->lpm.trees = kcalloc(mlxsw_sp->router->lpm.tree_count,
 698                                              sizeof(struct mlxsw_sp_lpm_tree),
 699                                              GFP_KERNEL);
 700         if (!mlxsw_sp->router->lpm.trees)
 701                 return -ENOMEM;
 702
 703         for (i = 0; i < mlxsw_sp->router->lpm.tree_count; i++) {
 704                 lpm_tree = &mlxsw_sp->router->lpm.trees[i];
 705                 lpm_tree->id = i + MLXSW_SP_LPM_TREE_MIN;
 706         }
 707
 708         lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
 709                                          MLXSW_SP_L3_PROTO_IPV4);
 710         if (IS_ERR(lpm_tree)) {
 711                 err = PTR_ERR(lpm_tree);
 712                 goto err_ipv4_tree_get;
 713         }
 714         mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4] = lpm_tree;
 715
 716         lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
 717                                          MLXSW_SP_L3_PROTO_IPV6);
 718         if (IS_ERR(lpm_tree)) {
 719                 err = PTR_ERR(lpm_tree);
 720                 goto err_ipv6_tree_get;
 721         }
 722         mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6] = lpm_tree;
 723
 724         return 0;
 725
 726 err_ipv6_tree_get:
 727         lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
 728         mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
 729 err_ipv4_tree_get:
 730         kfree(mlxsw_sp->router->lpm.trees);
 731         return err;
 732 }
 733
 734 static void mlxsw_sp_lpm_fini(struct mlxsw_sp *mlxsw_sp)
 735 {
 736         struct mlxsw_sp_lpm_tree *lpm_tree;
 737
 738         lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV6];
 739         mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
 740
 741         lpm_tree = mlxsw_sp->router->lpm.proto_trees[MLXSW_SP_L3_PROTO_IPV4];
 742         mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
 743
 744         kfree(mlxsw_sp->router->lpm.trees);
 745 }
 746
 747 static bool mlxsw_sp_vr_is_used(const struct mlxsw_sp_vr *vr)
 748 {
 749         return !!vr->fib4 || !!vr->fib6 ||
 750                !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] ||
 751                !!vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
 752 }
 753
 754 static struct mlxsw_sp_vr *mlxsw_sp_vr_find_unused(struct mlxsw_sp *mlxsw_sp)
 755 {
 756         struct mlxsw_sp_vr *vr;
 757         int i;
 758
 759         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
 760                 vr = &mlxsw_sp->router->vrs[i];
 761                 if (!mlxsw_sp_vr_is_used(vr))
 762                         return vr;
 763         }
 764         return NULL;
 765 }
 766
 767 static int mlxsw_sp_vr_lpm_tree_bind(struct mlxsw_sp *mlxsw_sp,
 768                                      const struct mlxsw_sp_fib *fib, u8 tree_id)
 769 {
 770         char xraltb_pl[MLXSW_REG_XRALTB_LEN];
 771
 772         mlxsw_reg_xraltb_pack(xraltb_pl, fib->vr->id,
 773                               (enum mlxsw_reg_ralxx_protocol) fib->proto,
 774                               tree_id);
 775         return fib->ll_ops->raltb_write(mlxsw_sp, xraltb_pl);
 776 }
 777
 778 static int mlxsw_sp_vr_lpm_tree_unbind(struct mlxsw_sp *mlxsw_sp,
 779                                        const struct mlxsw_sp_fib *fib)
 780 {
 781         char xraltb_pl[MLXSW_REG_XRALTB_LEN];
 782
 783         /* Bind to tree 0 which is default */
 784         mlxsw_reg_xraltb_pack(xraltb_pl, fib->vr->id,
 785                               (enum mlxsw_reg_ralxx_protocol) fib->proto, 0);
 786         return fib->ll_ops->raltb_write(mlxsw_sp, xraltb_pl);
 787 }
 788
 789 static u32 mlxsw_sp_fix_tb_id(u32 tb_id)
 790 {
 791         /* For our purpose, squash main, default and local tables into one */
 792         if (tb_id == RT_TABLE_LOCAL || tb_id == RT_TABLE_DEFAULT)
 793                 tb_id = RT_TABLE_MAIN;
 794         return tb_id;
 795 }
 796
 797 static struct mlxsw_sp_vr *mlxsw_sp_vr_find(struct mlxsw_sp *mlxsw_sp,
 798                                             u32 tb_id)
 799 {
 800         struct mlxsw_sp_vr *vr;
 801         int i;
 802
 803         tb_id = mlxsw_sp_fix_tb_id(tb_id);
 804
 805         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
 806                 vr = &mlxsw_sp->router->vrs[i];
 807                 if (mlxsw_sp_vr_is_used(vr) && vr->tb_id == tb_id)
 808                         return vr;
 809         }
 810         return NULL;
 811 }
 812
 813 int mlxsw_sp_router_tb_id_vr_id(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
 814                                 u16 *vr_id)
 815 {
 816         struct mlxsw_sp_vr *vr;
 817         int err = 0;
 818
 819         mutex_lock(&mlxsw_sp->router->lock);
 820         vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
 821         if (!vr) {
 822                 err = -ESRCH;
 823                 goto out;
 824         }
 825         *vr_id = vr->id;
 826 out:
 827         mutex_unlock(&mlxsw_sp->router->lock);
 828         return err;
 829 }
 830
 831 static struct mlxsw_sp_fib *mlxsw_sp_vr_fib(const struct mlxsw_sp_vr *vr,
 832                                             enum mlxsw_sp_l3proto proto)
 833 {
 834         switch (proto) {
 835         case MLXSW_SP_L3_PROTO_IPV4:
 836                 return vr->fib4;
 837         case MLXSW_SP_L3_PROTO_IPV6:
 838                 return vr->fib6;
 839         }
 840         return NULL;
 841 }
 842
 843 static struct mlxsw_sp_vr *mlxsw_sp_vr_create(struct mlxsw_sp *mlxsw_sp,
 844                                               u32 tb_id,
 845                                               struct netlink_ext_ack *extack)
 846 {
 847         struct mlxsw_sp_mr_table *mr4_table, *mr6_table;
 848         struct mlxsw_sp_fib *fib4;
 849         struct mlxsw_sp_fib *fib6;
 850         struct mlxsw_sp_vr *vr;
 851         int err;
 852
 853         vr = mlxsw_sp_vr_find_unused(mlxsw_sp);
 854         if (!vr) {
 855                 NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported virtual routers");
 856                 return ERR_PTR(-EBUSY);
 857         }
 858         fib4 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
 859         if (IS_ERR(fib4))
 860                 return ERR_CAST(fib4);
 861         fib6 = mlxsw_sp_fib_create(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
 862         if (IS_ERR(fib6)) {
 863                 err = PTR_ERR(fib6);
 864                 goto err_fib6_create;
 865         }
 866         mr4_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
 867                                              MLXSW_SP_L3_PROTO_IPV4);
 868         if (IS_ERR(mr4_table)) {
 869                 err = PTR_ERR(mr4_table);
 870                 goto err_mr4_table_create;
 871         }
 872         mr6_table = mlxsw_sp_mr_table_create(mlxsw_sp, vr->id,
 873                                              MLXSW_SP_L3_PROTO_IPV6);
 874         if (IS_ERR(mr6_table)) {
 875                 err = PTR_ERR(mr6_table);
 876                 goto err_mr6_table_create;
 877         }
 878
 879         vr->fib4 = fib4;
 880         vr->fib6 = fib6;
 881         vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = mr4_table;
 882         vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = mr6_table;
 883         vr->tb_id = tb_id;
 884         return vr;
 885
 886 err_mr6_table_create:
 887         mlxsw_sp_mr_table_destroy(mr4_table);
 888 err_mr4_table_create:
 889         mlxsw_sp_fib_destroy(mlxsw_sp, fib6);
 890 err_fib6_create:
 891         mlxsw_sp_fib_destroy(mlxsw_sp, fib4);
 892         return ERR_PTR(err);
 893 }
 894
 895 static void mlxsw_sp_vr_destroy(struct mlxsw_sp *mlxsw_sp,
 896                                 struct mlxsw_sp_vr *vr)
 897 {
 898         mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]);
 899         vr->mr_table[MLXSW_SP_L3_PROTO_IPV6] = NULL;
 900         mlxsw_sp_mr_table_destroy(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]);
 901         vr->mr_table[MLXSW_SP_L3_PROTO_IPV4] = NULL;
 902         mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib6);
 903         vr->fib6 = NULL;
 904         mlxsw_sp_fib_destroy(mlxsw_sp, vr->fib4);
 905         vr->fib4 = NULL;
 906 }
 907
 908 static struct mlxsw_sp_vr *mlxsw_sp_vr_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
 909                                            struct netlink_ext_ack *extack)
 910 {
 911         struct mlxsw_sp_vr *vr;
 912
 913         tb_id = mlxsw_sp_fix_tb_id(tb_id);
 914         vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
 915         if (!vr)
 916                 vr = mlxsw_sp_vr_create(mlxsw_sp, tb_id, extack);
 917         return vr;
 918 }
 919
 920 static void mlxsw_sp_vr_put(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr)
 921 {
 922         if (!vr->rif_count && list_empty(&vr->fib4->node_list) &&
 923             list_empty(&vr->fib6->node_list) &&
 924             mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV4]) &&
 925             mlxsw_sp_mr_table_empty(vr->mr_table[MLXSW_SP_L3_PROTO_IPV6]))
 926                 mlxsw_sp_vr_destroy(mlxsw_sp, vr);
 927 }
 928
 929 static bool
 930 mlxsw_sp_vr_lpm_tree_should_replace(struct mlxsw_sp_vr *vr,
 931                                     enum mlxsw_sp_l3proto proto, u8 tree_id)
 932 {
 933         struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
 934
 935         if (!mlxsw_sp_vr_is_used(vr))
 936                 return false;
 937         if (fib->lpm_tree->id == tree_id)
 938                 return true;
 939         return false;
 940 }
 941
 942 static int mlxsw_sp_vr_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
 943                                         struct mlxsw_sp_fib *fib,
 944                                         struct mlxsw_sp_lpm_tree *new_tree)
 945 {
 946         struct mlxsw_sp_lpm_tree *old_tree = fib->lpm_tree;
 947         int err;
 948
 949         fib->lpm_tree = new_tree;
 950         mlxsw_sp_lpm_tree_hold(new_tree);
 951         err = mlxsw_sp_vr_lpm_tree_bind(mlxsw_sp, fib, new_tree->id);
 952         if (err)
 953                 goto err_tree_bind;
 954         mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
 955         return 0;
 956
 957 err_tree_bind:
 958         mlxsw_sp_lpm_tree_put(mlxsw_sp, new_tree);
 959         fib->lpm_tree = old_tree;
 960         return err;
 961 }
 962
 963 static int mlxsw_sp_vrs_lpm_tree_replace(struct mlxsw_sp *mlxsw_sp,
 964                                          struct mlxsw_sp_fib *fib,
 965                                          struct mlxsw_sp_lpm_tree *new_tree)
 966 {
 967         enum mlxsw_sp_l3proto proto = fib->proto;
 968         struct mlxsw_sp_lpm_tree *old_tree;
 969         u8 old_id, new_id = new_tree->id;
 970         struct mlxsw_sp_vr *vr;
 971         int i, err;
 972
 973         old_tree = mlxsw_sp->router->lpm.proto_trees[proto];
 974         old_id = old_tree->id;
 975
 976         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
 977                 vr = &mlxsw_sp->router->vrs[i];
 978                 if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, old_id))
 979                         continue;
 980                 err = mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
 981                                                    mlxsw_sp_vr_fib(vr, proto),
 982                                                    new_tree);
 983                 if (err)
 984                         goto err_tree_replace;
 985         }
 986
 987         memcpy(new_tree->prefix_ref_count, old_tree->prefix_ref_count,
 988                sizeof(new_tree->prefix_ref_count));
 989         mlxsw_sp->router->lpm.proto_trees[proto] = new_tree;
 990         mlxsw_sp_lpm_tree_put(mlxsw_sp, old_tree);
 991
 992         return 0;
 993
 994 err_tree_replace:
 995         for (i--; i >= 0; i--) {
 996                 if (!mlxsw_sp_vr_lpm_tree_should_replace(vr, proto, new_id))
 997                         continue;
 998                 mlxsw_sp_vr_lpm_tree_replace(mlxsw_sp,
 999                                              mlxsw_sp_vr_fib(vr, proto),
1000                                              old_tree);
1001         }
1002         return err;
1003 }
1004
1005 static int mlxsw_sp_vrs_init(struct mlxsw_sp *mlxsw_sp)
1006 {
1007         struct mlxsw_sp_vr *vr;
1008         u64 max_vrs;
1009         int i;
1010
1011         if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_VRS))
1012                 return -EIO;
1013
1014         max_vrs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS);
1015         mlxsw_sp->router->vrs = kcalloc(max_vrs, sizeof(struct mlxsw_sp_vr),
1016                                         GFP_KERNEL);
1017         if (!mlxsw_sp->router->vrs)
1018                 return -ENOMEM;
1019
1020         for (i = 0; i < max_vrs; i++) {
1021                 vr = &mlxsw_sp->router->vrs[i];
1022                 vr->id = i;
1023         }
1024
1025         return 0;
1026 }
1027
1028 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp);
1029
1030 static void mlxsw_sp_vrs_fini(struct mlxsw_sp *mlxsw_sp)
1031 {
1032         /* At this stage we're guaranteed not to have new incoming
1033          * FIB notifications and the work queue is free from FIBs
1034          * sitting on top of mlxsw netdevs. However, we can still
1035          * have other FIBs queued. Flush the queue before flushing
1036          * the device's tables. No need for locks, as we're the only
1037          * writer.
1038          */
1039         mlxsw_core_flush_owq();
1040         mlxsw_sp_router_fib_flush(mlxsw_sp);
1041         kfree(mlxsw_sp->router->vrs);
1042 }
1043
1044 static struct net_device *
1045 __mlxsw_sp_ipip_netdev_ul_dev_get(const struct net_device *ol_dev)
1046 {
1047         struct ip_tunnel *tun = netdev_priv(ol_dev);
1048         struct net *net = dev_net(ol_dev);
1049
1050         return dev_get_by_index_rcu(net, tun->parms.link);
1051 }
1052
1053 u32 mlxsw_sp_ipip_dev_ul_tb_id(const struct net_device *ol_dev)
1054 {
1055         struct net_device *d;
1056         u32 tb_id;
1057
1058         rcu_read_lock();
1059         d = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1060         if (d)
1061                 tb_id = l3mdev_fib_table(d) ? : RT_TABLE_MAIN;
1062         else
1063                 tb_id = RT_TABLE_MAIN;
1064         rcu_read_unlock();
1065
1066         return tb_id;
1067 }
1068
1069 static struct mlxsw_sp_rif *
1070 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
1071                     const struct mlxsw_sp_rif_params *params,
1072                     struct netlink_ext_ack *extack);
1073
1074 static struct mlxsw_sp_rif_ipip_lb *
1075 mlxsw_sp_ipip_ol_ipip_lb_create(struct mlxsw_sp *mlxsw_sp,
1076                                 enum mlxsw_sp_ipip_type ipipt,
1077                                 struct net_device *ol_dev,
1078                                 struct netlink_ext_ack *extack)
1079 {
1080         struct mlxsw_sp_rif_params_ipip_lb lb_params;
1081         const struct mlxsw_sp_ipip_ops *ipip_ops;
1082         struct mlxsw_sp_rif *rif;
1083
1084         ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1085         lb_params = (struct mlxsw_sp_rif_params_ipip_lb) {
1086                 .common.dev = ol_dev,
1087                 .common.lag = false,
1088                 .lb_config = ipip_ops->ol_loopback_config(mlxsw_sp, ol_dev),
1089         };
1090
1091         rif = mlxsw_sp_rif_create(mlxsw_sp, &lb_params.common, extack);
1092         if (IS_ERR(rif))
1093                 return ERR_CAST(rif);
1094         return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
1095 }
1096
1097 static struct mlxsw_sp_ipip_entry *
1098 mlxsw_sp_ipip_entry_alloc(struct mlxsw_sp *mlxsw_sp,
1099                           enum mlxsw_sp_ipip_type ipipt,
1100                           struct net_device *ol_dev)
1101 {
1102         const struct mlxsw_sp_ipip_ops *ipip_ops;
1103         struct mlxsw_sp_ipip_entry *ipip_entry;
1104         struct mlxsw_sp_ipip_entry *ret = NULL;
1105
1106         ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipipt];
1107         ipip_entry = kzalloc(sizeof(*ipip_entry), GFP_KERNEL);
1108         if (!ipip_entry)
1109                 return ERR_PTR(-ENOMEM);
1110
1111         ipip_entry->ol_lb = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp, ipipt,
1112                                                             ol_dev, NULL);
1113         if (IS_ERR(ipip_entry->ol_lb)) {
1114                 ret = ERR_CAST(ipip_entry->ol_lb);
1115                 goto err_ol_ipip_lb_create;
1116         }
1117
1118         ipip_entry->ipipt = ipipt;
1119         ipip_entry->ol_dev = ol_dev;
1120
1121         switch (ipip_ops->ul_proto) {
1122         case MLXSW_SP_L3_PROTO_IPV4:
1123                 ipip_entry->parms4 = mlxsw_sp_ipip_netdev_parms4(ol_dev);
1124                 break;
1125         case MLXSW_SP_L3_PROTO_IPV6:
1126                 WARN_ON(1);
1127                 break;
1128         }
1129
1130         return ipip_entry;
1131
1132 err_ol_ipip_lb_create:
1133         kfree(ipip_entry);
1134         return ret;
1135 }
1136
1137 static void
1138 mlxsw_sp_ipip_entry_dealloc(struct mlxsw_sp_ipip_entry *ipip_entry)
1139 {
1140         mlxsw_sp_rif_destroy(&ipip_entry->ol_lb->common);
1141         kfree(ipip_entry);
1142 }
1143
1144 static bool
1145 mlxsw_sp_ipip_entry_saddr_matches(struct mlxsw_sp *mlxsw_sp,
1146                                   const enum mlxsw_sp_l3proto ul_proto,
1147                                   union mlxsw_sp_l3addr saddr,
1148                                   u32 ul_tb_id,
1149                                   struct mlxsw_sp_ipip_entry *ipip_entry)
1150 {
1151         u32 tun_ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1152         enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1153         union mlxsw_sp_l3addr tun_saddr;
1154
1155         if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1156                 return false;
1157
1158         tun_saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1159         return tun_ul_tb_id == ul_tb_id &&
1160                mlxsw_sp_l3addr_eq(&tun_saddr, &saddr);
1161 }
1162
1163 static int
1164 mlxsw_sp_fib_entry_decap_init(struct mlxsw_sp *mlxsw_sp,
1165                               struct mlxsw_sp_fib_entry *fib_entry,
1166                               struct mlxsw_sp_ipip_entry *ipip_entry)
1167 {
1168         u32 tunnel_index;
1169         int err;
1170
1171         err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1172                                   1, &tunnel_index);
1173         if (err)
1174                 return err;
1175
1176         ipip_entry->decap_fib_entry = fib_entry;
1177         fib_entry->decap.ipip_entry = ipip_entry;
1178         fib_entry->decap.tunnel_index = tunnel_index;
1179         return 0;
1180 }
1181
1182 static void mlxsw_sp_fib_entry_decap_fini(struct mlxsw_sp *mlxsw_sp,
1183                                           struct mlxsw_sp_fib_entry *fib_entry)
1184 {
1185         /* Unlink this node from the IPIP entry that it's the decap entry of. */
1186         fib_entry->decap.ipip_entry->decap_fib_entry = NULL;
1187         fib_entry->decap.ipip_entry = NULL;
1188         mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
1189                            1, fib_entry->decap.tunnel_index);
1190 }
1191
1192 static struct mlxsw_sp_fib_node *
1193 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
1194                          size_t addr_len, unsigned char prefix_len);
1195 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
1196                                      struct mlxsw_sp_fib_entry *fib_entry);
1197
1198 static void
1199 mlxsw_sp_ipip_entry_demote_decap(struct mlxsw_sp *mlxsw_sp,
1200                                  struct mlxsw_sp_ipip_entry *ipip_entry)
1201 {
1202         struct mlxsw_sp_fib_entry *fib_entry = ipip_entry->decap_fib_entry;
1203
1204         mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
1205         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1206
1207         mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1208 }
1209
1210 static void
1211 mlxsw_sp_ipip_entry_promote_decap(struct mlxsw_sp *mlxsw_sp,
1212                                   struct mlxsw_sp_ipip_entry *ipip_entry,
1213                                   struct mlxsw_sp_fib_entry *decap_fib_entry)
1214 {
1215         if (mlxsw_sp_fib_entry_decap_init(mlxsw_sp, decap_fib_entry,
1216                                           ipip_entry))
1217                 return;
1218         decap_fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
1219
1220         if (mlxsw_sp_fib_entry_update(mlxsw_sp, decap_fib_entry))
1221                 mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1222 }
1223
1224 static struct mlxsw_sp_fib_entry *
1225 mlxsw_sp_router_ip2me_fib_entry_find(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
1226                                      enum mlxsw_sp_l3proto proto,
1227                                      const union mlxsw_sp_l3addr *addr,
1228                                      enum mlxsw_sp_fib_entry_type type)
1229 {
1230         struct mlxsw_sp_fib_node *fib_node;
1231         unsigned char addr_prefix_len;
1232         struct mlxsw_sp_fib *fib;
1233         struct mlxsw_sp_vr *vr;
1234         const void *addrp;
1235         size_t addr_len;
1236         u32 addr4;
1237
1238         vr = mlxsw_sp_vr_find(mlxsw_sp, tb_id);
1239         if (!vr)
1240                 return NULL;
1241         fib = mlxsw_sp_vr_fib(vr, proto);
1242
1243         switch (proto) {
1244         case MLXSW_SP_L3_PROTO_IPV4:
1245                 addr4 = be32_to_cpu(addr->addr4);
1246                 addrp = &addr4;
1247                 addr_len = 4;
1248                 addr_prefix_len = 32;
1249                 break;
1250         case MLXSW_SP_L3_PROTO_IPV6:
1251         default:
1252                 WARN_ON(1);
1253                 return NULL;
1254         }
1255
1256         fib_node = mlxsw_sp_fib_node_lookup(fib, addrp, addr_len,
1257                                             addr_prefix_len);
1258         if (!fib_node || fib_node->fib_entry->type != type)
1259                 return NULL;
1260
1261         return fib_node->fib_entry;
1262 }
1263
1264 /* Given an IPIP entry, find the corresponding decap route. */
1265 static struct mlxsw_sp_fib_entry *
1266 mlxsw_sp_ipip_entry_find_decap(struct mlxsw_sp *mlxsw_sp,
1267                                struct mlxsw_sp_ipip_entry *ipip_entry)
1268 {
1269         static struct mlxsw_sp_fib_node *fib_node;
1270         const struct mlxsw_sp_ipip_ops *ipip_ops;
1271         unsigned char saddr_prefix_len;
1272         union mlxsw_sp_l3addr saddr;
1273         struct mlxsw_sp_fib *ul_fib;
1274         struct mlxsw_sp_vr *ul_vr;
1275         const void *saddrp;
1276         size_t saddr_len;
1277         u32 ul_tb_id;
1278         u32 saddr4;
1279
1280         ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1281
1282         ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
1283         ul_vr = mlxsw_sp_vr_find(mlxsw_sp, ul_tb_id);
1284         if (!ul_vr)
1285                 return NULL;
1286
1287         ul_fib = mlxsw_sp_vr_fib(ul_vr, ipip_ops->ul_proto);
1288         saddr = mlxsw_sp_ipip_netdev_saddr(ipip_ops->ul_proto,
1289                                            ipip_entry->ol_dev);
1290
1291         switch (ipip_ops->ul_proto) {
1292         case MLXSW_SP_L3_PROTO_IPV4:
1293                 saddr4 = be32_to_cpu(saddr.addr4);
1294                 saddrp = &saddr4;
1295                 saddr_len = 4;
1296                 saddr_prefix_len = 32;
1297                 break;
1298         default:
1299                 WARN_ON(1);
1300                 return NULL;
1301         }
1302
1303         fib_node = mlxsw_sp_fib_node_lookup(ul_fib, saddrp, saddr_len,
1304                                             saddr_prefix_len);
1305         if (!fib_node ||
1306             fib_node->fib_entry->type != MLXSW_SP_FIB_ENTRY_TYPE_TRAP)
1307                 return NULL;
1308
1309         return fib_node->fib_entry;
1310 }
1311
1312 static struct mlxsw_sp_ipip_entry *
1313 mlxsw_sp_ipip_entry_create(struct mlxsw_sp *mlxsw_sp,
1314                            enum mlxsw_sp_ipip_type ipipt,
1315                            struct net_device *ol_dev)
1316 {
1317         struct mlxsw_sp_ipip_entry *ipip_entry;
1318
1319         ipip_entry = mlxsw_sp_ipip_entry_alloc(mlxsw_sp, ipipt, ol_dev);
1320         if (IS_ERR(ipip_entry))
1321                 return ipip_entry;
1322
1323         list_add_tail(&ipip_entry->ipip_list_node,
1324                       &mlxsw_sp->router->ipip_list);
1325
1326         return ipip_entry;
1327 }
1328
1329 static void
1330 mlxsw_sp_ipip_entry_destroy(struct mlxsw_sp *mlxsw_sp,
1331                             struct mlxsw_sp_ipip_entry *ipip_entry)
1332 {
1333         list_del(&ipip_entry->ipip_list_node);
1334         mlxsw_sp_ipip_entry_dealloc(ipip_entry);
1335 }
1336
1337 static bool
1338 mlxsw_sp_ipip_entry_matches_decap(struct mlxsw_sp *mlxsw_sp,
1339                                   const struct net_device *ul_dev,
1340                                   enum mlxsw_sp_l3proto ul_proto,
1341                                   union mlxsw_sp_l3addr ul_dip,
1342                                   struct mlxsw_sp_ipip_entry *ipip_entry)
1343 {
1344         u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1345         enum mlxsw_sp_ipip_type ipipt = ipip_entry->ipipt;
1346
1347         if (mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto != ul_proto)
1348                 return false;
1349
1350         return mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, ul_dip,
1351                                                  ul_tb_id, ipip_entry);
1352 }
1353
1354 /* Given decap parameters, find the corresponding IPIP entry. */
1355 static struct mlxsw_sp_ipip_entry *
1356 mlxsw_sp_ipip_entry_find_by_decap(struct mlxsw_sp *mlxsw_sp,
1357                                   const struct net_device *ul_dev,
1358                                   enum mlxsw_sp_l3proto ul_proto,
1359                                   union mlxsw_sp_l3addr ul_dip)
1360 {
1361         struct mlxsw_sp_ipip_entry *ipip_entry;
1362
1363         list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1364                             ipip_list_node)
1365                 if (mlxsw_sp_ipip_entry_matches_decap(mlxsw_sp, ul_dev,
1366                                                       ul_proto, ul_dip,
1367                                                       ipip_entry))
1368                         return ipip_entry;
1369
1370         return NULL;
1371 }
1372
1373 static bool mlxsw_sp_netdev_ipip_type(const struct mlxsw_sp *mlxsw_sp,
1374                                       const struct net_device *dev,
1375                                       enum mlxsw_sp_ipip_type *p_type)
1376 {
1377         struct mlxsw_sp_router *router = mlxsw_sp->router;
1378         const struct mlxsw_sp_ipip_ops *ipip_ops;
1379         enum mlxsw_sp_ipip_type ipipt;
1380
1381         for (ipipt = 0; ipipt < MLXSW_SP_IPIP_TYPE_MAX; ++ipipt) {
1382                 ipip_ops = router->ipip_ops_arr[ipipt];
1383                 if (dev->type == ipip_ops->dev_type) {
1384                         if (p_type)
1385                                 *p_type = ipipt;
1386                         return true;
1387                 }
1388         }
1389         return false;
1390 }
1391
1392 bool mlxsw_sp_netdev_is_ipip_ol(const struct mlxsw_sp *mlxsw_sp,
1393                                 const struct net_device *dev)
1394 {
1395         return mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL);
1396 }
1397
1398 static struct mlxsw_sp_ipip_entry *
1399 mlxsw_sp_ipip_entry_find_by_ol_dev(struct mlxsw_sp *mlxsw_sp,
1400                                    const struct net_device *ol_dev)
1401 {
1402         struct mlxsw_sp_ipip_entry *ipip_entry;
1403
1404         list_for_each_entry(ipip_entry, &mlxsw_sp->router->ipip_list,
1405                             ipip_list_node)
1406                 if (ipip_entry->ol_dev == ol_dev)
1407                         return ipip_entry;
1408
1409         return NULL;
1410 }
1411
1412 static struct mlxsw_sp_ipip_entry *
1413 mlxsw_sp_ipip_entry_find_by_ul_dev(const struct mlxsw_sp *mlxsw_sp,
1414                                    const struct net_device *ul_dev,
1415                                    struct mlxsw_sp_ipip_entry *start)
1416 {
1417         struct mlxsw_sp_ipip_entry *ipip_entry;
1418
1419         ipip_entry = list_prepare_entry(start, &mlxsw_sp->router->ipip_list,
1420                                         ipip_list_node);
1421         list_for_each_entry_continue(ipip_entry, &mlxsw_sp->router->ipip_list,
1422                                      ipip_list_node) {
1423                 struct net_device *ol_dev = ipip_entry->ol_dev;
1424                 struct net_device *ipip_ul_dev;
1425
1426                 rcu_read_lock();
1427                 ipip_ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1428                 rcu_read_unlock();
1429
1430                 if (ipip_ul_dev == ul_dev)
1431                         return ipip_entry;
1432         }
1433
1434         return NULL;
1435 }
1436
1437 bool mlxsw_sp_netdev_is_ipip_ul(struct mlxsw_sp *mlxsw_sp,
1438                                 const struct net_device *dev)
1439 {
1440         bool is_ipip_ul;
1441
1442         mutex_lock(&mlxsw_sp->router->lock);
1443         is_ipip_ul = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp, dev, NULL);
1444         mutex_unlock(&mlxsw_sp->router->lock);
1445
1446         return is_ipip_ul;
1447 }
1448
1449 static bool mlxsw_sp_netdevice_ipip_can_offload(struct mlxsw_sp *mlxsw_sp,
1450                                                 const struct net_device *ol_dev,
1451                                                 enum mlxsw_sp_ipip_type ipipt)
1452 {
1453         const struct mlxsw_sp_ipip_ops *ops
1454                 = mlxsw_sp->router->ipip_ops_arr[ipipt];
1455
1456         return ops->can_offload(mlxsw_sp, ol_dev);
1457 }
1458
1459 static int mlxsw_sp_netdevice_ipip_ol_reg_event(struct mlxsw_sp *mlxsw_sp,
1460                                                 struct net_device *ol_dev)
1461 {
1462         enum mlxsw_sp_ipip_type ipipt = MLXSW_SP_IPIP_TYPE_MAX;
1463         struct mlxsw_sp_ipip_entry *ipip_entry;
1464         enum mlxsw_sp_l3proto ul_proto;
1465         union mlxsw_sp_l3addr saddr;
1466         u32 ul_tb_id;
1467
1468         mlxsw_sp_netdev_ipip_type(mlxsw_sp, ol_dev, &ipipt);
1469         if (mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev, ipipt)) {
1470                 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ol_dev);
1471                 ul_proto = mlxsw_sp->router->ipip_ops_arr[ipipt]->ul_proto;
1472                 saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ol_dev);
1473                 if (!mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1474                                                           saddr, ul_tb_id,
1475                                                           NULL)) {
1476                         ipip_entry = mlxsw_sp_ipip_entry_create(mlxsw_sp, ipipt,
1477                                                                 ol_dev);
1478                         if (IS_ERR(ipip_entry))
1479                                 return PTR_ERR(ipip_entry);
1480                 }
1481         }
1482
1483         return 0;
1484 }
1485
1486 static void mlxsw_sp_netdevice_ipip_ol_unreg_event(struct mlxsw_sp *mlxsw_sp,
1487                                                    struct net_device *ol_dev)
1488 {
1489         struct mlxsw_sp_ipip_entry *ipip_entry;
1490
1491         ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1492         if (ipip_entry)
1493                 mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1494 }
1495
1496 static void
1497 mlxsw_sp_ipip_entry_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1498                                 struct mlxsw_sp_ipip_entry *ipip_entry)
1499 {
1500         struct mlxsw_sp_fib_entry *decap_fib_entry;
1501
1502         decap_fib_entry = mlxsw_sp_ipip_entry_find_decap(mlxsw_sp, ipip_entry);
1503         if (decap_fib_entry)
1504                 mlxsw_sp_ipip_entry_promote_decap(mlxsw_sp, ipip_entry,
1505                                                   decap_fib_entry);
1506 }
1507
1508 static int
1509 mlxsw_sp_rif_ipip_lb_op(struct mlxsw_sp_rif_ipip_lb *lb_rif, u16 ul_vr_id,
1510                         u16 ul_rif_id, bool enable)
1511 {
1512         struct mlxsw_sp_rif_ipip_lb_config lb_cf = lb_rif->lb_config;
1513         struct mlxsw_sp_rif *rif = &lb_rif->common;
1514         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
1515         char ritr_pl[MLXSW_REG_RITR_LEN];
1516         u32 saddr4;
1517
1518         switch (lb_cf.ul_protocol) {
1519         case MLXSW_SP_L3_PROTO_IPV4:
1520                 saddr4 = be32_to_cpu(lb_cf.saddr.addr4);
1521                 mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
1522                                     rif->rif_index, rif->vr_id, rif->dev->mtu);
1523                 mlxsw_reg_ritr_loopback_ipip4_pack(ritr_pl, lb_cf.lb_ipipt,
1524                             MLXSW_REG_RITR_LOOPBACK_IPIP_OPTIONS_GRE_KEY_PRESET,
1525                             ul_vr_id, ul_rif_id, saddr4, lb_cf.okey);
1526                 break;
1527
1528         case MLXSW_SP_L3_PROTO_IPV6:
1529                 return -EAFNOSUPPORT;
1530         }
1531
1532         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
1533 }
1534
1535 static int mlxsw_sp_netdevice_ipip_ol_update_mtu(struct mlxsw_sp *mlxsw_sp,
1536                                                  struct net_device *ol_dev)
1537 {
1538         struct mlxsw_sp_ipip_entry *ipip_entry;
1539         struct mlxsw_sp_rif_ipip_lb *lb_rif;
1540         int err = 0;
1541
1542         ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1543         if (ipip_entry) {
1544                 lb_rif = ipip_entry->ol_lb;
1545                 err = mlxsw_sp_rif_ipip_lb_op(lb_rif, lb_rif->ul_vr_id,
1546                                               lb_rif->ul_rif_id, true);
1547                 if (err)
1548                         goto out;
1549                 lb_rif->common.mtu = ol_dev->mtu;
1550         }
1551
1552 out:
1553         return err;
1554 }
1555
1556 static void mlxsw_sp_netdevice_ipip_ol_up_event(struct mlxsw_sp *mlxsw_sp,
1557                                                 struct net_device *ol_dev)
1558 {
1559         struct mlxsw_sp_ipip_entry *ipip_entry;
1560
1561         ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1562         if (ipip_entry)
1563                 mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1564 }
1565
1566 static void
1567 mlxsw_sp_ipip_entry_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1568                                   struct mlxsw_sp_ipip_entry *ipip_entry)
1569 {
1570         if (ipip_entry->decap_fib_entry)
1571                 mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1572 }
1573
1574 static void mlxsw_sp_netdevice_ipip_ol_down_event(struct mlxsw_sp *mlxsw_sp,
1575                                                   struct net_device *ol_dev)
1576 {
1577         struct mlxsw_sp_ipip_entry *ipip_entry;
1578
1579         ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1580         if (ipip_entry)
1581                 mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1582 }
1583
1584 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
1585                                          struct mlxsw_sp_rif *old_rif,
1586                                          struct mlxsw_sp_rif *new_rif);
1587 static int
1588 mlxsw_sp_ipip_entry_ol_lb_update(struct mlxsw_sp *mlxsw_sp,
1589                                  struct mlxsw_sp_ipip_entry *ipip_entry,
1590                                  bool keep_encap,
1591                                  struct netlink_ext_ack *extack)
1592 {
1593         struct mlxsw_sp_rif_ipip_lb *old_lb_rif = ipip_entry->ol_lb;
1594         struct mlxsw_sp_rif_ipip_lb *new_lb_rif;
1595
1596         new_lb_rif = mlxsw_sp_ipip_ol_ipip_lb_create(mlxsw_sp,
1597                                                      ipip_entry->ipipt,
1598                                                      ipip_entry->ol_dev,
1599                                                      extack);
1600         if (IS_ERR(new_lb_rif))
1601                 return PTR_ERR(new_lb_rif);
1602         ipip_entry->ol_lb = new_lb_rif;
1603
1604         if (keep_encap)
1605                 mlxsw_sp_nexthop_rif_migrate(mlxsw_sp, &old_lb_rif->common,
1606                                              &new_lb_rif->common);
1607
1608         mlxsw_sp_rif_destroy(&old_lb_rif->common);
1609
1610         return 0;
1611 }
1612
1613 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
1614                                         struct mlxsw_sp_rif *rif);
1615
1616 /**
1617  * __mlxsw_sp_ipip_entry_update_tunnel - Update offload related to IPIP entry.
1618  * @mlxsw_sp: mlxsw_sp.
1619  * @ipip_entry: IPIP entry.
1620  * @recreate_loopback: Recreates the associated loopback RIF.
1621  * @keep_encap: Updates next hops that use the tunnel netdevice. This is only
1622  *              relevant when recreate_loopback is true.
1623  * @update_nexthops: Updates next hops, keeping the current loopback RIF. This
1624  *                   is only relevant when recreate_loopback is false.
1625  * @extack: extack.
1626  *
1627  * Return: Non-zero value on failure.
1628  */
1629 int __mlxsw_sp_ipip_entry_update_tunnel(struct mlxsw_sp *mlxsw_sp,
1630                                         struct mlxsw_sp_ipip_entry *ipip_entry,
1631                                         bool recreate_loopback,
1632                                         bool keep_encap,
1633                                         bool update_nexthops,
1634                                         struct netlink_ext_ack *extack)
1635 {
1636         int err;
1637
1638         /* RIFs can't be edited, so to update loopback, we need to destroy and
1639          * recreate it. That creates a window of opportunity where RALUE and
1640          * RATR registers end up referencing a RIF that's already gone. RATRs
1641          * are handled in mlxsw_sp_ipip_entry_ol_lb_update(), and to take care
1642          * of RALUE, demote the decap route back.
1643          */
1644         if (ipip_entry->decap_fib_entry)
1645                 mlxsw_sp_ipip_entry_demote_decap(mlxsw_sp, ipip_entry);
1646
1647         if (recreate_loopback) {
1648                 err = mlxsw_sp_ipip_entry_ol_lb_update(mlxsw_sp, ipip_entry,
1649                                                        keep_encap, extack);
1650                 if (err)
1651                         return err;
1652         } else if (update_nexthops) {
1653                 mlxsw_sp_nexthop_rif_update(mlxsw_sp,
1654                                             &ipip_entry->ol_lb->common);
1655         }
1656
1657         if (ipip_entry->ol_dev->flags & IFF_UP)
1658                 mlxsw_sp_ipip_entry_ol_up_event(mlxsw_sp, ipip_entry);
1659
1660         return 0;
1661 }
1662
1663 static int mlxsw_sp_netdevice_ipip_ol_vrf_event(struct mlxsw_sp *mlxsw_sp,
1664                                                 struct net_device *ol_dev,
1665                                                 struct netlink_ext_ack *extack)
1666 {
1667         struct mlxsw_sp_ipip_entry *ipip_entry =
1668                 mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1669
1670         if (!ipip_entry)
1671                 return 0;
1672
1673         return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1674                                                    true, false, false, extack);
1675 }
1676
1677 static int
1678 mlxsw_sp_netdevice_ipip_ul_vrf_event(struct mlxsw_sp *mlxsw_sp,
1679                                      struct mlxsw_sp_ipip_entry *ipip_entry,
1680                                      struct net_device *ul_dev,
1681                                      bool *demote_this,
1682                                      struct netlink_ext_ack *extack)
1683 {
1684         u32 ul_tb_id = l3mdev_fib_table(ul_dev) ? : RT_TABLE_MAIN;
1685         enum mlxsw_sp_l3proto ul_proto;
1686         union mlxsw_sp_l3addr saddr;
1687
1688         /* Moving underlay to a different VRF might cause local address
1689          * conflict, and the conflicting tunnels need to be demoted.
1690          */
1691         ul_proto = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt]->ul_proto;
1692         saddr = mlxsw_sp_ipip_netdev_saddr(ul_proto, ipip_entry->ol_dev);
1693         if (mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp, ul_proto,
1694                                                  saddr, ul_tb_id,
1695                                                  ipip_entry)) {
1696                 *demote_this = true;
1697                 return 0;
1698         }
1699
1700         return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1701                                                    true, true, false, extack);
1702 }
1703
1704 static int
1705 mlxsw_sp_netdevice_ipip_ul_up_event(struct mlxsw_sp *mlxsw_sp,
1706                                     struct mlxsw_sp_ipip_entry *ipip_entry,
1707                                     struct net_device *ul_dev)
1708 {
1709         return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1710                                                    false, false, true, NULL);
1711 }
1712
1713 static int
1714 mlxsw_sp_netdevice_ipip_ul_down_event(struct mlxsw_sp *mlxsw_sp,
1715                                       struct mlxsw_sp_ipip_entry *ipip_entry,
1716                                       struct net_device *ul_dev)
1717 {
1718         /* A down underlay device causes encapsulated packets to not be
1719          * forwarded, but decap still works. So refresh next hops without
1720          * touching anything else.
1721          */
1722         return __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
1723                                                    false, false, true, NULL);
1724 }
1725
1726 static int
1727 mlxsw_sp_netdevice_ipip_ol_change_event(struct mlxsw_sp *mlxsw_sp,
1728                                         struct net_device *ol_dev,
1729                                         struct netlink_ext_ack *extack)
1730 {
1731         const struct mlxsw_sp_ipip_ops *ipip_ops;
1732         struct mlxsw_sp_ipip_entry *ipip_entry;
1733         int err;
1734
1735         ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
1736         if (!ipip_entry)
1737                 /* A change might make a tunnel eligible for offloading, but
1738                  * that is currently not implemented. What falls to slow path
1739                  * stays there.
1740                  */
1741                 return 0;
1742
1743         /* A change might make a tunnel not eligible for offloading. */
1744         if (!mlxsw_sp_netdevice_ipip_can_offload(mlxsw_sp, ol_dev,
1745                                                  ipip_entry->ipipt)) {
1746                 mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1747                 return 0;
1748         }
1749
1750         ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
1751         err = ipip_ops->ol_netdev_change(mlxsw_sp, ipip_entry, extack);
1752         return err;
1753 }
1754
1755 void mlxsw_sp_ipip_entry_demote_tunnel(struct mlxsw_sp *mlxsw_sp,
1756                                        struct mlxsw_sp_ipip_entry *ipip_entry)
1757 {
1758         struct net_device *ol_dev = ipip_entry->ol_dev;
1759
1760         if (ol_dev->flags & IFF_UP)
1761                 mlxsw_sp_ipip_entry_ol_down_event(mlxsw_sp, ipip_entry);
1762         mlxsw_sp_ipip_entry_destroy(mlxsw_sp, ipip_entry);
1763 }
1764
1765 /* The configuration where several tunnels have the same local address in the
1766  * same underlay table needs special treatment in the HW. That is currently not
1767  * implemented in the driver. This function finds and demotes the first tunnel
1768  * with a given source address, except the one passed in in the argument
1769  * `except'.
1770  */
1771 bool
1772 mlxsw_sp_ipip_demote_tunnel_by_saddr(struct mlxsw_sp *mlxsw_sp,
1773                                      enum mlxsw_sp_l3proto ul_proto,
1774                                      union mlxsw_sp_l3addr saddr,
1775                                      u32 ul_tb_id,
1776                                      const struct mlxsw_sp_ipip_entry *except)
1777 {
1778         struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1779
1780         list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1781                                  ipip_list_node) {
1782                 if (ipip_entry != except &&
1783                     mlxsw_sp_ipip_entry_saddr_matches(mlxsw_sp, ul_proto, saddr,
1784                                                       ul_tb_id, ipip_entry)) {
1785                         mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1786                         return true;
1787                 }
1788         }
1789
1790         return false;
1791 }
1792
1793 static void mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(struct mlxsw_sp *mlxsw_sp,
1794                                                      struct net_device *ul_dev)
1795 {
1796         struct mlxsw_sp_ipip_entry *ipip_entry, *tmp;
1797
1798         list_for_each_entry_safe(ipip_entry, tmp, &mlxsw_sp->router->ipip_list,
1799                                  ipip_list_node) {
1800                 struct net_device *ol_dev = ipip_entry->ol_dev;
1801                 struct net_device *ipip_ul_dev;
1802
1803                 rcu_read_lock();
1804                 ipip_ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
1805                 rcu_read_unlock();
1806                 if (ipip_ul_dev == ul_dev)
1807                         mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1808         }
1809 }
1810
1811 int mlxsw_sp_netdevice_ipip_ol_event(struct mlxsw_sp *mlxsw_sp,
1812                                      struct net_device *ol_dev,
1813                                      unsigned long event,
1814                                      struct netdev_notifier_info *info)
1815 {
1816         struct netdev_notifier_changeupper_info *chup;
1817         struct netlink_ext_ack *extack;
1818         int err = 0;
1819
1820         mutex_lock(&mlxsw_sp->router->lock);
1821         switch (event) {
1822         case NETDEV_REGISTER:
1823                 err = mlxsw_sp_netdevice_ipip_ol_reg_event(mlxsw_sp, ol_dev);
1824                 break;
1825         case NETDEV_UNREGISTER:
1826                 mlxsw_sp_netdevice_ipip_ol_unreg_event(mlxsw_sp, ol_dev);
1827                 break;
1828         case NETDEV_UP:
1829                 mlxsw_sp_netdevice_ipip_ol_up_event(mlxsw_sp, ol_dev);
1830                 break;
1831         case NETDEV_DOWN:
1832                 mlxsw_sp_netdevice_ipip_ol_down_event(mlxsw_sp, ol_dev);
1833                 break;
1834         case NETDEV_CHANGEUPPER:
1835                 chup = container_of(info, typeof(*chup), info);
1836                 extack = info->extack;
1837                 if (netif_is_l3_master(chup->upper_dev))
1838                         err = mlxsw_sp_netdevice_ipip_ol_vrf_event(mlxsw_sp,
1839                                                                    ol_dev,
1840                                                                    extack);
1841                 break;
1842         case NETDEV_CHANGE:
1843                 extack = info->extack;
1844                 err = mlxsw_sp_netdevice_ipip_ol_change_event(mlxsw_sp,
1845                                                               ol_dev, extack);
1846                 break;
1847         case NETDEV_CHANGEMTU:
1848                 err = mlxsw_sp_netdevice_ipip_ol_update_mtu(mlxsw_sp, ol_dev);
1849                 break;
1850         }
1851         mutex_unlock(&mlxsw_sp->router->lock);
1852         return err;
1853 }
1854
1855 static int
1856 __mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1857                                    struct mlxsw_sp_ipip_entry *ipip_entry,
1858                                    struct net_device *ul_dev,
1859                                    bool *demote_this,
1860                                    unsigned long event,
1861                                    struct netdev_notifier_info *info)
1862 {
1863         struct netdev_notifier_changeupper_info *chup;
1864         struct netlink_ext_ack *extack;
1865
1866         switch (event) {
1867         case NETDEV_CHANGEUPPER:
1868                 chup = container_of(info, typeof(*chup), info);
1869                 extack = info->extack;
1870                 if (netif_is_l3_master(chup->upper_dev))
1871                         return mlxsw_sp_netdevice_ipip_ul_vrf_event(mlxsw_sp,
1872                                                                     ipip_entry,
1873                                                                     ul_dev,
1874                                                                     demote_this,
1875                                                                     extack);
1876                 break;
1877
1878         case NETDEV_UP:
1879                 return mlxsw_sp_netdevice_ipip_ul_up_event(mlxsw_sp, ipip_entry,
1880                                                            ul_dev);
1881         case NETDEV_DOWN:
1882                 return mlxsw_sp_netdevice_ipip_ul_down_event(mlxsw_sp,
1883                                                              ipip_entry,
1884                                                              ul_dev);
1885         }
1886         return 0;
1887 }
1888
1889 int
1890 mlxsw_sp_netdevice_ipip_ul_event(struct mlxsw_sp *mlxsw_sp,
1891                                  struct net_device *ul_dev,
1892                                  unsigned long event,
1893                                  struct netdev_notifier_info *info)
1894 {
1895         struct mlxsw_sp_ipip_entry *ipip_entry = NULL;
1896         int err = 0;
1897
1898         mutex_lock(&mlxsw_sp->router->lock);
1899         while ((ipip_entry = mlxsw_sp_ipip_entry_find_by_ul_dev(mlxsw_sp,
1900                                                                 ul_dev,
1901                                                                 ipip_entry))) {
1902                 struct mlxsw_sp_ipip_entry *prev;
1903                 bool demote_this = false;
1904
1905                 err = __mlxsw_sp_netdevice_ipip_ul_event(mlxsw_sp, ipip_entry,
1906                                                          ul_dev, &demote_this,
1907                                                          event, info);
1908                 if (err) {
1909                         mlxsw_sp_ipip_demote_tunnel_by_ul_netdev(mlxsw_sp,
1910                                                                  ul_dev);
1911                         break;
1912                 }
1913
1914                 if (demote_this) {
1915                         if (list_is_first(&ipip_entry->ipip_list_node,
1916                                           &mlxsw_sp->router->ipip_list))
1917                                 prev = NULL;
1918                         else
1919                                 /* This can't be cached from previous iteration,
1920                                  * because that entry could be gone now.
1921                                  */
1922                                 prev = list_prev_entry(ipip_entry,
1923                                                        ipip_list_node);
1924                         mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
1925                         ipip_entry = prev;
1926                 }
1927         }
1928         mutex_unlock(&mlxsw_sp->router->lock);
1929
1930         return err;
1931 }
1932
1933 int mlxsw_sp_router_nve_promote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
1934                                       enum mlxsw_sp_l3proto ul_proto,
1935                                       const union mlxsw_sp_l3addr *ul_sip,
1936                                       u32 tunnel_index)
1937 {
1938         enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1939         struct mlxsw_sp_router *router = mlxsw_sp->router;
1940         struct mlxsw_sp_fib_entry *fib_entry;
1941         int err = 0;
1942
1943         mutex_lock(&mlxsw_sp->router->lock);
1944
1945         if (WARN_ON_ONCE(router->nve_decap_config.valid)) {
1946                 err = -EINVAL;
1947                 goto out;
1948         }
1949
1950         router->nve_decap_config.ul_tb_id = ul_tb_id;
1951         router->nve_decap_config.tunnel_index = tunnel_index;
1952         router->nve_decap_config.ul_proto = ul_proto;
1953         router->nve_decap_config.ul_sip = *ul_sip;
1954         router->nve_decap_config.valid = true;
1955
1956         /* It is valid to create a tunnel with a local IP and only later
1957          * assign this IP address to a local interface
1958          */
1959         fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
1960                                                          ul_proto, ul_sip,
1961                                                          type);
1962         if (!fib_entry)
1963                 goto out;
1964
1965         fib_entry->decap.tunnel_index = tunnel_index;
1966         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
1967
1968         err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1969         if (err)
1970                 goto err_fib_entry_update;
1971
1972         goto out;
1973
1974 err_fib_entry_update:
1975         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
1976         mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
1977 out:
1978         mutex_unlock(&mlxsw_sp->router->lock);
1979         return err;
1980 }
1981
1982 void mlxsw_sp_router_nve_demote_decap(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
1983                                       enum mlxsw_sp_l3proto ul_proto,
1984                                       const union mlxsw_sp_l3addr *ul_sip)
1985 {
1986         enum mlxsw_sp_fib_entry_type type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
1987         struct mlxsw_sp_router *router = mlxsw_sp->router;
1988         struct mlxsw_sp_fib_entry *fib_entry;
1989
1990         mutex_lock(&mlxsw_sp->router->lock);
1991
1992         if (WARN_ON_ONCE(!router->nve_decap_config.valid))
1993                 goto out;
1994
1995         router->nve_decap_config.valid = false;
1996
1997         fib_entry = mlxsw_sp_router_ip2me_fib_entry_find(mlxsw_sp, ul_tb_id,
1998                                                          ul_proto, ul_sip,
1999                                                          type);
2000         if (!fib_entry)
2001                 goto out;
2002
2003         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
2004         mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
2005 out:
2006         mutex_unlock(&mlxsw_sp->router->lock);
2007 }
2008
2009 static bool mlxsw_sp_router_nve_is_decap(struct mlxsw_sp *mlxsw_sp,
2010                                          u32 ul_tb_id,
2011                                          enum mlxsw_sp_l3proto ul_proto,
2012                                          const union mlxsw_sp_l3addr *ul_sip)
2013 {
2014         struct mlxsw_sp_router *router = mlxsw_sp->router;
2015
2016         return router->nve_decap_config.valid &&
2017                router->nve_decap_config.ul_tb_id == ul_tb_id &&
2018                router->nve_decap_config.ul_proto == ul_proto &&
2019                !memcmp(&router->nve_decap_config.ul_sip, ul_sip,
2020                        sizeof(*ul_sip));
2021 }
2022
2023 struct mlxsw_sp_neigh_key {
2024         struct neighbour *n;
2025 };
2026
2027 struct mlxsw_sp_neigh_entry {
2028         struct list_head rif_list_node;
2029         struct rhash_head ht_node;
2030         struct mlxsw_sp_neigh_key key;
2031         u16 rif;
2032         bool connected;
2033         unsigned char ha[ETH_ALEN];
2034         struct list_head nexthop_list; /* list of nexthops using
2035                                         * this neigh entry
2036                                         */
2037         struct list_head nexthop_neighs_list_node;
2038         unsigned int counter_index;
2039         bool counter_valid;
2040 };
2041
2042 static const struct rhashtable_params mlxsw_sp_neigh_ht_params = {
2043         .key_offset = offsetof(struct mlxsw_sp_neigh_entry, key),
2044         .head_offset = offsetof(struct mlxsw_sp_neigh_entry, ht_node),
2045         .key_len = sizeof(struct mlxsw_sp_neigh_key),
2046 };
2047
2048 struct mlxsw_sp_neigh_entry *
2049 mlxsw_sp_rif_neigh_next(struct mlxsw_sp_rif *rif,
2050                         struct mlxsw_sp_neigh_entry *neigh_entry)
2051 {
2052         if (!neigh_entry) {
2053                 if (list_empty(&rif->neigh_list))
2054                         return NULL;
2055                 else
2056                         return list_first_entry(&rif->neigh_list,
2057                                                 typeof(*neigh_entry),
2058                                                 rif_list_node);
2059         }
2060         if (list_is_last(&neigh_entry->rif_list_node, &rif->neigh_list))
2061                 return NULL;
2062         return list_next_entry(neigh_entry, rif_list_node);
2063 }
2064
2065 int mlxsw_sp_neigh_entry_type(struct mlxsw_sp_neigh_entry *neigh_entry)
2066 {
2067         return neigh_entry->key.n->tbl->family;
2068 }
2069
2070 unsigned char *
2071 mlxsw_sp_neigh_entry_ha(struct mlxsw_sp_neigh_entry *neigh_entry)
2072 {
2073         return neigh_entry->ha;
2074 }
2075
2076 u32 mlxsw_sp_neigh4_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
2077 {
2078         struct neighbour *n;
2079
2080         n = neigh_entry->key.n;
2081         return ntohl(*((__be32 *) n->primary_key));
2082 }
2083
2084 struct in6_addr *
2085 mlxsw_sp_neigh6_entry_dip(struct mlxsw_sp_neigh_entry *neigh_entry)
2086 {
2087         struct neighbour *n;
2088
2089         n = neigh_entry->key.n;
2090         return (struct in6_addr *) &n->primary_key;
2091 }
2092
2093 int mlxsw_sp_neigh_counter_get(struct mlxsw_sp *mlxsw_sp,
2094                                struct mlxsw_sp_neigh_entry *neigh_entry,
2095                                u64 *p_counter)
2096 {
2097         if (!neigh_entry->counter_valid)
2098                 return -EINVAL;
2099
2100         return mlxsw_sp_flow_counter_get(mlxsw_sp, neigh_entry->counter_index,
2101                                          p_counter, NULL);
2102 }
2103
2104 static struct mlxsw_sp_neigh_entry *
2105 mlxsw_sp_neigh_entry_alloc(struct mlxsw_sp *mlxsw_sp, struct neighbour *n,
2106                            u16 rif)
2107 {
2108         struct mlxsw_sp_neigh_entry *neigh_entry;
2109
2110         neigh_entry = kzalloc(sizeof(*neigh_entry), GFP_KERNEL);
2111         if (!neigh_entry)
2112                 return NULL;
2113
2114         neigh_entry->key.n = n;
2115         neigh_entry->rif = rif;
2116         INIT_LIST_HEAD(&neigh_entry->nexthop_list);
2117
2118         return neigh_entry;
2119 }
2120
2121 static void mlxsw_sp_neigh_entry_free(struct mlxsw_sp_neigh_entry *neigh_entry)
2122 {
2123         kfree(neigh_entry);
2124 }
2125
2126 static int
2127 mlxsw_sp_neigh_entry_insert(struct mlxsw_sp *mlxsw_sp,
2128                             struct mlxsw_sp_neigh_entry *neigh_entry)
2129 {
2130         return rhashtable_insert_fast(&mlxsw_sp->router->neigh_ht,
2131                                       &neigh_entry->ht_node,
2132                                       mlxsw_sp_neigh_ht_params);
2133 }
2134
2135 static void
2136 mlxsw_sp_neigh_entry_remove(struct mlxsw_sp *mlxsw_sp,
2137                             struct mlxsw_sp_neigh_entry *neigh_entry)
2138 {
2139         rhashtable_remove_fast(&mlxsw_sp->router->neigh_ht,
2140                                &neigh_entry->ht_node,
2141                                mlxsw_sp_neigh_ht_params);
2142 }
2143
2144 static bool
2145 mlxsw_sp_neigh_counter_should_alloc(struct mlxsw_sp *mlxsw_sp,
2146                                     struct mlxsw_sp_neigh_entry *neigh_entry)
2147 {
2148         struct devlink *devlink;
2149         const char *table_name;
2150
2151         switch (mlxsw_sp_neigh_entry_type(neigh_entry)) {
2152         case AF_INET:
2153                 table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST4;
2154                 break;
2155         case AF_INET6:
2156                 table_name = MLXSW_SP_DPIPE_TABLE_NAME_HOST6;
2157                 break;
2158         default:
2159                 WARN_ON(1);
2160                 return false;
2161         }
2162
2163         devlink = priv_to_devlink(mlxsw_sp->core);
2164         return devlink_dpipe_table_counter_enabled(devlink, table_name);
2165 }
2166
2167 static void
2168 mlxsw_sp_neigh_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2169                              struct mlxsw_sp_neigh_entry *neigh_entry)
2170 {
2171         if (!mlxsw_sp_neigh_counter_should_alloc(mlxsw_sp, neigh_entry))
2172                 return;
2173
2174         if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &neigh_entry->counter_index))
2175                 return;
2176
2177         neigh_entry->counter_valid = true;
2178 }
2179
2180 static void
2181 mlxsw_sp_neigh_counter_free(struct mlxsw_sp *mlxsw_sp,
2182                             struct mlxsw_sp_neigh_entry *neigh_entry)
2183 {
2184         if (!neigh_entry->counter_valid)
2185                 return;
2186         mlxsw_sp_flow_counter_free(mlxsw_sp,
2187                                    neigh_entry->counter_index);
2188         neigh_entry->counter_valid = false;
2189 }
2190
2191 static struct mlxsw_sp_neigh_entry *
2192 mlxsw_sp_neigh_entry_create(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2193 {
2194         struct mlxsw_sp_neigh_entry *neigh_entry;
2195         struct mlxsw_sp_rif *rif;
2196         int err;
2197
2198         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, n->dev);
2199         if (!rif)
2200                 return ERR_PTR(-EINVAL);
2201
2202         neigh_entry = mlxsw_sp_neigh_entry_alloc(mlxsw_sp, n, rif->rif_index);
2203         if (!neigh_entry)
2204                 return ERR_PTR(-ENOMEM);
2205
2206         err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
2207         if (err)
2208                 goto err_neigh_entry_insert;
2209
2210         mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2211         list_add(&neigh_entry->rif_list_node, &rif->neigh_list);
2212
2213         return neigh_entry;
2214
2215 err_neigh_entry_insert:
2216         mlxsw_sp_neigh_entry_free(neigh_entry);
2217         return ERR_PTR(err);
2218 }
2219
2220 static void
2221 mlxsw_sp_neigh_entry_destroy(struct mlxsw_sp *mlxsw_sp,
2222                              struct mlxsw_sp_neigh_entry *neigh_entry)
2223 {
2224         list_del(&neigh_entry->rif_list_node);
2225         mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2226         mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
2227         mlxsw_sp_neigh_entry_free(neigh_entry);
2228 }
2229
2230 static struct mlxsw_sp_neigh_entry *
2231 mlxsw_sp_neigh_entry_lookup(struct mlxsw_sp *mlxsw_sp, struct neighbour *n)
2232 {
2233         struct mlxsw_sp_neigh_key key;
2234
2235         key.n = n;
2236         return rhashtable_lookup_fast(&mlxsw_sp->router->neigh_ht,
2237                                       &key, mlxsw_sp_neigh_ht_params);
2238 }
2239
2240 static void
2241 mlxsw_sp_router_neighs_update_interval_init(struct mlxsw_sp *mlxsw_sp)
2242 {
2243         unsigned long interval;
2244
2245 #if IS_ENABLED(CONFIG_IPV6)
2246         interval = min_t(unsigned long,
2247                          NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME),
2248                          NEIGH_VAR(&nd_tbl.parms, DELAY_PROBE_TIME));
2249 #else
2250         interval = NEIGH_VAR(&arp_tbl.parms, DELAY_PROBE_TIME);
2251 #endif
2252         mlxsw_sp->router->neighs_update.interval = jiffies_to_msecs(interval);
2253 }
2254
2255 static void mlxsw_sp_router_neigh_ent_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2256                                                    char *rauhtd_pl,
2257                                                    int ent_index)
2258 {
2259         struct net_device *dev;
2260         struct neighbour *n;
2261         __be32 dipn;
2262         u32 dip;
2263         u16 rif;
2264
2265         mlxsw_reg_rauhtd_ent_ipv4_unpack(rauhtd_pl, ent_index, &rif, &dip);
2266
2267         if (!mlxsw_sp->router->rifs[rif]) {
2268                 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2269                 return;
2270         }
2271
2272         dipn = htonl(dip);
2273         dev = mlxsw_sp->router->rifs[rif]->dev;
2274         n = neigh_lookup(&arp_tbl, &dipn, dev);
2275         if (!n)
2276                 return;
2277
2278         netdev_dbg(dev, "Updating neighbour with IP=%pI4h\n", &dip);
2279         neigh_event_send(n, NULL);
2280         neigh_release(n);
2281 }
2282
2283 #if IS_ENABLED(CONFIG_IPV6)
2284 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2285                                                    char *rauhtd_pl,
2286                                                    int rec_index)
2287 {
2288         struct net_device *dev;
2289         struct neighbour *n;
2290         struct in6_addr dip;
2291         u16 rif;
2292
2293         mlxsw_reg_rauhtd_ent_ipv6_unpack(rauhtd_pl, rec_index, &rif,
2294                                          (char *) &dip);
2295
2296         if (!mlxsw_sp->router->rifs[rif]) {
2297                 dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Incorrect RIF in neighbour entry\n");
2298                 return;
2299         }
2300
2301         dev = mlxsw_sp->router->rifs[rif]->dev;
2302         n = neigh_lookup(&nd_tbl, &dip, dev);
2303         if (!n)
2304                 return;
2305
2306         netdev_dbg(dev, "Updating neighbour with IP=%pI6c\n", &dip);
2307         neigh_event_send(n, NULL);
2308         neigh_release(n);
2309 }
2310 #else
2311 static void mlxsw_sp_router_neigh_ent_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2312                                                    char *rauhtd_pl,
2313                                                    int rec_index)
2314 {
2315 }
2316 #endif
2317
2318 static void mlxsw_sp_router_neigh_rec_ipv4_process(struct mlxsw_sp *mlxsw_sp,
2319                                                    char *rauhtd_pl,
2320                                                    int rec_index)
2321 {
2322         u8 num_entries;
2323         int i;
2324
2325         num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2326                                                                 rec_index);
2327         /* Hardware starts counting at 0, so add 1. */
2328         num_entries++;
2329
2330         /* Each record consists of several neighbour entries. */
2331         for (i = 0; i < num_entries; i++) {
2332                 int ent_index;
2333
2334                 ent_index = rec_index * MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC + i;
2335                 mlxsw_sp_router_neigh_ent_ipv4_process(mlxsw_sp, rauhtd_pl,
2336                                                        ent_index);
2337         }
2338
2339 }
2340
2341 static void mlxsw_sp_router_neigh_rec_ipv6_process(struct mlxsw_sp *mlxsw_sp,
2342                                                    char *rauhtd_pl,
2343                                                    int rec_index)
2344 {
2345         /* One record contains one entry. */
2346         mlxsw_sp_router_neigh_ent_ipv6_process(mlxsw_sp, rauhtd_pl,
2347                                                rec_index);
2348 }
2349
2350 static void mlxsw_sp_router_neigh_rec_process(struct mlxsw_sp *mlxsw_sp,
2351                                               char *rauhtd_pl, int rec_index)
2352 {
2353         switch (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, rec_index)) {
2354         case MLXSW_REG_RAUHTD_TYPE_IPV4:
2355                 mlxsw_sp_router_neigh_rec_ipv4_process(mlxsw_sp, rauhtd_pl,
2356                                                        rec_index);
2357                 break;
2358         case MLXSW_REG_RAUHTD_TYPE_IPV6:
2359                 mlxsw_sp_router_neigh_rec_ipv6_process(mlxsw_sp, rauhtd_pl,
2360                                                        rec_index);
2361                 break;
2362         }
2363 }
2364
2365 static bool mlxsw_sp_router_rauhtd_is_full(char *rauhtd_pl)
2366 {
2367         u8 num_rec, last_rec_index, num_entries;
2368
2369         num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2370         last_rec_index = num_rec - 1;
2371
2372         if (num_rec < MLXSW_REG_RAUHTD_REC_MAX_NUM)
2373                 return false;
2374         if (mlxsw_reg_rauhtd_rec_type_get(rauhtd_pl, last_rec_index) ==
2375             MLXSW_REG_RAUHTD_TYPE_IPV6)
2376                 return true;
2377
2378         num_entries = mlxsw_reg_rauhtd_ipv4_rec_num_entries_get(rauhtd_pl,
2379                                                                 last_rec_index);
2380         if (++num_entries == MLXSW_REG_RAUHTD_IPV4_ENT_PER_REC)
2381                 return true;
2382         return false;
2383 }
2384
2385 static int
2386 __mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp,
2387                                        char *rauhtd_pl,
2388                                        enum mlxsw_reg_rauhtd_type type)
2389 {
2390         int i, num_rec;
2391         int err;
2392
2393         /* Ensure the RIF we read from the device does not change mid-dump. */
2394         mutex_lock(&mlxsw_sp->router->lock);
2395         do {
2396                 mlxsw_reg_rauhtd_pack(rauhtd_pl, type);
2397                 err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(rauhtd),
2398                                       rauhtd_pl);
2399                 if (err) {
2400                         dev_err_ratelimited(mlxsw_sp->bus_info->dev, "Failed to dump neighbour table\n");
2401                         break;
2402                 }
2403                 num_rec = mlxsw_reg_rauhtd_num_rec_get(rauhtd_pl);
2404                 for (i = 0; i < num_rec; i++)
2405                         mlxsw_sp_router_neigh_rec_process(mlxsw_sp, rauhtd_pl,
2406                                                           i);
2407         } while (mlxsw_sp_router_rauhtd_is_full(rauhtd_pl));
2408         mutex_unlock(&mlxsw_sp->router->lock);
2409
2410         return err;
2411 }
2412
2413 static int mlxsw_sp_router_neighs_update_rauhtd(struct mlxsw_sp *mlxsw_sp)
2414 {
2415         enum mlxsw_reg_rauhtd_type type;
2416         char *rauhtd_pl;
2417         int err;
2418
2419         rauhtd_pl = kmalloc(MLXSW_REG_RAUHTD_LEN, GFP_KERNEL);
2420         if (!rauhtd_pl)
2421                 return -ENOMEM;
2422
2423         type = MLXSW_REG_RAUHTD_TYPE_IPV4;
2424         err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2425         if (err)
2426                 goto out;
2427
2428         type = MLXSW_REG_RAUHTD_TYPE_IPV6;
2429         err = __mlxsw_sp_router_neighs_update_rauhtd(mlxsw_sp, rauhtd_pl, type);
2430 out:
2431         kfree(rauhtd_pl);
2432         return err;
2433 }
2434
2435 static void mlxsw_sp_router_neighs_update_nh(struct mlxsw_sp *mlxsw_sp)
2436 {
2437         struct mlxsw_sp_neigh_entry *neigh_entry;
2438
2439         mutex_lock(&mlxsw_sp->router->lock);
2440         list_for_each_entry(neigh_entry, &mlxsw_sp->router->nexthop_neighs_list,
2441                             nexthop_neighs_list_node)
2442                 /* If this neigh have nexthops, make the kernel think this neigh
2443                  * is active regardless of the traffic.
2444                  */
2445                 neigh_event_send(neigh_entry->key.n, NULL);
2446         mutex_unlock(&mlxsw_sp->router->lock);
2447 }
2448
2449 static void
2450 mlxsw_sp_router_neighs_update_work_schedule(struct mlxsw_sp *mlxsw_sp)
2451 {
2452         unsigned long interval = mlxsw_sp->router->neighs_update.interval;
2453
2454         mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw,
2455                                msecs_to_jiffies(interval));
2456 }
2457
2458 static void mlxsw_sp_router_neighs_update_work(struct work_struct *work)
2459 {
2460         struct mlxsw_sp_router *router;
2461         int err;
2462
2463         router = container_of(work, struct mlxsw_sp_router,
2464                               neighs_update.dw.work);
2465         err = mlxsw_sp_router_neighs_update_rauhtd(router->mlxsw_sp);
2466         if (err)
2467                 dev_err(router->mlxsw_sp->bus_info->dev, "Could not update kernel for neigh activity");
2468
2469         mlxsw_sp_router_neighs_update_nh(router->mlxsw_sp);
2470
2471         mlxsw_sp_router_neighs_update_work_schedule(router->mlxsw_sp);
2472 }
2473
2474 static void mlxsw_sp_router_probe_unresolved_nexthops(struct work_struct *work)
2475 {
2476         struct mlxsw_sp_neigh_entry *neigh_entry;
2477         struct mlxsw_sp_router *router;
2478
2479         router = container_of(work, struct mlxsw_sp_router,
2480                               nexthop_probe_dw.work);
2481         /* Iterate over nexthop neighbours, find those who are unresolved and
2482          * send arp on them. This solves the chicken-egg problem when
2483          * the nexthop wouldn't get offloaded until the neighbor is resolved
2484          * but it wouldn't get resolved ever in case traffic is flowing in HW
2485          * using different nexthop.
2486          */
2487         mutex_lock(&router->lock);
2488         list_for_each_entry(neigh_entry, &router->nexthop_neighs_list,
2489                             nexthop_neighs_list_node)
2490                 if (!neigh_entry->connected)
2491                         neigh_event_send(neigh_entry->key.n, NULL);
2492         mutex_unlock(&router->lock);
2493
2494         mlxsw_core_schedule_dw(&router->nexthop_probe_dw,
2495                                MLXSW_SP_UNRESOLVED_NH_PROBE_INTERVAL);
2496 }
2497
2498 static void
2499 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
2500                               struct mlxsw_sp_neigh_entry *neigh_entry,
2501                               bool removing, bool dead);
2502
2503 static enum mlxsw_reg_rauht_op mlxsw_sp_rauht_op(bool adding)
2504 {
2505         return adding ? MLXSW_REG_RAUHT_OP_WRITE_ADD :
2506                         MLXSW_REG_RAUHT_OP_WRITE_DELETE;
2507 }
2508
2509 static int
2510 mlxsw_sp_router_neigh_entry_op4(struct mlxsw_sp *mlxsw_sp,
2511                                 struct mlxsw_sp_neigh_entry *neigh_entry,
2512                                 enum mlxsw_reg_rauht_op op)
2513 {
2514         struct neighbour *n = neigh_entry->key.n;
2515         u32 dip = ntohl(*((__be32 *) n->primary_key));
2516         char rauht_pl[MLXSW_REG_RAUHT_LEN];
2517
2518         mlxsw_reg_rauht_pack4(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2519                               dip);
2520         if (neigh_entry->counter_valid)
2521                 mlxsw_reg_rauht_pack_counter(rauht_pl,
2522                                              neigh_entry->counter_index);
2523         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2524 }
2525
2526 static int
2527 mlxsw_sp_router_neigh_entry_op6(struct mlxsw_sp *mlxsw_sp,
2528                                 struct mlxsw_sp_neigh_entry *neigh_entry,
2529                                 enum mlxsw_reg_rauht_op op)
2530 {
2531         struct neighbour *n = neigh_entry->key.n;
2532         char rauht_pl[MLXSW_REG_RAUHT_LEN];
2533         const char *dip = n->primary_key;
2534
2535         mlxsw_reg_rauht_pack6(rauht_pl, op, neigh_entry->rif, neigh_entry->ha,
2536                               dip);
2537         if (neigh_entry->counter_valid)
2538                 mlxsw_reg_rauht_pack_counter(rauht_pl,
2539                                              neigh_entry->counter_index);
2540         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rauht), rauht_pl);
2541 }
2542
2543 bool mlxsw_sp_neigh_ipv6_ignore(struct mlxsw_sp_neigh_entry *neigh_entry)
2544 {
2545         struct neighbour *n = neigh_entry->key.n;
2546
2547         /* Packets with a link-local destination address are trapped
2548          * after LPM lookup and never reach the neighbour table, so
2549          * there is no need to program such neighbours to the device.
2550          */
2551         if (ipv6_addr_type((struct in6_addr *) &n->primary_key) &
2552             IPV6_ADDR_LINKLOCAL)
2553                 return true;
2554         return false;
2555 }
2556
2557 static void
2558 mlxsw_sp_neigh_entry_update(struct mlxsw_sp *mlxsw_sp,
2559                             struct mlxsw_sp_neigh_entry *neigh_entry,
2560                             bool adding)
2561 {
2562         enum mlxsw_reg_rauht_op op = mlxsw_sp_rauht_op(adding);
2563         int err;
2564
2565         if (!adding && !neigh_entry->connected)
2566                 return;
2567         neigh_entry->connected = adding;
2568         if (neigh_entry->key.n->tbl->family == AF_INET) {
2569                 err = mlxsw_sp_router_neigh_entry_op4(mlxsw_sp, neigh_entry,
2570                                                       op);
2571                 if (err)
2572                         return;
2573         } else if (neigh_entry->key.n->tbl->family == AF_INET6) {
2574                 if (mlxsw_sp_neigh_ipv6_ignore(neigh_entry))
2575                         return;
2576                 err = mlxsw_sp_router_neigh_entry_op6(mlxsw_sp, neigh_entry,
2577                                                       op);
2578                 if (err)
2579                         return;
2580         } else {
2581                 WARN_ON_ONCE(1);
2582                 return;
2583         }
2584
2585         if (adding)
2586                 neigh_entry->key.n->flags |= NTF_OFFLOADED;
2587         else
2588                 neigh_entry->key.n->flags &= ~NTF_OFFLOADED;
2589 }
2590
2591 void
2592 mlxsw_sp_neigh_entry_counter_update(struct mlxsw_sp *mlxsw_sp,
2593                                     struct mlxsw_sp_neigh_entry *neigh_entry,
2594                                     bool adding)
2595 {
2596         if (adding)
2597                 mlxsw_sp_neigh_counter_alloc(mlxsw_sp, neigh_entry);
2598         else
2599                 mlxsw_sp_neigh_counter_free(mlxsw_sp, neigh_entry);
2600         mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, true);
2601 }
2602
2603 struct mlxsw_sp_netevent_work {
2604         struct work_struct work;
2605         struct mlxsw_sp *mlxsw_sp;
2606         struct neighbour *n;
2607 };
2608
2609 static void mlxsw_sp_router_neigh_event_work(struct work_struct *work)
2610 {
2611         struct mlxsw_sp_netevent_work *net_work =
2612                 container_of(work, struct mlxsw_sp_netevent_work, work);
2613         struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2614         struct mlxsw_sp_neigh_entry *neigh_entry;
2615         struct neighbour *n = net_work->n;
2616         unsigned char ha[ETH_ALEN];
2617         bool entry_connected;
2618         u8 nud_state, dead;
2619
2620         /* If these parameters are changed after we release the lock,
2621          * then we are guaranteed to receive another event letting us
2622          * know about it.
2623          */
2624         read_lock_bh(&n->lock);
2625         memcpy(ha, n->ha, ETH_ALEN);
2626         nud_state = n->nud_state;
2627         dead = n->dead;
2628         read_unlock_bh(&n->lock);
2629
2630         mutex_lock(&mlxsw_sp->router->lock);
2631         mlxsw_sp_span_respin(mlxsw_sp);
2632
2633         entry_connected = nud_state & NUD_VALID && !dead;
2634         neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
2635         if (!entry_connected && !neigh_entry)
2636                 goto out;
2637         if (!neigh_entry) {
2638                 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
2639                 if (IS_ERR(neigh_entry))
2640                         goto out;
2641         }
2642
2643         memcpy(neigh_entry->ha, ha, ETH_ALEN);
2644         mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, entry_connected);
2645         mlxsw_sp_nexthop_neigh_update(mlxsw_sp, neigh_entry, !entry_connected,
2646                                       dead);
2647
2648         if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
2649                 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2650
2651 out:
2652         mutex_unlock(&mlxsw_sp->router->lock);
2653         neigh_release(n);
2654         kfree(net_work);
2655 }
2656
2657 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp);
2658
2659 static void mlxsw_sp_router_mp_hash_event_work(struct work_struct *work)
2660 {
2661         struct mlxsw_sp_netevent_work *net_work =
2662                 container_of(work, struct mlxsw_sp_netevent_work, work);
2663         struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2664
2665         mlxsw_sp_mp_hash_init(mlxsw_sp);
2666         kfree(net_work);
2667 }
2668
2669 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp);
2670
2671 static void mlxsw_sp_router_update_priority_work(struct work_struct *work)
2672 {
2673         struct mlxsw_sp_netevent_work *net_work =
2674                 container_of(work, struct mlxsw_sp_netevent_work, work);
2675         struct mlxsw_sp *mlxsw_sp = net_work->mlxsw_sp;
2676
2677         __mlxsw_sp_router_init(mlxsw_sp);
2678         kfree(net_work);
2679 }
2680
2681 static int mlxsw_sp_router_schedule_work(struct net *net,
2682                                          struct notifier_block *nb,
2683                                          void (*cb)(struct work_struct *))
2684 {
2685         struct mlxsw_sp_netevent_work *net_work;
2686         struct mlxsw_sp_router *router;
2687
2688         router = container_of(nb, struct mlxsw_sp_router, netevent_nb);
2689         if (!net_eq(net, mlxsw_sp_net(router->mlxsw_sp)))
2690                 return NOTIFY_DONE;
2691
2692         net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2693         if (!net_work)
2694                 return NOTIFY_BAD;
2695
2696         INIT_WORK(&net_work->work, cb);
2697         net_work->mlxsw_sp = router->mlxsw_sp;
2698         mlxsw_core_schedule_work(&net_work->work);
2699         return NOTIFY_DONE;
2700 }
2701
2702 static int mlxsw_sp_router_netevent_event(struct notifier_block *nb,
2703                                           unsigned long event, void *ptr)
2704 {
2705         struct mlxsw_sp_netevent_work *net_work;
2706         struct mlxsw_sp_port *mlxsw_sp_port;
2707         struct mlxsw_sp *mlxsw_sp;
2708         unsigned long interval;
2709         struct neigh_parms *p;
2710         struct neighbour *n;
2711
2712         switch (event) {
2713         case NETEVENT_DELAY_PROBE_TIME_UPDATE:
2714                 p = ptr;
2715
2716                 /* We don't care about changes in the default table. */
2717                 if (!p->dev || (p->tbl->family != AF_INET &&
2718                                 p->tbl->family != AF_INET6))
2719                         return NOTIFY_DONE;
2720
2721                 /* We are in atomic context and can't take RTNL mutex,
2722                  * so use RCU variant to walk the device chain.
2723                  */
2724                 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(p->dev);
2725                 if (!mlxsw_sp_port)
2726                         return NOTIFY_DONE;
2727
2728                 mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2729                 interval = jiffies_to_msecs(NEIGH_VAR(p, DELAY_PROBE_TIME));
2730                 mlxsw_sp->router->neighs_update.interval = interval;
2731
2732                 mlxsw_sp_port_dev_put(mlxsw_sp_port);
2733                 break;
2734         case NETEVENT_NEIGH_UPDATE:
2735                 n = ptr;
2736
2737                 if (n->tbl->family != AF_INET && n->tbl->family != AF_INET6)
2738                         return NOTIFY_DONE;
2739
2740                 mlxsw_sp_port = mlxsw_sp_port_lower_dev_hold(n->dev);
2741                 if (!mlxsw_sp_port)
2742                         return NOTIFY_DONE;
2743
2744                 net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
2745                 if (!net_work) {
2746                         mlxsw_sp_port_dev_put(mlxsw_sp_port);
2747                         return NOTIFY_BAD;
2748                 }
2749
2750                 INIT_WORK(&net_work->work, mlxsw_sp_router_neigh_event_work);
2751                 net_work->mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
2752                 net_work->n = n;
2753
2754                 /* Take a reference to ensure the neighbour won't be
2755                  * destructed until we drop the reference in delayed
2756                  * work.
2757                  */
2758                 neigh_clone(n);
2759                 mlxsw_core_schedule_work(&net_work->work);
2760                 mlxsw_sp_port_dev_put(mlxsw_sp_port);
2761                 break;
2762         case NETEVENT_IPV4_MPATH_HASH_UPDATE:
2763         case NETEVENT_IPV6_MPATH_HASH_UPDATE:
2764                 return mlxsw_sp_router_schedule_work(ptr, nb,
2765                                 mlxsw_sp_router_mp_hash_event_work);
2766
2767         case NETEVENT_IPV4_FWD_UPDATE_PRIORITY_UPDATE:
2768                 return mlxsw_sp_router_schedule_work(ptr, nb,
2769                                 mlxsw_sp_router_update_priority_work);
2770         }
2771
2772         return NOTIFY_DONE;
2773 }
2774
2775 static int mlxsw_sp_neigh_init(struct mlxsw_sp *mlxsw_sp)
2776 {
2777         int err;
2778
2779         err = rhashtable_init(&mlxsw_sp->router->neigh_ht,
2780                               &mlxsw_sp_neigh_ht_params);
2781         if (err)
2782                 return err;
2783
2784         /* Initialize the polling interval according to the default
2785          * table.
2786          */
2787         mlxsw_sp_router_neighs_update_interval_init(mlxsw_sp);
2788
2789         /* Create the delayed works for the activity_update */
2790         INIT_DELAYED_WORK(&mlxsw_sp->router->neighs_update.dw,
2791                           mlxsw_sp_router_neighs_update_work);
2792         INIT_DELAYED_WORK(&mlxsw_sp->router->nexthop_probe_dw,
2793                           mlxsw_sp_router_probe_unresolved_nexthops);
2794         mlxsw_core_schedule_dw(&mlxsw_sp->router->neighs_update.dw, 0);
2795         mlxsw_core_schedule_dw(&mlxsw_sp->router->nexthop_probe_dw, 0);
2796         return 0;
2797 }
2798
2799 static void mlxsw_sp_neigh_fini(struct mlxsw_sp *mlxsw_sp)
2800 {
2801         cancel_delayed_work_sync(&mlxsw_sp->router->neighs_update.dw);
2802         cancel_delayed_work_sync(&mlxsw_sp->router->nexthop_probe_dw);
2803         rhashtable_destroy(&mlxsw_sp->router->neigh_ht);
2804 }
2805
2806 static void mlxsw_sp_neigh_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
2807                                          struct mlxsw_sp_rif *rif)
2808 {
2809         struct mlxsw_sp_neigh_entry *neigh_entry, *tmp;
2810
2811         list_for_each_entry_safe(neigh_entry, tmp, &rif->neigh_list,
2812                                  rif_list_node) {
2813                 mlxsw_sp_neigh_entry_update(mlxsw_sp, neigh_entry, false);
2814                 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
2815         }
2816 }
2817
2818 enum mlxsw_sp_nexthop_type {
2819         MLXSW_SP_NEXTHOP_TYPE_ETH,
2820         MLXSW_SP_NEXTHOP_TYPE_IPIP,
2821 };
2822
2823 struct mlxsw_sp_nexthop_key {
2824         struct fib_nh *fib_nh;
2825 };
2826
2827 struct mlxsw_sp_nexthop {
2828         struct list_head neigh_list_node; /* member of neigh entry list */
2829         struct list_head rif_list_node;
2830         struct list_head router_list_node;
2831         struct mlxsw_sp_nexthop_group_info *nhgi; /* pointer back to the group
2832                                                    * this nexthop belongs to
2833                                                    */
2834         struct rhash_head ht_node;
2835         struct neigh_table *neigh_tbl;
2836         struct mlxsw_sp_nexthop_key key;
2837         unsigned char gw_addr[sizeof(struct in6_addr)];
2838         int ifindex;
2839         int nh_weight;
2840         int norm_nh_weight;
2841         int num_adj_entries;
2842         struct mlxsw_sp_rif *rif;
2843         u8 should_offload:1, /* set indicates this neigh is connected and
2844                               * should be put to KVD linear area of this group.
2845                               */
2846            offloaded:1, /* set in case the neigh is actually put into
2847                          * KVD linear area of this group.
2848                          */
2849            update:1; /* set indicates that MAC of this neigh should be
2850                       * updated in HW
2851                       */
2852         enum mlxsw_sp_nexthop_type type;
2853         union {
2854                 struct mlxsw_sp_neigh_entry *neigh_entry;
2855                 struct mlxsw_sp_ipip_entry *ipip_entry;
2856         };
2857         unsigned int counter_index;
2858         bool counter_valid;
2859 };
2860
2861 enum mlxsw_sp_nexthop_group_type {
2862         MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4,
2863         MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6,
2864 };
2865
2866 struct mlxsw_sp_nexthop_group_info {
2867         struct mlxsw_sp_nexthop_group *nh_grp;
2868         u32 adj_index;
2869         u16 ecmp_size;
2870         u16 count;
2871         int sum_norm_weight;
2872         u8 adj_index_valid:1,
2873            gateway:1; /* routes using the group use a gateway */
2874         struct mlxsw_sp_nexthop nexthops[0];
2875 #define nh_rif  nexthops[0].rif
2876 };
2877
2878 struct mlxsw_sp_nexthop_group {
2879         struct rhash_head ht_node;
2880         struct list_head fib_list; /* list of fib entries that use this group */
2881         union {
2882                 struct {
2883                         struct fib_info *fi;
2884                 } ipv4;
2885         };
2886         struct mlxsw_sp_nexthop_group_info *nhgi;
2887         enum mlxsw_sp_nexthop_group_type type;
2888 };
2889
2890 void mlxsw_sp_nexthop_counter_alloc(struct mlxsw_sp *mlxsw_sp,
2891                                     struct mlxsw_sp_nexthop *nh)
2892 {
2893         struct devlink *devlink;
2894
2895         devlink = priv_to_devlink(mlxsw_sp->core);
2896         if (!devlink_dpipe_table_counter_enabled(devlink,
2897                                                  MLXSW_SP_DPIPE_TABLE_NAME_ADJ))
2898                 return;
2899
2900         if (mlxsw_sp_flow_counter_alloc(mlxsw_sp, &nh->counter_index))
2901                 return;
2902
2903         nh->counter_valid = true;
2904 }
2905
2906 void mlxsw_sp_nexthop_counter_free(struct mlxsw_sp *mlxsw_sp,
2907                                    struct mlxsw_sp_nexthop *nh)
2908 {
2909         if (!nh->counter_valid)
2910                 return;
2911         mlxsw_sp_flow_counter_free(mlxsw_sp, nh->counter_index);
2912         nh->counter_valid = false;
2913 }
2914
2915 int mlxsw_sp_nexthop_counter_get(struct mlxsw_sp *mlxsw_sp,
2916                                  struct mlxsw_sp_nexthop *nh, u64 *p_counter)
2917 {
2918         if (!nh->counter_valid)
2919                 return -EINVAL;
2920
2921         return mlxsw_sp_flow_counter_get(mlxsw_sp, nh->counter_index,
2922                                          p_counter, NULL);
2923 }
2924
2925 struct mlxsw_sp_nexthop *mlxsw_sp_nexthop_next(struct mlxsw_sp_router *router,
2926                                                struct mlxsw_sp_nexthop *nh)
2927 {
2928         if (!nh) {
2929                 if (list_empty(&router->nexthop_list))
2930                         return NULL;
2931                 else
2932                         return list_first_entry(&router->nexthop_list,
2933                                                 typeof(*nh), router_list_node);
2934         }
2935         if (list_is_last(&nh->router_list_node, &router->nexthop_list))
2936                 return NULL;
2937         return list_next_entry(nh, router_list_node);
2938 }
2939
2940 bool mlxsw_sp_nexthop_offload(struct mlxsw_sp_nexthop *nh)
2941 {
2942         return nh->offloaded;
2943 }
2944
2945 unsigned char *mlxsw_sp_nexthop_ha(struct mlxsw_sp_nexthop *nh)
2946 {
2947         if (!nh->offloaded)
2948                 return NULL;
2949         return nh->neigh_entry->ha;
2950 }
2951
2952 int mlxsw_sp_nexthop_indexes(struct mlxsw_sp_nexthop *nh, u32 *p_adj_index,
2953                              u32 *p_adj_size, u32 *p_adj_hash_index)
2954 {
2955         struct mlxsw_sp_nexthop_group_info *nhgi = nh->nhgi;
2956         u32 adj_hash_index = 0;
2957         int i;
2958
2959         if (!nh->offloaded || !nhgi->adj_index_valid)
2960                 return -EINVAL;
2961
2962         *p_adj_index = nhgi->adj_index;
2963         *p_adj_size = nhgi->ecmp_size;
2964
2965         for (i = 0; i < nhgi->count; i++) {
2966                 struct mlxsw_sp_nexthop *nh_iter = &nhgi->nexthops[i];
2967
2968                 if (nh_iter == nh)
2969                         break;
2970                 if (nh_iter->offloaded)
2971                         adj_hash_index += nh_iter->num_adj_entries;
2972         }
2973
2974         *p_adj_hash_index = adj_hash_index;
2975         return 0;
2976 }
2977
2978 struct mlxsw_sp_rif *mlxsw_sp_nexthop_rif(struct mlxsw_sp_nexthop *nh)
2979 {
2980         return nh->rif;
2981 }
2982
2983 bool mlxsw_sp_nexthop_group_has_ipip(struct mlxsw_sp_nexthop *nh)
2984 {
2985         struct mlxsw_sp_nexthop_group_info *nhgi = nh->nhgi;
2986         int i;
2987
2988         for (i = 0; i < nhgi->count; i++) {
2989                 struct mlxsw_sp_nexthop *nh_iter = &nhgi->nexthops[i];
2990
2991                 if (nh_iter->type == MLXSW_SP_NEXTHOP_TYPE_IPIP)
2992                         return true;
2993         }
2994         return false;
2995 }
2996
2997 struct mlxsw_sp_nexthop_group_cmp_arg {
2998         enum mlxsw_sp_nexthop_group_type type;
2999         union {
3000                 struct fib_info *fi;
3001                 struct mlxsw_sp_fib6_entry *fib6_entry;
3002         };
3003 };
3004
3005 static bool
3006 mlxsw_sp_nexthop6_group_has_nexthop(const struct mlxsw_sp_nexthop_group *nh_grp,
3007                                     const struct in6_addr *gw, int ifindex,
3008                                     int weight)
3009 {
3010         int i;
3011
3012         for (i = 0; i < nh_grp->nhgi->count; i++) {
3013                 const struct mlxsw_sp_nexthop *nh;
3014
3015                 nh = &nh_grp->nhgi->nexthops[i];
3016                 if (nh->ifindex == ifindex && nh->nh_weight == weight &&
3017                     ipv6_addr_equal(gw, (struct in6_addr *) nh->gw_addr))
3018                         return true;
3019         }
3020
3021         return false;
3022 }
3023
3024 static bool
3025 mlxsw_sp_nexthop6_group_cmp(const struct mlxsw_sp_nexthop_group *nh_grp,
3026                             const struct mlxsw_sp_fib6_entry *fib6_entry)
3027 {
3028         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3029
3030         if (nh_grp->nhgi->count != fib6_entry->nrt6)
3031                 return false;
3032
3033         list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3034                 struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3035                 struct in6_addr *gw;
3036                 int ifindex, weight;
3037
3038                 ifindex = fib6_nh->fib_nh_dev->ifindex;
3039                 weight = fib6_nh->fib_nh_weight;
3040                 gw = &fib6_nh->fib_nh_gw6;
3041                 if (!mlxsw_sp_nexthop6_group_has_nexthop(nh_grp, gw, ifindex,
3042                                                          weight))
3043                         return false;
3044         }
3045
3046         return true;
3047 }
3048
3049 static int
3050 mlxsw_sp_nexthop_group_cmp(struct rhashtable_compare_arg *arg, const void *ptr)
3051 {
3052         const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = arg->key;
3053         const struct mlxsw_sp_nexthop_group *nh_grp = ptr;
3054
3055         if (nh_grp->type != cmp_arg->type)
3056                 return 1;
3057
3058         switch (cmp_arg->type) {
3059         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3060                 return cmp_arg->fi != nh_grp->ipv4.fi;
3061         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3062                 return !mlxsw_sp_nexthop6_group_cmp(nh_grp,
3063                                                     cmp_arg->fib6_entry);
3064         default:
3065                 WARN_ON(1);
3066                 return 1;
3067         }
3068 }
3069
3070 static u32 mlxsw_sp_nexthop_group_hash_obj(const void *data, u32 len, u32 seed)
3071 {
3072         const struct mlxsw_sp_nexthop_group *nh_grp = data;
3073         const struct mlxsw_sp_nexthop *nh;
3074         struct fib_info *fi;
3075         unsigned int val;
3076         int i;
3077
3078         switch (nh_grp->type) {
3079         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3080                 fi = nh_grp->ipv4.fi;
3081                 return jhash(&fi, sizeof(fi), seed);
3082         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3083                 val = nh_grp->nhgi->count;
3084                 for (i = 0; i < nh_grp->nhgi->count; i++) {
3085                         nh = &nh_grp->nhgi->nexthops[i];
3086                         val ^= jhash(&nh->ifindex, sizeof(nh->ifindex), seed);
3087                         val ^= jhash(&nh->gw_addr, sizeof(nh->gw_addr), seed);
3088                 }
3089                 return jhash(&val, sizeof(val), seed);
3090         default:
3091                 WARN_ON(1);
3092                 return 0;
3093         }
3094 }
3095
3096 static u32
3097 mlxsw_sp_nexthop6_group_hash(struct mlxsw_sp_fib6_entry *fib6_entry, u32 seed)
3098 {
3099         unsigned int val = fib6_entry->nrt6;
3100         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3101
3102         list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3103                 struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3104                 struct net_device *dev = fib6_nh->fib_nh_dev;
3105                 struct in6_addr *gw = &fib6_nh->fib_nh_gw6;
3106
3107                 val ^= jhash(&dev->ifindex, sizeof(dev->ifindex), seed);
3108                 val ^= jhash(gw, sizeof(*gw), seed);
3109         }
3110
3111         return jhash(&val, sizeof(val), seed);
3112 }
3113
3114 static u32
3115 mlxsw_sp_nexthop_group_hash(const void *data, u32 len, u32 seed)
3116 {
3117         const struct mlxsw_sp_nexthop_group_cmp_arg *cmp_arg = data;
3118
3119         switch (cmp_arg->type) {
3120         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3121                 return jhash(&cmp_arg->fi, sizeof(cmp_arg->fi), seed);
3122         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3123                 return mlxsw_sp_nexthop6_group_hash(cmp_arg->fib6_entry, seed);
3124         default:
3125                 WARN_ON(1);
3126                 return 0;
3127         }
3128 }
3129
3130 static const struct rhashtable_params mlxsw_sp_nexthop_group_ht_params = {
3131         .head_offset = offsetof(struct mlxsw_sp_nexthop_group, ht_node),
3132         .hashfn      = mlxsw_sp_nexthop_group_hash,
3133         .obj_hashfn  = mlxsw_sp_nexthop_group_hash_obj,
3134         .obj_cmpfn   = mlxsw_sp_nexthop_group_cmp,
3135 };
3136
3137 static int mlxsw_sp_nexthop_group_insert(struct mlxsw_sp *mlxsw_sp,
3138                                          struct mlxsw_sp_nexthop_group *nh_grp)
3139 {
3140         if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6 &&
3141             !nh_grp->nhgi->gateway)
3142                 return 0;
3143
3144         return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_group_ht,
3145                                       &nh_grp->ht_node,
3146                                       mlxsw_sp_nexthop_group_ht_params);
3147 }
3148
3149 static void mlxsw_sp_nexthop_group_remove(struct mlxsw_sp *mlxsw_sp,
3150                                           struct mlxsw_sp_nexthop_group *nh_grp)
3151 {
3152         if (nh_grp->type == MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6 &&
3153             !nh_grp->nhgi->gateway)
3154                 return;
3155
3156         rhashtable_remove_fast(&mlxsw_sp->router->nexthop_group_ht,
3157                                &nh_grp->ht_node,
3158                                mlxsw_sp_nexthop_group_ht_params);
3159 }
3160
3161 static struct mlxsw_sp_nexthop_group *
3162 mlxsw_sp_nexthop4_group_lookup(struct mlxsw_sp *mlxsw_sp,
3163                                struct fib_info *fi)
3164 {
3165         struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3166
3167         cmp_arg.type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4;
3168         cmp_arg.fi = fi;
3169         return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3170                                       &cmp_arg,
3171                                       mlxsw_sp_nexthop_group_ht_params);
3172 }
3173
3174 static struct mlxsw_sp_nexthop_group *
3175 mlxsw_sp_nexthop6_group_lookup(struct mlxsw_sp *mlxsw_sp,
3176                                struct mlxsw_sp_fib6_entry *fib6_entry)
3177 {
3178         struct mlxsw_sp_nexthop_group_cmp_arg cmp_arg;
3179
3180         cmp_arg.type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6;
3181         cmp_arg.fib6_entry = fib6_entry;
3182         return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_group_ht,
3183                                       &cmp_arg,
3184                                       mlxsw_sp_nexthop_group_ht_params);
3185 }
3186
3187 static const struct rhashtable_params mlxsw_sp_nexthop_ht_params = {
3188         .key_offset = offsetof(struct mlxsw_sp_nexthop, key),
3189         .head_offset = offsetof(struct mlxsw_sp_nexthop, ht_node),
3190         .key_len = sizeof(struct mlxsw_sp_nexthop_key),
3191 };
3192
3193 static int mlxsw_sp_nexthop_insert(struct mlxsw_sp *mlxsw_sp,
3194                                    struct mlxsw_sp_nexthop *nh)
3195 {
3196         return rhashtable_insert_fast(&mlxsw_sp->router->nexthop_ht,
3197                                       &nh->ht_node, mlxsw_sp_nexthop_ht_params);
3198 }
3199
3200 static void mlxsw_sp_nexthop_remove(struct mlxsw_sp *mlxsw_sp,
3201                                     struct mlxsw_sp_nexthop *nh)
3202 {
3203         rhashtable_remove_fast(&mlxsw_sp->router->nexthop_ht, &nh->ht_node,
3204                                mlxsw_sp_nexthop_ht_params);
3205 }
3206
3207 static struct mlxsw_sp_nexthop *
3208 mlxsw_sp_nexthop_lookup(struct mlxsw_sp *mlxsw_sp,
3209                         struct mlxsw_sp_nexthop_key key)
3210 {
3211         return rhashtable_lookup_fast(&mlxsw_sp->router->nexthop_ht, &key,
3212                                       mlxsw_sp_nexthop_ht_params);
3213 }
3214
3215 static int mlxsw_sp_adj_index_mass_update_vr(struct mlxsw_sp *mlxsw_sp,
3216                                              const struct mlxsw_sp_fib *fib,
3217                                              u32 adj_index, u16 ecmp_size,
3218                                              u32 new_adj_index,
3219                                              u16 new_ecmp_size)
3220 {
3221         char raleu_pl[MLXSW_REG_RALEU_LEN];
3222
3223         mlxsw_reg_raleu_pack(raleu_pl,
3224                              (enum mlxsw_reg_ralxx_protocol) fib->proto,
3225                              fib->vr->id, adj_index, ecmp_size, new_adj_index,
3226                              new_ecmp_size);
3227         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(raleu), raleu_pl);
3228 }
3229
3230 static int mlxsw_sp_adj_index_mass_update(struct mlxsw_sp *mlxsw_sp,
3231                                           struct mlxsw_sp_nexthop_group *nh_grp,
3232                                           u32 old_adj_index, u16 old_ecmp_size)
3233 {
3234         struct mlxsw_sp_fib_entry *fib_entry;
3235         struct mlxsw_sp_fib *fib = NULL;
3236         int err;
3237
3238         list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3239                 struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
3240
3241                 if (fib == fib_entry->fib_node->fib)
3242                         continue;
3243                 fib = fib_entry->fib_node->fib;
3244                 err = mlxsw_sp_adj_index_mass_update_vr(mlxsw_sp, fib,
3245                                                         old_adj_index,
3246                                                         old_ecmp_size,
3247                                                         nhgi->adj_index,
3248                                                         nhgi->ecmp_size);
3249                 if (err)
3250                         return err;
3251         }
3252         return 0;
3253 }
3254
3255 static int __mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3256                                      struct mlxsw_sp_nexthop *nh)
3257 {
3258         struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3259         char ratr_pl[MLXSW_REG_RATR_LEN];
3260
3261         mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
3262                             true, MLXSW_REG_RATR_TYPE_ETHERNET,
3263                             adj_index, neigh_entry->rif);
3264         mlxsw_reg_ratr_eth_entry_pack(ratr_pl, neigh_entry->ha);
3265         if (nh->counter_valid)
3266                 mlxsw_reg_ratr_counter_pack(ratr_pl, nh->counter_index, true);
3267         else
3268                 mlxsw_reg_ratr_counter_pack(ratr_pl, 0, false);
3269
3270         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
3271 }
3272
3273 int mlxsw_sp_nexthop_update(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
3274                             struct mlxsw_sp_nexthop *nh)
3275 {
3276         int i;
3277
3278         for (i = 0; i < nh->num_adj_entries; i++) {
3279                 int err;
3280
3281                 err = __mlxsw_sp_nexthop_update(mlxsw_sp, adj_index + i, nh);
3282                 if (err)
3283                         return err;
3284         }
3285
3286         return 0;
3287 }
3288
3289 static int __mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3290                                           u32 adj_index,
3291                                           struct mlxsw_sp_nexthop *nh)
3292 {
3293         const struct mlxsw_sp_ipip_ops *ipip_ops;
3294
3295         ipip_ops = mlxsw_sp->router->ipip_ops_arr[nh->ipip_entry->ipipt];
3296         return ipip_ops->nexthop_update(mlxsw_sp, adj_index, nh->ipip_entry);
3297 }
3298
3299 static int mlxsw_sp_nexthop_ipip_update(struct mlxsw_sp *mlxsw_sp,
3300                                         u32 adj_index,
3301                                         struct mlxsw_sp_nexthop *nh)
3302 {
3303         int i;
3304
3305         for (i = 0; i < nh->num_adj_entries; i++) {
3306                 int err;
3307
3308                 err = __mlxsw_sp_nexthop_ipip_update(mlxsw_sp, adj_index + i,
3309                                                      nh);
3310                 if (err)
3311                         return err;
3312         }
3313
3314         return 0;
3315 }
3316
3317 static int
3318 mlxsw_sp_nexthop_group_update(struct mlxsw_sp *mlxsw_sp,
3319                               struct mlxsw_sp_nexthop_group_info *nhgi,
3320                               bool reallocate)
3321 {
3322         u32 adj_index = nhgi->adj_index; /* base */
3323         struct mlxsw_sp_nexthop *nh;
3324         int i;
3325
3326         for (i = 0; i < nhgi->count; i++) {
3327                 nh = &nhgi->nexthops[i];
3328
3329                 if (!nh->should_offload) {
3330                         nh->offloaded = 0;
3331                         continue;
3332                 }
3333
3334                 if (nh->update || reallocate) {
3335                         int err = 0;
3336
3337                         switch (nh->type) {
3338                         case MLXSW_SP_NEXTHOP_TYPE_ETH:
3339                                 err = mlxsw_sp_nexthop_update
3340                                             (mlxsw_sp, adj_index, nh);
3341                                 break;
3342                         case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3343                                 err = mlxsw_sp_nexthop_ipip_update
3344                                             (mlxsw_sp, adj_index, nh);
3345                                 break;
3346                         }
3347                         if (err)
3348                                 return err;
3349                         nh->update = 0;
3350                         nh->offloaded = 1;
3351                 }
3352                 adj_index += nh->num_adj_entries;
3353         }
3354         return 0;
3355 }
3356
3357 static int
3358 mlxsw_sp_nexthop_fib_entries_update(struct mlxsw_sp *mlxsw_sp,
3359                                     struct mlxsw_sp_nexthop_group *nh_grp)
3360 {
3361         struct mlxsw_sp_fib_entry *fib_entry;
3362         int err;
3363
3364         list_for_each_entry(fib_entry, &nh_grp->fib_list, nexthop_group_node) {
3365                 err = mlxsw_sp_fib_entry_update(mlxsw_sp, fib_entry);
3366                 if (err)
3367                         return err;
3368         }
3369         return 0;
3370 }
3371
3372 static void mlxsw_sp_adj_grp_size_round_up(u16 *p_adj_grp_size)
3373 {
3374         /* Valid sizes for an adjacency group are:
3375          * 1-64, 512, 1024, 2048 and 4096.
3376          */
3377         if (*p_adj_grp_size <= 64)
3378                 return;
3379         else if (*p_adj_grp_size <= 512)
3380                 *p_adj_grp_size = 512;
3381         else if (*p_adj_grp_size <= 1024)
3382                 *p_adj_grp_size = 1024;
3383         else if (*p_adj_grp_size <= 2048)
3384                 *p_adj_grp_size = 2048;
3385         else
3386                 *p_adj_grp_size = 4096;
3387 }
3388
3389 static void mlxsw_sp_adj_grp_size_round_down(u16 *p_adj_grp_size,
3390                                              unsigned int alloc_size)
3391 {
3392         if (alloc_size >= 4096)
3393                 *p_adj_grp_size = 4096;
3394         else if (alloc_size >= 2048)
3395                 *p_adj_grp_size = 2048;
3396         else if (alloc_size >= 1024)
3397                 *p_adj_grp_size = 1024;
3398         else if (alloc_size >= 512)
3399                 *p_adj_grp_size = 512;
3400 }
3401
3402 static int mlxsw_sp_fix_adj_grp_size(struct mlxsw_sp *mlxsw_sp,
3403                                      u16 *p_adj_grp_size)
3404 {
3405         unsigned int alloc_size;
3406         int err;
3407
3408         /* Round up the requested group size to the next size supported
3409          * by the device and make sure the request can be satisfied.
3410          */
3411         mlxsw_sp_adj_grp_size_round_up(p_adj_grp_size);
3412         err = mlxsw_sp_kvdl_alloc_count_query(mlxsw_sp,
3413                                               MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3414                                               *p_adj_grp_size, &alloc_size);
3415         if (err)
3416                 return err;
3417         /* It is possible the allocation results in more allocated
3418          * entries than requested. Try to use as much of them as
3419          * possible.
3420          */
3421         mlxsw_sp_adj_grp_size_round_down(p_adj_grp_size, alloc_size);
3422
3423         return 0;
3424 }
3425
3426 static void
3427 mlxsw_sp_nexthop_group_normalize(struct mlxsw_sp_nexthop_group_info *nhgi)
3428 {
3429         int i, g = 0, sum_norm_weight = 0;
3430         struct mlxsw_sp_nexthop *nh;
3431
3432         for (i = 0; i < nhgi->count; i++) {
3433                 nh = &nhgi->nexthops[i];
3434
3435                 if (!nh->should_offload)
3436                         continue;
3437                 if (g > 0)
3438                         g = gcd(nh->nh_weight, g);
3439                 else
3440                         g = nh->nh_weight;
3441         }
3442
3443         for (i = 0; i < nhgi->count; i++) {
3444                 nh = &nhgi->nexthops[i];
3445
3446                 if (!nh->should_offload)
3447                         continue;
3448                 nh->norm_nh_weight = nh->nh_weight / g;
3449                 sum_norm_weight += nh->norm_nh_weight;
3450         }
3451
3452         nhgi->sum_norm_weight = sum_norm_weight;
3453 }
3454
3455 static void
3456 mlxsw_sp_nexthop_group_rebalance(struct mlxsw_sp_nexthop_group_info *nhgi)
3457 {
3458         int i, weight = 0, lower_bound = 0;
3459         int total = nhgi->sum_norm_weight;
3460         u16 ecmp_size = nhgi->ecmp_size;
3461
3462         for (i = 0; i < nhgi->count; i++) {
3463                 struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
3464                 int upper_bound;
3465
3466                 if (!nh->should_offload)
3467                         continue;
3468                 weight += nh->norm_nh_weight;
3469                 upper_bound = DIV_ROUND_CLOSEST(ecmp_size * weight, total);
3470                 nh->num_adj_entries = upper_bound - lower_bound;
3471                 lower_bound = upper_bound;
3472         }
3473 }
3474
3475 static struct mlxsw_sp_nexthop *
3476 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
3477                      const struct mlxsw_sp_rt6 *mlxsw_sp_rt6);
3478
3479 static void
3480 mlxsw_sp_nexthop4_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3481                                         struct mlxsw_sp_nexthop_group *nh_grp)
3482 {
3483         int i;
3484
3485         for (i = 0; i < nh_grp->nhgi->count; i++) {
3486                 struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
3487
3488                 if (nh->offloaded)
3489                         nh->key.fib_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3490                 else
3491                         nh->key.fib_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3492         }
3493 }
3494
3495 static void
3496 __mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp_nexthop_group *nh_grp,
3497                                           struct mlxsw_sp_fib6_entry *fib6_entry)
3498 {
3499         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
3500
3501         list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
3502                 struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
3503                 struct mlxsw_sp_nexthop *nh;
3504
3505                 nh = mlxsw_sp_rt6_nexthop(nh_grp, mlxsw_sp_rt6);
3506                 if (nh && nh->offloaded)
3507                         fib6_nh->fib_nh_flags |= RTNH_F_OFFLOAD;
3508                 else
3509                         fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
3510         }
3511 }
3512
3513 static void
3514 mlxsw_sp_nexthop6_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3515                                         struct mlxsw_sp_nexthop_group *nh_grp)
3516 {
3517         struct mlxsw_sp_fib6_entry *fib6_entry;
3518
3519         /* Unfortunately, in IPv6 the route and the nexthop are described by
3520          * the same struct, so we need to iterate over all the routes using the
3521          * nexthop group and set / clear the offload indication for them.
3522          */
3523         list_for_each_entry(fib6_entry, &nh_grp->fib_list,
3524                             common.nexthop_group_node)
3525                 __mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
3526 }
3527
3528 static void
3529 mlxsw_sp_nexthop_group_offload_refresh(struct mlxsw_sp *mlxsw_sp,
3530                                        struct mlxsw_sp_nexthop_group *nh_grp)
3531 {
3532         switch (nh_grp->type) {
3533         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4:
3534                 mlxsw_sp_nexthop4_group_offload_refresh(mlxsw_sp, nh_grp);
3535                 break;
3536         case MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6:
3537                 mlxsw_sp_nexthop6_group_offload_refresh(mlxsw_sp, nh_grp);
3538                 break;
3539         }
3540 }
3541
3542 static void
3543 mlxsw_sp_nexthop_group_refresh(struct mlxsw_sp *mlxsw_sp,
3544                                struct mlxsw_sp_nexthop_group *nh_grp)
3545 {
3546         struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
3547         u16 ecmp_size, old_ecmp_size;
3548         struct mlxsw_sp_nexthop *nh;
3549         bool offload_change = false;
3550         u32 adj_index;
3551         bool old_adj_index_valid;
3552         u32 old_adj_index;
3553         int i;
3554         int err;
3555
3556         if (!nhgi->gateway) {
3557                 mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3558                 return;
3559         }
3560
3561         for (i = 0; i < nhgi->count; i++) {
3562                 nh = &nhgi->nexthops[i];
3563
3564                 if (nh->should_offload != nh->offloaded) {
3565                         offload_change = true;
3566                         if (nh->should_offload)
3567                                 nh->update = 1;
3568                 }
3569         }
3570         if (!offload_change) {
3571                 /* Nothing was added or removed, so no need to reallocate. Just
3572                  * update MAC on existing adjacency indexes.
3573                  */
3574                 err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nhgi, false);
3575                 if (err) {
3576                         dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3577                         goto set_trap;
3578                 }
3579                 return;
3580         }
3581         mlxsw_sp_nexthop_group_normalize(nhgi);
3582         if (!nhgi->sum_norm_weight)
3583                 /* No neigh of this group is connected so we just set
3584                  * the trap and let everthing flow through kernel.
3585                  */
3586                 goto set_trap;
3587
3588         ecmp_size = nhgi->sum_norm_weight;
3589         err = mlxsw_sp_fix_adj_grp_size(mlxsw_sp, &ecmp_size);
3590         if (err)
3591                 /* No valid allocation size available. */
3592                 goto set_trap;
3593
3594         err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3595                                   ecmp_size, &adj_index);
3596         if (err) {
3597                 /* We ran out of KVD linear space, just set the
3598                  * trap and let everything flow through kernel.
3599                  */
3600                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to allocate KVD linear area for nexthop group.\n");
3601                 goto set_trap;
3602         }
3603         old_adj_index_valid = nhgi->adj_index_valid;
3604         old_adj_index = nhgi->adj_index;
3605         old_ecmp_size = nhgi->ecmp_size;
3606         nhgi->adj_index_valid = 1;
3607         nhgi->adj_index = adj_index;
3608         nhgi->ecmp_size = ecmp_size;
3609         mlxsw_sp_nexthop_group_rebalance(nhgi);
3610         err = mlxsw_sp_nexthop_group_update(mlxsw_sp, nhgi, true);
3611         if (err) {
3612                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to update neigh MAC in adjacency table.\n");
3613                 goto set_trap;
3614         }
3615
3616         mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3617
3618         if (!old_adj_index_valid) {
3619                 /* The trap was set for fib entries, so we have to call
3620                  * fib entry update to unset it and use adjacency index.
3621                  */
3622                 err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3623                 if (err) {
3624                         dev_warn(mlxsw_sp->bus_info->dev, "Failed to add adjacency index to fib entries.\n");
3625                         goto set_trap;
3626                 }
3627                 return;
3628         }
3629
3630         err = mlxsw_sp_adj_index_mass_update(mlxsw_sp, nh_grp,
3631                                              old_adj_index, old_ecmp_size);
3632         mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3633                            old_ecmp_size, old_adj_index);
3634         if (err) {
3635                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to mass-update adjacency index for nexthop group.\n");
3636                 goto set_trap;
3637         }
3638
3639         return;
3640
3641 set_trap:
3642         old_adj_index_valid = nhgi->adj_index_valid;
3643         nhgi->adj_index_valid = 0;
3644         for (i = 0; i < nhgi->count; i++) {
3645                 nh = &nhgi->nexthops[i];
3646                 nh->offloaded = 0;
3647         }
3648         err = mlxsw_sp_nexthop_fib_entries_update(mlxsw_sp, nh_grp);
3649         if (err)
3650                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set traps for fib entries.\n");
3651         mlxsw_sp_nexthop_group_offload_refresh(mlxsw_sp, nh_grp);
3652         if (old_adj_index_valid)
3653                 mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ,
3654                                    nhgi->ecmp_size, nhgi->adj_index);
3655 }
3656
3657 static void __mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp_nexthop *nh,
3658                                             bool removing)
3659 {
3660         if (!removing)
3661                 nh->should_offload = 1;
3662         else
3663                 nh->should_offload = 0;
3664         nh->update = 1;
3665 }
3666
3667 static int
3668 mlxsw_sp_nexthop_dead_neigh_replace(struct mlxsw_sp *mlxsw_sp,
3669                                     struct mlxsw_sp_neigh_entry *neigh_entry)
3670 {
3671         struct neighbour *n, *old_n = neigh_entry->key.n;
3672         struct mlxsw_sp_nexthop *nh;
3673         bool entry_connected;
3674         u8 nud_state, dead;
3675         int err;
3676
3677         nh = list_first_entry(&neigh_entry->nexthop_list,
3678                               struct mlxsw_sp_nexthop, neigh_list_node);
3679
3680         n = neigh_lookup(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3681         if (!n) {
3682                 n = neigh_create(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3683                 if (IS_ERR(n))
3684                         return PTR_ERR(n);
3685                 neigh_event_send(n, NULL);
3686         }
3687
3688         mlxsw_sp_neigh_entry_remove(mlxsw_sp, neigh_entry);
3689         neigh_entry->key.n = n;
3690         err = mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
3691         if (err)
3692                 goto err_neigh_entry_insert;
3693
3694         read_lock_bh(&n->lock);
3695         nud_state = n->nud_state;
3696         dead = n->dead;
3697         read_unlock_bh(&n->lock);
3698         entry_connected = nud_state & NUD_VALID && !dead;
3699
3700         list_for_each_entry(nh, &neigh_entry->nexthop_list,
3701                             neigh_list_node) {
3702                 neigh_release(old_n);
3703                 neigh_clone(n);
3704                 __mlxsw_sp_nexthop_neigh_update(nh, !entry_connected);
3705                 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
3706         }
3707
3708         neigh_release(n);
3709
3710         return 0;
3711
3712 err_neigh_entry_insert:
3713         neigh_entry->key.n = old_n;
3714         mlxsw_sp_neigh_entry_insert(mlxsw_sp, neigh_entry);
3715         neigh_release(n);
3716         return err;
3717 }
3718
3719 static void
3720 mlxsw_sp_nexthop_neigh_update(struct mlxsw_sp *mlxsw_sp,
3721                               struct mlxsw_sp_neigh_entry *neigh_entry,
3722                               bool removing, bool dead)
3723 {
3724         struct mlxsw_sp_nexthop *nh;
3725
3726         if (list_empty(&neigh_entry->nexthop_list))
3727                 return;
3728
3729         if (dead) {
3730                 int err;
3731
3732                 err = mlxsw_sp_nexthop_dead_neigh_replace(mlxsw_sp,
3733                                                           neigh_entry);
3734                 if (err)
3735                         dev_err(mlxsw_sp->bus_info->dev, "Failed to replace dead neigh\n");
3736                 return;
3737         }
3738
3739         list_for_each_entry(nh, &neigh_entry->nexthop_list,
3740                             neigh_list_node) {
3741                 __mlxsw_sp_nexthop_neigh_update(nh, removing);
3742                 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
3743         }
3744 }
3745
3746 static void mlxsw_sp_nexthop_rif_init(struct mlxsw_sp_nexthop *nh,
3747                                       struct mlxsw_sp_rif *rif)
3748 {
3749         if (nh->rif)
3750                 return;
3751
3752         nh->rif = rif;
3753         list_add(&nh->rif_list_node, &rif->nexthop_list);
3754 }
3755
3756 static void mlxsw_sp_nexthop_rif_fini(struct mlxsw_sp_nexthop *nh)
3757 {
3758         if (!nh->rif)
3759                 return;
3760
3761         list_del(&nh->rif_list_node);
3762         nh->rif = NULL;
3763 }
3764
3765 static int mlxsw_sp_nexthop_neigh_init(struct mlxsw_sp *mlxsw_sp,
3766                                        struct mlxsw_sp_nexthop *nh)
3767 {
3768         struct mlxsw_sp_neigh_entry *neigh_entry;
3769         struct neighbour *n;
3770         u8 nud_state, dead;
3771         int err;
3772
3773         if (!nh->nhgi->gateway || nh->neigh_entry)
3774                 return 0;
3775
3776         /* Take a reference of neigh here ensuring that neigh would
3777          * not be destructed before the nexthop entry is finished.
3778          * The reference is taken either in neigh_lookup() or
3779          * in neigh_create() in case n is not found.
3780          */
3781         n = neigh_lookup(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3782         if (!n) {
3783                 n = neigh_create(nh->neigh_tbl, &nh->gw_addr, nh->rif->dev);
3784                 if (IS_ERR(n))
3785                         return PTR_ERR(n);
3786                 neigh_event_send(n, NULL);
3787         }
3788         neigh_entry = mlxsw_sp_neigh_entry_lookup(mlxsw_sp, n);
3789         if (!neigh_entry) {
3790                 neigh_entry = mlxsw_sp_neigh_entry_create(mlxsw_sp, n);
3791                 if (IS_ERR(neigh_entry)) {
3792                         err = -EINVAL;
3793                         goto err_neigh_entry_create;
3794                 }
3795         }
3796
3797         /* If that is the first nexthop connected to that neigh, add to
3798          * nexthop_neighs_list
3799          */
3800         if (list_empty(&neigh_entry->nexthop_list))
3801                 list_add_tail(&neigh_entry->nexthop_neighs_list_node,
3802                               &mlxsw_sp->router->nexthop_neighs_list);
3803
3804         nh->neigh_entry = neigh_entry;
3805         list_add_tail(&nh->neigh_list_node, &neigh_entry->nexthop_list);
3806         read_lock_bh(&n->lock);
3807         nud_state = n->nud_state;
3808         dead = n->dead;
3809         read_unlock_bh(&n->lock);
3810         __mlxsw_sp_nexthop_neigh_update(nh, !(nud_state & NUD_VALID && !dead));
3811
3812         return 0;
3813
3814 err_neigh_entry_create:
3815         neigh_release(n);
3816         return err;
3817 }
3818
3819 static void mlxsw_sp_nexthop_neigh_fini(struct mlxsw_sp *mlxsw_sp,
3820                                         struct mlxsw_sp_nexthop *nh)
3821 {
3822         struct mlxsw_sp_neigh_entry *neigh_entry = nh->neigh_entry;
3823         struct neighbour *n;
3824
3825         if (!neigh_entry)
3826                 return;
3827         n = neigh_entry->key.n;
3828
3829         __mlxsw_sp_nexthop_neigh_update(nh, true);
3830         list_del(&nh->neigh_list_node);
3831         nh->neigh_entry = NULL;
3832
3833         /* If that is the last nexthop connected to that neigh, remove from
3834          * nexthop_neighs_list
3835          */
3836         if (list_empty(&neigh_entry->nexthop_list))
3837                 list_del(&neigh_entry->nexthop_neighs_list_node);
3838
3839         if (!neigh_entry->connected && list_empty(&neigh_entry->nexthop_list))
3840                 mlxsw_sp_neigh_entry_destroy(mlxsw_sp, neigh_entry);
3841
3842         neigh_release(n);
3843 }
3844
3845 static bool mlxsw_sp_ipip_netdev_ul_up(struct net_device *ol_dev)
3846 {
3847         struct net_device *ul_dev;
3848         bool is_up;
3849
3850         rcu_read_lock();
3851         ul_dev = __mlxsw_sp_ipip_netdev_ul_dev_get(ol_dev);
3852         is_up = ul_dev ? (ul_dev->flags & IFF_UP) : true;
3853         rcu_read_unlock();
3854
3855         return is_up;
3856 }
3857
3858 static void mlxsw_sp_nexthop_ipip_init(struct mlxsw_sp *mlxsw_sp,
3859                                        struct mlxsw_sp_nexthop *nh,
3860                                        struct mlxsw_sp_ipip_entry *ipip_entry)
3861 {
3862         bool removing;
3863
3864         if (!nh->nhgi->gateway || nh->ipip_entry)
3865                 return;
3866
3867         nh->ipip_entry = ipip_entry;
3868         removing = !mlxsw_sp_ipip_netdev_ul_up(ipip_entry->ol_dev);
3869         __mlxsw_sp_nexthop_neigh_update(nh, removing);
3870         mlxsw_sp_nexthop_rif_init(nh, &ipip_entry->ol_lb->common);
3871 }
3872
3873 static void mlxsw_sp_nexthop_ipip_fini(struct mlxsw_sp *mlxsw_sp,
3874                                        struct mlxsw_sp_nexthop *nh)
3875 {
3876         struct mlxsw_sp_ipip_entry *ipip_entry = nh->ipip_entry;
3877
3878         if (!ipip_entry)
3879                 return;
3880
3881         __mlxsw_sp_nexthop_neigh_update(nh, true);
3882         nh->ipip_entry = NULL;
3883 }
3884
3885 static bool mlxsw_sp_nexthop4_ipip_type(const struct mlxsw_sp *mlxsw_sp,
3886                                         const struct fib_nh *fib_nh,
3887                                         enum mlxsw_sp_ipip_type *p_ipipt)
3888 {
3889         struct net_device *dev = fib_nh->fib_nh_dev;
3890
3891         return dev &&
3892                fib_nh->nh_parent->fib_type == RTN_UNICAST &&
3893                mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, p_ipipt);
3894 }
3895
3896 static int mlxsw_sp_nexthop_type_init(struct mlxsw_sp *mlxsw_sp,
3897                                       struct mlxsw_sp_nexthop *nh,
3898                                       const struct net_device *dev)
3899 {
3900         const struct mlxsw_sp_ipip_ops *ipip_ops;
3901         struct mlxsw_sp_ipip_entry *ipip_entry;
3902         struct mlxsw_sp_rif *rif;
3903         int err;
3904
3905         ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
3906         if (ipip_entry) {
3907                 ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
3908                 if (ipip_ops->can_offload(mlxsw_sp, dev)) {
3909                         nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
3910                         mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
3911                         return 0;
3912                 }
3913         }
3914
3915         nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
3916         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
3917         if (!rif)
3918                 return 0;
3919
3920         mlxsw_sp_nexthop_rif_init(nh, rif);
3921         err = mlxsw_sp_nexthop_neigh_init(mlxsw_sp, nh);
3922         if (err)
3923                 goto err_neigh_init;
3924
3925         return 0;
3926
3927 err_neigh_init:
3928         mlxsw_sp_nexthop_rif_fini(nh);
3929         return err;
3930 }
3931
3932 static void mlxsw_sp_nexthop_type_fini(struct mlxsw_sp *mlxsw_sp,
3933                                        struct mlxsw_sp_nexthop *nh)
3934 {
3935         switch (nh->type) {
3936         case MLXSW_SP_NEXTHOP_TYPE_ETH:
3937                 mlxsw_sp_nexthop_neigh_fini(mlxsw_sp, nh);
3938                 mlxsw_sp_nexthop_rif_fini(nh);
3939                 break;
3940         case MLXSW_SP_NEXTHOP_TYPE_IPIP:
3941                 mlxsw_sp_nexthop_rif_fini(nh);
3942                 mlxsw_sp_nexthop_ipip_fini(mlxsw_sp, nh);
3943                 break;
3944         }
3945 }
3946
3947 static int mlxsw_sp_nexthop4_init(struct mlxsw_sp *mlxsw_sp,
3948                                   struct mlxsw_sp_nexthop_group *nh_grp,
3949                                   struct mlxsw_sp_nexthop *nh,
3950                                   struct fib_nh *fib_nh)
3951 {
3952         struct net_device *dev = fib_nh->fib_nh_dev;
3953         struct in_device *in_dev;
3954         int err;
3955
3956         nh->nhgi = nh_grp->nhgi;
3957         nh->key.fib_nh = fib_nh;
3958 #ifdef CONFIG_IP_ROUTE_MULTIPATH
3959         nh->nh_weight = fib_nh->fib_nh_weight;
3960 #else
3961         nh->nh_weight = 1;
3962 #endif
3963         memcpy(&nh->gw_addr, &fib_nh->fib_nh_gw4, sizeof(fib_nh->fib_nh_gw4));
3964         nh->neigh_tbl = &arp_tbl;
3965         err = mlxsw_sp_nexthop_insert(mlxsw_sp, nh);
3966         if (err)
3967                 return err;
3968
3969         mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
3970         list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
3971
3972         if (!dev)
3973                 return 0;
3974         nh->ifindex = dev->ifindex;
3975
3976         rcu_read_lock();
3977         in_dev = __in_dev_get_rcu(dev);
3978         if (in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
3979             fib_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
3980                 rcu_read_unlock();
3981                 return 0;
3982         }
3983         rcu_read_unlock();
3984
3985         err = mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, dev);
3986         if (err)
3987                 goto err_nexthop_neigh_init;
3988
3989         return 0;
3990
3991 err_nexthop_neigh_init:
3992         mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
3993         return err;
3994 }
3995
3996 static void mlxsw_sp_nexthop4_fini(struct mlxsw_sp *mlxsw_sp,
3997                                    struct mlxsw_sp_nexthop *nh)
3998 {
3999         mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4000         list_del(&nh->router_list_node);
4001         mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
4002         mlxsw_sp_nexthop_remove(mlxsw_sp, nh);
4003 }
4004
4005 static void mlxsw_sp_nexthop4_event(struct mlxsw_sp *mlxsw_sp,
4006                                     unsigned long event, struct fib_nh *fib_nh)
4007 {
4008         struct mlxsw_sp_nexthop_key key;
4009         struct mlxsw_sp_nexthop *nh;
4010
4011         if (mlxsw_sp->router->aborted)
4012                 return;
4013
4014         key.fib_nh = fib_nh;
4015         nh = mlxsw_sp_nexthop_lookup(mlxsw_sp, key);
4016         if (!nh)
4017                 return;
4018
4019         switch (event) {
4020         case FIB_EVENT_NH_ADD:
4021                 mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, fib_nh->fib_nh_dev);
4022                 break;
4023         case FIB_EVENT_NH_DEL:
4024                 mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4025                 break;
4026         }
4027
4028         mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4029 }
4030
4031 static void mlxsw_sp_nexthop_rif_update(struct mlxsw_sp *mlxsw_sp,
4032                                         struct mlxsw_sp_rif *rif)
4033 {
4034         struct mlxsw_sp_nexthop *nh;
4035         bool removing;
4036
4037         list_for_each_entry(nh, &rif->nexthop_list, rif_list_node) {
4038                 switch (nh->type) {
4039                 case MLXSW_SP_NEXTHOP_TYPE_ETH:
4040                         removing = false;
4041                         break;
4042                 case MLXSW_SP_NEXTHOP_TYPE_IPIP:
4043                         removing = !mlxsw_sp_ipip_netdev_ul_up(rif->dev);
4044                         break;
4045                 default:
4046                         WARN_ON(1);
4047                         continue;
4048                 }
4049
4050                 __mlxsw_sp_nexthop_neigh_update(nh, removing);
4051                 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4052         }
4053 }
4054
4055 static void mlxsw_sp_nexthop_rif_migrate(struct mlxsw_sp *mlxsw_sp,
4056                                          struct mlxsw_sp_rif *old_rif,
4057                                          struct mlxsw_sp_rif *new_rif)
4058 {
4059         struct mlxsw_sp_nexthop *nh;
4060
4061         list_splice_init(&old_rif->nexthop_list, &new_rif->nexthop_list);
4062         list_for_each_entry(nh, &new_rif->nexthop_list, rif_list_node)
4063                 nh->rif = new_rif;
4064         mlxsw_sp_nexthop_rif_update(mlxsw_sp, new_rif);
4065 }
4066
4067 static void mlxsw_sp_nexthop_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
4068                                            struct mlxsw_sp_rif *rif)
4069 {
4070         struct mlxsw_sp_nexthop *nh, *tmp;
4071
4072         list_for_each_entry_safe(nh, tmp, &rif->nexthop_list, rif_list_node) {
4073                 mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
4074                 mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh->nhgi->nh_grp);
4075         }
4076 }
4077
4078 static bool mlxsw_sp_fi_is_gateway(const struct mlxsw_sp *mlxsw_sp,
4079                                    struct fib_info *fi)
4080 {
4081         const struct fib_nh *nh = fib_info_nh(fi, 0);
4082
4083         return nh->fib_nh_scope == RT_SCOPE_LINK ||
4084                mlxsw_sp_nexthop4_ipip_type(mlxsw_sp, nh, NULL);
4085 }
4086
4087 static int
4088 mlxsw_sp_nexthop4_group_info_init(struct mlxsw_sp *mlxsw_sp,
4089                                   struct mlxsw_sp_nexthop_group *nh_grp)
4090 {
4091         unsigned int nhs = fib_info_num_path(nh_grp->ipv4.fi);
4092         struct mlxsw_sp_nexthop_group_info *nhgi;
4093         struct mlxsw_sp_nexthop *nh;
4094         int err, i;
4095
4096         nhgi = kzalloc(struct_size(nhgi, nexthops, nhs), GFP_KERNEL);
4097         if (!nhgi)
4098                 return -ENOMEM;
4099         nh_grp->nhgi = nhgi;
4100         nhgi->nh_grp = nh_grp;
4101         nhgi->gateway = mlxsw_sp_fi_is_gateway(mlxsw_sp, nh_grp->ipv4.fi);
4102         nhgi->count = nhs;
4103         for (i = 0; i < nhgi->count; i++) {
4104                 struct fib_nh *fib_nh;
4105
4106                 nh = &nhgi->nexthops[i];
4107                 fib_nh = fib_info_nh(nh_grp->ipv4.fi, i);
4108                 err = mlxsw_sp_nexthop4_init(mlxsw_sp, nh_grp, nh, fib_nh);
4109                 if (err)
4110                         goto err_nexthop4_init;
4111         }
4112         mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4113
4114         return 0;
4115
4116 err_nexthop4_init:
4117         for (i--; i >= 0; i--) {
4118                 nh = &nhgi->nexthops[i];
4119                 mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
4120         }
4121         kfree(nhgi);
4122         return err;
4123 }
4124
4125 static void
4126 mlxsw_sp_nexthop4_group_info_fini(struct mlxsw_sp *mlxsw_sp,
4127                                   struct mlxsw_sp_nexthop_group *nh_grp)
4128 {
4129         struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
4130         int i;
4131
4132         for (i = nhgi->count - 1; i >= 0; i--) {
4133                 struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
4134
4135                 mlxsw_sp_nexthop4_fini(mlxsw_sp, nh);
4136         }
4137         mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
4138         WARN_ON_ONCE(nhgi->adj_index_valid);
4139         kfree(nhgi);
4140 }
4141
4142 static struct mlxsw_sp_nexthop_group *
4143 mlxsw_sp_nexthop4_group_create(struct mlxsw_sp *mlxsw_sp, struct fib_info *fi)
4144 {
4145         struct mlxsw_sp_nexthop_group *nh_grp;
4146         int err;
4147
4148         nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
4149         if (!nh_grp)
4150                 return ERR_PTR(-ENOMEM);
4151         INIT_LIST_HEAD(&nh_grp->fib_list);
4152         nh_grp->type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV4;
4153         nh_grp->ipv4.fi = fi;
4154         fib_info_hold(fi);
4155
4156         err = mlxsw_sp_nexthop4_group_info_init(mlxsw_sp, nh_grp);
4157         if (err)
4158                 goto err_nexthop_group_info_init;
4159
4160         err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
4161         if (err)
4162                 goto err_nexthop_group_insert;
4163
4164         return nh_grp;
4165
4166 err_nexthop_group_insert:
4167         mlxsw_sp_nexthop4_group_info_fini(mlxsw_sp, nh_grp);
4168 err_nexthop_group_info_init:
4169         fib_info_put(fi);
4170         kfree(nh_grp);
4171         return ERR_PTR(err);
4172 }
4173
4174 static void
4175 mlxsw_sp_nexthop4_group_destroy(struct mlxsw_sp *mlxsw_sp,
4176                                 struct mlxsw_sp_nexthop_group *nh_grp)
4177 {
4178         mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
4179         mlxsw_sp_nexthop4_group_info_fini(mlxsw_sp, nh_grp);
4180         fib_info_put(nh_grp->ipv4.fi);
4181         kfree(nh_grp);
4182 }
4183
4184 static int mlxsw_sp_nexthop4_group_get(struct mlxsw_sp *mlxsw_sp,
4185                                        struct mlxsw_sp_fib_entry *fib_entry,
4186                                        struct fib_info *fi)
4187 {
4188         struct mlxsw_sp_nexthop_group *nh_grp;
4189
4190         nh_grp = mlxsw_sp_nexthop4_group_lookup(mlxsw_sp, fi);
4191         if (!nh_grp) {
4192                 nh_grp = mlxsw_sp_nexthop4_group_create(mlxsw_sp, fi);
4193                 if (IS_ERR(nh_grp))
4194                         return PTR_ERR(nh_grp);
4195         }
4196         list_add_tail(&fib_entry->nexthop_group_node, &nh_grp->fib_list);
4197         fib_entry->nh_group = nh_grp;
4198         return 0;
4199 }
4200
4201 static void mlxsw_sp_nexthop4_group_put(struct mlxsw_sp *mlxsw_sp,
4202                                         struct mlxsw_sp_fib_entry *fib_entry)
4203 {
4204         struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
4205
4206         list_del(&fib_entry->nexthop_group_node);
4207         if (!list_empty(&nh_grp->fib_list))
4208                 return;
4209         mlxsw_sp_nexthop4_group_destroy(mlxsw_sp, nh_grp);
4210 }
4211
4212 static bool
4213 mlxsw_sp_fib4_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
4214 {
4215         struct mlxsw_sp_fib4_entry *fib4_entry;
4216
4217         fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4218                                   common);
4219         return !fib4_entry->tos;
4220 }
4221
4222 static bool
4223 mlxsw_sp_fib_entry_should_offload(const struct mlxsw_sp_fib_entry *fib_entry)
4224 {
4225         struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
4226
4227         switch (fib_entry->fib_node->fib->proto) {
4228         case MLXSW_SP_L3_PROTO_IPV4:
4229                 if (!mlxsw_sp_fib4_entry_should_offload(fib_entry))
4230                         return false;
4231                 break;
4232         case MLXSW_SP_L3_PROTO_IPV6:
4233                 break;
4234         }
4235
4236         switch (fib_entry->type) {
4237         case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
4238                 return !!nh_group->nhgi->adj_index_valid;
4239         case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
4240                 return !!nh_group->nhgi->nh_rif;
4241         case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
4242         case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4243         case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
4244                 return true;
4245         default:
4246                 return false;
4247         }
4248 }
4249
4250 static struct mlxsw_sp_nexthop *
4251 mlxsw_sp_rt6_nexthop(struct mlxsw_sp_nexthop_group *nh_grp,
4252                      const struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
4253 {
4254         int i;
4255
4256         for (i = 0; i < nh_grp->nhgi->count; i++) {
4257                 struct mlxsw_sp_nexthop *nh = &nh_grp->nhgi->nexthops[i];
4258                 struct fib6_info *rt = mlxsw_sp_rt6->rt;
4259
4260                 if (nh->rif && nh->rif->dev == rt->fib6_nh->fib_nh_dev &&
4261                     ipv6_addr_equal((const struct in6_addr *) &nh->gw_addr,
4262                                     &rt->fib6_nh->fib_nh_gw6))
4263                         return nh;
4264                 continue;
4265         }
4266
4267         return NULL;
4268 }
4269
4270 static void
4271 mlxsw_sp_fib4_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4272                                  struct mlxsw_sp_fib_entry *fib_entry)
4273 {
4274         u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
4275         int dst_len = fib_entry->fib_node->key.prefix_len;
4276         struct mlxsw_sp_fib4_entry *fib4_entry;
4277         struct fib_rt_info fri;
4278         bool should_offload;
4279
4280         should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
4281         fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4282                                   common);
4283         fri.fi = fib4_entry->fi;
4284         fri.tb_id = fib4_entry->tb_id;
4285         fri.dst = cpu_to_be32(*p_dst);
4286         fri.dst_len = dst_len;
4287         fri.tos = fib4_entry->tos;
4288         fri.type = fib4_entry->type;
4289         fri.offload = should_offload;
4290         fri.trap = !should_offload;
4291         fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
4292 }
4293
4294 static void
4295 mlxsw_sp_fib4_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4296                                    struct mlxsw_sp_fib_entry *fib_entry)
4297 {
4298         u32 *p_dst = (u32 *) fib_entry->fib_node->key.addr;
4299         int dst_len = fib_entry->fib_node->key.prefix_len;
4300         struct mlxsw_sp_fib4_entry *fib4_entry;
4301         struct fib_rt_info fri;
4302
4303         fib4_entry = container_of(fib_entry, struct mlxsw_sp_fib4_entry,
4304                                   common);
4305         fri.fi = fib4_entry->fi;
4306         fri.tb_id = fib4_entry->tb_id;
4307         fri.dst = cpu_to_be32(*p_dst);
4308         fri.dst_len = dst_len;
4309         fri.tos = fib4_entry->tos;
4310         fri.type = fib4_entry->type;
4311         fri.offload = false;
4312         fri.trap = false;
4313         fib_alias_hw_flags_set(mlxsw_sp_net(mlxsw_sp), &fri);
4314 }
4315
4316 static void
4317 mlxsw_sp_fib6_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4318                                  struct mlxsw_sp_fib_entry *fib_entry)
4319 {
4320         struct mlxsw_sp_fib6_entry *fib6_entry;
4321         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4322         bool should_offload;
4323
4324         should_offload = mlxsw_sp_fib_entry_should_offload(fib_entry);
4325
4326         /* In IPv6 a multipath route is represented using multiple routes, so
4327          * we need to set the flags on all of them.
4328          */
4329         fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
4330                                   common);
4331         list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
4332                 fib6_info_hw_flags_set(mlxsw_sp_rt6->rt, should_offload,
4333                                        !should_offload);
4334 }
4335
4336 static void
4337 mlxsw_sp_fib6_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4338                                    struct mlxsw_sp_fib_entry *fib_entry)
4339 {
4340         struct mlxsw_sp_fib6_entry *fib6_entry;
4341         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
4342
4343         fib6_entry = container_of(fib_entry, struct mlxsw_sp_fib6_entry,
4344                                   common);
4345         list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list)
4346                 fib6_info_hw_flags_set(mlxsw_sp_rt6->rt, false, false);
4347 }
4348
4349 static void
4350 mlxsw_sp_fib_entry_hw_flags_set(struct mlxsw_sp *mlxsw_sp,
4351                                 struct mlxsw_sp_fib_entry *fib_entry)
4352 {
4353         switch (fib_entry->fib_node->fib->proto) {
4354         case MLXSW_SP_L3_PROTO_IPV4:
4355                 mlxsw_sp_fib4_entry_hw_flags_set(mlxsw_sp, fib_entry);
4356                 break;
4357         case MLXSW_SP_L3_PROTO_IPV6:
4358                 mlxsw_sp_fib6_entry_hw_flags_set(mlxsw_sp, fib_entry);
4359                 break;
4360         }
4361 }
4362
4363 static void
4364 mlxsw_sp_fib_entry_hw_flags_clear(struct mlxsw_sp *mlxsw_sp,
4365                                   struct mlxsw_sp_fib_entry *fib_entry)
4366 {
4367         switch (fib_entry->fib_node->fib->proto) {
4368         case MLXSW_SP_L3_PROTO_IPV4:
4369                 mlxsw_sp_fib4_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4370                 break;
4371         case MLXSW_SP_L3_PROTO_IPV6:
4372                 mlxsw_sp_fib6_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4373                 break;
4374         }
4375 }
4376
4377 static void
4378 mlxsw_sp_fib_entry_hw_flags_refresh(struct mlxsw_sp *mlxsw_sp,
4379                                     struct mlxsw_sp_fib_entry *fib_entry,
4380                                     enum mlxsw_sp_fib_entry_op op)
4381 {
4382         switch (op) {
4383         case MLXSW_SP_FIB_ENTRY_OP_WRITE:
4384         case MLXSW_SP_FIB_ENTRY_OP_UPDATE:
4385                 mlxsw_sp_fib_entry_hw_flags_set(mlxsw_sp, fib_entry);
4386                 break;
4387         case MLXSW_SP_FIB_ENTRY_OP_DELETE:
4388                 mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, fib_entry);
4389                 break;
4390         default:
4391                 break;
4392         }
4393 }
4394
4395 struct mlxsw_sp_fib_entry_op_ctx_basic {
4396         char ralue_pl[MLXSW_REG_RALUE_LEN];
4397 };
4398
4399 static void
4400 mlxsw_sp_router_ll_basic_fib_entry_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4401                                         enum mlxsw_sp_l3proto proto,
4402                                         enum mlxsw_sp_fib_entry_op op,
4403                                         u16 virtual_router, u8 prefix_len,
4404                                         unsigned char *addr,
4405                                         struct mlxsw_sp_fib_entry_priv *priv)
4406 {
4407         struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4408         enum mlxsw_reg_ralxx_protocol ralxx_proto;
4409         char *ralue_pl = op_ctx_basic->ralue_pl;
4410         enum mlxsw_reg_ralue_op ralue_op;
4411
4412         ralxx_proto = (enum mlxsw_reg_ralxx_protocol) proto;
4413
4414         switch (op) {
4415         case MLXSW_SP_FIB_ENTRY_OP_WRITE:
4416         case MLXSW_SP_FIB_ENTRY_OP_UPDATE:
4417                 ralue_op = MLXSW_REG_RALUE_OP_WRITE_WRITE;
4418                 break;
4419         case MLXSW_SP_FIB_ENTRY_OP_DELETE:
4420                 ralue_op = MLXSW_REG_RALUE_OP_WRITE_DELETE;
4421                 break;
4422         default:
4423                 WARN_ON_ONCE(1);
4424                 return;
4425         }
4426
4427         switch (proto) {
4428         case MLXSW_SP_L3_PROTO_IPV4:
4429                 mlxsw_reg_ralue_pack4(ralue_pl, ralxx_proto, ralue_op,
4430                                       virtual_router, prefix_len, (u32 *) addr);
4431                 break;
4432         case MLXSW_SP_L3_PROTO_IPV6:
4433                 mlxsw_reg_ralue_pack6(ralue_pl, ralxx_proto, ralue_op,
4434                                       virtual_router, prefix_len, addr);
4435                 break;
4436         }
4437 }
4438
4439 static void
4440 mlxsw_sp_router_ll_basic_fib_entry_act_remote_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4441                                                    enum mlxsw_reg_ralue_trap_action trap_action,
4442                                                    u16 trap_id, u32 adjacency_index, u16 ecmp_size)
4443 {
4444         struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4445
4446         mlxsw_reg_ralue_act_remote_pack(op_ctx_basic->ralue_pl, trap_action,
4447                                         trap_id, adjacency_index, ecmp_size);
4448 }
4449
4450 static void
4451 mlxsw_sp_router_ll_basic_fib_entry_act_local_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4452                                                   enum mlxsw_reg_ralue_trap_action trap_action,
4453                                                   u16 trap_id, u16 local_erif)
4454 {
4455         struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4456
4457         mlxsw_reg_ralue_act_local_pack(op_ctx_basic->ralue_pl, trap_action,
4458                                        trap_id, local_erif);
4459 }
4460
4461 static void
4462 mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx)
4463 {
4464         struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4465
4466         mlxsw_reg_ralue_act_ip2me_pack(op_ctx_basic->ralue_pl);
4467 }
4468
4469 static void
4470 mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_tun_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4471                                                       u32 tunnel_ptr)
4472 {
4473         struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4474
4475         mlxsw_reg_ralue_act_ip2me_tun_pack(op_ctx_basic->ralue_pl, tunnel_ptr);
4476 }
4477
4478 static int
4479 mlxsw_sp_router_ll_basic_fib_entry_commit(struct mlxsw_sp *mlxsw_sp,
4480                                           struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4481                                           bool *postponed_for_bulk)
4482 {
4483         struct mlxsw_sp_fib_entry_op_ctx_basic *op_ctx_basic = (void *) op_ctx->ll_priv;
4484
4485         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue),
4486                                op_ctx_basic->ralue_pl);
4487 }
4488
4489 static bool
4490 mlxsw_sp_router_ll_basic_fib_entry_is_committed(struct mlxsw_sp_fib_entry_priv *priv)
4491 {
4492         return true;
4493 }
4494
4495 static void mlxsw_sp_fib_entry_pack(struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4496                                     struct mlxsw_sp_fib_entry *fib_entry,
4497                                     enum mlxsw_sp_fib_entry_op op)
4498 {
4499         struct mlxsw_sp_fib *fib = fib_entry->fib_node->fib;
4500
4501         mlxsw_sp_fib_entry_op_ctx_priv_hold(op_ctx, fib_entry->priv);
4502         fib->ll_ops->fib_entry_pack(op_ctx, fib->proto, op, fib->vr->id,
4503                                     fib_entry->fib_node->key.prefix_len,
4504                                     fib_entry->fib_node->key.addr,
4505                                     fib_entry->priv);
4506 }
4507
4508 int mlxsw_sp_fib_entry_commit(struct mlxsw_sp *mlxsw_sp,
4509                               struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4510                               const struct mlxsw_sp_router_ll_ops *ll_ops)
4511 {
4512         bool postponed_for_bulk = false;
4513         int err;
4514
4515         err = ll_ops->fib_entry_commit(mlxsw_sp, op_ctx, &postponed_for_bulk);
4516         if (!postponed_for_bulk)
4517                 mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
4518         return err;
4519 }
4520
4521 static int mlxsw_sp_adj_discard_write(struct mlxsw_sp *mlxsw_sp, u16 rif_index)
4522 {
4523         enum mlxsw_reg_ratr_trap_action trap_action;
4524         char ratr_pl[MLXSW_REG_RATR_LEN];
4525         int err;
4526
4527         if (mlxsw_sp->router->adj_discard_index_valid)
4528                 return 0;
4529
4530         err = mlxsw_sp_kvdl_alloc(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4531                                   &mlxsw_sp->router->adj_discard_index);
4532         if (err)
4533                 return err;
4534
4535         trap_action = MLXSW_REG_RATR_TRAP_ACTION_DISCARD_ERRORS;
4536         mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY, true,
4537                             MLXSW_REG_RATR_TYPE_ETHERNET,
4538                             mlxsw_sp->router->adj_discard_index, rif_index);
4539         mlxsw_reg_ratr_trap_action_set(ratr_pl, trap_action);
4540         err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
4541         if (err)
4542                 goto err_ratr_write;
4543
4544         mlxsw_sp->router->adj_discard_index_valid = true;
4545
4546         return 0;
4547
4548 err_ratr_write:
4549         mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
4550                            mlxsw_sp->router->adj_discard_index);
4551         return err;
4552 }
4553
4554 static int mlxsw_sp_fib_entry_op_remote(struct mlxsw_sp *mlxsw_sp,
4555                                         struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4556                                         struct mlxsw_sp_fib_entry *fib_entry,
4557                                         enum mlxsw_sp_fib_entry_op op)
4558 {
4559         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4560         struct mlxsw_sp_nexthop_group *nh_group = fib_entry->nh_group;
4561         struct mlxsw_sp_nexthop_group_info *nhgi = nh_group->nhgi;
4562         enum mlxsw_reg_ralue_trap_action trap_action;
4563         u16 trap_id = 0;
4564         u32 adjacency_index = 0;
4565         u16 ecmp_size = 0;
4566         int err;
4567
4568         /* In case the nexthop group adjacency index is valid, use it
4569          * with provided ECMP size. Otherwise, setup trap and pass
4570          * traffic to kernel.
4571          */
4572         if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
4573                 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4574                 adjacency_index = nhgi->adj_index;
4575                 ecmp_size = nhgi->ecmp_size;
4576         } else if (!nhgi->adj_index_valid && nhgi->count && nhgi->nh_rif) {
4577                 err = mlxsw_sp_adj_discard_write(mlxsw_sp,
4578                                                  nhgi->nh_rif->rif_index);
4579                 if (err)
4580                         return err;
4581                 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4582                 adjacency_index = mlxsw_sp->router->adj_discard_index;
4583                 ecmp_size = 1;
4584         } else {
4585                 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4586                 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
4587         }
4588
4589         mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4590         ll_ops->fib_entry_act_remote_pack(op_ctx, trap_action, trap_id,
4591                                           adjacency_index, ecmp_size);
4592         return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4593 }
4594
4595 static int mlxsw_sp_fib_entry_op_local(struct mlxsw_sp *mlxsw_sp,
4596                                        struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4597                                        struct mlxsw_sp_fib_entry *fib_entry,
4598                                        enum mlxsw_sp_fib_entry_op op)
4599 {
4600         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4601         struct mlxsw_sp_rif *rif = fib_entry->nh_group->nhgi->nh_rif;
4602         enum mlxsw_reg_ralue_trap_action trap_action;
4603         u16 trap_id = 0;
4604         u16 rif_index = 0;
4605
4606         if (mlxsw_sp_fib_entry_should_offload(fib_entry)) {
4607                 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_NOP;
4608                 rif_index = rif->rif_index;
4609         } else {
4610                 trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4611                 trap_id = MLXSW_TRAP_ID_RTR_INGRESS0;
4612         }
4613
4614         mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4615         ll_ops->fib_entry_act_local_pack(op_ctx, trap_action, trap_id, rif_index);
4616         return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4617 }
4618
4619 static int mlxsw_sp_fib_entry_op_trap(struct mlxsw_sp *mlxsw_sp,
4620                                       struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4621                                       struct mlxsw_sp_fib_entry *fib_entry,
4622                                       enum mlxsw_sp_fib_entry_op op)
4623 {
4624         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4625
4626         mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4627         ll_ops->fib_entry_act_ip2me_pack(op_ctx);
4628         return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4629 }
4630
4631 static int mlxsw_sp_fib_entry_op_blackhole(struct mlxsw_sp *mlxsw_sp,
4632                                            struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4633                                            struct mlxsw_sp_fib_entry *fib_entry,
4634                                            enum mlxsw_sp_fib_entry_op op)
4635 {
4636         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4637         enum mlxsw_reg_ralue_trap_action trap_action;
4638
4639         trap_action = MLXSW_REG_RALUE_TRAP_ACTION_DISCARD_ERROR;
4640         mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4641         ll_ops->fib_entry_act_local_pack(op_ctx, trap_action, 0, 0);
4642         return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4643 }
4644
4645 static int
4646 mlxsw_sp_fib_entry_op_unreachable(struct mlxsw_sp *mlxsw_sp,
4647                                   struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4648                                   struct mlxsw_sp_fib_entry *fib_entry,
4649                                   enum mlxsw_sp_fib_entry_op op)
4650 {
4651         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4652         enum mlxsw_reg_ralue_trap_action trap_action;
4653         u16 trap_id;
4654
4655         trap_action = MLXSW_REG_RALUE_TRAP_ACTION_TRAP;
4656         trap_id = MLXSW_TRAP_ID_RTR_INGRESS1;
4657
4658         mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4659         ll_ops->fib_entry_act_local_pack(op_ctx, trap_action, trap_id, 0);
4660         return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4661 }
4662
4663 static int
4664 mlxsw_sp_fib_entry_op_ipip_decap(struct mlxsw_sp *mlxsw_sp,
4665                                  struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4666                                  struct mlxsw_sp_fib_entry *fib_entry,
4667                                  enum mlxsw_sp_fib_entry_op op)
4668 {
4669         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4670         struct mlxsw_sp_ipip_entry *ipip_entry = fib_entry->decap.ipip_entry;
4671         const struct mlxsw_sp_ipip_ops *ipip_ops;
4672
4673         if (WARN_ON(!ipip_entry))
4674                 return -EINVAL;
4675
4676         ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
4677         return ipip_ops->fib_entry_op(mlxsw_sp, ll_ops, op_ctx, ipip_entry, op,
4678                                       fib_entry->decap.tunnel_index, fib_entry->priv);
4679 }
4680
4681 static int mlxsw_sp_fib_entry_op_nve_decap(struct mlxsw_sp *mlxsw_sp,
4682                                            struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4683                                            struct mlxsw_sp_fib_entry *fib_entry,
4684                                            enum mlxsw_sp_fib_entry_op op)
4685 {
4686         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4687
4688         mlxsw_sp_fib_entry_pack(op_ctx, fib_entry, op);
4689         ll_ops->fib_entry_act_ip2me_tun_pack(op_ctx,
4690                                              fib_entry->decap.tunnel_index);
4691         return mlxsw_sp_fib_entry_commit(mlxsw_sp, op_ctx, ll_ops);
4692 }
4693
4694 static int __mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
4695                                    struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4696                                    struct mlxsw_sp_fib_entry *fib_entry,
4697                                    enum mlxsw_sp_fib_entry_op op)
4698 {
4699         switch (fib_entry->type) {
4700         case MLXSW_SP_FIB_ENTRY_TYPE_REMOTE:
4701                 return mlxsw_sp_fib_entry_op_remote(mlxsw_sp, op_ctx, fib_entry, op);
4702         case MLXSW_SP_FIB_ENTRY_TYPE_LOCAL:
4703                 return mlxsw_sp_fib_entry_op_local(mlxsw_sp, op_ctx, fib_entry, op);
4704         case MLXSW_SP_FIB_ENTRY_TYPE_TRAP:
4705                 return mlxsw_sp_fib_entry_op_trap(mlxsw_sp, op_ctx, fib_entry, op);
4706         case MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE:
4707                 return mlxsw_sp_fib_entry_op_blackhole(mlxsw_sp, op_ctx, fib_entry, op);
4708         case MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE:
4709                 return mlxsw_sp_fib_entry_op_unreachable(mlxsw_sp, op_ctx, fib_entry, op);
4710         case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4711                 return mlxsw_sp_fib_entry_op_ipip_decap(mlxsw_sp, op_ctx, fib_entry, op);
4712         case MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP:
4713                 return mlxsw_sp_fib_entry_op_nve_decap(mlxsw_sp, op_ctx, fib_entry, op);
4714         }
4715         return -EINVAL;
4716 }
4717
4718 static int mlxsw_sp_fib_entry_op(struct mlxsw_sp *mlxsw_sp,
4719                                  struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4720                                  struct mlxsw_sp_fib_entry *fib_entry,
4721                                  enum mlxsw_sp_fib_entry_op op)
4722 {
4723         int err = __mlxsw_sp_fib_entry_op(mlxsw_sp, op_ctx, fib_entry, op);
4724
4725         if (err)
4726                 return err;
4727
4728         mlxsw_sp_fib_entry_hw_flags_refresh(mlxsw_sp, fib_entry, op);
4729
4730         return err;
4731 }
4732
4733 static int __mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
4734                                        struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4735                                        struct mlxsw_sp_fib_entry *fib_entry,
4736                                        bool is_new)
4737 {
4738         return mlxsw_sp_fib_entry_op(mlxsw_sp, op_ctx, fib_entry,
4739                                      is_new ? MLXSW_SP_FIB_ENTRY_OP_WRITE :
4740                                               MLXSW_SP_FIB_ENTRY_OP_UPDATE);
4741 }
4742
4743 static int mlxsw_sp_fib_entry_update(struct mlxsw_sp *mlxsw_sp,
4744                                      struct mlxsw_sp_fib_entry *fib_entry)
4745 {
4746         struct mlxsw_sp_fib_entry_op_ctx *op_ctx = mlxsw_sp->router->ll_op_ctx;
4747
4748         mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
4749         return __mlxsw_sp_fib_entry_update(mlxsw_sp, op_ctx, fib_entry, false);
4750 }
4751
4752 static int mlxsw_sp_fib_entry_del(struct mlxsw_sp *mlxsw_sp,
4753                                   struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
4754                                   struct mlxsw_sp_fib_entry *fib_entry)
4755 {
4756         const struct mlxsw_sp_router_ll_ops *ll_ops = fib_entry->fib_node->fib->ll_ops;
4757
4758         if (!ll_ops->fib_entry_is_committed(fib_entry->priv))
4759                 return 0;
4760         return mlxsw_sp_fib_entry_op(mlxsw_sp, op_ctx, fib_entry,
4761                                      MLXSW_SP_FIB_ENTRY_OP_DELETE);
4762 }
4763
4764 static int
4765 mlxsw_sp_fib4_entry_type_set(struct mlxsw_sp *mlxsw_sp,
4766                              const struct fib_entry_notifier_info *fen_info,
4767                              struct mlxsw_sp_fib_entry *fib_entry)
4768 {
4769         struct net_device *dev = fib_info_nh(fen_info->fi, 0)->fib_nh_dev;
4770         union mlxsw_sp_l3addr dip = { .addr4 = htonl(fen_info->dst) };
4771         struct mlxsw_sp_router *router = mlxsw_sp->router;
4772         u32 tb_id = mlxsw_sp_fix_tb_id(fen_info->tb_id);
4773         struct mlxsw_sp_ipip_entry *ipip_entry;
4774         struct fib_info *fi = fen_info->fi;
4775
4776         switch (fen_info->type) {
4777         case RTN_LOCAL:
4778                 ipip_entry = mlxsw_sp_ipip_entry_find_by_decap(mlxsw_sp, dev,
4779                                                  MLXSW_SP_L3_PROTO_IPV4, dip);
4780                 if (ipip_entry && ipip_entry->ol_dev->flags & IFF_UP) {
4781                         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP;
4782                         return mlxsw_sp_fib_entry_decap_init(mlxsw_sp,
4783                                                              fib_entry,
4784                                                              ipip_entry);
4785                 }
4786                 if (mlxsw_sp_router_nve_is_decap(mlxsw_sp, tb_id,
4787                                                  MLXSW_SP_L3_PROTO_IPV4,
4788                                                  &dip)) {
4789                         u32 tunnel_index;
4790
4791                         tunnel_index = router->nve_decap_config.tunnel_index;
4792                         fib_entry->decap.tunnel_index = tunnel_index;
4793                         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_NVE_DECAP;
4794                         return 0;
4795                 }
4796                 fallthrough;
4797         case RTN_BROADCAST:
4798                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
4799                 return 0;
4800         case RTN_BLACKHOLE:
4801                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
4802                 return 0;
4803         case RTN_UNREACHABLE:
4804         case RTN_PROHIBIT:
4805                 /* Packets hitting these routes need to be trapped, but
4806                  * can do so with a lower priority than packets directed
4807                  * at the host, so use action type local instead of trap.
4808                  */
4809                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
4810                 return 0;
4811         case RTN_UNICAST:
4812                 if (mlxsw_sp_fi_is_gateway(mlxsw_sp, fi))
4813                         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
4814                 else
4815                         fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
4816                 return 0;
4817         default:
4818                 return -EINVAL;
4819         }
4820 }
4821
4822 static void
4823 mlxsw_sp_fib4_entry_type_unset(struct mlxsw_sp *mlxsw_sp,
4824                                struct mlxsw_sp_fib_entry *fib_entry)
4825 {
4826         switch (fib_entry->type) {
4827         case MLXSW_SP_FIB_ENTRY_TYPE_IPIP_DECAP:
4828                 mlxsw_sp_fib_entry_decap_fini(mlxsw_sp, fib_entry);
4829                 break;
4830         default:
4831                 break;
4832         }
4833 }
4834
4835 static struct mlxsw_sp_fib4_entry *
4836 mlxsw_sp_fib4_entry_create(struct mlxsw_sp *mlxsw_sp,
4837                            struct mlxsw_sp_fib_node *fib_node,
4838                            const struct fib_entry_notifier_info *fen_info)
4839 {
4840         struct mlxsw_sp_fib4_entry *fib4_entry;
4841         struct mlxsw_sp_fib_entry *fib_entry;
4842         int err;
4843
4844         fib4_entry = kzalloc(sizeof(*fib4_entry), GFP_KERNEL);
4845         if (!fib4_entry)
4846                 return ERR_PTR(-ENOMEM);
4847         fib_entry = &fib4_entry->common;
4848
4849         fib_entry->priv = mlxsw_sp_fib_entry_priv_create(fib_node->fib->ll_ops);
4850         if (IS_ERR(fib_entry->priv)) {
4851                 err = PTR_ERR(fib_entry->priv);
4852                 goto err_fib_entry_priv_create;
4853         }
4854
4855         err = mlxsw_sp_fib4_entry_type_set(mlxsw_sp, fen_info, fib_entry);
4856         if (err)
4857                 goto err_fib4_entry_type_set;
4858
4859         err = mlxsw_sp_nexthop4_group_get(mlxsw_sp, fib_entry, fen_info->fi);
4860         if (err)
4861                 goto err_nexthop4_group_get;
4862
4863         fib4_entry->fi = fen_info->fi;
4864         fib_info_hold(fib4_entry->fi);
4865         fib4_entry->tb_id = fen_info->tb_id;
4866         fib4_entry->type = fen_info->type;
4867         fib4_entry->tos = fen_info->tos;
4868
4869         fib_entry->fib_node = fib_node;
4870
4871         return fib4_entry;
4872
4873 err_nexthop4_group_get:
4874         mlxsw_sp_fib4_entry_type_unset(mlxsw_sp, fib_entry);
4875 err_fib4_entry_type_set:
4876         mlxsw_sp_fib_entry_priv_put(fib_entry->priv);
4877 err_fib_entry_priv_create:
4878         kfree(fib4_entry);
4879         return ERR_PTR(err);
4880 }
4881
4882 static void mlxsw_sp_fib4_entry_destroy(struct mlxsw_sp *mlxsw_sp,
4883                                         struct mlxsw_sp_fib4_entry *fib4_entry)
4884 {
4885         fib_info_put(fib4_entry->fi);
4886         mlxsw_sp_nexthop4_group_put(mlxsw_sp, &fib4_entry->common);
4887         mlxsw_sp_fib4_entry_type_unset(mlxsw_sp, &fib4_entry->common);
4888         mlxsw_sp_fib_entry_priv_put(fib4_entry->common.priv);
4889         kfree(fib4_entry);
4890 }
4891
4892 static struct mlxsw_sp_fib4_entry *
4893 mlxsw_sp_fib4_entry_lookup(struct mlxsw_sp *mlxsw_sp,
4894                            const struct fib_entry_notifier_info *fen_info)
4895 {
4896         struct mlxsw_sp_fib4_entry *fib4_entry;
4897         struct mlxsw_sp_fib_node *fib_node;
4898         struct mlxsw_sp_fib *fib;
4899         struct mlxsw_sp_vr *vr;
4900
4901         vr = mlxsw_sp_vr_find(mlxsw_sp, fen_info->tb_id);
4902         if (!vr)
4903                 return NULL;
4904         fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV4);
4905
4906         fib_node = mlxsw_sp_fib_node_lookup(fib, &fen_info->dst,
4907                                             sizeof(fen_info->dst),
4908                                             fen_info->dst_len);
4909         if (!fib_node)
4910                 return NULL;
4911
4912         fib4_entry = container_of(fib_node->fib_entry,
4913                                   struct mlxsw_sp_fib4_entry, common);
4914         if (fib4_entry->tb_id == fen_info->tb_id &&
4915             fib4_entry->tos == fen_info->tos &&
4916             fib4_entry->type == fen_info->type &&
4917             fib4_entry->fi == fen_info->fi)
4918                 return fib4_entry;
4919
4920         return NULL;
4921 }
4922
4923 static const struct rhashtable_params mlxsw_sp_fib_ht_params = {
4924         .key_offset = offsetof(struct mlxsw_sp_fib_node, key),
4925         .head_offset = offsetof(struct mlxsw_sp_fib_node, ht_node),
4926         .key_len = sizeof(struct mlxsw_sp_fib_key),
4927         .automatic_shrinking = true,
4928 };
4929
4930 static int mlxsw_sp_fib_node_insert(struct mlxsw_sp_fib *fib,
4931                                     struct mlxsw_sp_fib_node *fib_node)
4932 {
4933         return rhashtable_insert_fast(&fib->ht, &fib_node->ht_node,
4934                                       mlxsw_sp_fib_ht_params);
4935 }
4936
4937 static void mlxsw_sp_fib_node_remove(struct mlxsw_sp_fib *fib,
4938                                      struct mlxsw_sp_fib_node *fib_node)
4939 {
4940         rhashtable_remove_fast(&fib->ht, &fib_node->ht_node,
4941                                mlxsw_sp_fib_ht_params);
4942 }
4943
4944 static struct mlxsw_sp_fib_node *
4945 mlxsw_sp_fib_node_lookup(struct mlxsw_sp_fib *fib, const void *addr,
4946                          size_t addr_len, unsigned char prefix_len)
4947 {
4948         struct mlxsw_sp_fib_key key;
4949
4950         memset(&key, 0, sizeof(key));
4951         memcpy(key.addr, addr, addr_len);
4952         key.prefix_len = prefix_len;
4953         return rhashtable_lookup_fast(&fib->ht, &key, mlxsw_sp_fib_ht_params);
4954 }
4955
4956 static struct mlxsw_sp_fib_node *
4957 mlxsw_sp_fib_node_create(struct mlxsw_sp_fib *fib, const void *addr,
4958                          size_t addr_len, unsigned char prefix_len)
4959 {
4960         struct mlxsw_sp_fib_node *fib_node;
4961
4962         fib_node = kzalloc(sizeof(*fib_node), GFP_KERNEL);
4963         if (!fib_node)
4964                 return NULL;
4965
4966         list_add(&fib_node->list, &fib->node_list);
4967         memcpy(fib_node->key.addr, addr, addr_len);
4968         fib_node->key.prefix_len = prefix_len;
4969
4970         return fib_node;
4971 }
4972
4973 static void mlxsw_sp_fib_node_destroy(struct mlxsw_sp_fib_node *fib_node)
4974 {
4975         list_del(&fib_node->list);
4976         kfree(fib_node);
4977 }
4978
4979 static int mlxsw_sp_fib_lpm_tree_link(struct mlxsw_sp *mlxsw_sp,
4980                                       struct mlxsw_sp_fib_node *fib_node)
4981 {
4982         struct mlxsw_sp_prefix_usage req_prefix_usage;
4983         struct mlxsw_sp_fib *fib = fib_node->fib;
4984         struct mlxsw_sp_lpm_tree *lpm_tree;
4985         int err;
4986
4987         lpm_tree = mlxsw_sp->router->lpm.proto_trees[fib->proto];
4988         if (lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
4989                 goto out;
4990
4991         mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
4992         mlxsw_sp_prefix_usage_set(&req_prefix_usage, fib_node->key.prefix_len);
4993         lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
4994                                          fib->proto);
4995         if (IS_ERR(lpm_tree))
4996                 return PTR_ERR(lpm_tree);
4997
4998         err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
4999         if (err)
5000                 goto err_lpm_tree_replace;
5001
5002 out:
5003         lpm_tree->prefix_ref_count[fib_node->key.prefix_len]++;
5004         return 0;
5005
5006 err_lpm_tree_replace:
5007         mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
5008         return err;
5009 }
5010
5011 static void mlxsw_sp_fib_lpm_tree_unlink(struct mlxsw_sp *mlxsw_sp,
5012                                          struct mlxsw_sp_fib_node *fib_node)
5013 {
5014         struct mlxsw_sp_lpm_tree *lpm_tree = fib_node->fib->lpm_tree;
5015         struct mlxsw_sp_prefix_usage req_prefix_usage;
5016         struct mlxsw_sp_fib *fib = fib_node->fib;
5017         int err;
5018
5019         if (--lpm_tree->prefix_ref_count[fib_node->key.prefix_len] != 0)
5020                 return;
5021         /* Try to construct a new LPM tree from the current prefix usage
5022          * minus the unused one. If we fail, continue using the old one.
5023          */
5024         mlxsw_sp_prefix_usage_cpy(&req_prefix_usage, &lpm_tree->prefix_usage);
5025         mlxsw_sp_prefix_usage_clear(&req_prefix_usage,
5026                                     fib_node->key.prefix_len);
5027         lpm_tree = mlxsw_sp_lpm_tree_get(mlxsw_sp, &req_prefix_usage,
5028                                          fib->proto);
5029         if (IS_ERR(lpm_tree))
5030                 return;
5031
5032         err = mlxsw_sp_vrs_lpm_tree_replace(mlxsw_sp, fib, lpm_tree);
5033         if (err)
5034                 goto err_lpm_tree_replace;
5035
5036         return;
5037
5038 err_lpm_tree_replace:
5039         mlxsw_sp_lpm_tree_put(mlxsw_sp, lpm_tree);
5040 }
5041
5042 static int mlxsw_sp_fib_node_init(struct mlxsw_sp *mlxsw_sp,
5043                                   struct mlxsw_sp_fib_node *fib_node,
5044                                   struct mlxsw_sp_fib *fib)
5045 {
5046         int err;
5047
5048         err = mlxsw_sp_fib_node_insert(fib, fib_node);
5049         if (err)
5050                 return err;
5051         fib_node->fib = fib;
5052
5053         err = mlxsw_sp_fib_lpm_tree_link(mlxsw_sp, fib_node);
5054         if (err)
5055                 goto err_fib_lpm_tree_link;
5056
5057         return 0;
5058
5059 err_fib_lpm_tree_link:
5060         fib_node->fib = NULL;
5061         mlxsw_sp_fib_node_remove(fib, fib_node);
5062         return err;
5063 }
5064
5065 static void mlxsw_sp_fib_node_fini(struct mlxsw_sp *mlxsw_sp,
5066                                    struct mlxsw_sp_fib_node *fib_node)
5067 {
5068         struct mlxsw_sp_fib *fib = fib_node->fib;
5069
5070         mlxsw_sp_fib_lpm_tree_unlink(mlxsw_sp, fib_node);
5071         fib_node->fib = NULL;
5072         mlxsw_sp_fib_node_remove(fib, fib_node);
5073 }
5074
5075 static struct mlxsw_sp_fib_node *
5076 mlxsw_sp_fib_node_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id, const void *addr,
5077                       size_t addr_len, unsigned char prefix_len,
5078                       enum mlxsw_sp_l3proto proto)
5079 {
5080         struct mlxsw_sp_fib_node *fib_node;
5081         struct mlxsw_sp_fib *fib;
5082         struct mlxsw_sp_vr *vr;
5083         int err;
5084
5085         vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, NULL);
5086         if (IS_ERR(vr))
5087                 return ERR_CAST(vr);
5088         fib = mlxsw_sp_vr_fib(vr, proto);
5089
5090         fib_node = mlxsw_sp_fib_node_lookup(fib, addr, addr_len, prefix_len);
5091         if (fib_node)
5092                 return fib_node;
5093
5094         fib_node = mlxsw_sp_fib_node_create(fib, addr, addr_len, prefix_len);
5095         if (!fib_node) {
5096                 err = -ENOMEM;
5097                 goto err_fib_node_create;
5098         }
5099
5100         err = mlxsw_sp_fib_node_init(mlxsw_sp, fib_node, fib);
5101         if (err)
5102                 goto err_fib_node_init;
5103
5104         return fib_node;
5105
5106 err_fib_node_init:
5107         mlxsw_sp_fib_node_destroy(fib_node);
5108 err_fib_node_create:
5109         mlxsw_sp_vr_put(mlxsw_sp, vr);
5110         return ERR_PTR(err);
5111 }
5112
5113 static void mlxsw_sp_fib_node_put(struct mlxsw_sp *mlxsw_sp,
5114                                   struct mlxsw_sp_fib_node *fib_node)
5115 {
5116         struct mlxsw_sp_vr *vr = fib_node->fib->vr;
5117
5118         if (fib_node->fib_entry)
5119                 return;
5120         mlxsw_sp_fib_node_fini(mlxsw_sp, fib_node);
5121         mlxsw_sp_fib_node_destroy(fib_node);
5122         mlxsw_sp_vr_put(mlxsw_sp, vr);
5123 }
5124
5125 static int mlxsw_sp_fib_node_entry_link(struct mlxsw_sp *mlxsw_sp,
5126                                         struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5127                                         struct mlxsw_sp_fib_entry *fib_entry)
5128 {
5129         struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
5130         bool is_new = !fib_node->fib_entry;
5131         int err;
5132
5133         fib_node->fib_entry = fib_entry;
5134
5135         err = __mlxsw_sp_fib_entry_update(mlxsw_sp, op_ctx, fib_entry, is_new);
5136         if (err)
5137                 goto err_fib_entry_update;
5138
5139         return 0;
5140
5141 err_fib_entry_update:
5142         fib_node->fib_entry = NULL;
5143         return err;
5144 }
5145
5146 static int __mlxsw_sp_fib_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
5147                                             struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5148                                             struct mlxsw_sp_fib_entry *fib_entry)
5149 {
5150         struct mlxsw_sp_fib_node *fib_node = fib_entry->fib_node;
5151         int err;
5152
5153         err = mlxsw_sp_fib_entry_del(mlxsw_sp, op_ctx, fib_entry);
5154         fib_node->fib_entry = NULL;
5155         return err;
5156 }
5157
5158 static void mlxsw_sp_fib_node_entry_unlink(struct mlxsw_sp *mlxsw_sp,
5159                                            struct mlxsw_sp_fib_entry *fib_entry)
5160 {
5161         struct mlxsw_sp_fib_entry_op_ctx *op_ctx = mlxsw_sp->router->ll_op_ctx;
5162
5163         mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
5164         __mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, op_ctx, fib_entry);
5165 }
5166
5167 static bool mlxsw_sp_fib4_allow_replace(struct mlxsw_sp_fib4_entry *fib4_entry)
5168 {
5169         struct mlxsw_sp_fib_node *fib_node = fib4_entry->common.fib_node;
5170         struct mlxsw_sp_fib4_entry *fib4_replaced;
5171
5172         if (!fib_node->fib_entry)
5173                 return true;
5174
5175         fib4_replaced = container_of(fib_node->fib_entry,
5176                                      struct mlxsw_sp_fib4_entry, common);
5177         if (fib4_entry->tb_id == RT_TABLE_MAIN &&
5178             fib4_replaced->tb_id == RT_TABLE_LOCAL)
5179                 return false;
5180
5181         return true;
5182 }
5183
5184 static int
5185 mlxsw_sp_router_fib4_replace(struct mlxsw_sp *mlxsw_sp,
5186                              struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5187                              const struct fib_entry_notifier_info *fen_info)
5188 {
5189         struct mlxsw_sp_fib4_entry *fib4_entry, *fib4_replaced;
5190         struct mlxsw_sp_fib_entry *replaced;
5191         struct mlxsw_sp_fib_node *fib_node;
5192         int err;
5193
5194         if (mlxsw_sp->router->aborted)
5195                 return 0;
5196
5197         fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, fen_info->tb_id,
5198                                          &fen_info->dst, sizeof(fen_info->dst),
5199                                          fen_info->dst_len,
5200                                          MLXSW_SP_L3_PROTO_IPV4);
5201         if (IS_ERR(fib_node)) {
5202                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to get FIB node\n");
5203                 return PTR_ERR(fib_node);
5204         }
5205
5206         fib4_entry = mlxsw_sp_fib4_entry_create(mlxsw_sp, fib_node, fen_info);
5207         if (IS_ERR(fib4_entry)) {
5208                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to create FIB entry\n");
5209                 err = PTR_ERR(fib4_entry);
5210                 goto err_fib4_entry_create;
5211         }
5212
5213         if (!mlxsw_sp_fib4_allow_replace(fib4_entry)) {
5214                 mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5215                 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5216                 return 0;
5217         }
5218
5219         replaced = fib_node->fib_entry;
5220         err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, op_ctx, &fib4_entry->common);
5221         if (err) {
5222                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to link FIB entry to node\n");
5223                 goto err_fib_node_entry_link;
5224         }
5225
5226         /* Nothing to replace */
5227         if (!replaced)
5228                 return 0;
5229
5230         mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
5231         fib4_replaced = container_of(replaced, struct mlxsw_sp_fib4_entry,
5232                                      common);
5233         mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_replaced);
5234
5235         return 0;
5236
5237 err_fib_node_entry_link:
5238         fib_node->fib_entry = replaced;
5239         mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5240 err_fib4_entry_create:
5241         mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5242         return err;
5243 }
5244
5245 static int mlxsw_sp_router_fib4_del(struct mlxsw_sp *mlxsw_sp,
5246                                     struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5247                                     struct fib_entry_notifier_info *fen_info)
5248 {
5249         struct mlxsw_sp_fib4_entry *fib4_entry;
5250         struct mlxsw_sp_fib_node *fib_node;
5251         int err;
5252
5253         if (mlxsw_sp->router->aborted)
5254                 return 0;
5255
5256         fib4_entry = mlxsw_sp_fib4_entry_lookup(mlxsw_sp, fen_info);
5257         if (!fib4_entry)
5258                 return 0;
5259         fib_node = fib4_entry->common.fib_node;
5260
5261         err = __mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, op_ctx, &fib4_entry->common);
5262         mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
5263         mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5264         return err;
5265 }
5266
5267 static bool mlxsw_sp_fib6_rt_should_ignore(const struct fib6_info *rt)
5268 {
5269         /* Multicast routes aren't supported, so ignore them. Neighbour
5270          * Discovery packets are specifically trapped.
5271          */
5272         if (ipv6_addr_type(&rt->fib6_dst.addr) & IPV6_ADDR_MULTICAST)
5273                 return true;
5274
5275         /* Cloned routes are irrelevant in the forwarding path. */
5276         if (rt->fib6_flags & RTF_CACHE)
5277                 return true;
5278
5279         return false;
5280 }
5281
5282 static struct mlxsw_sp_rt6 *mlxsw_sp_rt6_create(struct fib6_info *rt)
5283 {
5284         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5285
5286         mlxsw_sp_rt6 = kzalloc(sizeof(*mlxsw_sp_rt6), GFP_KERNEL);
5287         if (!mlxsw_sp_rt6)
5288                 return ERR_PTR(-ENOMEM);
5289
5290         /* In case of route replace, replaced route is deleted with
5291          * no notification. Take reference to prevent accessing freed
5292          * memory.
5293          */
5294         mlxsw_sp_rt6->rt = rt;
5295         fib6_info_hold(rt);
5296
5297         return mlxsw_sp_rt6;
5298 }
5299
5300 #if IS_ENABLED(CONFIG_IPV6)
5301 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
5302 {
5303         fib6_info_release(rt);
5304 }
5305 #else
5306 static void mlxsw_sp_rt6_release(struct fib6_info *rt)
5307 {
5308 }
5309 #endif
5310
5311 static void mlxsw_sp_rt6_destroy(struct mlxsw_sp_rt6 *mlxsw_sp_rt6)
5312 {
5313         struct fib6_nh *fib6_nh = mlxsw_sp_rt6->rt->fib6_nh;
5314
5315         fib6_nh->fib_nh_flags &= ~RTNH_F_OFFLOAD;
5316         mlxsw_sp_rt6_release(mlxsw_sp_rt6->rt);
5317         kfree(mlxsw_sp_rt6);
5318 }
5319
5320 static struct fib6_info *
5321 mlxsw_sp_fib6_entry_rt(const struct mlxsw_sp_fib6_entry *fib6_entry)
5322 {
5323         return list_first_entry(&fib6_entry->rt6_list, struct mlxsw_sp_rt6,
5324                                 list)->rt;
5325 }
5326
5327 static struct mlxsw_sp_rt6 *
5328 mlxsw_sp_fib6_entry_rt_find(const struct mlxsw_sp_fib6_entry *fib6_entry,
5329                             const struct fib6_info *rt)
5330 {
5331         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5332
5333         list_for_each_entry(mlxsw_sp_rt6, &fib6_entry->rt6_list, list) {
5334                 if (mlxsw_sp_rt6->rt == rt)
5335                         return mlxsw_sp_rt6;
5336         }
5337
5338         return NULL;
5339 }
5340
5341 static bool mlxsw_sp_nexthop6_ipip_type(const struct mlxsw_sp *mlxsw_sp,
5342                                         const struct fib6_info *rt,
5343                                         enum mlxsw_sp_ipip_type *ret)
5344 {
5345         return rt->fib6_nh->fib_nh_dev &&
5346                mlxsw_sp_netdev_ipip_type(mlxsw_sp, rt->fib6_nh->fib_nh_dev, ret);
5347 }
5348
5349 static int mlxsw_sp_nexthop6_init(struct mlxsw_sp *mlxsw_sp,
5350                                   struct mlxsw_sp_nexthop_group *nh_grp,
5351                                   struct mlxsw_sp_nexthop *nh,
5352                                   const struct fib6_info *rt)
5353 {
5354         struct net_device *dev = rt->fib6_nh->fib_nh_dev;
5355
5356         nh->nhgi = nh_grp->nhgi;
5357         nh->nh_weight = rt->fib6_nh->fib_nh_weight;
5358         memcpy(&nh->gw_addr, &rt->fib6_nh->fib_nh_gw6, sizeof(nh->gw_addr));
5359 #if IS_ENABLED(CONFIG_IPV6)
5360         nh->neigh_tbl = &nd_tbl;
5361 #endif
5362         mlxsw_sp_nexthop_counter_alloc(mlxsw_sp, nh);
5363
5364         list_add_tail(&nh->router_list_node, &mlxsw_sp->router->nexthop_list);
5365
5366         if (!dev)
5367                 return 0;
5368         nh->ifindex = dev->ifindex;
5369
5370         return mlxsw_sp_nexthop_type_init(mlxsw_sp, nh, dev);
5371 }
5372
5373 static void mlxsw_sp_nexthop6_fini(struct mlxsw_sp *mlxsw_sp,
5374                                    struct mlxsw_sp_nexthop *nh)
5375 {
5376         mlxsw_sp_nexthop_type_fini(mlxsw_sp, nh);
5377         list_del(&nh->router_list_node);
5378         mlxsw_sp_nexthop_counter_free(mlxsw_sp, nh);
5379 }
5380
5381 static bool mlxsw_sp_rt6_is_gateway(const struct mlxsw_sp *mlxsw_sp,
5382                                     const struct fib6_info *rt)
5383 {
5384         return rt->fib6_nh->fib_nh_gw_family ||
5385                mlxsw_sp_nexthop6_ipip_type(mlxsw_sp, rt, NULL);
5386 }
5387
5388 static int
5389 mlxsw_sp_nexthop6_group_info_init(struct mlxsw_sp *mlxsw_sp,
5390                                   struct mlxsw_sp_nexthop_group *nh_grp,
5391                                   struct mlxsw_sp_fib6_entry *fib6_entry)
5392 {
5393         struct mlxsw_sp_nexthop_group_info *nhgi;
5394         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5395         struct mlxsw_sp_nexthop *nh;
5396         int err, i;
5397
5398         nhgi = kzalloc(struct_size(nhgi, nexthops, fib6_entry->nrt6),
5399                        GFP_KERNEL);
5400         if (!nhgi)
5401                 return -ENOMEM;
5402         nh_grp->nhgi = nhgi;
5403         nhgi->nh_grp = nh_grp;
5404         mlxsw_sp_rt6 = list_first_entry(&fib6_entry->rt6_list,
5405                                         struct mlxsw_sp_rt6, list);
5406         nhgi->gateway = mlxsw_sp_rt6_is_gateway(mlxsw_sp, mlxsw_sp_rt6->rt);
5407         nhgi->count = fib6_entry->nrt6;
5408         for (i = 0; i < nhgi->count; i++) {
5409                 struct fib6_info *rt = mlxsw_sp_rt6->rt;
5410
5411                 nh = &nhgi->nexthops[i];
5412                 err = mlxsw_sp_nexthop6_init(mlxsw_sp, nh_grp, nh, rt);
5413                 if (err)
5414                         goto err_nexthop6_init;
5415                 mlxsw_sp_rt6 = list_next_entry(mlxsw_sp_rt6, list);
5416         }
5417         nh_grp->nhgi = nhgi;
5418         mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5419
5420         return 0;
5421
5422 err_nexthop6_init:
5423         for (i--; i >= 0; i--) {
5424                 nh = &nhgi->nexthops[i];
5425                 mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
5426         }
5427         kfree(nhgi);
5428         return err;
5429 }
5430
5431 static void
5432 mlxsw_sp_nexthop6_group_info_fini(struct mlxsw_sp *mlxsw_sp,
5433                                   struct mlxsw_sp_nexthop_group *nh_grp)
5434 {
5435         struct mlxsw_sp_nexthop_group_info *nhgi = nh_grp->nhgi;
5436         int i;
5437
5438         for (i = nhgi->count - 1; i >= 0; i--) {
5439                 struct mlxsw_sp_nexthop *nh = &nhgi->nexthops[i];
5440
5441                 mlxsw_sp_nexthop6_fini(mlxsw_sp, nh);
5442         }
5443         mlxsw_sp_nexthop_group_refresh(mlxsw_sp, nh_grp);
5444         WARN_ON_ONCE(nhgi->adj_index_valid);
5445         kfree(nhgi);
5446 }
5447
5448 static struct mlxsw_sp_nexthop_group *
5449 mlxsw_sp_nexthop6_group_create(struct mlxsw_sp *mlxsw_sp,
5450                                struct mlxsw_sp_fib6_entry *fib6_entry)
5451 {
5452         struct mlxsw_sp_nexthop_group *nh_grp;
5453         int err;
5454
5455         nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
5456         if (!nh_grp)
5457                 return ERR_PTR(-ENOMEM);
5458         INIT_LIST_HEAD(&nh_grp->fib_list);
5459         nh_grp->type = MLXSW_SP_NEXTHOP_GROUP_TYPE_IPV6;
5460
5461         err = mlxsw_sp_nexthop6_group_info_init(mlxsw_sp, nh_grp, fib6_entry);
5462         if (err)
5463                 goto err_nexthop_group_info_init;
5464
5465         err = mlxsw_sp_nexthop_group_insert(mlxsw_sp, nh_grp);
5466         if (err)
5467                 goto err_nexthop_group_insert;
5468
5469         return nh_grp;
5470
5471 err_nexthop_group_insert:
5472         mlxsw_sp_nexthop6_group_info_fini(mlxsw_sp, nh_grp);
5473 err_nexthop_group_info_init:
5474         kfree(nh_grp);
5475         return ERR_PTR(err);
5476 }
5477
5478 static void
5479 mlxsw_sp_nexthop6_group_destroy(struct mlxsw_sp *mlxsw_sp,
5480                                 struct mlxsw_sp_nexthop_group *nh_grp)
5481 {
5482         mlxsw_sp_nexthop_group_remove(mlxsw_sp, nh_grp);
5483         mlxsw_sp_nexthop6_group_info_fini(mlxsw_sp, nh_grp);
5484         kfree(nh_grp);
5485 }
5486
5487 static int mlxsw_sp_nexthop6_group_get(struct mlxsw_sp *mlxsw_sp,
5488                                        struct mlxsw_sp_fib6_entry *fib6_entry)
5489 {
5490         struct mlxsw_sp_nexthop_group *nh_grp;
5491
5492         nh_grp = mlxsw_sp_nexthop6_group_lookup(mlxsw_sp, fib6_entry);
5493         if (!nh_grp) {
5494                 nh_grp = mlxsw_sp_nexthop6_group_create(mlxsw_sp, fib6_entry);
5495                 if (IS_ERR(nh_grp))
5496                         return PTR_ERR(nh_grp);
5497         }
5498
5499         list_add_tail(&fib6_entry->common.nexthop_group_node,
5500                       &nh_grp->fib_list);
5501         fib6_entry->common.nh_group = nh_grp;
5502
5503         /* The route and the nexthop are described by the same struct, so we
5504          * need to the update the nexthop offload indication for the new route.
5505          */
5506         __mlxsw_sp_nexthop6_group_offload_refresh(nh_grp, fib6_entry);
5507
5508         return 0;
5509 }
5510
5511 static void mlxsw_sp_nexthop6_group_put(struct mlxsw_sp *mlxsw_sp,
5512                                         struct mlxsw_sp_fib_entry *fib_entry)
5513 {
5514         struct mlxsw_sp_nexthop_group *nh_grp = fib_entry->nh_group;
5515
5516         list_del(&fib_entry->nexthop_group_node);
5517         if (!list_empty(&nh_grp->fib_list))
5518                 return;
5519         mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, nh_grp);
5520 }
5521
5522 static int mlxsw_sp_nexthop6_group_update(struct mlxsw_sp *mlxsw_sp,
5523                                           struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5524                                           struct mlxsw_sp_fib6_entry *fib6_entry)
5525 {
5526         struct mlxsw_sp_nexthop_group *old_nh_grp = fib6_entry->common.nh_group;
5527         int err;
5528
5529         fib6_entry->common.nh_group = NULL;
5530         list_del(&fib6_entry->common.nexthop_group_node);
5531
5532         err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
5533         if (err)
5534                 goto err_nexthop6_group_get;
5535
5536         /* In case this entry is offloaded, then the adjacency index
5537          * currently associated with it in the device's table is that
5538          * of the old group. Start using the new one instead.
5539          */
5540         err = __mlxsw_sp_fib_entry_update(mlxsw_sp, op_ctx,
5541                                           &fib6_entry->common, false);
5542         if (err)
5543                 goto err_fib_entry_update;
5544
5545         if (list_empty(&old_nh_grp->fib_list))
5546                 mlxsw_sp_nexthop6_group_destroy(mlxsw_sp, old_nh_grp);
5547
5548         return 0;
5549
5550 err_fib_entry_update:
5551         mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
5552 err_nexthop6_group_get:
5553         list_add_tail(&fib6_entry->common.nexthop_group_node,
5554                       &old_nh_grp->fib_list);
5555         fib6_entry->common.nh_group = old_nh_grp;
5556         return err;
5557 }
5558
5559 static int
5560 mlxsw_sp_fib6_entry_nexthop_add(struct mlxsw_sp *mlxsw_sp,
5561                                 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5562                                 struct mlxsw_sp_fib6_entry *fib6_entry,
5563                                 struct fib6_info **rt_arr, unsigned int nrt6)
5564 {
5565         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5566         int err, i;
5567
5568         for (i = 0; i < nrt6; i++) {
5569                 mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
5570                 if (IS_ERR(mlxsw_sp_rt6)) {
5571                         err = PTR_ERR(mlxsw_sp_rt6);
5572                         goto err_rt6_create;
5573                 }
5574
5575                 list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
5576                 fib6_entry->nrt6++;
5577         }
5578
5579         err = mlxsw_sp_nexthop6_group_update(mlxsw_sp, op_ctx, fib6_entry);
5580         if (err)
5581                 goto err_nexthop6_group_update;
5582
5583         return 0;
5584
5585 err_nexthop6_group_update:
5586         i = nrt6;
5587 err_rt6_create:
5588         for (i--; i >= 0; i--) {
5589                 fib6_entry->nrt6--;
5590                 mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
5591                                                struct mlxsw_sp_rt6, list);
5592                 list_del(&mlxsw_sp_rt6->list);
5593                 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5594         }
5595         return err;
5596 }
5597
5598 static void
5599 mlxsw_sp_fib6_entry_nexthop_del(struct mlxsw_sp *mlxsw_sp,
5600                                 struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5601                                 struct mlxsw_sp_fib6_entry *fib6_entry,
5602                                 struct fib6_info **rt_arr, unsigned int nrt6)
5603 {
5604         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5605         int i;
5606
5607         for (i = 0; i < nrt6; i++) {
5608                 mlxsw_sp_rt6 = mlxsw_sp_fib6_entry_rt_find(fib6_entry,
5609                                                            rt_arr[i]);
5610                 if (WARN_ON_ONCE(!mlxsw_sp_rt6))
5611                         continue;
5612
5613                 fib6_entry->nrt6--;
5614                 list_del(&mlxsw_sp_rt6->list);
5615                 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5616         }
5617
5618         mlxsw_sp_nexthop6_group_update(mlxsw_sp, op_ctx, fib6_entry);
5619 }
5620
5621 static void mlxsw_sp_fib6_entry_type_set(struct mlxsw_sp *mlxsw_sp,
5622                                          struct mlxsw_sp_fib_entry *fib_entry,
5623                                          const struct fib6_info *rt)
5624 {
5625         if (rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
5626                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_TRAP;
5627         else if (rt->fib6_type == RTN_BLACKHOLE)
5628                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_BLACKHOLE;
5629         else if (rt->fib6_flags & RTF_REJECT)
5630                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_UNREACHABLE;
5631         else if (mlxsw_sp_rt6_is_gateway(mlxsw_sp, rt))
5632                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_REMOTE;
5633         else
5634                 fib_entry->type = MLXSW_SP_FIB_ENTRY_TYPE_LOCAL;
5635 }
5636
5637 static void
5638 mlxsw_sp_fib6_entry_rt_destroy_all(struct mlxsw_sp_fib6_entry *fib6_entry)
5639 {
5640         struct mlxsw_sp_rt6 *mlxsw_sp_rt6, *tmp;
5641
5642         list_for_each_entry_safe(mlxsw_sp_rt6, tmp, &fib6_entry->rt6_list,
5643                                  list) {
5644                 fib6_entry->nrt6--;
5645                 list_del(&mlxsw_sp_rt6->list);
5646                 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5647         }
5648 }
5649
5650 static struct mlxsw_sp_fib6_entry *
5651 mlxsw_sp_fib6_entry_create(struct mlxsw_sp *mlxsw_sp,
5652                            struct mlxsw_sp_fib_node *fib_node,
5653                            struct fib6_info **rt_arr, unsigned int nrt6)
5654 {
5655         struct mlxsw_sp_fib6_entry *fib6_entry;
5656         struct mlxsw_sp_fib_entry *fib_entry;
5657         struct mlxsw_sp_rt6 *mlxsw_sp_rt6;
5658         int err, i;
5659
5660         fib6_entry = kzalloc(sizeof(*fib6_entry), GFP_KERNEL);
5661         if (!fib6_entry)
5662                 return ERR_PTR(-ENOMEM);
5663         fib_entry = &fib6_entry->common;
5664
5665         fib_entry->priv = mlxsw_sp_fib_entry_priv_create(fib_node->fib->ll_ops);
5666         if (IS_ERR(fib_entry->priv)) {
5667                 err = PTR_ERR(fib_entry->priv);
5668                 goto err_fib_entry_priv_create;
5669         }
5670
5671         INIT_LIST_HEAD(&fib6_entry->rt6_list);
5672
5673         for (i = 0; i < nrt6; i++) {
5674                 mlxsw_sp_rt6 = mlxsw_sp_rt6_create(rt_arr[i]);
5675                 if (IS_ERR(mlxsw_sp_rt6)) {
5676                         err = PTR_ERR(mlxsw_sp_rt6);
5677                         goto err_rt6_create;
5678                 }
5679                 list_add_tail(&mlxsw_sp_rt6->list, &fib6_entry->rt6_list);
5680                 fib6_entry->nrt6++;
5681         }
5682
5683         mlxsw_sp_fib6_entry_type_set(mlxsw_sp, fib_entry, rt_arr[0]);
5684
5685         err = mlxsw_sp_nexthop6_group_get(mlxsw_sp, fib6_entry);
5686         if (err)
5687                 goto err_nexthop6_group_get;
5688
5689         fib_entry->fib_node = fib_node;
5690
5691         return fib6_entry;
5692
5693 err_nexthop6_group_get:
5694         i = nrt6;
5695 err_rt6_create:
5696         for (i--; i >= 0; i--) {
5697                 fib6_entry->nrt6--;
5698                 mlxsw_sp_rt6 = list_last_entry(&fib6_entry->rt6_list,
5699                                                struct mlxsw_sp_rt6, list);
5700                 list_del(&mlxsw_sp_rt6->list);
5701                 mlxsw_sp_rt6_destroy(mlxsw_sp_rt6);
5702         }
5703         mlxsw_sp_fib_entry_priv_put(fib_entry->priv);
5704 err_fib_entry_priv_create:
5705         kfree(fib6_entry);
5706         return ERR_PTR(err);
5707 }
5708
5709 static void mlxsw_sp_fib6_entry_destroy(struct mlxsw_sp *mlxsw_sp,
5710                                         struct mlxsw_sp_fib6_entry *fib6_entry)
5711 {
5712         mlxsw_sp_nexthop6_group_put(mlxsw_sp, &fib6_entry->common);
5713         mlxsw_sp_fib6_entry_rt_destroy_all(fib6_entry);
5714         WARN_ON(fib6_entry->nrt6);
5715         mlxsw_sp_fib_entry_priv_put(fib6_entry->common.priv);
5716         kfree(fib6_entry);
5717 }
5718
5719 static struct mlxsw_sp_fib6_entry *
5720 mlxsw_sp_fib6_entry_lookup(struct mlxsw_sp *mlxsw_sp,
5721                            const struct fib6_info *rt)
5722 {
5723         struct mlxsw_sp_fib6_entry *fib6_entry;
5724         struct mlxsw_sp_fib_node *fib_node;
5725         struct mlxsw_sp_fib *fib;
5726         struct fib6_info *cmp_rt;
5727         struct mlxsw_sp_vr *vr;
5728
5729         vr = mlxsw_sp_vr_find(mlxsw_sp, rt->fib6_table->tb6_id);
5730         if (!vr)
5731                 return NULL;
5732         fib = mlxsw_sp_vr_fib(vr, MLXSW_SP_L3_PROTO_IPV6);
5733
5734         fib_node = mlxsw_sp_fib_node_lookup(fib, &rt->fib6_dst.addr,
5735                                             sizeof(rt->fib6_dst.addr),
5736                                             rt->fib6_dst.plen);
5737         if (!fib_node)
5738                 return NULL;
5739
5740         fib6_entry = container_of(fib_node->fib_entry,
5741                                   struct mlxsw_sp_fib6_entry, common);
5742         cmp_rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5743         if (rt->fib6_table->tb6_id == cmp_rt->fib6_table->tb6_id &&
5744             rt->fib6_metric == cmp_rt->fib6_metric &&
5745             mlxsw_sp_fib6_entry_rt_find(fib6_entry, rt))
5746                 return fib6_entry;
5747
5748         return NULL;
5749 }
5750
5751 static bool mlxsw_sp_fib6_allow_replace(struct mlxsw_sp_fib6_entry *fib6_entry)
5752 {
5753         struct mlxsw_sp_fib_node *fib_node = fib6_entry->common.fib_node;
5754         struct mlxsw_sp_fib6_entry *fib6_replaced;
5755         struct fib6_info *rt, *rt_replaced;
5756
5757         if (!fib_node->fib_entry)
5758                 return true;
5759
5760         fib6_replaced = container_of(fib_node->fib_entry,
5761                                      struct mlxsw_sp_fib6_entry,
5762                                      common);
5763         rt = mlxsw_sp_fib6_entry_rt(fib6_entry);
5764         rt_replaced = mlxsw_sp_fib6_entry_rt(fib6_replaced);
5765         if (rt->fib6_table->tb6_id == RT_TABLE_MAIN &&
5766             rt_replaced->fib6_table->tb6_id == RT_TABLE_LOCAL)
5767                 return false;
5768
5769         return true;
5770 }
5771
5772 static int mlxsw_sp_router_fib6_replace(struct mlxsw_sp *mlxsw_sp,
5773                                         struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5774                                         struct fib6_info **rt_arr, unsigned int nrt6)
5775 {
5776         struct mlxsw_sp_fib6_entry *fib6_entry, *fib6_replaced;
5777         struct mlxsw_sp_fib_entry *replaced;
5778         struct mlxsw_sp_fib_node *fib_node;
5779         struct fib6_info *rt = rt_arr[0];
5780         int err;
5781
5782         if (mlxsw_sp->router->aborted)
5783                 return 0;
5784
5785         if (rt->fib6_src.plen)
5786                 return -EINVAL;
5787
5788         if (mlxsw_sp_fib6_rt_should_ignore(rt))
5789                 return 0;
5790
5791         fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
5792                                          &rt->fib6_dst.addr,
5793                                          sizeof(rt->fib6_dst.addr),
5794                                          rt->fib6_dst.plen,
5795                                          MLXSW_SP_L3_PROTO_IPV6);
5796         if (IS_ERR(fib_node))
5797                 return PTR_ERR(fib_node);
5798
5799         fib6_entry = mlxsw_sp_fib6_entry_create(mlxsw_sp, fib_node, rt_arr,
5800                                                 nrt6);
5801         if (IS_ERR(fib6_entry)) {
5802                 err = PTR_ERR(fib6_entry);
5803                 goto err_fib6_entry_create;
5804         }
5805
5806         if (!mlxsw_sp_fib6_allow_replace(fib6_entry)) {
5807                 mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5808                 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5809                 return 0;
5810         }
5811
5812         replaced = fib_node->fib_entry;
5813         err = mlxsw_sp_fib_node_entry_link(mlxsw_sp, op_ctx, &fib6_entry->common);
5814         if (err)
5815                 goto err_fib_node_entry_link;
5816
5817         /* Nothing to replace */
5818         if (!replaced)
5819                 return 0;
5820
5821         mlxsw_sp_fib_entry_hw_flags_clear(mlxsw_sp, replaced);
5822         fib6_replaced = container_of(replaced, struct mlxsw_sp_fib6_entry,
5823                                      common);
5824         mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_replaced);
5825
5826         return 0;
5827
5828 err_fib_node_entry_link:
5829         fib_node->fib_entry = replaced;
5830         mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5831 err_fib6_entry_create:
5832         mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5833         return err;
5834 }
5835
5836 static int mlxsw_sp_router_fib6_append(struct mlxsw_sp *mlxsw_sp,
5837                                        struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5838                                        struct fib6_info **rt_arr, unsigned int nrt6)
5839 {
5840         struct mlxsw_sp_fib6_entry *fib6_entry;
5841         struct mlxsw_sp_fib_node *fib_node;
5842         struct fib6_info *rt = rt_arr[0];
5843         int err;
5844
5845         if (mlxsw_sp->router->aborted)
5846                 return 0;
5847
5848         if (rt->fib6_src.plen)
5849                 return -EINVAL;
5850
5851         if (mlxsw_sp_fib6_rt_should_ignore(rt))
5852                 return 0;
5853
5854         fib_node = mlxsw_sp_fib_node_get(mlxsw_sp, rt->fib6_table->tb6_id,
5855                                          &rt->fib6_dst.addr,
5856                                          sizeof(rt->fib6_dst.addr),
5857                                          rt->fib6_dst.plen,
5858                                          MLXSW_SP_L3_PROTO_IPV6);
5859         if (IS_ERR(fib_node))
5860                 return PTR_ERR(fib_node);
5861
5862         if (WARN_ON_ONCE(!fib_node->fib_entry)) {
5863                 mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5864                 return -EINVAL;
5865         }
5866
5867         fib6_entry = container_of(fib_node->fib_entry,
5868                                   struct mlxsw_sp_fib6_entry, common);
5869         err = mlxsw_sp_fib6_entry_nexthop_add(mlxsw_sp, op_ctx, fib6_entry, rt_arr, nrt6);
5870         if (err)
5871                 goto err_fib6_entry_nexthop_add;
5872
5873         return 0;
5874
5875 err_fib6_entry_nexthop_add:
5876         mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5877         return err;
5878 }
5879
5880 static int mlxsw_sp_router_fib6_del(struct mlxsw_sp *mlxsw_sp,
5881                                     struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
5882                                     struct fib6_info **rt_arr, unsigned int nrt6)
5883 {
5884         struct mlxsw_sp_fib6_entry *fib6_entry;
5885         struct mlxsw_sp_fib_node *fib_node;
5886         struct fib6_info *rt = rt_arr[0];
5887         int err;
5888
5889         if (mlxsw_sp->router->aborted)
5890                 return 0;
5891
5892         if (mlxsw_sp_fib6_rt_should_ignore(rt))
5893                 return 0;
5894
5895         /* Multipath routes are first added to the FIB trie and only then
5896          * notified. If we vetoed the addition, we will get a delete
5897          * notification for a route we do not have. Therefore, do not warn if
5898          * route was not found.
5899          */
5900         fib6_entry = mlxsw_sp_fib6_entry_lookup(mlxsw_sp, rt);
5901         if (!fib6_entry)
5902                 return 0;
5903
5904         /* If not all the nexthops are deleted, then only reduce the nexthop
5905          * group.
5906          */
5907         if (nrt6 != fib6_entry->nrt6) {
5908                 mlxsw_sp_fib6_entry_nexthop_del(mlxsw_sp, op_ctx, fib6_entry, rt_arr, nrt6);
5909                 return 0;
5910         }
5911
5912         fib_node = fib6_entry->common.fib_node;
5913
5914         err = __mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, op_ctx, &fib6_entry->common);
5915         mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
5916         mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
5917         return err;
5918 }
5919
5920 static int __mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp,
5921                                             enum mlxsw_sp_l3proto proto,
5922                                             u8 tree_id)
5923 {
5924         const struct mlxsw_sp_router_ll_ops *ll_ops = mlxsw_sp->router->proto_ll_ops[proto];
5925         enum mlxsw_reg_ralxx_protocol ralxx_proto =
5926                                 (enum mlxsw_reg_ralxx_protocol) proto;
5927         struct mlxsw_sp_fib_entry_priv *priv;
5928         char xralta_pl[MLXSW_REG_XRALTA_LEN];
5929         char xralst_pl[MLXSW_REG_XRALST_LEN];
5930         int i, err;
5931
5932         mlxsw_reg_xralta_pack(xralta_pl, true, ralxx_proto, tree_id);
5933         err = ll_ops->ralta_write(mlxsw_sp, xralta_pl);
5934         if (err)
5935                 return err;
5936
5937         mlxsw_reg_xralst_pack(xralst_pl, 0xff, tree_id);
5938         err = ll_ops->ralst_write(mlxsw_sp, xralst_pl);
5939         if (err)
5940                 return err;
5941
5942         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
5943                 struct mlxsw_sp_fib_entry_op_ctx *op_ctx = mlxsw_sp->router->ll_op_ctx;
5944                 struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
5945                 char xraltb_pl[MLXSW_REG_XRALTB_LEN];
5946
5947                 mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
5948                 mlxsw_reg_xraltb_pack(xraltb_pl, vr->id, ralxx_proto, tree_id);
5949                 err = ll_ops->raltb_write(mlxsw_sp, xraltb_pl);
5950                 if (err)
5951                         return err;
5952
5953                 priv = mlxsw_sp_fib_entry_priv_create(ll_ops);
5954                 if (IS_ERR(priv))
5955                         return PTR_ERR(priv);
5956
5957                 ll_ops->fib_entry_pack(op_ctx, proto, MLXSW_SP_FIB_ENTRY_OP_WRITE,
5958                                        vr->id, 0, NULL, priv);
5959                 ll_ops->fib_entry_act_ip2me_pack(op_ctx);
5960                 err = ll_ops->fib_entry_commit(mlxsw_sp, op_ctx, NULL);
5961                 mlxsw_sp_fib_entry_priv_put(priv);
5962                 if (err)
5963                         return err;
5964         }
5965
5966         return 0;
5967 }
5968
5969 static struct mlxsw_sp_mr_table *
5970 mlxsw_sp_router_fibmr_family_to_table(struct mlxsw_sp_vr *vr, int family)
5971 {
5972         if (family == RTNL_FAMILY_IPMR)
5973                 return vr->mr_table[MLXSW_SP_L3_PROTO_IPV4];
5974         else
5975                 return vr->mr_table[MLXSW_SP_L3_PROTO_IPV6];
5976 }
5977
5978 static int mlxsw_sp_router_fibmr_add(struct mlxsw_sp *mlxsw_sp,
5979                                      struct mfc_entry_notifier_info *men_info,
5980                                      bool replace)
5981 {
5982         struct mlxsw_sp_mr_table *mrt;
5983         struct mlxsw_sp_vr *vr;
5984
5985         if (mlxsw_sp->router->aborted)
5986                 return 0;
5987
5988         vr = mlxsw_sp_vr_get(mlxsw_sp, men_info->tb_id, NULL);
5989         if (IS_ERR(vr))
5990                 return PTR_ERR(vr);
5991
5992         mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
5993         return mlxsw_sp_mr_route_add(mrt, men_info->mfc, replace);
5994 }
5995
5996 static void mlxsw_sp_router_fibmr_del(struct mlxsw_sp *mlxsw_sp,
5997                                       struct mfc_entry_notifier_info *men_info)
5998 {
5999         struct mlxsw_sp_mr_table *mrt;
6000         struct mlxsw_sp_vr *vr;
6001
6002         if (mlxsw_sp->router->aborted)
6003                 return;
6004
6005         vr = mlxsw_sp_vr_find(mlxsw_sp, men_info->tb_id);
6006         if (WARN_ON(!vr))
6007                 return;
6008
6009         mrt = mlxsw_sp_router_fibmr_family_to_table(vr, men_info->info.family);
6010         mlxsw_sp_mr_route_del(mrt, men_info->mfc);
6011         mlxsw_sp_vr_put(mlxsw_sp, vr);
6012 }
6013
6014 static int
6015 mlxsw_sp_router_fibmr_vif_add(struct mlxsw_sp *mlxsw_sp,
6016                               struct vif_entry_notifier_info *ven_info)
6017 {
6018         struct mlxsw_sp_mr_table *mrt;
6019         struct mlxsw_sp_rif *rif;
6020         struct mlxsw_sp_vr *vr;
6021
6022         if (mlxsw_sp->router->aborted)
6023                 return 0;
6024
6025         vr = mlxsw_sp_vr_get(mlxsw_sp, ven_info->tb_id, NULL);
6026         if (IS_ERR(vr))
6027                 return PTR_ERR(vr);
6028
6029         mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
6030         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, ven_info->dev);
6031         return mlxsw_sp_mr_vif_add(mrt, ven_info->dev,
6032                                    ven_info->vif_index,
6033                                    ven_info->vif_flags, rif);
6034 }
6035
6036 static void
6037 mlxsw_sp_router_fibmr_vif_del(struct mlxsw_sp *mlxsw_sp,
6038                               struct vif_entry_notifier_info *ven_info)
6039 {
6040         struct mlxsw_sp_mr_table *mrt;
6041         struct mlxsw_sp_vr *vr;
6042
6043         if (mlxsw_sp->router->aborted)
6044                 return;
6045
6046         vr = mlxsw_sp_vr_find(mlxsw_sp, ven_info->tb_id);
6047         if (WARN_ON(!vr))
6048                 return;
6049
6050         mrt = mlxsw_sp_router_fibmr_family_to_table(vr, ven_info->info.family);
6051         mlxsw_sp_mr_vif_del(mrt, ven_info->vif_index);
6052         mlxsw_sp_vr_put(mlxsw_sp, vr);
6053 }
6054
6055 static int mlxsw_sp_router_set_abort_trap(struct mlxsw_sp *mlxsw_sp)
6056 {
6057         enum mlxsw_sp_l3proto proto = MLXSW_SP_L3_PROTO_IPV4;
6058         int err;
6059
6060         err = __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
6061                                                MLXSW_SP_LPM_TREE_MIN);
6062         if (err)
6063                 return err;
6064
6065         /* The multicast router code does not need an abort trap as by default,
6066          * packets that don't match any routes are trapped to the CPU.
6067          */
6068
6069         proto = MLXSW_SP_L3_PROTO_IPV6;
6070         return __mlxsw_sp_router_set_abort_trap(mlxsw_sp, proto,
6071                                                 MLXSW_SP_LPM_TREE_MIN + 1);
6072 }
6073
6074 static void mlxsw_sp_fib4_node_flush(struct mlxsw_sp *mlxsw_sp,
6075                                      struct mlxsw_sp_fib_node *fib_node)
6076 {
6077         struct mlxsw_sp_fib4_entry *fib4_entry;
6078
6079         fib4_entry = container_of(fib_node->fib_entry,
6080                                   struct mlxsw_sp_fib4_entry, common);
6081         mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
6082         mlxsw_sp_fib4_entry_destroy(mlxsw_sp, fib4_entry);
6083         mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
6084 }
6085
6086 static void mlxsw_sp_fib6_node_flush(struct mlxsw_sp *mlxsw_sp,
6087                                      struct mlxsw_sp_fib_node *fib_node)
6088 {
6089         struct mlxsw_sp_fib6_entry *fib6_entry;
6090
6091         fib6_entry = container_of(fib_node->fib_entry,
6092                                   struct mlxsw_sp_fib6_entry, common);
6093         mlxsw_sp_fib_node_entry_unlink(mlxsw_sp, fib_node->fib_entry);
6094         mlxsw_sp_fib6_entry_destroy(mlxsw_sp, fib6_entry);
6095         mlxsw_sp_fib_node_put(mlxsw_sp, fib_node);
6096 }
6097
6098 static void mlxsw_sp_fib_node_flush(struct mlxsw_sp *mlxsw_sp,
6099                                     struct mlxsw_sp_fib_node *fib_node)
6100 {
6101         switch (fib_node->fib->proto) {
6102         case MLXSW_SP_L3_PROTO_IPV4:
6103                 mlxsw_sp_fib4_node_flush(mlxsw_sp, fib_node);
6104                 break;
6105         case MLXSW_SP_L3_PROTO_IPV6:
6106                 mlxsw_sp_fib6_node_flush(mlxsw_sp, fib_node);
6107                 break;
6108         }
6109 }
6110
6111 static void mlxsw_sp_vr_fib_flush(struct mlxsw_sp *mlxsw_sp,
6112                                   struct mlxsw_sp_vr *vr,
6113                                   enum mlxsw_sp_l3proto proto)
6114 {
6115         struct mlxsw_sp_fib *fib = mlxsw_sp_vr_fib(vr, proto);
6116         struct mlxsw_sp_fib_node *fib_node, *tmp;
6117
6118         list_for_each_entry_safe(fib_node, tmp, &fib->node_list, list) {
6119                 bool do_break = &tmp->list == &fib->node_list;
6120
6121                 mlxsw_sp_fib_node_flush(mlxsw_sp, fib_node);
6122                 if (do_break)
6123                         break;
6124         }
6125 }
6126
6127 static void mlxsw_sp_router_fib_flush(struct mlxsw_sp *mlxsw_sp)
6128 {
6129         int i, j;
6130
6131         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_VRS); i++) {
6132                 struct mlxsw_sp_vr *vr = &mlxsw_sp->router->vrs[i];
6133
6134                 if (!mlxsw_sp_vr_is_used(vr))
6135                         continue;
6136
6137                 for (j = 0; j < MLXSW_SP_L3_PROTO_MAX; j++)
6138                         mlxsw_sp_mr_table_flush(vr->mr_table[j]);
6139                 mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV4);
6140
6141                 /* If virtual router was only used for IPv4, then it's no
6142                  * longer used.
6143                  */
6144                 if (!mlxsw_sp_vr_is_used(vr))
6145                         continue;
6146                 mlxsw_sp_vr_fib_flush(mlxsw_sp, vr, MLXSW_SP_L3_PROTO_IPV6);
6147         }
6148
6149         /* After flushing all the routes, it is not possible anyone is still
6150          * using the adjacency index that is discarding packets, so free it in
6151          * case it was allocated.
6152          */
6153         if (!mlxsw_sp->router->adj_discard_index_valid)
6154                 return;
6155         mlxsw_sp_kvdl_free(mlxsw_sp, MLXSW_SP_KVDL_ENTRY_TYPE_ADJ, 1,
6156                            mlxsw_sp->router->adj_discard_index);
6157         mlxsw_sp->router->adj_discard_index_valid = false;
6158 }
6159
6160 static void mlxsw_sp_router_fib_abort(struct mlxsw_sp *mlxsw_sp)
6161 {
6162         int err;
6163
6164         if (mlxsw_sp->router->aborted)
6165                 return;
6166         dev_warn(mlxsw_sp->bus_info->dev, "FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.\n");
6167         mlxsw_sp_router_fib_flush(mlxsw_sp);
6168         mlxsw_sp->router->aborted = true;
6169         err = mlxsw_sp_router_set_abort_trap(mlxsw_sp);
6170         if (err)
6171                 dev_warn(mlxsw_sp->bus_info->dev, "Failed to set abort trap.\n");
6172 }
6173
6174 struct mlxsw_sp_fib6_event {
6175         struct fib6_info **rt_arr;
6176         unsigned int nrt6;
6177 };
6178
6179 struct mlxsw_sp_fib_event {
6180         struct list_head list; /* node in fib queue */
6181         union {
6182                 struct mlxsw_sp_fib6_event fib6_event;
6183                 struct fib_entry_notifier_info fen_info;
6184                 struct fib_rule_notifier_info fr_info;
6185                 struct fib_nh_notifier_info fnh_info;
6186                 struct mfc_entry_notifier_info men_info;
6187                 struct vif_entry_notifier_info ven_info;
6188         };
6189         struct mlxsw_sp *mlxsw_sp;
6190         unsigned long event;
6191         int family;
6192 };
6193
6194 static int
6195 mlxsw_sp_router_fib6_event_init(struct mlxsw_sp_fib6_event *fib6_event,
6196                                 struct fib6_entry_notifier_info *fen6_info)
6197 {
6198         struct fib6_info *rt = fen6_info->rt;
6199         struct fib6_info **rt_arr;
6200         struct fib6_info *iter;
6201         unsigned int nrt6;
6202         int i = 0;
6203
6204         nrt6 = fen6_info->nsiblings + 1;
6205
6206         rt_arr = kcalloc(nrt6, sizeof(struct fib6_info *), GFP_ATOMIC);
6207         if (!rt_arr)
6208                 return -ENOMEM;
6209
6210         fib6_event->rt_arr = rt_arr;
6211         fib6_event->nrt6 = nrt6;
6212
6213         rt_arr[0] = rt;
6214         fib6_info_hold(rt);
6215
6216         if (!fen6_info->nsiblings)
6217                 return 0;
6218
6219         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
6220                 if (i == fen6_info->nsiblings)
6221                         break;
6222
6223                 rt_arr[i + 1] = iter;
6224                 fib6_info_hold(iter);
6225                 i++;
6226         }
6227         WARN_ON_ONCE(i != fen6_info->nsiblings);
6228
6229         return 0;
6230 }
6231
6232 static void
6233 mlxsw_sp_router_fib6_event_fini(struct mlxsw_sp_fib6_event *fib6_event)
6234 {
6235         int i;
6236
6237         for (i = 0; i < fib6_event->nrt6; i++)
6238                 mlxsw_sp_rt6_release(fib6_event->rt_arr[i]);
6239         kfree(fib6_event->rt_arr);
6240 }
6241
6242 static void mlxsw_sp_router_fib4_event_process(struct mlxsw_sp *mlxsw_sp,
6243                                                struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
6244                                                struct mlxsw_sp_fib_event *fib_event)
6245 {
6246         int err;
6247
6248         mlxsw_sp_span_respin(mlxsw_sp);
6249
6250         switch (fib_event->event) {
6251         case FIB_EVENT_ENTRY_REPLACE:
6252                 err = mlxsw_sp_router_fib4_replace(mlxsw_sp, op_ctx, &fib_event->fen_info);
6253                 if (err) {
6254                         mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6255                         mlxsw_sp_router_fib_abort(mlxsw_sp);
6256                 }
6257                 fib_info_put(fib_event->fen_info.fi);
6258                 break;
6259         case FIB_EVENT_ENTRY_DEL:
6260                 err = mlxsw_sp_router_fib4_del(mlxsw_sp, op_ctx, &fib_event->fen_info);
6261                 if (err)
6262                         mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6263                 fib_info_put(fib_event->fen_info.fi);
6264                 break;
6265         case FIB_EVENT_NH_ADD:
6266         case FIB_EVENT_NH_DEL:
6267                 mlxsw_sp_nexthop4_event(mlxsw_sp, fib_event->event, fib_event->fnh_info.fib_nh);
6268                 fib_info_put(fib_event->fnh_info.fib_nh->nh_parent);
6269                 break;
6270         }
6271 }
6272
6273 static void mlxsw_sp_router_fib6_event_process(struct mlxsw_sp *mlxsw_sp,
6274                                                struct mlxsw_sp_fib_entry_op_ctx *op_ctx,
6275                                                struct mlxsw_sp_fib_event *fib_event)
6276 {
6277         int err;
6278
6279         mlxsw_sp_span_respin(mlxsw_sp);
6280
6281         switch (fib_event->event) {
6282         case FIB_EVENT_ENTRY_REPLACE:
6283                 err = mlxsw_sp_router_fib6_replace(mlxsw_sp, op_ctx, fib_event->fib6_event.rt_arr,
6284                                                    fib_event->fib6_event.nrt6);
6285                 if (err) {
6286                         mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6287                         mlxsw_sp_router_fib_abort(mlxsw_sp);
6288                 }
6289                 mlxsw_sp_router_fib6_event_fini(&fib_event->fib6_event);
6290                 break;
6291         case FIB_EVENT_ENTRY_APPEND:
6292                 err = mlxsw_sp_router_fib6_append(mlxsw_sp, op_ctx, fib_event->fib6_event.rt_arr,
6293                                                   fib_event->fib6_event.nrt6);
6294                 if (err) {
6295                         mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6296                         mlxsw_sp_router_fib_abort(mlxsw_sp);
6297                 }
6298                 mlxsw_sp_router_fib6_event_fini(&fib_event->fib6_event);
6299                 break;
6300         case FIB_EVENT_ENTRY_DEL:
6301                 err = mlxsw_sp_router_fib6_del(mlxsw_sp, op_ctx, fib_event->fib6_event.rt_arr,
6302                                                fib_event->fib6_event.nrt6);
6303                 if (err)
6304                         mlxsw_sp_fib_entry_op_ctx_priv_put_all(op_ctx);
6305                 mlxsw_sp_router_fib6_event_fini(&fib_event->fib6_event);
6306                 break;
6307         }
6308 }
6309
6310 static void mlxsw_sp_router_fibmr_event_process(struct mlxsw_sp *mlxsw_sp,
6311                                                 struct mlxsw_sp_fib_event *fib_event)
6312 {
6313         bool replace;
6314         int err;
6315
6316         rtnl_lock();
6317         mutex_lock(&mlxsw_sp->router->lock);
6318         switch (fib_event->event) {
6319         case FIB_EVENT_ENTRY_REPLACE:
6320         case FIB_EVENT_ENTRY_ADD:
6321                 replace = fib_event->event == FIB_EVENT_ENTRY_REPLACE;
6322
6323                 err = mlxsw_sp_router_fibmr_add(mlxsw_sp, &fib_event->men_info, replace);
6324                 if (err)
6325                         mlxsw_sp_router_fib_abort(mlxsw_sp);
6326                 mr_cache_put(fib_event->men_info.mfc);
6327                 break;
6328         case FIB_EVENT_ENTRY_DEL:
6329                 mlxsw_sp_router_fibmr_del(mlxsw_sp, &fib_event->men_info);
6330                 mr_cache_put(fib_event->men_info.mfc);
6331                 break;
6332         case FIB_EVENT_VIF_ADD:
6333                 err = mlxsw_sp_router_fibmr_vif_add(mlxsw_sp,
6334                                                     &fib_event->ven_info);
6335                 if (err)
6336                         mlxsw_sp_router_fib_abort(mlxsw_sp);
6337                 dev_put(fib_event->ven_info.dev);
6338                 break;
6339         case FIB_EVENT_VIF_DEL:
6340                 mlxsw_sp_router_fibmr_vif_del(mlxsw_sp, &fib_event->ven_info);
6341                 dev_put(fib_event->ven_info.dev);
6342                 break;
6343         }
6344         mutex_unlock(&mlxsw_sp->router->lock);
6345         rtnl_unlock();
6346 }
6347
6348 static void mlxsw_sp_router_fib_event_work(struct work_struct *work)
6349 {
6350         struct mlxsw_sp_router *router = container_of(work, struct mlxsw_sp_router, fib_event_work);
6351         struct mlxsw_sp_fib_entry_op_ctx *op_ctx = router->ll_op_ctx;
6352         struct mlxsw_sp *mlxsw_sp = router->mlxsw_sp;
6353         struct mlxsw_sp_fib_event *next_fib_event;
6354         struct mlxsw_sp_fib_event *fib_event;
6355         int last_family = AF_UNSPEC;
6356         LIST_HEAD(fib_event_queue);
6357
6358         spin_lock_bh(&router->fib_event_queue_lock);
6359         list_splice_init(&router->fib_event_queue, &fib_event_queue);
6360         spin_unlock_bh(&router->fib_event_queue_lock);
6361
6362         /* Router lock is held here to make sure per-instance
6363          * operation context is not used in between FIB4/6 events
6364          * processing.
6365          */
6366         mutex_lock(&router->lock);
6367         mlxsw_sp_fib_entry_op_ctx_clear(op_ctx);
6368         list_for_each_entry_safe(fib_event, next_fib_event,
6369                                  &fib_event_queue, list) {
6370                 /* Check if the next entry in the queue exists and it is
6371                  * of the same type (family and event) as the currect one.
6372                  * In that case it is permitted to do the bulking
6373                  * of multiple FIB entries to a single register write.
6374                  */
6375                 op_ctx->bulk_ok = !list_is_last(&fib_event->list, &fib_event_queue) &&
6376                                   fib_event->family == next_fib_event->family &&
6377                                   fib_event->event == next_fib_event->event;
6378
6379                 /* In case family of this and the previous entry are different, context
6380                  * reinitialization is going to be needed now, indicate that.
6381                  * Note that since last_family is initialized to AF_UNSPEC, this is always
6382                  * going to happen for the first entry processed in the work.
6383                  */
6384                 if (fib_event->family != last_family)
6385                         op_ctx->initialized = false;
6386
6387                 switch (fib_event->family) {
6388                 case AF_INET:
6389                         mlxsw_sp_router_fib4_event_process(mlxsw_sp, op_ctx,
6390                                                            fib_event);
6391                         break;
6392                 case AF_INET6:
6393                         mlxsw_sp_router_fib6_event_process(mlxsw_sp, op_ctx,
6394                                                            fib_event);
6395                         break;
6396                 case RTNL_FAMILY_IP6MR:
6397                 case RTNL_FAMILY_IPMR:
6398                         /* Unlock here as inside FIBMR the lock is taken again
6399                          * under RTNL. The per-instance operation context
6400                          * is not used by FIBMR.
6401                          */
6402                         mutex_unlock(&router->lock);
6403                         mlxsw_sp_router_fibmr_event_process(mlxsw_sp,
6404                                                             fib_event);
6405                         mutex_lock(&router->lock);
6406                         break;
6407                 default:
6408                         WARN_ON_ONCE(1);
6409                 }
6410                 last_family = fib_event->family;
6411                 kfree(fib_event);
6412                 cond_resched();
6413         }
6414         WARN_ON_ONCE(!list_empty(&router->ll_op_ctx->fib_entry_priv_list));
6415         mutex_unlock(&router->lock);
6416 }
6417
6418 static void mlxsw_sp_router_fib4_event(struct mlxsw_sp_fib_event *fib_event,
6419                                        struct fib_notifier_info *info)
6420 {
6421         struct fib_entry_notifier_info *fen_info;
6422         struct fib_nh_notifier_info *fnh_info;
6423
6424         switch (fib_event->event) {
6425         case FIB_EVENT_ENTRY_REPLACE:
6426         case FIB_EVENT_ENTRY_DEL:
6427                 fen_info = container_of(info, struct fib_entry_notifier_info,
6428                                         info);
6429                 fib_event->fen_info = *fen_info;
6430                 /* Take reference on fib_info to prevent it from being
6431                  * freed while event is queued. Release it afterwards.
6432                  */
6433                 fib_info_hold(fib_event->fen_info.fi);
6434                 break;
6435         case FIB_EVENT_NH_ADD:
6436         case FIB_EVENT_NH_DEL:
6437                 fnh_info = container_of(info, struct fib_nh_notifier_info,
6438                                         info);
6439                 fib_event->fnh_info = *fnh_info;
6440                 fib_info_hold(fib_event->fnh_info.fib_nh->nh_parent);
6441                 break;
6442         }
6443 }
6444
6445 static int mlxsw_sp_router_fib6_event(struct mlxsw_sp_fib_event *fib_event,
6446                                       struct fib_notifier_info *info)
6447 {
6448         struct fib6_entry_notifier_info *fen6_info;
6449         int err;
6450
6451         switch (fib_event->event) {
6452         case FIB_EVENT_ENTRY_REPLACE:
6453         case FIB_EVENT_ENTRY_APPEND:
6454         case FIB_EVENT_ENTRY_DEL:
6455                 fen6_info = container_of(info, struct fib6_entry_notifier_info,
6456                                          info);
6457                 err = mlxsw_sp_router_fib6_event_init(&fib_event->fib6_event,
6458                                                       fen6_info);
6459                 if (err)
6460                         return err;
6461                 break;
6462         }
6463
6464         return 0;
6465 }
6466
6467 static void
6468 mlxsw_sp_router_fibmr_event(struct mlxsw_sp_fib_event *fib_event,
6469                             struct fib_notifier_info *info)
6470 {
6471         switch (fib_event->event) {
6472         case FIB_EVENT_ENTRY_REPLACE:
6473         case FIB_EVENT_ENTRY_ADD:
6474         case FIB_EVENT_ENTRY_DEL:
6475                 memcpy(&fib_event->men_info, info, sizeof(fib_event->men_info));
6476                 mr_cache_hold(fib_event->men_info.mfc);
6477                 break;
6478         case FIB_EVENT_VIF_ADD:
6479         case FIB_EVENT_VIF_DEL:
6480                 memcpy(&fib_event->ven_info, info, sizeof(fib_event->ven_info));
6481                 dev_hold(fib_event->ven_info.dev);
6482                 break;
6483         }
6484 }
6485
6486 static int mlxsw_sp_router_fib_rule_event(unsigned long event,
6487                                           struct fib_notifier_info *info,
6488                                           struct mlxsw_sp *mlxsw_sp)
6489 {
6490         struct netlink_ext_ack *extack = info->extack;
6491         struct fib_rule_notifier_info *fr_info;
6492         struct fib_rule *rule;
6493         int err = 0;
6494
6495         /* nothing to do at the moment */
6496         if (event == FIB_EVENT_RULE_DEL)
6497                 return 0;
6498
6499         if (mlxsw_sp->router->aborted)
6500                 return 0;
6501
6502         fr_info = container_of(info, struct fib_rule_notifier_info, info);
6503         rule = fr_info->rule;
6504
6505         /* Rule only affects locally generated traffic */
6506         if (rule->iifindex == mlxsw_sp_net(mlxsw_sp)->loopback_dev->ifindex)
6507                 return 0;
6508
6509         switch (info->family) {
6510         case AF_INET:
6511                 if (!fib4_rule_default(rule) && !rule->l3mdev)
6512                         err = -EOPNOTSUPP;
6513                 break;
6514         case AF_INET6:
6515                 if (!fib6_rule_default(rule) && !rule->l3mdev)
6516                         err = -EOPNOTSUPP;
6517                 break;
6518         case RTNL_FAMILY_IPMR:
6519                 if (!ipmr_rule_default(rule) && !rule->l3mdev)
6520                         err = -EOPNOTSUPP;
6521                 break;
6522         case RTNL_FAMILY_IP6MR:
6523                 if (!ip6mr_rule_default(rule) && !rule->l3mdev)
6524                         err = -EOPNOTSUPP;
6525                 break;
6526         }
6527
6528         if (err < 0)
6529                 NL_SET_ERR_MSG_MOD(extack, "FIB rules not supported");
6530
6531         return err;
6532 }
6533
6534 /* Called with rcu_read_lock() */
6535 static int mlxsw_sp_router_fib_event(struct notifier_block *nb,
6536                                      unsigned long event, void *ptr)
6537 {
6538         struct mlxsw_sp_fib_event *fib_event;
6539         struct fib_notifier_info *info = ptr;
6540         struct mlxsw_sp_router *router;
6541         int err;
6542
6543         if ((info->family != AF_INET && info->family != AF_INET6 &&
6544              info->family != RTNL_FAMILY_IPMR &&
6545              info->family != RTNL_FAMILY_IP6MR))
6546                 return NOTIFY_DONE;
6547
6548         router = container_of(nb, struct mlxsw_sp_router, fib_nb);
6549
6550         switch (event) {
6551         case FIB_EVENT_RULE_ADD:
6552         case FIB_EVENT_RULE_DEL:
6553                 err = mlxsw_sp_router_fib_rule_event(event, info,
6554                                                      router->mlxsw_sp);
6555                 return notifier_from_errno(err);
6556         case FIB_EVENT_ENTRY_ADD:
6557         case FIB_EVENT_ENTRY_REPLACE:
6558         case FIB_EVENT_ENTRY_APPEND:
6559                 if (router->aborted) {
6560                         NL_SET_ERR_MSG_MOD(info->extack, "FIB offload was aborted. Not configuring route");
6561                         return notifier_from_errno(-EINVAL);
6562                 }
6563                 if (info->family == AF_INET) {
6564                         struct fib_entry_notifier_info *fen_info = ptr;
6565
6566                         if (fen_info->fi->fib_nh_is_v6) {
6567                                 NL_SET_ERR_MSG_MOD(info->extack, "IPv6 gateway with IPv4 route is not supported");
6568                                 return notifier_from_errno(-EINVAL);
6569                         }
6570                         if (fen_info->fi->nh) {
6571                                 NL_SET_ERR_MSG_MOD(info->extack, "IPv4 route with nexthop objects is not supported");
6572                                 return notifier_from_errno(-EINVAL);
6573                         }
6574                 } else if (info->family == AF_INET6) {
6575                         struct fib6_entry_notifier_info *fen6_info;
6576
6577                         fen6_info = container_of(info,
6578                                                  struct fib6_entry_notifier_info,
6579                                                  info);
6580                         if (fen6_info->rt->nh) {
6581                                 NL_SET_ERR_MSG_MOD(info->extack, "IPv6 route with nexthop objects is not supported");
6582                                 return notifier_from_errno(-EINVAL);
6583                         }
6584                 }
6585                 break;
6586         }
6587
6588         fib_event = kzalloc(sizeof(*fib_event), GFP_ATOMIC);
6589         if (!fib_event)
6590                 return NOTIFY_BAD;
6591
6592         fib_event->mlxsw_sp = router->mlxsw_sp;
6593         fib_event->event = event;
6594         fib_event->family = info->family;
6595
6596         switch (info->family) {
6597         case AF_INET:
6598                 mlxsw_sp_router_fib4_event(fib_event, info);
6599                 break;
6600         case AF_INET6:
6601                 err = mlxsw_sp_router_fib6_event(fib_event, info);
6602                 if (err)
6603                         goto err_fib_event;
6604                 break;
6605         case RTNL_FAMILY_IP6MR:
6606         case RTNL_FAMILY_IPMR:
6607                 mlxsw_sp_router_fibmr_event(fib_event, info);
6608                 break;
6609         }
6610
6611         /* Enqueue the event and trigger the work */
6612         spin_lock_bh(&router->fib_event_queue_lock);
6613         list_add_tail(&fib_event->list, &router->fib_event_queue);
6614         spin_unlock_bh(&router->fib_event_queue_lock);
6615         mlxsw_core_schedule_work(&router->fib_event_work);
6616
6617         return NOTIFY_DONE;
6618
6619 err_fib_event:
6620         kfree(fib_event);
6621         return NOTIFY_BAD;
6622 }
6623
6624 static struct mlxsw_sp_rif *
6625 mlxsw_sp_rif_find_by_dev(const struct mlxsw_sp *mlxsw_sp,
6626                          const struct net_device *dev)
6627 {
6628         int i;
6629
6630         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
6631                 if (mlxsw_sp->router->rifs[i] &&
6632                     mlxsw_sp->router->rifs[i]->dev == dev)
6633                         return mlxsw_sp->router->rifs[i];
6634
6635         return NULL;
6636 }
6637
6638 bool mlxsw_sp_rif_exists(struct mlxsw_sp *mlxsw_sp,
6639                          const struct net_device *dev)
6640 {
6641         struct mlxsw_sp_rif *rif;
6642
6643         mutex_lock(&mlxsw_sp->router->lock);
6644         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6645         mutex_unlock(&mlxsw_sp->router->lock);
6646
6647         return rif;
6648 }
6649
6650 u16 mlxsw_sp_rif_vid(struct mlxsw_sp *mlxsw_sp, const struct net_device *dev)
6651 {
6652         struct mlxsw_sp_rif *rif;
6653         u16 vid = 0;
6654
6655         mutex_lock(&mlxsw_sp->router->lock);
6656         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6657         if (!rif)
6658                 goto out;
6659
6660         /* We only return the VID for VLAN RIFs. Otherwise we return an
6661          * invalid value (0).
6662          */
6663         if (rif->ops->type != MLXSW_SP_RIF_TYPE_VLAN)
6664                 goto out;
6665
6666         vid = mlxsw_sp_fid_8021q_vid(rif->fid);
6667
6668 out:
6669         mutex_unlock(&mlxsw_sp->router->lock);
6670         return vid;
6671 }
6672
6673 static int mlxsw_sp_router_rif_disable(struct mlxsw_sp *mlxsw_sp, u16 rif)
6674 {
6675         char ritr_pl[MLXSW_REG_RITR_LEN];
6676         int err;
6677
6678         mlxsw_reg_ritr_rif_pack(ritr_pl, rif);
6679         err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6680         if (err)
6681                 return err;
6682
6683         mlxsw_reg_ritr_enable_set(ritr_pl, false);
6684         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
6685 }
6686
6687 static void mlxsw_sp_router_rif_gone_sync(struct mlxsw_sp *mlxsw_sp,
6688                                           struct mlxsw_sp_rif *rif)
6689 {
6690         mlxsw_sp_router_rif_disable(mlxsw_sp, rif->rif_index);
6691         mlxsw_sp_nexthop_rif_gone_sync(mlxsw_sp, rif);
6692         mlxsw_sp_neigh_rif_gone_sync(mlxsw_sp, rif);
6693 }
6694
6695 static bool
6696 mlxsw_sp_rif_should_config(struct mlxsw_sp_rif *rif, struct net_device *dev,
6697                            unsigned long event)
6698 {
6699         struct inet6_dev *inet6_dev;
6700         bool addr_list_empty = true;
6701         struct in_device *idev;
6702
6703         switch (event) {
6704         case NETDEV_UP:
6705                 return rif == NULL;
6706         case NETDEV_DOWN:
6707                 rcu_read_lock();
6708                 idev = __in_dev_get_rcu(dev);
6709                 if (idev && idev->ifa_list)
6710                         addr_list_empty = false;
6711
6712                 inet6_dev = __in6_dev_get(dev);
6713                 if (addr_list_empty && inet6_dev &&
6714                     !list_empty(&inet6_dev->addr_list))
6715                         addr_list_empty = false;
6716                 rcu_read_unlock();
6717
6718                 /* macvlans do not have a RIF, but rather piggy back on the
6719                  * RIF of their lower device.
6720                  */
6721                 if (netif_is_macvlan(dev) && addr_list_empty)
6722                         return true;
6723
6724                 if (rif && addr_list_empty &&
6725                     !netif_is_l3_slave(rif->dev))
6726                         return true;
6727                 /* It is possible we already removed the RIF ourselves
6728                  * if it was assigned to a netdev that is now a bridge
6729                  * or LAG slave.
6730                  */
6731                 return false;
6732         }
6733
6734         return false;
6735 }
6736
6737 static enum mlxsw_sp_rif_type
6738 mlxsw_sp_dev_rif_type(const struct mlxsw_sp *mlxsw_sp,
6739                       const struct net_device *dev)
6740 {
6741         enum mlxsw_sp_fid_type type;
6742
6743         if (mlxsw_sp_netdev_ipip_type(mlxsw_sp, dev, NULL))
6744                 return MLXSW_SP_RIF_TYPE_IPIP_LB;
6745
6746         /* Otherwise RIF type is derived from the type of the underlying FID. */
6747         if (is_vlan_dev(dev) && netif_is_bridge_master(vlan_dev_real_dev(dev)))
6748                 type = MLXSW_SP_FID_TYPE_8021Q;
6749         else if (netif_is_bridge_master(dev) && br_vlan_enabled(dev))
6750                 type = MLXSW_SP_FID_TYPE_8021Q;
6751         else if (netif_is_bridge_master(dev))
6752                 type = MLXSW_SP_FID_TYPE_8021D;
6753         else
6754                 type = MLXSW_SP_FID_TYPE_RFID;
6755
6756         return mlxsw_sp_fid_type_rif_type(mlxsw_sp, type);
6757 }
6758
6759 static int mlxsw_sp_rif_index_alloc(struct mlxsw_sp *mlxsw_sp, u16 *p_rif_index)
6760 {
6761         int i;
6762
6763         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
6764                 if (!mlxsw_sp->router->rifs[i]) {
6765                         *p_rif_index = i;
6766                         return 0;
6767                 }
6768         }
6769
6770         return -ENOBUFS;
6771 }
6772
6773 static struct mlxsw_sp_rif *mlxsw_sp_rif_alloc(size_t rif_size, u16 rif_index,
6774                                                u16 vr_id,
6775                                                struct net_device *l3_dev)
6776 {
6777         struct mlxsw_sp_rif *rif;
6778
6779         rif = kzalloc(rif_size, GFP_KERNEL);
6780         if (!rif)
6781                 return NULL;
6782
6783         INIT_LIST_HEAD(&rif->nexthop_list);
6784         INIT_LIST_HEAD(&rif->neigh_list);
6785         if (l3_dev) {
6786                 ether_addr_copy(rif->addr, l3_dev->dev_addr);
6787                 rif->mtu = l3_dev->mtu;
6788                 rif->dev = l3_dev;
6789         }
6790         rif->vr_id = vr_id;
6791         rif->rif_index = rif_index;
6792
6793         return rif;
6794 }
6795
6796 struct mlxsw_sp_rif *mlxsw_sp_rif_by_index(const struct mlxsw_sp *mlxsw_sp,
6797                                            u16 rif_index)
6798 {
6799         return mlxsw_sp->router->rifs[rif_index];
6800 }
6801
6802 u16 mlxsw_sp_rif_index(const struct mlxsw_sp_rif *rif)
6803 {
6804         return rif->rif_index;
6805 }
6806
6807 u16 mlxsw_sp_ipip_lb_rif_index(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6808 {
6809         return lb_rif->common.rif_index;
6810 }
6811
6812 u16 mlxsw_sp_ipip_lb_ul_vr_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6813 {
6814         u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(lb_rif->common.dev);
6815         struct mlxsw_sp_vr *ul_vr;
6816
6817         ul_vr = mlxsw_sp_vr_get(lb_rif->common.mlxsw_sp, ul_tb_id, NULL);
6818         if (WARN_ON(IS_ERR(ul_vr)))
6819                 return 0;
6820
6821         return ul_vr->id;
6822 }
6823
6824 u16 mlxsw_sp_ipip_lb_ul_rif_id(const struct mlxsw_sp_rif_ipip_lb *lb_rif)
6825 {
6826         return lb_rif->ul_rif_id;
6827 }
6828
6829 int mlxsw_sp_rif_dev_ifindex(const struct mlxsw_sp_rif *rif)
6830 {
6831         return rif->dev->ifindex;
6832 }
6833
6834 const struct net_device *mlxsw_sp_rif_dev(const struct mlxsw_sp_rif *rif)
6835 {
6836         return rif->dev;
6837 }
6838
6839 static struct mlxsw_sp_rif *
6840 mlxsw_sp_rif_create(struct mlxsw_sp *mlxsw_sp,
6841                     const struct mlxsw_sp_rif_params *params,
6842                     struct netlink_ext_ack *extack)
6843 {
6844         u32 tb_id = l3mdev_fib_table(params->dev);
6845         const struct mlxsw_sp_rif_ops *ops;
6846         struct mlxsw_sp_fid *fid = NULL;
6847         enum mlxsw_sp_rif_type type;
6848         struct mlxsw_sp_rif *rif;
6849         struct mlxsw_sp_vr *vr;
6850         u16 rif_index;
6851         int i, err;
6852
6853         type = mlxsw_sp_dev_rif_type(mlxsw_sp, params->dev);
6854         ops = mlxsw_sp->rif_ops_arr[type];
6855
6856         vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id ? : RT_TABLE_MAIN, extack);
6857         if (IS_ERR(vr))
6858                 return ERR_CAST(vr);
6859         vr->rif_count++;
6860
6861         err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
6862         if (err) {
6863                 NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
6864                 goto err_rif_index_alloc;
6865         }
6866
6867         rif = mlxsw_sp_rif_alloc(ops->rif_size, rif_index, vr->id, params->dev);
6868         if (!rif) {
6869                 err = -ENOMEM;
6870                 goto err_rif_alloc;
6871         }
6872         dev_hold(rif->dev);
6873         mlxsw_sp->router->rifs[rif_index] = rif;
6874         rif->mlxsw_sp = mlxsw_sp;
6875         rif->ops = ops;
6876
6877         if (ops->fid_get) {
6878                 fid = ops->fid_get(rif, extack);
6879                 if (IS_ERR(fid)) {
6880                         err = PTR_ERR(fid);
6881                         goto err_fid_get;
6882                 }
6883                 rif->fid = fid;
6884         }
6885
6886         if (ops->setup)
6887                 ops->setup(rif, params);
6888
6889         err = ops->configure(rif);
6890         if (err)
6891                 goto err_configure;
6892
6893         for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++) {
6894                 err = mlxsw_sp_mr_rif_add(vr->mr_table[i], rif);
6895                 if (err)
6896                         goto err_mr_rif_add;
6897         }
6898
6899         mlxsw_sp_rif_counters_alloc(rif);
6900
6901         return rif;
6902
6903 err_mr_rif_add:
6904         for (i--; i >= 0; i--)
6905                 mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
6906         ops->deconfigure(rif);
6907 err_configure:
6908         if (fid)
6909                 mlxsw_sp_fid_put(fid);
6910 err_fid_get:
6911         mlxsw_sp->router->rifs[rif_index] = NULL;
6912         dev_put(rif->dev);
6913         kfree(rif);
6914 err_rif_alloc:
6915 err_rif_index_alloc:
6916         vr->rif_count--;
6917         mlxsw_sp_vr_put(mlxsw_sp, vr);
6918         return ERR_PTR(err);
6919 }
6920
6921 static void mlxsw_sp_rif_destroy(struct mlxsw_sp_rif *rif)
6922 {
6923         const struct mlxsw_sp_rif_ops *ops = rif->ops;
6924         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
6925         struct mlxsw_sp_fid *fid = rif->fid;
6926         struct mlxsw_sp_vr *vr;
6927         int i;
6928
6929         mlxsw_sp_router_rif_gone_sync(mlxsw_sp, rif);
6930         vr = &mlxsw_sp->router->vrs[rif->vr_id];
6931
6932         mlxsw_sp_rif_counters_free(rif);
6933         for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
6934                 mlxsw_sp_mr_rif_del(vr->mr_table[i], rif);
6935         ops->deconfigure(rif);
6936         if (fid)
6937                 /* Loopback RIFs are not associated with a FID. */
6938                 mlxsw_sp_fid_put(fid);
6939         mlxsw_sp->router->rifs[rif->rif_index] = NULL;
6940         dev_put(rif->dev);
6941         kfree(rif);
6942         vr->rif_count--;
6943         mlxsw_sp_vr_put(mlxsw_sp, vr);
6944 }
6945
6946 void mlxsw_sp_rif_destroy_by_dev(struct mlxsw_sp *mlxsw_sp,
6947                                  struct net_device *dev)
6948 {
6949         struct mlxsw_sp_rif *rif;
6950
6951         mutex_lock(&mlxsw_sp->router->lock);
6952         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
6953         if (!rif)
6954                 goto out;
6955         mlxsw_sp_rif_destroy(rif);
6956 out:
6957         mutex_unlock(&mlxsw_sp->router->lock);
6958 }
6959
6960 static void
6961 mlxsw_sp_rif_subport_params_init(struct mlxsw_sp_rif_params *params,
6962                                  struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
6963 {
6964         struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
6965
6966         params->vid = mlxsw_sp_port_vlan->vid;
6967         params->lag = mlxsw_sp_port->lagged;
6968         if (params->lag)
6969                 params->lag_id = mlxsw_sp_port->lag_id;
6970         else
6971                 params->system_port = mlxsw_sp_port->local_port;
6972 }
6973
6974 static struct mlxsw_sp_rif_subport *
6975 mlxsw_sp_rif_subport_rif(const struct mlxsw_sp_rif *rif)
6976 {
6977         return container_of(rif, struct mlxsw_sp_rif_subport, common);
6978 }
6979
6980 static struct mlxsw_sp_rif *
6981 mlxsw_sp_rif_subport_get(struct mlxsw_sp *mlxsw_sp,
6982                          const struct mlxsw_sp_rif_params *params,
6983                          struct netlink_ext_ack *extack)
6984 {
6985         struct mlxsw_sp_rif_subport *rif_subport;
6986         struct mlxsw_sp_rif *rif;
6987
6988         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, params->dev);
6989         if (!rif)
6990                 return mlxsw_sp_rif_create(mlxsw_sp, params, extack);
6991
6992         rif_subport = mlxsw_sp_rif_subport_rif(rif);
6993         refcount_inc(&rif_subport->ref_count);
6994         return rif;
6995 }
6996
6997 static void mlxsw_sp_rif_subport_put(struct mlxsw_sp_rif *rif)
6998 {
6999         struct mlxsw_sp_rif_subport *rif_subport;
7000
7001         rif_subport = mlxsw_sp_rif_subport_rif(rif);
7002         if (!refcount_dec_and_test(&rif_subport->ref_count))
7003                 return;
7004
7005         mlxsw_sp_rif_destroy(rif);
7006 }
7007
7008 static int
7009 mlxsw_sp_port_vlan_router_join(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan,
7010                                struct net_device *l3_dev,
7011                                struct netlink_ext_ack *extack)
7012 {
7013         struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
7014         struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
7015         struct mlxsw_sp_rif_params params = {
7016                 .dev = l3_dev,
7017         };
7018         u16 vid = mlxsw_sp_port_vlan->vid;
7019         struct mlxsw_sp_rif *rif;
7020         struct mlxsw_sp_fid *fid;
7021         int err;
7022
7023         mlxsw_sp_rif_subport_params_init(&params, mlxsw_sp_port_vlan);
7024         rif = mlxsw_sp_rif_subport_get(mlxsw_sp, &params, extack);
7025         if (IS_ERR(rif))
7026                 return PTR_ERR(rif);
7027
7028         /* FID was already created, just take a reference */
7029         fid = rif->ops->fid_get(rif, extack);
7030         err = mlxsw_sp_fid_port_vid_map(fid, mlxsw_sp_port, vid);
7031         if (err)
7032                 goto err_fid_port_vid_map;
7033
7034         err = mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, false);
7035         if (err)
7036                 goto err_port_vid_learning_set;
7037
7038         err = mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid,
7039                                         BR_STATE_FORWARDING);
7040         if (err)
7041                 goto err_port_vid_stp_set;
7042
7043         mlxsw_sp_port_vlan->fid = fid;
7044
7045         return 0;
7046
7047 err_port_vid_stp_set:
7048         mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
7049 err_port_vid_learning_set:
7050         mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
7051 err_fid_port_vid_map:
7052         mlxsw_sp_fid_put(fid);
7053         mlxsw_sp_rif_subport_put(rif);
7054         return err;
7055 }
7056
7057 static void
7058 __mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
7059 {
7060         struct mlxsw_sp_port *mlxsw_sp_port = mlxsw_sp_port_vlan->mlxsw_sp_port;
7061         struct mlxsw_sp_fid *fid = mlxsw_sp_port_vlan->fid;
7062         struct mlxsw_sp_rif *rif = mlxsw_sp_fid_rif(fid);
7063         u16 vid = mlxsw_sp_port_vlan->vid;
7064
7065         if (WARN_ON(mlxsw_sp_fid_type(fid) != MLXSW_SP_FID_TYPE_RFID))
7066                 return;
7067
7068         mlxsw_sp_port_vlan->fid = NULL;
7069         mlxsw_sp_port_vid_stp_set(mlxsw_sp_port, vid, BR_STATE_BLOCKING);
7070         mlxsw_sp_port_vid_learning_set(mlxsw_sp_port, vid, true);
7071         mlxsw_sp_fid_port_vid_unmap(fid, mlxsw_sp_port, vid);
7072         mlxsw_sp_fid_put(fid);
7073         mlxsw_sp_rif_subport_put(rif);
7074 }
7075
7076 void
7077 mlxsw_sp_port_vlan_router_leave(struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan)
7078 {
7079         struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port_vlan->mlxsw_sp_port->mlxsw_sp;
7080
7081         mutex_lock(&mlxsw_sp->router->lock);
7082         __mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
7083         mutex_unlock(&mlxsw_sp->router->lock);
7084 }
7085
7086 static int mlxsw_sp_inetaddr_port_vlan_event(struct net_device *l3_dev,
7087                                              struct net_device *port_dev,
7088                                              unsigned long event, u16 vid,
7089                                              struct netlink_ext_ack *extack)
7090 {
7091         struct mlxsw_sp_port *mlxsw_sp_port = netdev_priv(port_dev);
7092         struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan;
7093
7094         mlxsw_sp_port_vlan = mlxsw_sp_port_vlan_find_by_vid(mlxsw_sp_port, vid);
7095         if (WARN_ON(!mlxsw_sp_port_vlan))
7096                 return -EINVAL;
7097
7098         switch (event) {
7099         case NETDEV_UP:
7100                 return mlxsw_sp_port_vlan_router_join(mlxsw_sp_port_vlan,
7101                                                       l3_dev, extack);
7102         case NETDEV_DOWN:
7103                 __mlxsw_sp_port_vlan_router_leave(mlxsw_sp_port_vlan);
7104                 break;
7105         }
7106
7107         return 0;
7108 }
7109
7110 static int mlxsw_sp_inetaddr_port_event(struct net_device *port_dev,
7111                                         unsigned long event,
7112                                         struct netlink_ext_ack *extack)
7113 {
7114         if (netif_is_bridge_port(port_dev) ||
7115             netif_is_lag_port(port_dev) ||
7116             netif_is_ovs_port(port_dev))
7117                 return 0;
7118
7119         return mlxsw_sp_inetaddr_port_vlan_event(port_dev, port_dev, event,
7120                                                  MLXSW_SP_DEFAULT_VID, extack);
7121 }
7122
7123 static int __mlxsw_sp_inetaddr_lag_event(struct net_device *l3_dev,
7124                                          struct net_device *lag_dev,
7125                                          unsigned long event, u16 vid,
7126                                          struct netlink_ext_ack *extack)
7127 {
7128         struct net_device *port_dev;
7129         struct list_head *iter;
7130         int err;
7131
7132         netdev_for_each_lower_dev(lag_dev, port_dev, iter) {
7133                 if (mlxsw_sp_port_dev_check(port_dev)) {
7134                         err = mlxsw_sp_inetaddr_port_vlan_event(l3_dev,
7135                                                                 port_dev,
7136                                                                 event, vid,
7137                                                                 extack);
7138                         if (err)
7139                                 return err;
7140                 }
7141         }
7142
7143         return 0;
7144 }
7145
7146 static int mlxsw_sp_inetaddr_lag_event(struct net_device *lag_dev,
7147                                        unsigned long event,
7148                                        struct netlink_ext_ack *extack)
7149 {
7150         if (netif_is_bridge_port(lag_dev))
7151                 return 0;
7152
7153         return __mlxsw_sp_inetaddr_lag_event(lag_dev, lag_dev, event,
7154                                              MLXSW_SP_DEFAULT_VID, extack);
7155 }
7156
7157 static int mlxsw_sp_inetaddr_bridge_event(struct mlxsw_sp *mlxsw_sp,
7158                                           struct net_device *l3_dev,
7159                                           unsigned long event,
7160                                           struct netlink_ext_ack *extack)
7161 {
7162         struct mlxsw_sp_rif_params params = {
7163                 .dev = l3_dev,
7164         };
7165         struct mlxsw_sp_rif *rif;
7166
7167         switch (event) {
7168         case NETDEV_UP:
7169                 rif = mlxsw_sp_rif_create(mlxsw_sp, &params, extack);
7170                 if (IS_ERR(rif))
7171                         return PTR_ERR(rif);
7172                 break;
7173         case NETDEV_DOWN:
7174                 rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7175                 mlxsw_sp_rif_destroy(rif);
7176                 break;
7177         }
7178
7179         return 0;
7180 }
7181
7182 static int mlxsw_sp_inetaddr_vlan_event(struct mlxsw_sp *mlxsw_sp,
7183                                         struct net_device *vlan_dev,
7184                                         unsigned long event,
7185                                         struct netlink_ext_ack *extack)
7186 {
7187         struct net_device *real_dev = vlan_dev_real_dev(vlan_dev);
7188         u16 vid = vlan_dev_vlan_id(vlan_dev);
7189
7190         if (netif_is_bridge_port(vlan_dev))
7191                 return 0;
7192
7193         if (mlxsw_sp_port_dev_check(real_dev))
7194                 return mlxsw_sp_inetaddr_port_vlan_event(vlan_dev, real_dev,
7195                                                          event, vid, extack);
7196         else if (netif_is_lag_master(real_dev))
7197                 return __mlxsw_sp_inetaddr_lag_event(vlan_dev, real_dev, event,
7198                                                      vid, extack);
7199         else if (netif_is_bridge_master(real_dev) && br_vlan_enabled(real_dev))
7200                 return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, vlan_dev, event,
7201                                                       extack);
7202
7203         return 0;
7204 }
7205
7206 static bool mlxsw_sp_rif_macvlan_is_vrrp4(const u8 *mac)
7207 {
7208         u8 vrrp4[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x01, 0x00 };
7209         u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
7210
7211         return ether_addr_equal_masked(mac, vrrp4, mask);
7212 }
7213
7214 static bool mlxsw_sp_rif_macvlan_is_vrrp6(const u8 *mac)
7215 {
7216         u8 vrrp6[ETH_ALEN] = { 0x00, 0x00, 0x5e, 0x00, 0x02, 0x00 };
7217         u8 mask[ETH_ALEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0x00 };
7218
7219         return ether_addr_equal_masked(mac, vrrp6, mask);
7220 }
7221
7222 static int mlxsw_sp_rif_vrrp_op(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
7223                                 const u8 *mac, bool adding)
7224 {
7225         char ritr_pl[MLXSW_REG_RITR_LEN];
7226         u8 vrrp_id = adding ? mac[5] : 0;
7227         int err;
7228
7229         if (!mlxsw_sp_rif_macvlan_is_vrrp4(mac) &&
7230             !mlxsw_sp_rif_macvlan_is_vrrp6(mac))
7231                 return 0;
7232
7233         mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
7234         err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7235         if (err)
7236                 return err;
7237
7238         if (mlxsw_sp_rif_macvlan_is_vrrp4(mac))
7239                 mlxsw_reg_ritr_if_vrrp_id_ipv4_set(ritr_pl, vrrp_id);
7240         else
7241                 mlxsw_reg_ritr_if_vrrp_id_ipv6_set(ritr_pl, vrrp_id);
7242
7243         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7244 }
7245
7246 static int mlxsw_sp_rif_macvlan_add(struct mlxsw_sp *mlxsw_sp,
7247                                     const struct net_device *macvlan_dev,
7248                                     struct netlink_ext_ack *extack)
7249 {
7250         struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
7251         struct mlxsw_sp_rif *rif;
7252         int err;
7253
7254         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
7255         if (!rif) {
7256                 NL_SET_ERR_MSG_MOD(extack, "macvlan is only supported on top of router interfaces");
7257                 return -EOPNOTSUPP;
7258         }
7259
7260         err = mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
7261                                   mlxsw_sp_fid_index(rif->fid), true);
7262         if (err)
7263                 return err;
7264
7265         err = mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index,
7266                                    macvlan_dev->dev_addr, true);
7267         if (err)
7268                 goto err_rif_vrrp_add;
7269
7270         /* Make sure the bridge driver does not have this MAC pointing at
7271          * some other port.
7272          */
7273         if (rif->ops->fdb_del)
7274                 rif->ops->fdb_del(rif, macvlan_dev->dev_addr);
7275
7276         return 0;
7277
7278 err_rif_vrrp_add:
7279         mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
7280                             mlxsw_sp_fid_index(rif->fid), false);
7281         return err;
7282 }
7283
7284 static void __mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
7285                                        const struct net_device *macvlan_dev)
7286 {
7287         struct macvlan_dev *vlan = netdev_priv(macvlan_dev);
7288         struct mlxsw_sp_rif *rif;
7289
7290         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, vlan->lowerdev);
7291         /* If we do not have a RIF, then we already took care of
7292          * removing the macvlan's MAC during RIF deletion.
7293          */
7294         if (!rif)
7295                 return;
7296         mlxsw_sp_rif_vrrp_op(mlxsw_sp, rif->rif_index, macvlan_dev->dev_addr,
7297                              false);
7298         mlxsw_sp_rif_fdb_op(mlxsw_sp, macvlan_dev->dev_addr,
7299                             mlxsw_sp_fid_index(rif->fid), false);
7300 }
7301
7302 void mlxsw_sp_rif_macvlan_del(struct mlxsw_sp *mlxsw_sp,
7303                               const struct net_device *macvlan_dev)
7304 {
7305         mutex_lock(&mlxsw_sp->router->lock);
7306         __mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
7307         mutex_unlock(&mlxsw_sp->router->lock);
7308 }
7309
7310 static int mlxsw_sp_inetaddr_macvlan_event(struct mlxsw_sp *mlxsw_sp,
7311                                            struct net_device *macvlan_dev,
7312                                            unsigned long event,
7313                                            struct netlink_ext_ack *extack)
7314 {
7315         switch (event) {
7316         case NETDEV_UP:
7317                 return mlxsw_sp_rif_macvlan_add(mlxsw_sp, macvlan_dev, extack);
7318         case NETDEV_DOWN:
7319                 __mlxsw_sp_rif_macvlan_del(mlxsw_sp, macvlan_dev);
7320                 break;
7321         }
7322
7323         return 0;
7324 }
7325
7326 static int mlxsw_sp_router_port_check_rif_addr(struct mlxsw_sp *mlxsw_sp,
7327                                                struct net_device *dev,
7328                                                const unsigned char *dev_addr,
7329                                                struct netlink_ext_ack *extack)
7330 {
7331         struct mlxsw_sp_rif *rif;
7332         int i;
7333
7334         /* A RIF is not created for macvlan netdevs. Their MAC is used to
7335          * populate the FDB
7336          */
7337         if (netif_is_macvlan(dev) || netif_is_l3_master(dev))
7338                 return 0;
7339
7340         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++) {
7341                 rif = mlxsw_sp->router->rifs[i];
7342                 if (rif && rif->ops &&
7343                     rif->ops->type == MLXSW_SP_RIF_TYPE_IPIP_LB)
7344                         continue;
7345                 if (rif && rif->dev && rif->dev != dev &&
7346                     !ether_addr_equal_masked(rif->dev->dev_addr, dev_addr,
7347                                              mlxsw_sp->mac_mask)) {
7348                         NL_SET_ERR_MSG_MOD(extack, "All router interface MAC addresses must have the same prefix");
7349                         return -EINVAL;
7350                 }
7351         }
7352
7353         return 0;
7354 }
7355
7356 static int __mlxsw_sp_inetaddr_event(struct mlxsw_sp *mlxsw_sp,
7357                                      struct net_device *dev,
7358                                      unsigned long event,
7359                                      struct netlink_ext_ack *extack)
7360 {
7361         if (mlxsw_sp_port_dev_check(dev))
7362                 return mlxsw_sp_inetaddr_port_event(dev, event, extack);
7363         else if (netif_is_lag_master(dev))
7364                 return mlxsw_sp_inetaddr_lag_event(dev, event, extack);
7365         else if (netif_is_bridge_master(dev))
7366                 return mlxsw_sp_inetaddr_bridge_event(mlxsw_sp, dev, event,
7367                                                       extack);
7368         else if (is_vlan_dev(dev))
7369                 return mlxsw_sp_inetaddr_vlan_event(mlxsw_sp, dev, event,
7370                                                     extack);
7371         else if (netif_is_macvlan(dev))
7372                 return mlxsw_sp_inetaddr_macvlan_event(mlxsw_sp, dev, event,
7373                                                        extack);
7374         else
7375                 return 0;
7376 }
7377
7378 static int mlxsw_sp_inetaddr_event(struct notifier_block *nb,
7379                                    unsigned long event, void *ptr)
7380 {
7381         struct in_ifaddr *ifa = (struct in_ifaddr *) ptr;
7382         struct net_device *dev = ifa->ifa_dev->dev;
7383         struct mlxsw_sp_router *router;
7384         struct mlxsw_sp_rif *rif;
7385         int err = 0;
7386
7387         /* NETDEV_UP event is handled by mlxsw_sp_inetaddr_valid_event */
7388         if (event == NETDEV_UP)
7389                 return NOTIFY_DONE;
7390
7391         router = container_of(nb, struct mlxsw_sp_router, inetaddr_nb);
7392         mutex_lock(&router->lock);
7393         rif = mlxsw_sp_rif_find_by_dev(router->mlxsw_sp, dev);
7394         if (!mlxsw_sp_rif_should_config(rif, dev, event))
7395                 goto out;
7396
7397         err = __mlxsw_sp_inetaddr_event(router->mlxsw_sp, dev, event, NULL);
7398 out:
7399         mutex_unlock(&router->lock);
7400         return notifier_from_errno(err);
7401 }
7402
7403 int mlxsw_sp_inetaddr_valid_event(struct notifier_block *unused,
7404                                   unsigned long event, void *ptr)
7405 {
7406         struct in_validator_info *ivi = (struct in_validator_info *) ptr;
7407         struct net_device *dev = ivi->ivi_dev->dev;
7408         struct mlxsw_sp *mlxsw_sp;
7409         struct mlxsw_sp_rif *rif;
7410         int err = 0;
7411
7412         mlxsw_sp = mlxsw_sp_lower_get(dev);
7413         if (!mlxsw_sp)
7414                 return NOTIFY_DONE;
7415
7416         mutex_lock(&mlxsw_sp->router->lock);
7417         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7418         if (!mlxsw_sp_rif_should_config(rif, dev, event))
7419                 goto out;
7420
7421         err = mlxsw_sp_router_port_check_rif_addr(mlxsw_sp, dev, dev->dev_addr,
7422                                                   ivi->extack);
7423         if (err)
7424                 goto out;
7425
7426         err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, ivi->extack);
7427 out:
7428         mutex_unlock(&mlxsw_sp->router->lock);
7429         return notifier_from_errno(err);
7430 }
7431
7432 struct mlxsw_sp_inet6addr_event_work {
7433         struct work_struct work;
7434         struct mlxsw_sp *mlxsw_sp;
7435         struct net_device *dev;
7436         unsigned long event;
7437 };
7438
7439 static void mlxsw_sp_inet6addr_event_work(struct work_struct *work)
7440 {
7441         struct mlxsw_sp_inet6addr_event_work *inet6addr_work =
7442                 container_of(work, struct mlxsw_sp_inet6addr_event_work, work);
7443         struct mlxsw_sp *mlxsw_sp = inet6addr_work->mlxsw_sp;
7444         struct net_device *dev = inet6addr_work->dev;
7445         unsigned long event = inet6addr_work->event;
7446         struct mlxsw_sp_rif *rif;
7447
7448         rtnl_lock();
7449         mutex_lock(&mlxsw_sp->router->lock);
7450
7451         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7452         if (!mlxsw_sp_rif_should_config(rif, dev, event))
7453                 goto out;
7454
7455         __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, NULL);
7456 out:
7457         mutex_unlock(&mlxsw_sp->router->lock);
7458         rtnl_unlock();
7459         dev_put(dev);
7460         kfree(inet6addr_work);
7461 }
7462
7463 /* Called with rcu_read_lock() */
7464 static int mlxsw_sp_inet6addr_event(struct notifier_block *nb,
7465                                     unsigned long event, void *ptr)
7466 {
7467         struct inet6_ifaddr *if6 = (struct inet6_ifaddr *) ptr;
7468         struct mlxsw_sp_inet6addr_event_work *inet6addr_work;
7469         struct net_device *dev = if6->idev->dev;
7470         struct mlxsw_sp_router *router;
7471
7472         /* NETDEV_UP event is handled by mlxsw_sp_inet6addr_valid_event */
7473         if (event == NETDEV_UP)
7474                 return NOTIFY_DONE;
7475
7476         inet6addr_work = kzalloc(sizeof(*inet6addr_work), GFP_ATOMIC);
7477         if (!inet6addr_work)
7478                 return NOTIFY_BAD;
7479
7480         router = container_of(nb, struct mlxsw_sp_router, inet6addr_nb);
7481         INIT_WORK(&inet6addr_work->work, mlxsw_sp_inet6addr_event_work);
7482         inet6addr_work->mlxsw_sp = router->mlxsw_sp;
7483         inet6addr_work->dev = dev;
7484         inet6addr_work->event = event;
7485         dev_hold(dev);
7486         mlxsw_core_schedule_work(&inet6addr_work->work);
7487
7488         return NOTIFY_DONE;
7489 }
7490
7491 int mlxsw_sp_inet6addr_valid_event(struct notifier_block *unused,
7492                                    unsigned long event, void *ptr)
7493 {
7494         struct in6_validator_info *i6vi = (struct in6_validator_info *) ptr;
7495         struct net_device *dev = i6vi->i6vi_dev->dev;
7496         struct mlxsw_sp *mlxsw_sp;
7497         struct mlxsw_sp_rif *rif;
7498         int err = 0;
7499
7500         mlxsw_sp = mlxsw_sp_lower_get(dev);
7501         if (!mlxsw_sp)
7502                 return NOTIFY_DONE;
7503
7504         mutex_lock(&mlxsw_sp->router->lock);
7505         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7506         if (!mlxsw_sp_rif_should_config(rif, dev, event))
7507                 goto out;
7508
7509         err = mlxsw_sp_router_port_check_rif_addr(mlxsw_sp, dev, dev->dev_addr,
7510                                                   i6vi->extack);
7511         if (err)
7512                 goto out;
7513
7514         err = __mlxsw_sp_inetaddr_event(mlxsw_sp, dev, event, i6vi->extack);
7515 out:
7516         mutex_unlock(&mlxsw_sp->router->lock);
7517         return notifier_from_errno(err);
7518 }
7519
7520 static int mlxsw_sp_rif_edit(struct mlxsw_sp *mlxsw_sp, u16 rif_index,
7521                              const char *mac, int mtu)
7522 {
7523         char ritr_pl[MLXSW_REG_RITR_LEN];
7524         int err;
7525
7526         mlxsw_reg_ritr_rif_pack(ritr_pl, rif_index);
7527         err = mlxsw_reg_query(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7528         if (err)
7529                 return err;
7530
7531         mlxsw_reg_ritr_mtu_set(ritr_pl, mtu);
7532         mlxsw_reg_ritr_if_mac_memcpy_to(ritr_pl, mac);
7533         mlxsw_reg_ritr_op_set(ritr_pl, MLXSW_REG_RITR_RIF_CREATE);
7534         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7535 }
7536
7537 static int
7538 mlxsw_sp_router_port_change_event(struct mlxsw_sp *mlxsw_sp,
7539                                   struct mlxsw_sp_rif *rif)
7540 {
7541         struct net_device *dev = rif->dev;
7542         u16 fid_index;
7543         int err;
7544
7545         fid_index = mlxsw_sp_fid_index(rif->fid);
7546
7547         err = mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, false);
7548         if (err)
7549                 return err;
7550
7551         err = mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, dev->dev_addr,
7552                                 dev->mtu);
7553         if (err)
7554                 goto err_rif_edit;
7555
7556         err = mlxsw_sp_rif_fdb_op(mlxsw_sp, dev->dev_addr, fid_index, true);
7557         if (err)
7558                 goto err_rif_fdb_op;
7559
7560         if (rif->mtu != dev->mtu) {
7561                 struct mlxsw_sp_vr *vr;
7562                 int i;
7563
7564                 /* The RIF is relevant only to its mr_table instance, as unlike
7565                  * unicast routing, in multicast routing a RIF cannot be shared
7566                  * between several multicast routing tables.
7567                  */
7568                 vr = &mlxsw_sp->router->vrs[rif->vr_id];
7569                 for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++)
7570                         mlxsw_sp_mr_rif_mtu_update(vr->mr_table[i],
7571                                                    rif, dev->mtu);
7572         }
7573
7574         ether_addr_copy(rif->addr, dev->dev_addr);
7575         rif->mtu = dev->mtu;
7576
7577         netdev_dbg(dev, "Updated RIF=%d\n", rif->rif_index);
7578
7579         return 0;
7580
7581 err_rif_fdb_op:
7582         mlxsw_sp_rif_edit(mlxsw_sp, rif->rif_index, rif->addr, rif->mtu);
7583 err_rif_edit:
7584         mlxsw_sp_rif_fdb_op(mlxsw_sp, rif->addr, fid_index, true);
7585         return err;
7586 }
7587
7588 static int mlxsw_sp_router_port_pre_changeaddr_event(struct mlxsw_sp_rif *rif,
7589                             struct netdev_notifier_pre_changeaddr_info *info)
7590 {
7591         struct netlink_ext_ack *extack;
7592
7593         extack = netdev_notifier_info_to_extack(&info->info);
7594         return mlxsw_sp_router_port_check_rif_addr(rif->mlxsw_sp, rif->dev,
7595                                                    info->dev_addr, extack);
7596 }
7597
7598 int mlxsw_sp_netdevice_router_port_event(struct net_device *dev,
7599                                          unsigned long event, void *ptr)
7600 {
7601         struct mlxsw_sp *mlxsw_sp;
7602         struct mlxsw_sp_rif *rif;
7603         int err = 0;
7604
7605         mlxsw_sp = mlxsw_sp_lower_get(dev);
7606         if (!mlxsw_sp)
7607                 return 0;
7608
7609         mutex_lock(&mlxsw_sp->router->lock);
7610         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, dev);
7611         if (!rif)
7612                 goto out;
7613
7614         switch (event) {
7615         case NETDEV_CHANGEMTU:
7616         case NETDEV_CHANGEADDR:
7617                 err = mlxsw_sp_router_port_change_event(mlxsw_sp, rif);
7618                 break;
7619         case NETDEV_PRE_CHANGEADDR:
7620                 err = mlxsw_sp_router_port_pre_changeaddr_event(rif, ptr);
7621                 break;
7622         }
7623
7624 out:
7625         mutex_unlock(&mlxsw_sp->router->lock);
7626         return err;
7627 }
7628
7629 static int mlxsw_sp_port_vrf_join(struct mlxsw_sp *mlxsw_sp,
7630                                   struct net_device *l3_dev,
7631                                   struct netlink_ext_ack *extack)
7632 {
7633         struct mlxsw_sp_rif *rif;
7634
7635         /* If netdev is already associated with a RIF, then we need to
7636          * destroy it and create a new one with the new virtual router ID.
7637          */
7638         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7639         if (rif)
7640                 __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN,
7641                                           extack);
7642
7643         return __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_UP, extack);
7644 }
7645
7646 static void mlxsw_sp_port_vrf_leave(struct mlxsw_sp *mlxsw_sp,
7647                                     struct net_device *l3_dev)
7648 {
7649         struct mlxsw_sp_rif *rif;
7650
7651         rif = mlxsw_sp_rif_find_by_dev(mlxsw_sp, l3_dev);
7652         if (!rif)
7653                 return;
7654         __mlxsw_sp_inetaddr_event(mlxsw_sp, l3_dev, NETDEV_DOWN, NULL);
7655 }
7656
7657 int mlxsw_sp_netdevice_vrf_event(struct net_device *l3_dev, unsigned long event,
7658                                  struct netdev_notifier_changeupper_info *info)
7659 {
7660         struct mlxsw_sp *mlxsw_sp = mlxsw_sp_lower_get(l3_dev);
7661         int err = 0;
7662
7663         /* We do not create a RIF for a macvlan, but only use it to
7664          * direct more MAC addresses to the router.
7665          */
7666         if (!mlxsw_sp || netif_is_macvlan(l3_dev))
7667                 return 0;
7668
7669         mutex_lock(&mlxsw_sp->router->lock);
7670         switch (event) {
7671         case NETDEV_PRECHANGEUPPER:
7672                 break;
7673         case NETDEV_CHANGEUPPER:
7674                 if (info->linking) {
7675                         struct netlink_ext_ack *extack;
7676
7677                         extack = netdev_notifier_info_to_extack(&info->info);
7678                         err = mlxsw_sp_port_vrf_join(mlxsw_sp, l3_dev, extack);
7679                 } else {
7680                         mlxsw_sp_port_vrf_leave(mlxsw_sp, l3_dev);
7681                 }
7682                 break;
7683         }
7684         mutex_unlock(&mlxsw_sp->router->lock);
7685
7686         return err;
7687 }
7688
7689 static int __mlxsw_sp_rif_macvlan_flush(struct net_device *dev,
7690                                         struct netdev_nested_priv *priv)
7691 {
7692         struct mlxsw_sp_rif *rif = (struct mlxsw_sp_rif *)priv->data;
7693
7694         if (!netif_is_macvlan(dev))
7695                 return 0;
7696
7697         return mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, dev->dev_addr,
7698                                    mlxsw_sp_fid_index(rif->fid), false);
7699 }
7700
7701 static int mlxsw_sp_rif_macvlan_flush(struct mlxsw_sp_rif *rif)
7702 {
7703         struct netdev_nested_priv priv = {
7704                 .data = (void *)rif,
7705         };
7706
7707         if (!netif_is_macvlan_port(rif->dev))
7708                 return 0;
7709
7710         netdev_warn(rif->dev, "Router interface is deleted. Upper macvlans will not work\n");
7711         return netdev_walk_all_upper_dev_rcu(rif->dev,
7712                                              __mlxsw_sp_rif_macvlan_flush, &priv);
7713 }
7714
7715 static void mlxsw_sp_rif_subport_setup(struct mlxsw_sp_rif *rif,
7716                                        const struct mlxsw_sp_rif_params *params)
7717 {
7718         struct mlxsw_sp_rif_subport *rif_subport;
7719
7720         rif_subport = mlxsw_sp_rif_subport_rif(rif);
7721         refcount_set(&rif_subport->ref_count, 1);
7722         rif_subport->vid = params->vid;
7723         rif_subport->lag = params->lag;
7724         if (params->lag)
7725                 rif_subport->lag_id = params->lag_id;
7726         else
7727                 rif_subport->system_port = params->system_port;
7728 }
7729
7730 static int mlxsw_sp_rif_subport_op(struct mlxsw_sp_rif *rif, bool enable)
7731 {
7732         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7733         struct mlxsw_sp_rif_subport *rif_subport;
7734         char ritr_pl[MLXSW_REG_RITR_LEN];
7735
7736         rif_subport = mlxsw_sp_rif_subport_rif(rif);
7737         mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_SP_IF,
7738                             rif->rif_index, rif->vr_id, rif->dev->mtu);
7739         mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
7740         mlxsw_reg_ritr_sp_if_pack(ritr_pl, rif_subport->lag,
7741                                   rif_subport->lag ? rif_subport->lag_id :
7742                                                      rif_subport->system_port,
7743                                   rif_subport->vid);
7744
7745         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7746 }
7747
7748 static int mlxsw_sp_rif_subport_configure(struct mlxsw_sp_rif *rif)
7749 {
7750         int err;
7751
7752         err = mlxsw_sp_rif_subport_op(rif, true);
7753         if (err)
7754                 return err;
7755
7756         err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7757                                   mlxsw_sp_fid_index(rif->fid), true);
7758         if (err)
7759                 goto err_rif_fdb_op;
7760
7761         mlxsw_sp_fid_rif_set(rif->fid, rif);
7762         return 0;
7763
7764 err_rif_fdb_op:
7765         mlxsw_sp_rif_subport_op(rif, false);
7766         return err;
7767 }
7768
7769 static void mlxsw_sp_rif_subport_deconfigure(struct mlxsw_sp_rif *rif)
7770 {
7771         struct mlxsw_sp_fid *fid = rif->fid;
7772
7773         mlxsw_sp_fid_rif_set(fid, NULL);
7774         mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7775                             mlxsw_sp_fid_index(fid), false);
7776         mlxsw_sp_rif_macvlan_flush(rif);
7777         mlxsw_sp_rif_subport_op(rif, false);
7778 }
7779
7780 static struct mlxsw_sp_fid *
7781 mlxsw_sp_rif_subport_fid_get(struct mlxsw_sp_rif *rif,
7782                              struct netlink_ext_ack *extack)
7783 {
7784         return mlxsw_sp_fid_rfid_get(rif->mlxsw_sp, rif->rif_index);
7785 }
7786
7787 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_subport_ops = {
7788         .type                   = MLXSW_SP_RIF_TYPE_SUBPORT,
7789         .rif_size               = sizeof(struct mlxsw_sp_rif_subport),
7790         .setup                  = mlxsw_sp_rif_subport_setup,
7791         .configure              = mlxsw_sp_rif_subport_configure,
7792         .deconfigure            = mlxsw_sp_rif_subport_deconfigure,
7793         .fid_get                = mlxsw_sp_rif_subport_fid_get,
7794 };
7795
7796 static int mlxsw_sp_rif_vlan_fid_op(struct mlxsw_sp_rif *rif,
7797                                     enum mlxsw_reg_ritr_if_type type,
7798                                     u16 vid_fid, bool enable)
7799 {
7800         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7801         char ritr_pl[MLXSW_REG_RITR_LEN];
7802
7803         mlxsw_reg_ritr_pack(ritr_pl, enable, type, rif->rif_index, rif->vr_id,
7804                             rif->dev->mtu);
7805         mlxsw_reg_ritr_mac_pack(ritr_pl, rif->dev->dev_addr);
7806         mlxsw_reg_ritr_fid_set(ritr_pl, type, vid_fid);
7807
7808         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
7809 }
7810
7811 u8 mlxsw_sp_router_port(const struct mlxsw_sp *mlxsw_sp)
7812 {
7813         return mlxsw_core_max_ports(mlxsw_sp->core) + 1;
7814 }
7815
7816 static int mlxsw_sp_rif_fid_configure(struct mlxsw_sp_rif *rif)
7817 {
7818         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7819         u16 fid_index = mlxsw_sp_fid_index(rif->fid);
7820         int err;
7821
7822         err = mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index,
7823                                        true);
7824         if (err)
7825                 return err;
7826
7827         err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7828                                      mlxsw_sp_router_port(mlxsw_sp), true);
7829         if (err)
7830                 goto err_fid_mc_flood_set;
7831
7832         err = mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7833                                      mlxsw_sp_router_port(mlxsw_sp), true);
7834         if (err)
7835                 goto err_fid_bc_flood_set;
7836
7837         err = mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7838                                   mlxsw_sp_fid_index(rif->fid), true);
7839         if (err)
7840                 goto err_rif_fdb_op;
7841
7842         mlxsw_sp_fid_rif_set(rif->fid, rif);
7843         return 0;
7844
7845 err_rif_fdb_op:
7846         mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7847                                mlxsw_sp_router_port(mlxsw_sp), false);
7848 err_fid_bc_flood_set:
7849         mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7850                                mlxsw_sp_router_port(mlxsw_sp), false);
7851 err_fid_mc_flood_set:
7852         mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
7853         return err;
7854 }
7855
7856 static void mlxsw_sp_rif_fid_deconfigure(struct mlxsw_sp_rif *rif)
7857 {
7858         u16 fid_index = mlxsw_sp_fid_index(rif->fid);
7859         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7860         struct mlxsw_sp_fid *fid = rif->fid;
7861
7862         mlxsw_sp_fid_rif_set(fid, NULL);
7863         mlxsw_sp_rif_fdb_op(rif->mlxsw_sp, rif->dev->dev_addr,
7864                             mlxsw_sp_fid_index(fid), false);
7865         mlxsw_sp_rif_macvlan_flush(rif);
7866         mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_BC,
7867                                mlxsw_sp_router_port(mlxsw_sp), false);
7868         mlxsw_sp_fid_flood_set(rif->fid, MLXSW_SP_FLOOD_TYPE_MC,
7869                                mlxsw_sp_router_port(mlxsw_sp), false);
7870         mlxsw_sp_rif_vlan_fid_op(rif, MLXSW_REG_RITR_FID_IF, fid_index, false);
7871 }
7872
7873 static struct mlxsw_sp_fid *
7874 mlxsw_sp_rif_fid_fid_get(struct mlxsw_sp_rif *rif,
7875                          struct netlink_ext_ack *extack)
7876 {
7877         return mlxsw_sp_fid_8021d_get(rif->mlxsw_sp, rif->dev->ifindex);
7878 }
7879
7880 static void mlxsw_sp_rif_fid_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
7881 {
7882         struct switchdev_notifier_fdb_info info;
7883         struct net_device *dev;
7884
7885         dev = br_fdb_find_port(rif->dev, mac, 0);
7886         if (!dev)
7887                 return;
7888
7889         info.addr = mac;
7890         info.vid = 0;
7891         call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
7892                                  NULL);
7893 }
7894
7895 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_fid_ops = {
7896         .type                   = MLXSW_SP_RIF_TYPE_FID,
7897         .rif_size               = sizeof(struct mlxsw_sp_rif),
7898         .configure              = mlxsw_sp_rif_fid_configure,
7899         .deconfigure            = mlxsw_sp_rif_fid_deconfigure,
7900         .fid_get                = mlxsw_sp_rif_fid_fid_get,
7901         .fdb_del                = mlxsw_sp_rif_fid_fdb_del,
7902 };
7903
7904 static struct mlxsw_sp_fid *
7905 mlxsw_sp_rif_vlan_fid_get(struct mlxsw_sp_rif *rif,
7906                           struct netlink_ext_ack *extack)
7907 {
7908         struct net_device *br_dev;
7909         u16 vid;
7910         int err;
7911
7912         if (is_vlan_dev(rif->dev)) {
7913                 vid = vlan_dev_vlan_id(rif->dev);
7914                 br_dev = vlan_dev_real_dev(rif->dev);
7915                 if (WARN_ON(!netif_is_bridge_master(br_dev)))
7916                         return ERR_PTR(-EINVAL);
7917         } else {
7918                 err = br_vlan_get_pvid(rif->dev, &vid);
7919                 if (err < 0 || !vid) {
7920                         NL_SET_ERR_MSG_MOD(extack, "Couldn't determine bridge PVID");
7921                         return ERR_PTR(-EINVAL);
7922                 }
7923         }
7924
7925         return mlxsw_sp_fid_8021q_get(rif->mlxsw_sp, vid);
7926 }
7927
7928 static void mlxsw_sp_rif_vlan_fdb_del(struct mlxsw_sp_rif *rif, const char *mac)
7929 {
7930         u16 vid = mlxsw_sp_fid_8021q_vid(rif->fid);
7931         struct switchdev_notifier_fdb_info info;
7932         struct net_device *br_dev;
7933         struct net_device *dev;
7934
7935         br_dev = is_vlan_dev(rif->dev) ? vlan_dev_real_dev(rif->dev) : rif->dev;
7936         dev = br_fdb_find_port(br_dev, mac, vid);
7937         if (!dev)
7938                 return;
7939
7940         info.addr = mac;
7941         info.vid = vid;
7942         call_switchdev_notifiers(SWITCHDEV_FDB_DEL_TO_BRIDGE, dev, &info.info,
7943                                  NULL);
7944 }
7945
7946 static const struct mlxsw_sp_rif_ops mlxsw_sp_rif_vlan_emu_ops = {
7947         .type                   = MLXSW_SP_RIF_TYPE_VLAN,
7948         .rif_size               = sizeof(struct mlxsw_sp_rif),
7949         .configure              = mlxsw_sp_rif_fid_configure,
7950         .deconfigure            = mlxsw_sp_rif_fid_deconfigure,
7951         .fid_get                = mlxsw_sp_rif_vlan_fid_get,
7952         .fdb_del                = mlxsw_sp_rif_vlan_fdb_del,
7953 };
7954
7955 static struct mlxsw_sp_rif_ipip_lb *
7956 mlxsw_sp_rif_ipip_lb_rif(struct mlxsw_sp_rif *rif)
7957 {
7958         return container_of(rif, struct mlxsw_sp_rif_ipip_lb, common);
7959 }
7960
7961 static void
7962 mlxsw_sp_rif_ipip_lb_setup(struct mlxsw_sp_rif *rif,
7963                            const struct mlxsw_sp_rif_params *params)
7964 {
7965         struct mlxsw_sp_rif_params_ipip_lb *params_lb;
7966         struct mlxsw_sp_rif_ipip_lb *rif_lb;
7967
7968         params_lb = container_of(params, struct mlxsw_sp_rif_params_ipip_lb,
7969                                  common);
7970         rif_lb = mlxsw_sp_rif_ipip_lb_rif(rif);
7971         rif_lb->lb_config = params_lb->lb_config;
7972 }
7973
7974 static int
7975 mlxsw_sp1_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif)
7976 {
7977         struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
7978         u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
7979         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
7980         struct mlxsw_sp_vr *ul_vr;
7981         int err;
7982
7983         ul_vr = mlxsw_sp_vr_get(mlxsw_sp, ul_tb_id, NULL);
7984         if (IS_ERR(ul_vr))
7985                 return PTR_ERR(ul_vr);
7986
7987         err = mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, true);
7988         if (err)
7989                 goto err_loopback_op;
7990
7991         lb_rif->ul_vr_id = ul_vr->id;
7992         lb_rif->ul_rif_id = 0;
7993         ++ul_vr->rif_count;
7994         return 0;
7995
7996 err_loopback_op:
7997         mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
7998         return err;
7999 }
8000
8001 static void mlxsw_sp1_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
8002 {
8003         struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
8004         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8005         struct mlxsw_sp_vr *ul_vr;
8006
8007         ul_vr = &mlxsw_sp->router->vrs[lb_rif->ul_vr_id];
8008         mlxsw_sp_rif_ipip_lb_op(lb_rif, ul_vr->id, 0, false);
8009
8010         --ul_vr->rif_count;
8011         mlxsw_sp_vr_put(mlxsw_sp, ul_vr);
8012 }
8013
8014 static const struct mlxsw_sp_rif_ops mlxsw_sp1_rif_ipip_lb_ops = {
8015         .type                   = MLXSW_SP_RIF_TYPE_IPIP_LB,
8016         .rif_size               = sizeof(struct mlxsw_sp_rif_ipip_lb),
8017         .setup                  = mlxsw_sp_rif_ipip_lb_setup,
8018         .configure              = mlxsw_sp1_rif_ipip_lb_configure,
8019         .deconfigure            = mlxsw_sp1_rif_ipip_lb_deconfigure,
8020 };
8021
8022 const struct mlxsw_sp_rif_ops *mlxsw_sp1_rif_ops_arr[] = {
8023         [MLXSW_SP_RIF_TYPE_SUBPORT]     = &mlxsw_sp_rif_subport_ops,
8024         [MLXSW_SP_RIF_TYPE_VLAN]        = &mlxsw_sp_rif_vlan_emu_ops,
8025         [MLXSW_SP_RIF_TYPE_FID]         = &mlxsw_sp_rif_fid_ops,
8026         [MLXSW_SP_RIF_TYPE_IPIP_LB]     = &mlxsw_sp1_rif_ipip_lb_ops,
8027 };
8028
8029 static int
8030 mlxsw_sp_rif_ipip_lb_ul_rif_op(struct mlxsw_sp_rif *ul_rif, bool enable)
8031 {
8032         struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
8033         char ritr_pl[MLXSW_REG_RITR_LEN];
8034
8035         mlxsw_reg_ritr_pack(ritr_pl, enable, MLXSW_REG_RITR_LOOPBACK_IF,
8036                             ul_rif->rif_index, ul_rif->vr_id, IP_MAX_MTU);
8037         mlxsw_reg_ritr_loopback_protocol_set(ritr_pl,
8038                                              MLXSW_REG_RITR_LOOPBACK_GENERIC);
8039
8040         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ritr), ritr_pl);
8041 }
8042
8043 static struct mlxsw_sp_rif *
8044 mlxsw_sp_ul_rif_create(struct mlxsw_sp *mlxsw_sp, struct mlxsw_sp_vr *vr,
8045                        struct netlink_ext_ack *extack)
8046 {
8047         struct mlxsw_sp_rif *ul_rif;
8048         u16 rif_index;
8049         int err;
8050
8051         err = mlxsw_sp_rif_index_alloc(mlxsw_sp, &rif_index);
8052         if (err) {
8053                 NL_SET_ERR_MSG_MOD(extack, "Exceeded number of supported router interfaces");
8054                 return ERR_PTR(err);
8055         }
8056
8057         ul_rif = mlxsw_sp_rif_alloc(sizeof(*ul_rif), rif_index, vr->id, NULL);
8058         if (!ul_rif)
8059                 return ERR_PTR(-ENOMEM);
8060
8061         mlxsw_sp->router->rifs[rif_index] = ul_rif;
8062         ul_rif->mlxsw_sp = mlxsw_sp;
8063         err = mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, true);
8064         if (err)
8065                 goto ul_rif_op_err;
8066
8067         return ul_rif;
8068
8069 ul_rif_op_err:
8070         mlxsw_sp->router->rifs[rif_index] = NULL;
8071         kfree(ul_rif);
8072         return ERR_PTR(err);
8073 }
8074
8075 static void mlxsw_sp_ul_rif_destroy(struct mlxsw_sp_rif *ul_rif)
8076 {
8077         struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
8078
8079         mlxsw_sp_rif_ipip_lb_ul_rif_op(ul_rif, false);
8080         mlxsw_sp->router->rifs[ul_rif->rif_index] = NULL;
8081         kfree(ul_rif);
8082 }
8083
8084 static struct mlxsw_sp_rif *
8085 mlxsw_sp_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 tb_id,
8086                     struct netlink_ext_ack *extack)
8087 {
8088         struct mlxsw_sp_vr *vr;
8089         int err;
8090
8091         vr = mlxsw_sp_vr_get(mlxsw_sp, tb_id, extack);
8092         if (IS_ERR(vr))
8093                 return ERR_CAST(vr);
8094
8095         if (refcount_inc_not_zero(&vr->ul_rif_refcnt))
8096                 return vr->ul_rif;
8097
8098         vr->ul_rif = mlxsw_sp_ul_rif_create(mlxsw_sp, vr, extack);
8099         if (IS_ERR(vr->ul_rif)) {
8100                 err = PTR_ERR(vr->ul_rif);
8101                 goto err_ul_rif_create;
8102         }
8103
8104         vr->rif_count++;
8105         refcount_set(&vr->ul_rif_refcnt, 1);
8106
8107         return vr->ul_rif;
8108
8109 err_ul_rif_create:
8110         mlxsw_sp_vr_put(mlxsw_sp, vr);
8111         return ERR_PTR(err);
8112 }
8113
8114 static void mlxsw_sp_ul_rif_put(struct mlxsw_sp_rif *ul_rif)
8115 {
8116         struct mlxsw_sp *mlxsw_sp = ul_rif->mlxsw_sp;
8117         struct mlxsw_sp_vr *vr;
8118
8119         vr = &mlxsw_sp->router->vrs[ul_rif->vr_id];
8120
8121         if (!refcount_dec_and_test(&vr->ul_rif_refcnt))
8122                 return;
8123
8124         vr->rif_count--;
8125         mlxsw_sp_ul_rif_destroy(ul_rif);
8126         mlxsw_sp_vr_put(mlxsw_sp, vr);
8127 }
8128
8129 int mlxsw_sp_router_ul_rif_get(struct mlxsw_sp *mlxsw_sp, u32 ul_tb_id,
8130                                u16 *ul_rif_index)
8131 {
8132         struct mlxsw_sp_rif *ul_rif;
8133         int err = 0;
8134
8135         mutex_lock(&mlxsw_sp->router->lock);
8136         ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
8137         if (IS_ERR(ul_rif)) {
8138                 err = PTR_ERR(ul_rif);
8139                 goto out;
8140         }
8141         *ul_rif_index = ul_rif->rif_index;
8142 out:
8143         mutex_unlock(&mlxsw_sp->router->lock);
8144         return err;
8145 }
8146
8147 void mlxsw_sp_router_ul_rif_put(struct mlxsw_sp *mlxsw_sp, u16 ul_rif_index)
8148 {
8149         struct mlxsw_sp_rif *ul_rif;
8150
8151         mutex_lock(&mlxsw_sp->router->lock);
8152         ul_rif = mlxsw_sp->router->rifs[ul_rif_index];
8153         if (WARN_ON(!ul_rif))
8154                 goto out;
8155
8156         mlxsw_sp_ul_rif_put(ul_rif);
8157 out:
8158         mutex_unlock(&mlxsw_sp->router->lock);
8159 }
8160
8161 static int
8162 mlxsw_sp2_rif_ipip_lb_configure(struct mlxsw_sp_rif *rif)
8163 {
8164         struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
8165         u32 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(rif->dev);
8166         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8167         struct mlxsw_sp_rif *ul_rif;
8168         int err;
8169
8170         ul_rif = mlxsw_sp_ul_rif_get(mlxsw_sp, ul_tb_id, NULL);
8171         if (IS_ERR(ul_rif))
8172                 return PTR_ERR(ul_rif);
8173
8174         err = mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, ul_rif->rif_index, true);
8175         if (err)
8176                 goto err_loopback_op;
8177
8178         lb_rif->ul_vr_id = 0;
8179         lb_rif->ul_rif_id = ul_rif->rif_index;
8180
8181         return 0;
8182
8183 err_loopback_op:
8184         mlxsw_sp_ul_rif_put(ul_rif);
8185         return err;
8186 }
8187
8188 static void mlxsw_sp2_rif_ipip_lb_deconfigure(struct mlxsw_sp_rif *rif)
8189 {
8190         struct mlxsw_sp_rif_ipip_lb *lb_rif = mlxsw_sp_rif_ipip_lb_rif(rif);
8191         struct mlxsw_sp *mlxsw_sp = rif->mlxsw_sp;
8192         struct mlxsw_sp_rif *ul_rif;
8193
8194         ul_rif = mlxsw_sp_rif_by_index(mlxsw_sp, lb_rif->ul_rif_id);
8195         mlxsw_sp_rif_ipip_lb_op(lb_rif, 0, lb_rif->ul_rif_id, false);
8196         mlxsw_sp_ul_rif_put(ul_rif);
8197 }
8198
8199 static const struct mlxsw_sp_rif_ops mlxsw_sp2_rif_ipip_lb_ops = {
8200         .type                   = MLXSW_SP_RIF_TYPE_IPIP_LB,
8201         .rif_size               = sizeof(struct mlxsw_sp_rif_ipip_lb),
8202         .setup                  = mlxsw_sp_rif_ipip_lb_setup,
8203         .configure              = mlxsw_sp2_rif_ipip_lb_configure,
8204         .deconfigure            = mlxsw_sp2_rif_ipip_lb_deconfigure,
8205 };
8206
8207 const struct mlxsw_sp_rif_ops *mlxsw_sp2_rif_ops_arr[] = {
8208         [MLXSW_SP_RIF_TYPE_SUBPORT]     = &mlxsw_sp_rif_subport_ops,
8209         [MLXSW_SP_RIF_TYPE_VLAN]        = &mlxsw_sp_rif_vlan_emu_ops,
8210         [MLXSW_SP_RIF_TYPE_FID]         = &mlxsw_sp_rif_fid_ops,
8211         [MLXSW_SP_RIF_TYPE_IPIP_LB]     = &mlxsw_sp2_rif_ipip_lb_ops,
8212 };
8213
8214 static int mlxsw_sp_rifs_init(struct mlxsw_sp *mlxsw_sp)
8215 {
8216         u64 max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
8217
8218         mlxsw_sp->router->rifs = kcalloc(max_rifs,
8219                                          sizeof(struct mlxsw_sp_rif *),
8220                                          GFP_KERNEL);
8221         if (!mlxsw_sp->router->rifs)
8222                 return -ENOMEM;
8223
8224         return 0;
8225 }
8226
8227 static void mlxsw_sp_rifs_fini(struct mlxsw_sp *mlxsw_sp)
8228 {
8229         int i;
8230
8231         for (i = 0; i < MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS); i++)
8232                 WARN_ON_ONCE(mlxsw_sp->router->rifs[i]);
8233
8234         kfree(mlxsw_sp->router->rifs);
8235 }
8236
8237 static int
8238 mlxsw_sp_ipip_config_tigcr(struct mlxsw_sp *mlxsw_sp)
8239 {
8240         char tigcr_pl[MLXSW_REG_TIGCR_LEN];
8241
8242         mlxsw_reg_tigcr_pack(tigcr_pl, true, 0);
8243         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tigcr), tigcr_pl);
8244 }
8245
8246 static int mlxsw_sp_ipips_init(struct mlxsw_sp *mlxsw_sp)
8247 {
8248         int err;
8249
8250         mlxsw_sp->router->ipip_ops_arr = mlxsw_sp_ipip_ops_arr;
8251         INIT_LIST_HEAD(&mlxsw_sp->router->ipip_list);
8252
8253         err = mlxsw_sp_ipip_ecn_encap_init(mlxsw_sp);
8254         if (err)
8255                 return err;
8256         err = mlxsw_sp_ipip_ecn_decap_init(mlxsw_sp);
8257         if (err)
8258                 return err;
8259
8260         return mlxsw_sp_ipip_config_tigcr(mlxsw_sp);
8261 }
8262
8263 static void mlxsw_sp_ipips_fini(struct mlxsw_sp *mlxsw_sp)
8264 {
8265         WARN_ON(!list_empty(&mlxsw_sp->router->ipip_list));
8266 }
8267
8268 static void mlxsw_sp_router_fib_dump_flush(struct notifier_block *nb)
8269 {
8270         struct mlxsw_sp_router *router;
8271
8272         /* Flush pending FIB notifications and then flush the device's
8273          * table before requesting another dump. The FIB notification
8274          * block is unregistered, so no need to take RTNL.
8275          */
8276         mlxsw_core_flush_owq();
8277         router = container_of(nb, struct mlxsw_sp_router, fib_nb);
8278         mlxsw_sp_router_fib_flush(router->mlxsw_sp);
8279 }
8280
8281 #ifdef CONFIG_IP_ROUTE_MULTIPATH
8282 static void mlxsw_sp_mp_hash_header_set(char *recr2_pl, int header)
8283 {
8284         mlxsw_reg_recr2_outer_header_enables_set(recr2_pl, header, true);
8285 }
8286
8287 static void mlxsw_sp_mp_hash_field_set(char *recr2_pl, int field)
8288 {
8289         mlxsw_reg_recr2_outer_header_fields_enable_set(recr2_pl, field, true);
8290 }
8291
8292 static void mlxsw_sp_mp4_hash_init(struct mlxsw_sp *mlxsw_sp, char *recr2_pl)
8293 {
8294         struct net *net = mlxsw_sp_net(mlxsw_sp);
8295         bool only_l3 = !net->ipv4.sysctl_fib_multipath_hash_policy;
8296
8297         mlxsw_sp_mp_hash_header_set(recr2_pl,
8298                                     MLXSW_REG_RECR2_IPV4_EN_NOT_TCP_NOT_UDP);
8299         mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV4_EN_TCP_UDP);
8300         mlxsw_reg_recr2_ipv4_sip_enable(recr2_pl);
8301         mlxsw_reg_recr2_ipv4_dip_enable(recr2_pl);
8302         if (only_l3)
8303                 return;
8304         mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_EN_IPV4);
8305         mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV4_PROTOCOL);
8306         mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_SPORT);
8307         mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_TCP_UDP_DPORT);
8308 }
8309
8310 static void mlxsw_sp_mp6_hash_init(struct mlxsw_sp *mlxsw_sp, char *recr2_pl)
8311 {
8312         bool only_l3 = !ip6_multipath_hash_policy(mlxsw_sp_net(mlxsw_sp));
8313
8314         mlxsw_sp_mp_hash_header_set(recr2_pl,
8315                                     MLXSW_REG_RECR2_IPV6_EN_NOT_TCP_NOT_UDP);
8316         mlxsw_sp_mp_hash_header_set(recr2_pl, MLXSW_REG_RECR2_IPV6_EN_TCP_UDP);
8317         mlxsw_reg_recr2_ipv6_sip_enable(recr2_pl);
8318         mlxsw_reg_recr2_ipv6_dip_enable(recr2_pl);
8319         mlxsw_sp_mp_hash_field_set(recr2_pl, MLXSW_REG_RECR2_IPV6_NEXT_HEADER);
8320         if (only_l3) {
8321                 mlxsw_sp_mp_hash_field_set(recr2_pl,
8322                                            MLXSW_REG_RECR2_IPV6_FLOW_LABEL);
8323         } else {
8324                 mlxsw_sp_mp_hash_header_set(recr2_pl,
8325                                             MLXSW_REG_RECR2_TCP_UDP_EN_IPV6);
8326                 mlxsw_sp_mp_hash_field_set(recr2_pl,
8327                                            MLXSW_REG_RECR2_TCP_UDP_SPORT);
8328                 mlxsw_sp_mp_hash_field_set(recr2_pl,
8329                                            MLXSW_REG_RECR2_TCP_UDP_DPORT);
8330         }
8331 }
8332
8333 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
8334 {
8335         char recr2_pl[MLXSW_REG_RECR2_LEN];
8336         u32 seed;
8337
8338         seed = jhash(mlxsw_sp->base_mac, sizeof(mlxsw_sp->base_mac), 0);
8339         mlxsw_reg_recr2_pack(recr2_pl, seed);
8340         mlxsw_sp_mp4_hash_init(mlxsw_sp, recr2_pl);
8341         mlxsw_sp_mp6_hash_init(mlxsw_sp, recr2_pl);
8342
8343         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(recr2), recr2_pl);
8344 }
8345 #else
8346 static int mlxsw_sp_mp_hash_init(struct mlxsw_sp *mlxsw_sp)
8347 {
8348         return 0;
8349 }
8350 #endif
8351
8352 static int mlxsw_sp_dscp_init(struct mlxsw_sp *mlxsw_sp)
8353 {
8354         char rdpm_pl[MLXSW_REG_RDPM_LEN];
8355         unsigned int i;
8356
8357         MLXSW_REG_ZERO(rdpm, rdpm_pl);
8358
8359         /* HW is determining switch priority based on DSCP-bits, but the
8360          * kernel is still doing that based on the ToS. Since there's a
8361          * mismatch in bits we need to make sure to translate the right
8362          * value ToS would observe, skipping the 2 least-significant ECN bits.
8363          */
8364         for (i = 0; i < MLXSW_REG_RDPM_DSCP_ENTRY_REC_MAX_COUNT; i++)
8365                 mlxsw_reg_rdpm_pack(rdpm_pl, i, rt_tos2priority(i << 2));
8366
8367         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rdpm), rdpm_pl);
8368 }
8369
8370 static int __mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp)
8371 {
8372         struct net *net = mlxsw_sp_net(mlxsw_sp);
8373         bool usp = net->ipv4.sysctl_ip_fwd_update_priority;
8374         char rgcr_pl[MLXSW_REG_RGCR_LEN];
8375         u64 max_rifs;
8376
8377         if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_RIFS))
8378                 return -EIO;
8379         max_rifs = MLXSW_CORE_RES_GET(mlxsw_sp->core, MAX_RIFS);
8380
8381         mlxsw_reg_rgcr_pack(rgcr_pl, true, true);
8382         mlxsw_reg_rgcr_max_router_interfaces_set(rgcr_pl, max_rifs);
8383         mlxsw_reg_rgcr_usp_set(rgcr_pl, usp);
8384         return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
8385 }
8386
8387 static void __mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
8388 {
8389         char rgcr_pl[MLXSW_REG_RGCR_LEN];
8390
8391         mlxsw_reg_rgcr_pack(rgcr_pl, false, false);
8392         mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rgcr), rgcr_pl);
8393 }
8394
8395 static const struct mlxsw_sp_router_ll_ops mlxsw_sp_router_ll_basic_ops = {
8396         .ralta_write = mlxsw_sp_router_ll_basic_ralta_write,
8397         .ralst_write = mlxsw_sp_router_ll_basic_ralst_write,
8398         .raltb_write = mlxsw_sp_router_ll_basic_raltb_write,
8399         .fib_entry_op_ctx_size = sizeof(struct mlxsw_sp_fib_entry_op_ctx_basic),
8400         .fib_entry_pack = mlxsw_sp_router_ll_basic_fib_entry_pack,
8401         .fib_entry_act_remote_pack = mlxsw_sp_router_ll_basic_fib_entry_act_remote_pack,
8402         .fib_entry_act_local_pack = mlxsw_sp_router_ll_basic_fib_entry_act_local_pack,
8403         .fib_entry_act_ip2me_pack = mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_pack,
8404         .fib_entry_act_ip2me_tun_pack = mlxsw_sp_router_ll_basic_fib_entry_act_ip2me_tun_pack,
8405         .fib_entry_commit = mlxsw_sp_router_ll_basic_fib_entry_commit,
8406         .fib_entry_is_committed = mlxsw_sp_router_ll_basic_fib_entry_is_committed,
8407 };
8408
8409 static int mlxsw_sp_router_ll_op_ctx_init(struct mlxsw_sp_router *router)
8410 {
8411         size_t max_size = 0;
8412         int i;
8413
8414         for (i = 0; i < MLXSW_SP_L3_PROTO_MAX; i++) {
8415                 size_t size = router->proto_ll_ops[i]->fib_entry_op_ctx_size;
8416
8417                 if (size > max_size)
8418                         max_size = size;
8419         }
8420         router->ll_op_ctx = kzalloc(sizeof(*router->ll_op_ctx) + max_size,
8421                                     GFP_KERNEL);
8422         if (!router->ll_op_ctx)
8423                 return -ENOMEM;
8424         INIT_LIST_HEAD(&router->ll_op_ctx->fib_entry_priv_list);
8425         return 0;
8426 }
8427
8428 static void mlxsw_sp_router_ll_op_ctx_fini(struct mlxsw_sp_router *router)
8429 {
8430         WARN_ON(!list_empty(&router->ll_op_ctx->fib_entry_priv_list));
8431         kfree(router->ll_op_ctx);
8432 }
8433
8434 int mlxsw_sp_router_init(struct mlxsw_sp *mlxsw_sp,
8435                          struct netlink_ext_ack *extack)
8436 {
8437         struct mlxsw_sp_router *router;
8438         int err;
8439
8440         router = kzalloc(sizeof(*mlxsw_sp->router), GFP_KERNEL);
8441         if (!router)
8442                 return -ENOMEM;
8443         mutex_init(&router->lock);
8444         mlxsw_sp->router = router;
8445         router->mlxsw_sp = mlxsw_sp;
8446
8447         router->proto_ll_ops[MLXSW_SP_L3_PROTO_IPV4] = &mlxsw_sp_router_ll_basic_ops;
8448         router->proto_ll_ops[MLXSW_SP_L3_PROTO_IPV6] = &mlxsw_sp_router_ll_basic_ops;
8449
8450         err = mlxsw_sp_router_ll_op_ctx_init(router);
8451         if (err)
8452                 goto err_ll_op_ctx_init;
8453
8454         INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_neighs_list);
8455         err = __mlxsw_sp_router_init(mlxsw_sp);
8456         if (err)
8457                 goto err_router_init;
8458
8459         err = mlxsw_sp_rifs_init(mlxsw_sp);
8460         if (err)
8461                 goto err_rifs_init;
8462
8463         err = mlxsw_sp_ipips_init(mlxsw_sp);
8464         if (err)
8465                 goto err_ipips_init;
8466
8467         err = rhashtable_init(&mlxsw_sp->router->nexthop_ht,
8468                               &mlxsw_sp_nexthop_ht_params);
8469         if (err)
8470                 goto err_nexthop_ht_init;
8471
8472         err = rhashtable_init(&mlxsw_sp->router->nexthop_group_ht,
8473                               &mlxsw_sp_nexthop_group_ht_params);
8474         if (err)
8475                 goto err_nexthop_group_ht_init;
8476
8477         INIT_LIST_HEAD(&mlxsw_sp->router->nexthop_list);
8478         err = mlxsw_sp_lpm_init(mlxsw_sp);
8479         if (err)
8480                 goto err_lpm_init;
8481
8482         err = mlxsw_sp_mr_init(mlxsw_sp, &mlxsw_sp_mr_tcam_ops);
8483         if (err)
8484                 goto err_mr_init;
8485
8486         err = mlxsw_sp_vrs_init(mlxsw_sp);
8487         if (err)
8488                 goto err_vrs_init;
8489
8490         err = mlxsw_sp_neigh_init(mlxsw_sp);
8491         if (err)
8492                 goto err_neigh_init;
8493
8494         err = mlxsw_sp_mp_hash_init(mlxsw_sp);
8495         if (err)
8496                 goto err_mp_hash_init;
8497
8498         err = mlxsw_sp_dscp_init(mlxsw_sp);
8499         if (err)
8500                 goto err_dscp_init;
8501
8502         INIT_WORK(&router->fib_event_work, mlxsw_sp_router_fib_event_work);
8503         INIT_LIST_HEAD(&router->fib_event_queue);
8504         spin_lock_init(&router->fib_event_queue_lock);
8505
8506         router->inetaddr_nb.notifier_call = mlxsw_sp_inetaddr_event;
8507         err = register_inetaddr_notifier(&router->inetaddr_nb);
8508         if (err)
8509                 goto err_register_inetaddr_notifier;
8510
8511         router->inet6addr_nb.notifier_call = mlxsw_sp_inet6addr_event;
8512         err = register_inet6addr_notifier(&router->inet6addr_nb);
8513         if (err)
8514                 goto err_register_inet6addr_notifier;
8515
8516         mlxsw_sp->router->netevent_nb.notifier_call =
8517                 mlxsw_sp_router_netevent_event;
8518         err = register_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8519         if (err)
8520                 goto err_register_netevent_notifier;
8521
8522         mlxsw_sp->router->fib_nb.notifier_call = mlxsw_sp_router_fib_event;
8523         err = register_fib_notifier(mlxsw_sp_net(mlxsw_sp),
8524                                     &mlxsw_sp->router->fib_nb,
8525                                     mlxsw_sp_router_fib_dump_flush, extack);
8526         if (err)
8527                 goto err_register_fib_notifier;
8528
8529         return 0;
8530
8531 err_register_fib_notifier:
8532         unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8533 err_register_netevent_notifier:
8534         unregister_inet6addr_notifier(&router->inet6addr_nb);
8535 err_register_inet6addr_notifier:
8536         unregister_inetaddr_notifier(&router->inetaddr_nb);
8537 err_register_inetaddr_notifier:
8538         mlxsw_core_flush_owq();
8539         WARN_ON(!list_empty(&router->fib_event_queue));
8540 err_dscp_init:
8541 err_mp_hash_init:
8542         mlxsw_sp_neigh_fini(mlxsw_sp);
8543 err_neigh_init:
8544         mlxsw_sp_vrs_fini(mlxsw_sp);
8545 err_vrs_init:
8546         mlxsw_sp_mr_fini(mlxsw_sp);
8547 err_mr_init:
8548         mlxsw_sp_lpm_fini(mlxsw_sp);
8549 err_lpm_init:
8550         rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
8551 err_nexthop_group_ht_init:
8552         rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
8553 err_nexthop_ht_init:
8554         mlxsw_sp_ipips_fini(mlxsw_sp);
8555 err_ipips_init:
8556         mlxsw_sp_rifs_fini(mlxsw_sp);
8557 err_rifs_init:
8558         __mlxsw_sp_router_fini(mlxsw_sp);
8559 err_router_init:
8560         mlxsw_sp_router_ll_op_ctx_fini(router);
8561 err_ll_op_ctx_init:
8562         mutex_destroy(&mlxsw_sp->router->lock);
8563         kfree(mlxsw_sp->router);
8564         return err;
8565 }
8566
8567 void mlxsw_sp_router_fini(struct mlxsw_sp *mlxsw_sp)
8568 {
8569         unregister_fib_notifier(mlxsw_sp_net(mlxsw_sp),
8570                                 &mlxsw_sp->router->fib_nb);
8571         unregister_netevent_notifier(&mlxsw_sp->router->netevent_nb);
8572         unregister_inet6addr_notifier(&mlxsw_sp->router->inet6addr_nb);
8573         unregister_inetaddr_notifier(&mlxsw_sp->router->inetaddr_nb);
8574         mlxsw_core_flush_owq();
8575         WARN_ON(!list_empty(&mlxsw_sp->router->fib_event_queue));
8576         mlxsw_sp_neigh_fini(mlxsw_sp);
8577         mlxsw_sp_vrs_fini(mlxsw_sp);
8578         mlxsw_sp_mr_fini(mlxsw_sp);
8579         mlxsw_sp_lpm_fini(mlxsw_sp);
8580         rhashtable_destroy(&mlxsw_sp->router->nexthop_group_ht);
8581         rhashtable_destroy(&mlxsw_sp->router->nexthop_ht);
8582         mlxsw_sp_ipips_fini(mlxsw_sp);
8583         mlxsw_sp_rifs_fini(mlxsw_sp);
8584         __mlxsw_sp_router_fini(mlxsw_sp);
8585         mlxsw_sp_router_ll_op_ctx_fini(mlxsw_sp->router);
8586         mutex_destroy(&mlxsw_sp->router->lock);
8587         kfree(mlxsw_sp->router);
8588 }