net/ipv4/nexthop.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Generic nexthop implementation
   3  *
   4  * Copyright (c) 2017-19 Cumulus Networks
   5  * Copyright (c) 2017-19 David Ahern <dsa@cumulusnetworks.com>
   6  */
   7
   8 #include <linux/nexthop.h>
   9 #include <linux/rtnetlink.h>
  10 #include <linux/slab.h>
  11 #include <linux/vmalloc.h>
  12 #include <net/arp.h>
  13 #include <net/ipv6_stubs.h>
  14 #include <net/lwtunnel.h>
  15 #include <net/ndisc.h>
  16 #include <net/nexthop.h>
  17 #include <net/route.h>
  18 #include <net/sock.h>
  19
  20 #define NH_RES_DEFAULT_IDLE_TIMER       (120 * HZ)
  21 #define NH_RES_DEFAULT_UNBALANCED_TIMER 0       /* No forced rebalancing. */
  22
  23 static void remove_nexthop(struct net *net, struct nexthop *nh,
  24                            struct nl_info *nlinfo);
  25
  26 #define NH_DEV_HASHBITS  8
  27 #define NH_DEV_HASHSIZE (1U << NH_DEV_HASHBITS)
  28
  29 static const struct nla_policy rtm_nh_policy_new[] = {
  30         [NHA_ID]                = { .type = NLA_U32 },
  31         [NHA_GROUP]             = { .type = NLA_BINARY },
  32         [NHA_GROUP_TYPE]        = { .type = NLA_U16 },
  33         [NHA_BLACKHOLE]         = { .type = NLA_FLAG },
  34         [NHA_OIF]               = { .type = NLA_U32 },
  35         [NHA_GATEWAY]           = { .type = NLA_BINARY },
  36         [NHA_ENCAP_TYPE]        = { .type = NLA_U16 },
  37         [NHA_ENCAP]             = { .type = NLA_NESTED },
  38         [NHA_FDB]               = { .type = NLA_FLAG },
  39         [NHA_RES_GROUP]         = { .type = NLA_NESTED },
  40 };
  41
  42 static const struct nla_policy rtm_nh_policy_get[] = {
  43         [NHA_ID]                = { .type = NLA_U32 },
  44 };
  45
  46 static const struct nla_policy rtm_nh_policy_dump[] = {
  47         [NHA_OIF]               = { .type = NLA_U32 },
  48         [NHA_GROUPS]            = { .type = NLA_FLAG },
  49         [NHA_MASTER]            = { .type = NLA_U32 },
  50         [NHA_FDB]               = { .type = NLA_FLAG },
  51 };
  52
  53 static const struct nla_policy rtm_nh_res_policy_new[] = {
  54         [NHA_RES_GROUP_BUCKETS]                 = { .type = NLA_U16 },
  55         [NHA_RES_GROUP_IDLE_TIMER]              = { .type = NLA_U32 },
  56         [NHA_RES_GROUP_UNBALANCED_TIMER]        = { .type = NLA_U32 },
  57 };
  58
  59 static const struct nla_policy rtm_nh_policy_dump_bucket[] = {
  60         [NHA_ID]                = { .type = NLA_U32 },
  61         [NHA_OIF]               = { .type = NLA_U32 },
  62         [NHA_MASTER]            = { .type = NLA_U32 },
  63         [NHA_RES_BUCKET]        = { .type = NLA_NESTED },
  64 };
  65
  66 static const struct nla_policy rtm_nh_res_bucket_policy_dump[] = {
  67         [NHA_RES_BUCKET_NH_ID]  = { .type = NLA_U32 },
  68 };
  69
  70 static const struct nla_policy rtm_nh_policy_get_bucket[] = {
  71         [NHA_ID]                = { .type = NLA_U32 },
  72         [NHA_RES_BUCKET]        = { .type = NLA_NESTED },
  73 };
  74
  75 static const struct nla_policy rtm_nh_res_bucket_policy_get[] = {
  76         [NHA_RES_BUCKET_INDEX]  = { .type = NLA_U16 },
  77 };
  78
  79 static bool nexthop_notifiers_is_empty(struct net *net)
  80 {
  81         return !net->nexthop.notifier_chain.head;
  82 }
  83
  84 static void
  85 __nh_notifier_single_info_init(struct nh_notifier_single_info *nh_info,
  86                                const struct nh_info *nhi)
  87 {
  88         nh_info->dev = nhi->fib_nhc.nhc_dev;
  89         nh_info->gw_family = nhi->fib_nhc.nhc_gw_family;
  90         if (nh_info->gw_family == AF_INET)
  91                 nh_info->ipv4 = nhi->fib_nhc.nhc_gw.ipv4;
  92         else if (nh_info->gw_family == AF_INET6)
  93                 nh_info->ipv6 = nhi->fib_nhc.nhc_gw.ipv6;
  94
  95         nh_info->is_reject = nhi->reject_nh;
  96         nh_info->is_fdb = nhi->fdb_nh;
  97         nh_info->has_encap = !!nhi->fib_nhc.nhc_lwtstate;
  98 }
  99
 100 static int nh_notifier_single_info_init(struct nh_notifier_info *info,
 101                                         const struct nexthop *nh)
 102 {
 103         struct nh_info *nhi = rtnl_dereference(nh->nh_info);
 104
 105         info->type = NH_NOTIFIER_INFO_TYPE_SINGLE;
 106         info->nh = kzalloc(sizeof(*info->nh), GFP_KERNEL);
 107         if (!info->nh)
 108                 return -ENOMEM;
 109
 110         __nh_notifier_single_info_init(info->nh, nhi);
 111
 112         return 0;
 113 }
 114
 115 static void nh_notifier_single_info_fini(struct nh_notifier_info *info)
 116 {
 117         kfree(info->nh);
 118 }
 119
 120 static int nh_notifier_mpath_info_init(struct nh_notifier_info *info,
 121                                        struct nh_group *nhg)
 122 {
 123         u16 num_nh = nhg->num_nh;
 124         int i;
 125
 126         info->type = NH_NOTIFIER_INFO_TYPE_GRP;
 127         info->nh_grp = kzalloc(struct_size(info->nh_grp, nh_entries, num_nh),
 128                                GFP_KERNEL);
 129         if (!info->nh_grp)
 130                 return -ENOMEM;
 131
 132         info->nh_grp->num_nh = num_nh;
 133         info->nh_grp->is_fdb = nhg->fdb_nh;
 134
 135         for (i = 0; i < num_nh; i++) {
 136                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 137                 struct nh_info *nhi;
 138
 139                 nhi = rtnl_dereference(nhge->nh->nh_info);
 140                 info->nh_grp->nh_entries[i].id = nhge->nh->id;
 141                 info->nh_grp->nh_entries[i].weight = nhge->weight;
 142                 __nh_notifier_single_info_init(&info->nh_grp->nh_entries[i].nh,
 143                                                nhi);
 144         }
 145
 146         return 0;
 147 }
 148
 149 static int nh_notifier_res_table_info_init(struct nh_notifier_info *info,
 150                                            struct nh_group *nhg)
 151 {
 152         struct nh_res_table *res_table = rtnl_dereference(nhg->res_table);
 153         u16 num_nh_buckets = res_table->num_nh_buckets;
 154         unsigned long size;
 155         u16 i;
 156
 157         info->type = NH_NOTIFIER_INFO_TYPE_RES_TABLE;
 158         size = struct_size(info->nh_res_table, nhs, num_nh_buckets);
 159         info->nh_res_table = __vmalloc(size, GFP_KERNEL | __GFP_ZERO |
 160                                        __GFP_NOWARN);
 161         if (!info->nh_res_table)
 162                 return -ENOMEM;
 163
 164         info->nh_res_table->num_nh_buckets = num_nh_buckets;
 165
 166         for (i = 0; i < num_nh_buckets; i++) {
 167                 struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
 168                 struct nh_grp_entry *nhge;
 169                 struct nh_info *nhi;
 170
 171                 nhge = rtnl_dereference(bucket->nh_entry);
 172                 nhi = rtnl_dereference(nhge->nh->nh_info);
 173                 __nh_notifier_single_info_init(&info->nh_res_table->nhs[i],
 174                                                nhi);
 175         }
 176
 177         return 0;
 178 }
 179
 180 static int nh_notifier_grp_info_init(struct nh_notifier_info *info,
 181                                      const struct nexthop *nh)
 182 {
 183         struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 184
 185         if (nhg->hash_threshold)
 186                 return nh_notifier_mpath_info_init(info, nhg);
 187         else if (nhg->resilient)
 188                 return nh_notifier_res_table_info_init(info, nhg);
 189         return -EINVAL;
 190 }
 191
 192 static void nh_notifier_grp_info_fini(struct nh_notifier_info *info,
 193                                       const struct nexthop *nh)
 194 {
 195         struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 196
 197         if (nhg->hash_threshold)
 198                 kfree(info->nh_grp);
 199         else if (nhg->resilient)
 200                 vfree(info->nh_res_table);
 201 }
 202
 203 static int nh_notifier_info_init(struct nh_notifier_info *info,
 204                                  const struct nexthop *nh)
 205 {
 206         info->id = nh->id;
 207
 208         if (nh->is_group)
 209                 return nh_notifier_grp_info_init(info, nh);
 210         else
 211                 return nh_notifier_single_info_init(info, nh);
 212 }
 213
 214 static void nh_notifier_info_fini(struct nh_notifier_info *info,
 215                                   const struct nexthop *nh)
 216 {
 217         if (nh->is_group)
 218                 nh_notifier_grp_info_fini(info, nh);
 219         else
 220                 nh_notifier_single_info_fini(info);
 221 }
 222
 223 static int call_nexthop_notifiers(struct net *net,
 224                                   enum nexthop_event_type event_type,
 225                                   struct nexthop *nh,
 226                                   struct netlink_ext_ack *extack)
 227 {
 228         struct nh_notifier_info info = {
 229                 .net = net,
 230                 .extack = extack,
 231         };
 232         int err;
 233
 234         ASSERT_RTNL();
 235
 236         if (nexthop_notifiers_is_empty(net))
 237                 return 0;
 238
 239         err = nh_notifier_info_init(&info, nh);
 240         if (err) {
 241                 NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info");
 242                 return err;
 243         }
 244
 245         err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
 246                                            event_type, &info);
 247         nh_notifier_info_fini(&info, nh);
 248
 249         return notifier_to_errno(err);
 250 }
 251
 252 static int
 253 nh_notifier_res_bucket_idle_timer_get(const struct nh_notifier_info *info,
 254                                       bool force, unsigned int *p_idle_timer_ms)
 255 {
 256         struct nh_res_table *res_table;
 257         struct nh_group *nhg;
 258         struct nexthop *nh;
 259         int err = 0;
 260
 261         /* When 'force' is false, nexthop bucket replacement is performed
 262          * because the bucket was deemed to be idle. In this case, capable
 263          * listeners can choose to perform an atomic replacement: The bucket is
 264          * only replaced if it is inactive. However, if the idle timer interval
 265          * is smaller than the interval in which a listener is querying
 266          * buckets' activity from the device, then atomic replacement should
 267          * not be tried. Pass the idle timer value to listeners, so that they
 268          * could determine which type of replacement to perform.
 269          */
 270         if (force) {
 271                 *p_idle_timer_ms = 0;
 272                 return 0;
 273         }
 274
 275         rcu_read_lock();
 276
 277         nh = nexthop_find_by_id(info->net, info->id);
 278         if (!nh) {
 279                 err = -EINVAL;
 280                 goto out;
 281         }
 282
 283         nhg = rcu_dereference(nh->nh_grp);
 284         res_table = rcu_dereference(nhg->res_table);
 285         *p_idle_timer_ms = jiffies_to_msecs(res_table->idle_timer);
 286
 287 out:
 288         rcu_read_unlock();
 289
 290         return err;
 291 }
 292
 293 static int nh_notifier_res_bucket_info_init(struct nh_notifier_info *info,
 294                                             u16 bucket_index, bool force,
 295                                             struct nh_info *oldi,
 296                                             struct nh_info *newi)
 297 {
 298         unsigned int idle_timer_ms;
 299         int err;
 300
 301         err = nh_notifier_res_bucket_idle_timer_get(info, force,
 302                                                     &idle_timer_ms);
 303         if (err)
 304                 return err;
 305
 306         info->type = NH_NOTIFIER_INFO_TYPE_RES_BUCKET;
 307         info->nh_res_bucket = kzalloc(sizeof(*info->nh_res_bucket),
 308                                       GFP_KERNEL);
 309         if (!info->nh_res_bucket)
 310                 return -ENOMEM;
 311
 312         info->nh_res_bucket->bucket_index = bucket_index;
 313         info->nh_res_bucket->idle_timer_ms = idle_timer_ms;
 314         info->nh_res_bucket->force = force;
 315         __nh_notifier_single_info_init(&info->nh_res_bucket->old_nh, oldi);
 316         __nh_notifier_single_info_init(&info->nh_res_bucket->new_nh, newi);
 317         return 0;
 318 }
 319
 320 static void nh_notifier_res_bucket_info_fini(struct nh_notifier_info *info)
 321 {
 322         kfree(info->nh_res_bucket);
 323 }
 324
 325 static int __call_nexthop_res_bucket_notifiers(struct net *net, u32 nhg_id,
 326                                                u16 bucket_index, bool force,
 327                                                struct nh_info *oldi,
 328                                                struct nh_info *newi,
 329                                                struct netlink_ext_ack *extack)
 330 {
 331         struct nh_notifier_info info = {
 332                 .net = net,
 333                 .extack = extack,
 334                 .id = nhg_id,
 335         };
 336         int err;
 337
 338         if (nexthop_notifiers_is_empty(net))
 339                 return 0;
 340
 341         err = nh_notifier_res_bucket_info_init(&info, bucket_index, force,
 342                                                oldi, newi);
 343         if (err)
 344                 return err;
 345
 346         err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
 347                                            NEXTHOP_EVENT_BUCKET_REPLACE, &info);
 348         nh_notifier_res_bucket_info_fini(&info);
 349
 350         return notifier_to_errno(err);
 351 }
 352
 353 /* There are three users of RES_TABLE, and NHs etc. referenced from there:
 354  *
 355  * 1) a collection of callbacks for NH maintenance. This operates under
 356  *    RTNL,
 357  * 2) the delayed work that gradually balances the resilient table,
 358  * 3) and nexthop_select_path(), operating under RCU.
 359  *
 360  * Both the delayed work and the RTNL block are writers, and need to
 361  * maintain mutual exclusion. Since there are only two and well-known
 362  * writers for each table, the RTNL code can make sure it has exclusive
 363  * access thus:
 364  *
 365  * - Have the DW operate without locking;
 366  * - synchronously cancel the DW;
 367  * - do the writing;
 368  * - if the write was not actually a delete, call upkeep, which schedules
 369  *   DW again if necessary.
 370  *
 371  * The functions that are always called from the RTNL context use
 372  * rtnl_dereference(). The functions that can also be called from the DW do
 373  * a raw dereference and rely on the above mutual exclusion scheme.
 374  */
 375 #define nh_res_dereference(p) (rcu_dereference_raw(p))
 376
 377 static int call_nexthop_res_bucket_notifiers(struct net *net, u32 nhg_id,
 378                                              u16 bucket_index, bool force,
 379                                              struct nexthop *old_nh,
 380                                              struct nexthop *new_nh,
 381                                              struct netlink_ext_ack *extack)
 382 {
 383         struct nh_info *oldi = nh_res_dereference(old_nh->nh_info);
 384         struct nh_info *newi = nh_res_dereference(new_nh->nh_info);
 385
 386         return __call_nexthop_res_bucket_notifiers(net, nhg_id, bucket_index,
 387                                                    force, oldi, newi, extack);
 388 }
 389
 390 static int call_nexthop_res_table_notifiers(struct net *net, struct nexthop *nh,
 391                                             struct netlink_ext_ack *extack)
 392 {
 393         struct nh_notifier_info info = {
 394                 .net = net,
 395                 .extack = extack,
 396         };
 397         struct nh_group *nhg;
 398         int err;
 399
 400         ASSERT_RTNL();
 401
 402         if (nexthop_notifiers_is_empty(net))
 403                 return 0;
 404
 405         /* At this point, the nexthop buckets are still not populated. Only
 406          * emit a notification with the logical nexthops, so that a listener
 407          * could potentially veto it in case of unsupported configuration.
 408          */
 409         nhg = rtnl_dereference(nh->nh_grp);
 410         err = nh_notifier_mpath_info_init(&info, nhg);
 411         if (err) {
 412                 NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info");
 413                 return err;
 414         }
 415
 416         err = blocking_notifier_call_chain(&net->nexthop.notifier_chain,
 417                                            NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE,
 418                                            &info);
 419         kfree(info.nh_grp);
 420
 421         return notifier_to_errno(err);
 422 }
 423
 424 static int call_nexthop_notifier(struct notifier_block *nb, struct net *net,
 425                                  enum nexthop_event_type event_type,
 426                                  struct nexthop *nh,
 427                                  struct netlink_ext_ack *extack)
 428 {
 429         struct nh_notifier_info info = {
 430                 .net = net,
 431                 .extack = extack,
 432         };
 433         int err;
 434
 435         err = nh_notifier_info_init(&info, nh);
 436         if (err)
 437                 return err;
 438
 439         err = nb->notifier_call(nb, event_type, &info);
 440         nh_notifier_info_fini(&info, nh);
 441
 442         return notifier_to_errno(err);
 443 }
 444
 445 static unsigned int nh_dev_hashfn(unsigned int val)
 446 {
 447         unsigned int mask = NH_DEV_HASHSIZE - 1;
 448
 449         return (val ^
 450                 (val >> NH_DEV_HASHBITS) ^
 451                 (val >> (NH_DEV_HASHBITS * 2))) & mask;
 452 }
 453
 454 static void nexthop_devhash_add(struct net *net, struct nh_info *nhi)
 455 {
 456         struct net_device *dev = nhi->fib_nhc.nhc_dev;
 457         struct hlist_head *head;
 458         unsigned int hash;
 459
 460         WARN_ON(!dev);
 461
 462         hash = nh_dev_hashfn(dev->ifindex);
 463         head = &net->nexthop.devhash[hash];
 464         hlist_add_head(&nhi->dev_hash, head);
 465 }
 466
 467 static void nexthop_free_group(struct nexthop *nh)
 468 {
 469         struct nh_group *nhg;
 470         int i;
 471
 472         nhg = rcu_dereference_raw(nh->nh_grp);
 473         for (i = 0; i < nhg->num_nh; ++i) {
 474                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 475
 476                 WARN_ON(!list_empty(&nhge->nh_list));
 477                 nexthop_put(nhge->nh);
 478         }
 479
 480         WARN_ON(nhg->spare == nhg);
 481
 482         if (nhg->resilient)
 483                 vfree(rcu_dereference_raw(nhg->res_table));
 484
 485         kfree(nhg->spare);
 486         kfree(nhg);
 487 }
 488
 489 static void nexthop_free_single(struct nexthop *nh)
 490 {
 491         struct nh_info *nhi;
 492
 493         nhi = rcu_dereference_raw(nh->nh_info);
 494         switch (nhi->family) {
 495         case AF_INET:
 496                 fib_nh_release(nh->net, &nhi->fib_nh);
 497                 break;
 498         case AF_INET6:
 499                 ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
 500                 break;
 501         }
 502         kfree(nhi);
 503 }
 504
 505 void nexthop_free_rcu(struct rcu_head *head)
 506 {
 507         struct nexthop *nh = container_of(head, struct nexthop, rcu);
 508
 509         if (nh->is_group)
 510                 nexthop_free_group(nh);
 511         else
 512                 nexthop_free_single(nh);
 513
 514         kfree(nh);
 515 }
 516 EXPORT_SYMBOL_GPL(nexthop_free_rcu);
 517
 518 static struct nexthop *nexthop_alloc(void)
 519 {
 520         struct nexthop *nh;
 521
 522         nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
 523         if (nh) {
 524                 INIT_LIST_HEAD(&nh->fi_list);
 525                 INIT_LIST_HEAD(&nh->f6i_list);
 526                 INIT_LIST_HEAD(&nh->grp_list);
 527                 INIT_LIST_HEAD(&nh->fdb_list);
 528         }
 529         return nh;
 530 }
 531
 532 static struct nh_group *nexthop_grp_alloc(u16 num_nh)
 533 {
 534         struct nh_group *nhg;
 535
 536         nhg = kzalloc(struct_size(nhg, nh_entries, num_nh), GFP_KERNEL);
 537         if (nhg)
 538                 nhg->num_nh = num_nh;
 539
 540         return nhg;
 541 }
 542
 543 static void nh_res_table_upkeep_dw(struct work_struct *work);
 544
 545 static struct nh_res_table *
 546 nexthop_res_table_alloc(struct net *net, u32 nhg_id, struct nh_config *cfg)
 547 {
 548         const u16 num_nh_buckets = cfg->nh_grp_res_num_buckets;
 549         struct nh_res_table *res_table;
 550         unsigned long size;
 551
 552         size = struct_size(res_table, nh_buckets, num_nh_buckets);
 553         res_table = __vmalloc(size, GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN);
 554         if (!res_table)
 555                 return NULL;
 556
 557         res_table->net = net;
 558         res_table->nhg_id = nhg_id;
 559         INIT_DELAYED_WORK(&res_table->upkeep_dw, &nh_res_table_upkeep_dw);
 560         INIT_LIST_HEAD(&res_table->uw_nh_entries);
 561         res_table->idle_timer = cfg->nh_grp_res_idle_timer;
 562         res_table->unbalanced_timer = cfg->nh_grp_res_unbalanced_timer;
 563         res_table->num_nh_buckets = num_nh_buckets;
 564         return res_table;
 565 }
 566
 567 static void nh_base_seq_inc(struct net *net)
 568 {
 569         while (++net->nexthop.seq == 0)
 570                 ;
 571 }
 572
 573 /* no reference taken; rcu lock or rtnl must be held */
 574 struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
 575 {
 576         struct rb_node **pp, *parent = NULL, *next;
 577
 578         pp = &net->nexthop.rb_root.rb_node;
 579         while (1) {
 580                 struct nexthop *nh;
 581
 582                 next = rcu_dereference_raw(*pp);
 583                 if (!next)
 584                         break;
 585                 parent = next;
 586
 587                 nh = rb_entry(parent, struct nexthop, rb_node);
 588                 if (id < nh->id)
 589                         pp = &next->rb_left;
 590                 else if (id > nh->id)
 591                         pp = &next->rb_right;
 592                 else
 593                         return nh;
 594         }
 595         return NULL;
 596 }
 597 EXPORT_SYMBOL_GPL(nexthop_find_by_id);
 598
 599 /* used for auto id allocation; called with rtnl held */
 600 static u32 nh_find_unused_id(struct net *net)
 601 {
 602         u32 id_start = net->nexthop.last_id_allocated;
 603
 604         while (1) {
 605                 net->nexthop.last_id_allocated++;
 606                 if (net->nexthop.last_id_allocated == id_start)
 607                         break;
 608
 609                 if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
 610                         return net->nexthop.last_id_allocated;
 611         }
 612         return 0;
 613 }
 614
 615 static void nh_res_time_set_deadline(unsigned long next_time,
 616                                      unsigned long *deadline)
 617 {
 618         if (time_before(next_time, *deadline))
 619                 *deadline = next_time;
 620 }
 621
 622 static clock_t nh_res_table_unbalanced_time(struct nh_res_table *res_table)
 623 {
 624         if (list_empty(&res_table->uw_nh_entries))
 625                 return 0;
 626         return jiffies_delta_to_clock_t(jiffies - res_table->unbalanced_since);
 627 }
 628
 629 static int nla_put_nh_group_res(struct sk_buff *skb, struct nh_group *nhg)
 630 {
 631         struct nh_res_table *res_table = rtnl_dereference(nhg->res_table);
 632         struct nlattr *nest;
 633
 634         nest = nla_nest_start(skb, NHA_RES_GROUP);
 635         if (!nest)
 636                 return -EMSGSIZE;
 637
 638         if (nla_put_u16(skb, NHA_RES_GROUP_BUCKETS,
 639                         res_table->num_nh_buckets) ||
 640             nla_put_u32(skb, NHA_RES_GROUP_IDLE_TIMER,
 641                         jiffies_to_clock_t(res_table->idle_timer)) ||
 642             nla_put_u32(skb, NHA_RES_GROUP_UNBALANCED_TIMER,
 643                         jiffies_to_clock_t(res_table->unbalanced_timer)) ||
 644             nla_put_u64_64bit(skb, NHA_RES_GROUP_UNBALANCED_TIME,
 645                               nh_res_table_unbalanced_time(res_table),
 646                               NHA_RES_GROUP_PAD))
 647                 goto nla_put_failure;
 648
 649         nla_nest_end(skb, nest);
 650         return 0;
 651
 652 nla_put_failure:
 653         nla_nest_cancel(skb, nest);
 654         return -EMSGSIZE;
 655 }
 656
 657 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
 658 {
 659         struct nexthop_grp *p;
 660         size_t len = nhg->num_nh * sizeof(*p);
 661         struct nlattr *nla;
 662         u16 group_type = 0;
 663         int i;
 664
 665         if (nhg->hash_threshold)
 666                 group_type = NEXTHOP_GRP_TYPE_MPATH;
 667         else if (nhg->resilient)
 668                 group_type = NEXTHOP_GRP_TYPE_RES;
 669
 670         if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
 671                 goto nla_put_failure;
 672
 673         nla = nla_reserve(skb, NHA_GROUP, len);
 674         if (!nla)
 675                 goto nla_put_failure;
 676
 677         p = nla_data(nla);
 678         for (i = 0; i < nhg->num_nh; ++i) {
 679                 p->id = nhg->nh_entries[i].nh->id;
 680                 p->weight = nhg->nh_entries[i].weight - 1;
 681                 p += 1;
 682         }
 683
 684         if (nhg->resilient && nla_put_nh_group_res(skb, nhg))
 685                 goto nla_put_failure;
 686
 687         return 0;
 688
 689 nla_put_failure:
 690         return -EMSGSIZE;
 691 }
 692
 693 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
 694                         int event, u32 portid, u32 seq, unsigned int nlflags)
 695 {
 696         struct fib6_nh *fib6_nh;
 697         struct fib_nh *fib_nh;
 698         struct nlmsghdr *nlh;
 699         struct nh_info *nhi;
 700         struct nhmsg *nhm;
 701
 702         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
 703         if (!nlh)
 704                 return -EMSGSIZE;
 705
 706         nhm = nlmsg_data(nlh);
 707         nhm->nh_family = AF_UNSPEC;
 708         nhm->nh_flags = nh->nh_flags;
 709         nhm->nh_protocol = nh->protocol;
 710         nhm->nh_scope = 0;
 711         nhm->resvd = 0;
 712
 713         if (nla_put_u32(skb, NHA_ID, nh->id))
 714                 goto nla_put_failure;
 715
 716         if (nh->is_group) {
 717                 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 718
 719                 if (nhg->fdb_nh && nla_put_flag(skb, NHA_FDB))
 720                         goto nla_put_failure;
 721                 if (nla_put_nh_group(skb, nhg))
 722                         goto nla_put_failure;
 723                 goto out;
 724         }
 725
 726         nhi = rtnl_dereference(nh->nh_info);
 727         nhm->nh_family = nhi->family;
 728         if (nhi->reject_nh) {
 729                 if (nla_put_flag(skb, NHA_BLACKHOLE))
 730                         goto nla_put_failure;
 731                 goto out;
 732         } else if (nhi->fdb_nh) {
 733                 if (nla_put_flag(skb, NHA_FDB))
 734                         goto nla_put_failure;
 735         } else {
 736                 const struct net_device *dev;
 737
 738                 dev = nhi->fib_nhc.nhc_dev;
 739                 if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
 740                         goto nla_put_failure;
 741         }
 742
 743         nhm->nh_scope = nhi->fib_nhc.nhc_scope;
 744         switch (nhi->family) {
 745         case AF_INET:
 746                 fib_nh = &nhi->fib_nh;
 747                 if (fib_nh->fib_nh_gw_family &&
 748                     nla_put_be32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
 749                         goto nla_put_failure;
 750                 break;
 751
 752         case AF_INET6:
 753                 fib6_nh = &nhi->fib6_nh;
 754                 if (fib6_nh->fib_nh_gw_family &&
 755                     nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
 756                         goto nla_put_failure;
 757                 break;
 758         }
 759
 760         if (nhi->fib_nhc.nhc_lwtstate &&
 761             lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
 762                                 NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
 763                 goto nla_put_failure;
 764
 765 out:
 766         nlmsg_end(skb, nlh);
 767         return 0;
 768
 769 nla_put_failure:
 770         nlmsg_cancel(skb, nlh);
 771         return -EMSGSIZE;
 772 }
 773
 774 static size_t nh_nlmsg_size_grp_res(struct nh_group *nhg)
 775 {
 776         return nla_total_size(0) +      /* NHA_RES_GROUP */
 777                 nla_total_size(2) +     /* NHA_RES_GROUP_BUCKETS */
 778                 nla_total_size(4) +     /* NHA_RES_GROUP_IDLE_TIMER */
 779                 nla_total_size(4) +     /* NHA_RES_GROUP_UNBALANCED_TIMER */
 780                 nla_total_size_64bit(8);/* NHA_RES_GROUP_UNBALANCED_TIME */
 781 }
 782
 783 static size_t nh_nlmsg_size_grp(struct nexthop *nh)
 784 {
 785         struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 786         size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
 787         size_t tot = nla_total_size(sz) +
 788                 nla_total_size(2); /* NHA_GROUP_TYPE */
 789
 790         if (nhg->resilient)
 791                 tot += nh_nlmsg_size_grp_res(nhg);
 792
 793         return tot;
 794 }
 795
 796 static size_t nh_nlmsg_size_single(struct nexthop *nh)
 797 {
 798         struct nh_info *nhi = rtnl_dereference(nh->nh_info);
 799         size_t sz;
 800
 801         /* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
 802          * are mutually exclusive
 803          */
 804         sz = nla_total_size(4);  /* NHA_OIF */
 805
 806         switch (nhi->family) {
 807         case AF_INET:
 808                 if (nhi->fib_nh.fib_nh_gw_family)
 809                         sz += nla_total_size(4);  /* NHA_GATEWAY */
 810                 break;
 811
 812         case AF_INET6:
 813                 /* NHA_GATEWAY */
 814                 if (nhi->fib6_nh.fib_nh_gw_family)
 815                         sz += nla_total_size(sizeof(const struct in6_addr));
 816                 break;
 817         }
 818
 819         if (nhi->fib_nhc.nhc_lwtstate) {
 820                 sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
 821                 sz += nla_total_size(2);  /* NHA_ENCAP_TYPE */
 822         }
 823
 824         return sz;
 825 }
 826
 827 static size_t nh_nlmsg_size(struct nexthop *nh)
 828 {
 829         size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg));
 830
 831         sz += nla_total_size(4); /* NHA_ID */
 832
 833         if (nh->is_group)
 834                 sz += nh_nlmsg_size_grp(nh);
 835         else
 836                 sz += nh_nlmsg_size_single(nh);
 837
 838         return sz;
 839 }
 840
 841 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
 842 {
 843         unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
 844         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 845         struct sk_buff *skb;
 846         int err = -ENOBUFS;
 847
 848         skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
 849         if (!skb)
 850                 goto errout;
 851
 852         err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
 853         if (err < 0) {
 854                 /* -EMSGSIZE implies BUG in nh_nlmsg_size() */
 855                 WARN_ON(err == -EMSGSIZE);
 856                 kfree_skb(skb);
 857                 goto errout;
 858         }
 859
 860         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
 861                     info->nlh, gfp_any());
 862         return;
 863 errout:
 864         if (err < 0)
 865                 rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
 866 }
 867
 868 static unsigned long nh_res_bucket_used_time(const struct nh_res_bucket *bucket)
 869 {
 870         return (unsigned long)atomic_long_read(&bucket->used_time);
 871 }
 872
 873 static unsigned long
 874 nh_res_bucket_idle_point(const struct nh_res_table *res_table,
 875                          const struct nh_res_bucket *bucket,
 876                          unsigned long now)
 877 {
 878         unsigned long time = nh_res_bucket_used_time(bucket);
 879
 880         /* Bucket was not used since it was migrated. The idle time is now. */
 881         if (time == bucket->migrated_time)
 882                 return now;
 883
 884         return time + res_table->idle_timer;
 885 }
 886
 887 static unsigned long
 888 nh_res_table_unb_point(const struct nh_res_table *res_table)
 889 {
 890         return res_table->unbalanced_since + res_table->unbalanced_timer;
 891 }
 892
 893 static void nh_res_bucket_set_idle(const struct nh_res_table *res_table,
 894                                    struct nh_res_bucket *bucket)
 895 {
 896         unsigned long now = jiffies;
 897
 898         atomic_long_set(&bucket->used_time, (long)now);
 899         bucket->migrated_time = now;
 900 }
 901
 902 static void nh_res_bucket_set_busy(struct nh_res_bucket *bucket)
 903 {
 904         atomic_long_set(&bucket->used_time, (long)jiffies);
 905 }
 906
 907 static clock_t nh_res_bucket_idle_time(const struct nh_res_bucket *bucket)
 908 {
 909         unsigned long used_time = nh_res_bucket_used_time(bucket);
 910
 911         return jiffies_delta_to_clock_t(jiffies - used_time);
 912 }
 913
 914 static int nh_fill_res_bucket(struct sk_buff *skb, struct nexthop *nh,
 915                               struct nh_res_bucket *bucket, u16 bucket_index,
 916                               int event, u32 portid, u32 seq,
 917                               unsigned int nlflags,
 918                               struct netlink_ext_ack *extack)
 919 {
 920         struct nh_grp_entry *nhge = nh_res_dereference(bucket->nh_entry);
 921         struct nlmsghdr *nlh;
 922         struct nlattr *nest;
 923         struct nhmsg *nhm;
 924
 925         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
 926         if (!nlh)
 927                 return -EMSGSIZE;
 928
 929         nhm = nlmsg_data(nlh);
 930         nhm->nh_family = AF_UNSPEC;
 931         nhm->nh_flags = bucket->nh_flags;
 932         nhm->nh_protocol = nh->protocol;
 933         nhm->nh_scope = 0;
 934         nhm->resvd = 0;
 935
 936         if (nla_put_u32(skb, NHA_ID, nh->id))
 937                 goto nla_put_failure;
 938
 939         nest = nla_nest_start(skb, NHA_RES_BUCKET);
 940         if (!nest)
 941                 goto nla_put_failure;
 942
 943         if (nla_put_u16(skb, NHA_RES_BUCKET_INDEX, bucket_index) ||
 944             nla_put_u32(skb, NHA_RES_BUCKET_NH_ID, nhge->nh->id) ||
 945             nla_put_u64_64bit(skb, NHA_RES_BUCKET_IDLE_TIME,
 946                               nh_res_bucket_idle_time(bucket),
 947                               NHA_RES_BUCKET_PAD))
 948                 goto nla_put_failure_nest;
 949
 950         nla_nest_end(skb, nest);
 951         nlmsg_end(skb, nlh);
 952         return 0;
 953
 954 nla_put_failure_nest:
 955         nla_nest_cancel(skb, nest);
 956 nla_put_failure:
 957         nlmsg_cancel(skb, nlh);
 958         return -EMSGSIZE;
 959 }
 960
 961 static void nexthop_bucket_notify(struct nh_res_table *res_table,
 962                                   u16 bucket_index)
 963 {
 964         struct nh_res_bucket *bucket = &res_table->nh_buckets[bucket_index];
 965         struct nh_grp_entry *nhge = nh_res_dereference(bucket->nh_entry);
 966         struct nexthop *nh = nhge->nh_parent;
 967         struct sk_buff *skb;
 968         int err = -ENOBUFS;
 969
 970         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
 971         if (!skb)
 972                 goto errout;
 973
 974         err = nh_fill_res_bucket(skb, nh, bucket, bucket_index,
 975                                  RTM_NEWNEXTHOPBUCKET, 0, 0, NLM_F_REPLACE,
 976                                  NULL);
 977         if (err < 0) {
 978                 kfree_skb(skb);
 979                 goto errout;
 980         }
 981
 982         rtnl_notify(skb, nh->net, 0, RTNLGRP_NEXTHOP, NULL, GFP_KERNEL);
 983         return;
 984 errout:
 985         if (err < 0)
 986                 rtnl_set_sk_err(nh->net, RTNLGRP_NEXTHOP, err);
 987 }
 988
 989 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
 990                            bool *is_fdb, struct netlink_ext_ack *extack)
 991 {
 992         if (nh->is_group) {
 993                 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 994
 995                 /* Nesting groups within groups is not supported. */
 996                 if (nhg->hash_threshold) {
 997                         NL_SET_ERR_MSG(extack,
 998                                        "Hash-threshold group can not be a nexthop within a group");
 999                         return false;
1000                 }
1001                 if (nhg->resilient) {
1002                         NL_SET_ERR_MSG(extack,
1003                                        "Resilient group can not be a nexthop within a group");
1004                         return false;
1005                 }
1006                 *is_fdb = nhg->fdb_nh;
1007         } else {
1008                 struct nh_info *nhi = rtnl_dereference(nh->nh_info);
1009
1010                 if (nhi->reject_nh && npaths > 1) {
1011                         NL_SET_ERR_MSG(extack,
1012                                        "Blackhole nexthop can not be used in a group with more than 1 path");
1013                         return false;
1014                 }
1015                 *is_fdb = nhi->fdb_nh;
1016         }
1017
1018         return true;
1019 }
1020
1021 static int nh_check_attr_fdb_group(struct nexthop *nh, u8 *nh_family,
1022                                    struct netlink_ext_ack *extack)
1023 {
1024         struct nh_info *nhi;
1025
1026         nhi = rtnl_dereference(nh->nh_info);
1027
1028         if (!nhi->fdb_nh) {
1029                 NL_SET_ERR_MSG(extack, "FDB nexthop group can only have fdb nexthops");
1030                 return -EINVAL;
1031         }
1032
1033         if (*nh_family == AF_UNSPEC) {
1034                 *nh_family = nhi->family;
1035         } else if (*nh_family != nhi->family) {
1036                 NL_SET_ERR_MSG(extack, "FDB nexthop group cannot have mixed family nexthops");
1037                 return -EINVAL;
1038         }
1039
1040         return 0;
1041 }
1042
1043 static int nh_check_attr_group(struct net *net,
1044                                struct nlattr *tb[], size_t tb_size,
1045                                u16 nh_grp_type, struct netlink_ext_ack *extack)
1046 {
1047         unsigned int len = nla_len(tb[NHA_GROUP]);
1048         u8 nh_family = AF_UNSPEC;
1049         struct nexthop_grp *nhg;
1050         unsigned int i, j;
1051         u8 nhg_fdb = 0;
1052
1053         if (!len || len & (sizeof(struct nexthop_grp) - 1)) {
1054                 NL_SET_ERR_MSG(extack,
1055                                "Invalid length for nexthop group attribute");
1056                 return -EINVAL;
1057         }
1058
1059         /* convert len to number of nexthop ids */
1060         len /= sizeof(*nhg);
1061
1062         nhg = nla_data(tb[NHA_GROUP]);
1063         for (i = 0; i < len; ++i) {
1064                 if (nhg[i].resvd1 || nhg[i].resvd2) {
1065                         NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
1066                         return -EINVAL;
1067                 }
1068                 if (nhg[i].weight > 254) {
1069                         NL_SET_ERR_MSG(extack, "Invalid value for weight");
1070                         return -EINVAL;
1071                 }
1072                 for (j = i + 1; j < len; ++j) {
1073                         if (nhg[i].id == nhg[j].id) {
1074                                 NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
1075                                 return -EINVAL;
1076                         }
1077                 }
1078         }
1079
1080         if (tb[NHA_FDB])
1081                 nhg_fdb = 1;
1082         nhg = nla_data(tb[NHA_GROUP]);
1083         for (i = 0; i < len; ++i) {
1084                 struct nexthop *nh;
1085                 bool is_fdb_nh;
1086
1087                 nh = nexthop_find_by_id(net, nhg[i].id);
1088                 if (!nh) {
1089                         NL_SET_ERR_MSG(extack, "Invalid nexthop id");
1090                         return -EINVAL;
1091                 }
1092                 if (!valid_group_nh(nh, len, &is_fdb_nh, extack))
1093                         return -EINVAL;
1094
1095                 if (nhg_fdb && nh_check_attr_fdb_group(nh, &nh_family, extack))
1096                         return -EINVAL;
1097
1098                 if (!nhg_fdb && is_fdb_nh) {
1099                         NL_SET_ERR_MSG(extack, "Non FDB nexthop group cannot have fdb nexthops");
1100                         return -EINVAL;
1101                 }
1102         }
1103         for (i = NHA_GROUP_TYPE + 1; i < tb_size; ++i) {
1104                 if (!tb[i])
1105                         continue;
1106                 switch (i) {
1107                 case NHA_FDB:
1108                         continue;
1109                 case NHA_RES_GROUP:
1110                         if (nh_grp_type == NEXTHOP_GRP_TYPE_RES)
1111                                 continue;
1112                         break;
1113                 }
1114                 NL_SET_ERR_MSG(extack,
1115                                "No other attributes can be set in nexthop groups");
1116                 return -EINVAL;
1117         }
1118
1119         return 0;
1120 }
1121
1122 static bool ipv6_good_nh(const struct fib6_nh *nh)
1123 {
1124         int state = NUD_REACHABLE;
1125         struct neighbour *n;
1126
1127         rcu_read_lock_bh();
1128
1129         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
1130         if (n)
1131                 state = n->nud_state;
1132
1133         rcu_read_unlock_bh();
1134
1135         return !!(state & NUD_VALID);
1136 }
1137
1138 static bool ipv4_good_nh(const struct fib_nh *nh)
1139 {
1140         int state = NUD_REACHABLE;
1141         struct neighbour *n;
1142
1143         rcu_read_lock_bh();
1144
1145         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
1146                                       (__force u32)nh->fib_nh_gw4);
1147         if (n)
1148                 state = n->nud_state;
1149
1150         rcu_read_unlock_bh();
1151
1152         return !!(state & NUD_VALID);
1153 }
1154
1155 static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash)
1156 {
1157         struct nexthop *rc = NULL;
1158         int i;
1159
1160         for (i = 0; i < nhg->num_nh; ++i) {
1161                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
1162                 struct nh_info *nhi;
1163
1164                 if (hash > atomic_read(&nhge->hthr.upper_bound))
1165                         continue;
1166
1167                 nhi = rcu_dereference(nhge->nh->nh_info);
1168                 if (nhi->fdb_nh)
1169                         return nhge->nh;
1170
1171                 /* nexthops always check if it is good and does
1172                  * not rely on a sysctl for this behavior
1173                  */
1174                 switch (nhi->family) {
1175                 case AF_INET:
1176                         if (ipv4_good_nh(&nhi->fib_nh))
1177                                 return nhge->nh;
1178                         break;
1179                 case AF_INET6:
1180                         if (ipv6_good_nh(&nhi->fib6_nh))
1181                                 return nhge->nh;
1182                         break;
1183                 }
1184
1185                 if (!rc)
1186                         rc = nhge->nh;
1187         }
1188
1189         return rc;
1190 }
1191
1192 static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash)
1193 {
1194         struct nh_res_table *res_table = rcu_dereference(nhg->res_table);
1195         u16 bucket_index = hash % res_table->num_nh_buckets;
1196         struct nh_res_bucket *bucket;
1197         struct nh_grp_entry *nhge;
1198
1199         /* nexthop_select_path() is expected to return a non-NULL value, so
1200          * skip protocol validation and just hand out whatever there is.
1201          */
1202         bucket = &res_table->nh_buckets[bucket_index];
1203         nh_res_bucket_set_busy(bucket);
1204         nhge = rcu_dereference(bucket->nh_entry);
1205         return nhge->nh;
1206 }
1207
1208 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
1209 {
1210         struct nh_group *nhg;
1211
1212         if (!nh->is_group)
1213                 return nh;
1214
1215         nhg = rcu_dereference(nh->nh_grp);
1216         if (nhg->hash_threshold)
1217                 return nexthop_select_path_hthr(nhg, hash);
1218         else if (nhg->resilient)
1219                 return nexthop_select_path_res(nhg, hash);
1220
1221         /* Unreachable. */
1222         return NULL;
1223 }
1224 EXPORT_SYMBOL_GPL(nexthop_select_path);
1225
1226 int nexthop_for_each_fib6_nh(struct nexthop *nh,
1227                              int (*cb)(struct fib6_nh *nh, void *arg),
1228                              void *arg)
1229 {
1230         struct nh_info *nhi;
1231         int err;
1232
1233         if (nh->is_group) {
1234                 struct nh_group *nhg;
1235                 int i;
1236
1237                 nhg = rcu_dereference_rtnl(nh->nh_grp);
1238                 for (i = 0; i < nhg->num_nh; i++) {
1239                         struct nh_grp_entry *nhge = &nhg->nh_entries[i];
1240
1241                         nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
1242                         err = cb(&nhi->fib6_nh, arg);
1243                         if (err)
1244                                 return err;
1245                 }
1246         } else {
1247                 nhi = rcu_dereference_rtnl(nh->nh_info);
1248                 err = cb(&nhi->fib6_nh, arg);
1249                 if (err)
1250                         return err;
1251         }
1252
1253         return 0;
1254 }
1255 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
1256
1257 static int check_src_addr(const struct in6_addr *saddr,
1258                           struct netlink_ext_ack *extack)
1259 {
1260         if (!ipv6_addr_any(saddr)) {
1261                 NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
1262                 return -EINVAL;
1263         }
1264         return 0;
1265 }
1266
1267 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
1268                        struct netlink_ext_ack *extack)
1269 {
1270         struct nh_info *nhi;
1271         bool is_fdb_nh;
1272
1273         /* fib6_src is unique to a fib6_info and limits the ability to cache
1274          * routes in fib6_nh within a nexthop that is potentially shared
1275          * across multiple fib entries. If the config wants to use source
1276          * routing it can not use nexthop objects. mlxsw also does not allow
1277          * fib6_src on routes.
1278          */
1279         if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
1280                 return -EINVAL;
1281
1282         if (nh->is_group) {
1283                 struct nh_group *nhg;
1284
1285                 nhg = rtnl_dereference(nh->nh_grp);
1286                 if (nhg->has_v4)
1287                         goto no_v4_nh;
1288                 is_fdb_nh = nhg->fdb_nh;
1289         } else {
1290                 nhi = rtnl_dereference(nh->nh_info);
1291                 if (nhi->family == AF_INET)
1292                         goto no_v4_nh;
1293                 is_fdb_nh = nhi->fdb_nh;
1294         }
1295
1296         if (is_fdb_nh) {
1297                 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
1298                 return -EINVAL;
1299         }
1300
1301         return 0;
1302 no_v4_nh:
1303         NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
1304         return -EINVAL;
1305 }
1306 EXPORT_SYMBOL_GPL(fib6_check_nexthop);
1307
1308 /* if existing nexthop has ipv6 routes linked to it, need
1309  * to verify this new spec works with ipv6
1310  */
1311 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
1312                               struct netlink_ext_ack *extack)
1313 {
1314         struct fib6_info *f6i;
1315
1316         if (list_empty(&old->f6i_list))
1317                 return 0;
1318
1319         list_for_each_entry(f6i, &old->f6i_list, nh_list) {
1320                 if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
1321                         return -EINVAL;
1322         }
1323
1324         return fib6_check_nexthop(new, NULL, extack);
1325 }
1326
1327 static int nexthop_check_scope(struct nh_info *nhi, u8 scope,
1328                                struct netlink_ext_ack *extack)
1329 {
1330         if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
1331                 NL_SET_ERR_MSG(extack,
1332                                "Route with host scope can not have a gateway");
1333                 return -EINVAL;
1334         }
1335
1336         if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
1337                 NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
1338                 return -EINVAL;
1339         }
1340
1341         return 0;
1342 }
1343
1344 /* Invoked by fib add code to verify nexthop by id is ok with
1345  * config for prefix; parts of fib_check_nh not done when nexthop
1346  * object is used.
1347  */
1348 int fib_check_nexthop(struct nexthop *nh, u8 scope,
1349                       struct netlink_ext_ack *extack)
1350 {
1351         struct nh_info *nhi;
1352         int err = 0;
1353
1354         if (nh->is_group) {
1355                 struct nh_group *nhg;
1356
1357                 nhg = rtnl_dereference(nh->nh_grp);
1358                 if (nhg->fdb_nh) {
1359                         NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
1360                         err = -EINVAL;
1361                         goto out;
1362                 }
1363
1364                 if (scope == RT_SCOPE_HOST) {
1365                         NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
1366                         err = -EINVAL;
1367                         goto out;
1368                 }
1369
1370                 /* all nexthops in a group have the same scope */
1371                 nhi = rtnl_dereference(nhg->nh_entries[0].nh->nh_info);
1372                 err = nexthop_check_scope(nhi, scope, extack);
1373         } else {
1374                 nhi = rtnl_dereference(nh->nh_info);
1375                 if (nhi->fdb_nh) {
1376                         NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
1377                         err = -EINVAL;
1378                         goto out;
1379                 }
1380                 err = nexthop_check_scope(nhi, scope, extack);
1381         }
1382
1383 out:
1384         return err;
1385 }
1386
1387 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
1388                              struct netlink_ext_ack *extack)
1389 {
1390         struct fib_info *fi;
1391
1392         list_for_each_entry(fi, &old->fi_list, nh_list) {
1393                 int err;
1394
1395                 err = fib_check_nexthop(new, fi->fib_scope, extack);
1396                 if (err)
1397                         return err;
1398         }
1399         return 0;
1400 }
1401
1402 static bool nh_res_nhge_is_balanced(const struct nh_grp_entry *nhge)
1403 {
1404         return nhge->res.count_buckets == nhge->res.wants_buckets;
1405 }
1406
1407 static bool nh_res_nhge_is_ow(const struct nh_grp_entry *nhge)
1408 {
1409         return nhge->res.count_buckets > nhge->res.wants_buckets;
1410 }
1411
1412 static bool nh_res_nhge_is_uw(const struct nh_grp_entry *nhge)
1413 {
1414         return nhge->res.count_buckets < nhge->res.wants_buckets;
1415 }
1416
1417 static bool nh_res_table_is_balanced(const struct nh_res_table *res_table)
1418 {
1419         return list_empty(&res_table->uw_nh_entries);
1420 }
1421
1422 static void nh_res_bucket_unset_nh(struct nh_res_bucket *bucket)
1423 {
1424         struct nh_grp_entry *nhge;
1425
1426         if (bucket->occupied) {
1427                 nhge = nh_res_dereference(bucket->nh_entry);
1428                 nhge->res.count_buckets--;
1429                 bucket->occupied = false;
1430         }
1431 }
1432
1433 static void nh_res_bucket_set_nh(struct nh_res_bucket *bucket,
1434                                  struct nh_grp_entry *nhge)
1435 {
1436         nh_res_bucket_unset_nh(bucket);
1437
1438         bucket->occupied = true;
1439         rcu_assign_pointer(bucket->nh_entry, nhge);
1440         nhge->res.count_buckets++;
1441 }
1442
1443 static bool nh_res_bucket_should_migrate(struct nh_res_table *res_table,
1444                                          struct nh_res_bucket *bucket,
1445                                          unsigned long *deadline, bool *force)
1446 {
1447         unsigned long now = jiffies;
1448         struct nh_grp_entry *nhge;
1449         unsigned long idle_point;
1450
1451         if (!bucket->occupied) {
1452                 /* The bucket is not occupied, its NHGE pointer is either
1453                  * NULL or obsolete. We _have to_ migrate: set force.
1454                  */
1455                 *force = true;
1456                 return true;
1457         }
1458
1459         nhge = nh_res_dereference(bucket->nh_entry);
1460
1461         /* If the bucket is populated by an underweight or balanced
1462          * nexthop, do not migrate.
1463          */
1464         if (!nh_res_nhge_is_ow(nhge))
1465                 return false;
1466
1467         /* At this point we know that the bucket is populated with an
1468          * overweight nexthop. It needs to be migrated to a new nexthop if
1469          * the idle timer of unbalanced timer expired.
1470          */
1471
1472         idle_point = nh_res_bucket_idle_point(res_table, bucket, now);
1473         if (time_after_eq(now, idle_point)) {
1474                 /* The bucket is idle. We _can_ migrate: unset force. */
1475                 *force = false;
1476                 return true;
1477         }
1478
1479         /* Unbalanced timer of 0 means "never force". */
1480         if (res_table->unbalanced_timer) {
1481                 unsigned long unb_point;
1482
1483                 unb_point = nh_res_table_unb_point(res_table);
1484                 if (time_after(now, unb_point)) {
1485                         /* The bucket is not idle, but the unbalanced timer
1486                          * expired. We _can_ migrate, but set force anyway,
1487                          * so that drivers know to ignore activity reports
1488                          * from the HW.
1489                          */
1490                         *force = true;
1491                         return true;
1492                 }
1493
1494                 nh_res_time_set_deadline(unb_point, deadline);
1495         }
1496
1497         nh_res_time_set_deadline(idle_point, deadline);
1498         return false;
1499 }
1500
1501 static bool nh_res_bucket_migrate(struct nh_res_table *res_table,
1502                                   u16 bucket_index, bool notify,
1503                                   bool notify_nl, bool force)
1504 {
1505         struct nh_res_bucket *bucket = &res_table->nh_buckets[bucket_index];
1506         struct nh_grp_entry *new_nhge;
1507         struct netlink_ext_ack extack;
1508         int err;
1509
1510         new_nhge = list_first_entry_or_null(&res_table->uw_nh_entries,
1511                                             struct nh_grp_entry,
1512                                             res.uw_nh_entry);
1513         if (WARN_ON_ONCE(!new_nhge))
1514                 /* If this function is called, "bucket" is either not
1515                  * occupied, or it belongs to a next hop that is
1516                  * overweight. In either case, there ought to be a
1517                  * corresponding underweight next hop.
1518                  */
1519                 return false;
1520
1521         if (notify) {
1522                 struct nh_grp_entry *old_nhge;
1523
1524                 old_nhge = nh_res_dereference(bucket->nh_entry);
1525                 err = call_nexthop_res_bucket_notifiers(res_table->net,
1526                                                         res_table->nhg_id,
1527                                                         bucket_index, force,
1528                                                         old_nhge->nh,
1529                                                         new_nhge->nh, &extack);
1530                 if (err) {
1531                         pr_err_ratelimited("%s\n", extack._msg);
1532                         if (!force)
1533                                 return false;
1534                         /* It is not possible to veto a forced replacement, so
1535                          * just clear the hardware flags from the nexthop
1536                          * bucket to indicate to user space that this bucket is
1537                          * not correctly populated in hardware.
1538                          */
1539                         bucket->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
1540                 }
1541         }
1542
1543         nh_res_bucket_set_nh(bucket, new_nhge);
1544         nh_res_bucket_set_idle(res_table, bucket);
1545
1546         if (notify_nl)
1547                 nexthop_bucket_notify(res_table, bucket_index);
1548
1549         if (nh_res_nhge_is_balanced(new_nhge))
1550                 list_del(&new_nhge->res.uw_nh_entry);
1551         return true;
1552 }
1553
1554 #define NH_RES_UPKEEP_DW_MINIMUM_INTERVAL (HZ / 2)
1555
1556 static void nh_res_table_upkeep(struct nh_res_table *res_table,
1557                                 bool notify, bool notify_nl)
1558 {
1559         unsigned long now = jiffies;
1560         unsigned long deadline;
1561         u16 i;
1562
1563         /* Deadline is the next time that upkeep should be run. It is the
1564          * earliest time at which one of the buckets might be migrated.
1565          * Start at the most pessimistic estimate: either unbalanced_timer
1566          * from now, or if there is none, idle_timer from now. For each
1567          * encountered time point, call nh_res_time_set_deadline() to
1568          * refine the estimate.
1569          */
1570         if (res_table->unbalanced_timer)
1571                 deadline = now + res_table->unbalanced_timer;
1572         else
1573                 deadline = now + res_table->idle_timer;
1574
1575         for (i = 0; i < res_table->num_nh_buckets; i++) {
1576                 struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
1577                 bool force;
1578
1579                 if (nh_res_bucket_should_migrate(res_table, bucket,
1580                                                  &deadline, &force)) {
1581                         if (!nh_res_bucket_migrate(res_table, i, notify,
1582                                                    notify_nl, force)) {
1583                                 unsigned long idle_point;
1584
1585                                 /* A driver can override the migration
1586                                  * decision if the HW reports that the
1587                                  * bucket is actually not idle. Therefore
1588                                  * remark the bucket as busy again and
1589                                  * update the deadline.
1590                                  */
1591                                 nh_res_bucket_set_busy(bucket);
1592                                 idle_point = nh_res_bucket_idle_point(res_table,
1593                                                                       bucket,
1594                                                                       now);
1595                                 nh_res_time_set_deadline(idle_point, &deadline);
1596                         }
1597                 }
1598         }
1599
1600         /* If the group is still unbalanced, schedule the next upkeep to
1601          * either the deadline computed above, or the minimum deadline,
1602          * whichever comes later.
1603          */
1604         if (!nh_res_table_is_balanced(res_table)) {
1605                 unsigned long now = jiffies;
1606                 unsigned long min_deadline;
1607
1608                 min_deadline = now + NH_RES_UPKEEP_DW_MINIMUM_INTERVAL;
1609                 if (time_before(deadline, min_deadline))
1610                         deadline = min_deadline;
1611
1612                 queue_delayed_work(system_power_efficient_wq,
1613                                    &res_table->upkeep_dw, deadline - now);
1614         }
1615 }
1616
1617 static void nh_res_table_upkeep_dw(struct work_struct *work)
1618 {
1619         struct delayed_work *dw = to_delayed_work(work);
1620         struct nh_res_table *res_table;
1621
1622         res_table = container_of(dw, struct nh_res_table, upkeep_dw);
1623         nh_res_table_upkeep(res_table, true, true);
1624 }
1625
1626 static void nh_res_table_cancel_upkeep(struct nh_res_table *res_table)
1627 {
1628         cancel_delayed_work_sync(&res_table->upkeep_dw);
1629 }
1630
1631 static void nh_res_group_rebalance(struct nh_group *nhg,
1632                                    struct nh_res_table *res_table)
1633 {
1634         int prev_upper_bound = 0;
1635         int total = 0;
1636         int w = 0;
1637         int i;
1638
1639         INIT_LIST_HEAD(&res_table->uw_nh_entries);
1640
1641         for (i = 0; i < nhg->num_nh; ++i)
1642                 total += nhg->nh_entries[i].weight;
1643
1644         for (i = 0; i < nhg->num_nh; ++i) {
1645                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
1646                 int upper_bound;
1647
1648                 w += nhge->weight;
1649                 upper_bound = DIV_ROUND_CLOSEST(res_table->num_nh_buckets * w,
1650                                                 total);
1651                 nhge->res.wants_buckets = upper_bound - prev_upper_bound;
1652                 prev_upper_bound = upper_bound;
1653
1654                 if (nh_res_nhge_is_uw(nhge)) {
1655                         if (list_empty(&res_table->uw_nh_entries))
1656                                 res_table->unbalanced_since = jiffies;
1657                         list_add(&nhge->res.uw_nh_entry,
1658                                  &res_table->uw_nh_entries);
1659                 }
1660         }
1661 }
1662
1663 /* Migrate buckets in res_table so that they reference NHGE's from NHG with
1664  * the right NH ID. Set those buckets that do not have a corresponding NHGE
1665  * entry in NHG as not occupied.
1666  */
1667 static void nh_res_table_migrate_buckets(struct nh_res_table *res_table,
1668                                          struct nh_group *nhg)
1669 {
1670         u16 i;
1671
1672         for (i = 0; i < res_table->num_nh_buckets; i++) {
1673                 struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
1674                 u32 id = rtnl_dereference(bucket->nh_entry)->nh->id;
1675                 bool found = false;
1676                 int j;
1677
1678                 for (j = 0; j < nhg->num_nh; j++) {
1679                         struct nh_grp_entry *nhge = &nhg->nh_entries[j];
1680
1681                         if (nhge->nh->id == id) {
1682                                 nh_res_bucket_set_nh(bucket, nhge);
1683                                 found = true;
1684                                 break;
1685                         }
1686                 }
1687
1688                 if (!found)
1689                         nh_res_bucket_unset_nh(bucket);
1690         }
1691 }
1692
1693 static void replace_nexthop_grp_res(struct nh_group *oldg,
1694                                     struct nh_group *newg)
1695 {
1696         /* For NH group replacement, the new NHG might only have a stub
1697          * hash table with 0 buckets, because the number of buckets was not
1698          * specified. For NH removal, oldg and newg both reference the same
1699          * res_table. So in any case, in the following, we want to work
1700          * with oldg->res_table.
1701          */
1702         struct nh_res_table *old_res_table = rtnl_dereference(oldg->res_table);
1703         unsigned long prev_unbalanced_since = old_res_table->unbalanced_since;
1704         bool prev_has_uw = !list_empty(&old_res_table->uw_nh_entries);
1705
1706         nh_res_table_cancel_upkeep(old_res_table);
1707         nh_res_table_migrate_buckets(old_res_table, newg);
1708         nh_res_group_rebalance(newg, old_res_table);
1709         if (prev_has_uw && !list_empty(&old_res_table->uw_nh_entries))
1710                 old_res_table->unbalanced_since = prev_unbalanced_since;
1711         nh_res_table_upkeep(old_res_table, true, false);
1712 }
1713
1714 static void nh_hthr_group_rebalance(struct nh_group *nhg)
1715 {
1716         int total = 0;
1717         int w = 0;
1718         int i;
1719
1720         for (i = 0; i < nhg->num_nh; ++i)
1721                 total += nhg->nh_entries[i].weight;
1722
1723         for (i = 0; i < nhg->num_nh; ++i) {
1724                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
1725                 int upper_bound;
1726
1727                 w += nhge->weight;
1728                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
1729                 atomic_set(&nhge->hthr.upper_bound, upper_bound);
1730         }
1731 }
1732
1733 static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
1734                                 struct nl_info *nlinfo)
1735 {
1736         struct nh_grp_entry *nhges, *new_nhges;
1737         struct nexthop *nhp = nhge->nh_parent;
1738         struct netlink_ext_ack extack;
1739         struct nexthop *nh = nhge->nh;
1740         struct nh_group *nhg, *newg;
1741         int i, j, err;
1742
1743         WARN_ON(!nh);
1744
1745         nhg = rtnl_dereference(nhp->nh_grp);
1746         newg = nhg->spare;
1747
1748         /* last entry, keep it visible and remove the parent */
1749         if (nhg->num_nh == 1) {
1750                 remove_nexthop(net, nhp, nlinfo);
1751                 return;
1752         }
1753
1754         newg->has_v4 = false;
1755         newg->is_multipath = nhg->is_multipath;
1756         newg->hash_threshold = nhg->hash_threshold;
1757         newg->resilient = nhg->resilient;
1758         newg->fdb_nh = nhg->fdb_nh;
1759         newg->num_nh = nhg->num_nh;
1760
1761         /* copy old entries to new except the one getting removed */
1762         nhges = nhg->nh_entries;
1763         new_nhges = newg->nh_entries;
1764         for (i = 0, j = 0; i < nhg->num_nh; ++i) {
1765                 struct nh_info *nhi;
1766
1767                 /* current nexthop getting removed */
1768                 if (nhg->nh_entries[i].nh == nh) {
1769                         newg->num_nh--;
1770                         continue;
1771                 }
1772
1773                 nhi = rtnl_dereference(nhges[i].nh->nh_info);
1774                 if (nhi->family == AF_INET)
1775                         newg->has_v4 = true;
1776
1777                 list_del(&nhges[i].nh_list);
1778                 new_nhges[j].nh_parent = nhges[i].nh_parent;
1779                 new_nhges[j].nh = nhges[i].nh;
1780                 new_nhges[j].weight = nhges[i].weight;
1781                 list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list);
1782                 j++;
1783         }
1784
1785         if (newg->hash_threshold)
1786                 nh_hthr_group_rebalance(newg);
1787         else if (newg->resilient)
1788                 replace_nexthop_grp_res(nhg, newg);
1789
1790         rcu_assign_pointer(nhp->nh_grp, newg);
1791
1792         list_del(&nhge->nh_list);
1793         nexthop_put(nhge->nh);
1794
1795         /* Removal of a NH from a resilient group is notified through
1796          * bucket notifications.
1797          */
1798         if (newg->hash_threshold) {
1799                 err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp,
1800                                              &extack);
1801                 if (err)
1802                         pr_err("%s\n", extack._msg);
1803         }
1804
1805         if (nlinfo)
1806                 nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo);
1807 }
1808
1809 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
1810                                        struct nl_info *nlinfo)
1811 {
1812         struct nh_grp_entry *nhge, *tmp;
1813
1814         list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list)
1815                 remove_nh_grp_entry(net, nhge, nlinfo);
1816
1817         /* make sure all see the newly published array before releasing rtnl */
1818         synchronize_net();
1819 }
1820
1821 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
1822 {
1823         struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
1824         struct nh_res_table *res_table;
1825         int i, num_nh = nhg->num_nh;
1826
1827         for (i = 0; i < num_nh; ++i) {
1828                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
1829
1830                 if (WARN_ON(!nhge->nh))
1831                         continue;
1832
1833                 list_del_init(&nhge->nh_list);
1834         }
1835
1836         if (nhg->resilient) {
1837                 res_table = rtnl_dereference(nhg->res_table);
1838                 nh_res_table_cancel_upkeep(res_table);
1839         }
1840 }
1841
1842 /* not called for nexthop replace */
1843 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
1844 {
1845         struct fib6_info *f6i, *tmp;
1846         bool do_flush = false;
1847         struct fib_info *fi;
1848
1849         list_for_each_entry(fi, &nh->fi_list, nh_list) {
1850                 fi->fib_flags |= RTNH_F_DEAD;
1851                 do_flush = true;
1852         }
1853         if (do_flush)
1854                 fib_flush(net);
1855
1856         /* ip6_del_rt removes the entry from this list hence the _safe */
1857         list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
1858                 /* __ip6_del_rt does a release, so do a hold here */
1859                 fib6_info_hold(f6i);
1860                 ipv6_stub->ip6_del_rt(net, f6i,
1861                                       !READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode));
1862         }
1863 }
1864
1865 static void __remove_nexthop(struct net *net, struct nexthop *nh,
1866                              struct nl_info *nlinfo)
1867 {
1868         __remove_nexthop_fib(net, nh);
1869
1870         if (nh->is_group) {
1871                 remove_nexthop_group(nh, nlinfo);
1872         } else {
1873                 struct nh_info *nhi;
1874
1875                 nhi = rtnl_dereference(nh->nh_info);
1876                 if (nhi->fib_nhc.nhc_dev)
1877                         hlist_del(&nhi->dev_hash);
1878
1879                 remove_nexthop_from_groups(net, nh, nlinfo);
1880         }
1881 }
1882
1883 static void remove_nexthop(struct net *net, struct nexthop *nh,
1884                            struct nl_info *nlinfo)
1885 {
1886         call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh, NULL);
1887
1888         /* remove from the tree */
1889         rb_erase(&nh->rb_node, &net->nexthop.rb_root);
1890
1891         if (nlinfo)
1892                 nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
1893
1894         __remove_nexthop(net, nh, nlinfo);
1895         nh_base_seq_inc(net);
1896
1897         nexthop_put(nh);
1898 }
1899
1900 /* if any FIB entries reference this nexthop, any dst entries
1901  * need to be regenerated
1902  */
1903 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh,
1904                               struct nexthop *replaced_nh)
1905 {
1906         struct fib6_info *f6i;
1907         struct nh_group *nhg;
1908         int i;
1909
1910         if (!list_empty(&nh->fi_list))
1911                 rt_cache_flush(net);
1912
1913         list_for_each_entry(f6i, &nh->f6i_list, nh_list)
1914                 ipv6_stub->fib6_update_sernum(net, f6i);
1915
1916         /* if an IPv6 group was replaced, we have to release all old
1917          * dsts to make sure all refcounts are released
1918          */
1919         if (!replaced_nh->is_group)
1920                 return;
1921
1922         nhg = rtnl_dereference(replaced_nh->nh_grp);
1923         for (i = 0; i < nhg->num_nh; i++) {
1924                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
1925                 struct nh_info *nhi = rtnl_dereference(nhge->nh->nh_info);
1926
1927                 if (nhi->family == AF_INET6)
1928                         ipv6_stub->fib6_nh_release_dsts(&nhi->fib6_nh);
1929         }
1930 }
1931
1932 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
1933                                struct nexthop *new, const struct nh_config *cfg,
1934                                struct netlink_ext_ack *extack)
1935 {
1936         struct nh_res_table *tmp_table = NULL;
1937         struct nh_res_table *new_res_table;
1938         struct nh_res_table *old_res_table;
1939         struct nh_group *oldg, *newg;
1940         int i, err;
1941
1942         if (!new->is_group) {
1943                 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
1944                 return -EINVAL;
1945         }
1946
1947         oldg = rtnl_dereference(old->nh_grp);
1948         newg = rtnl_dereference(new->nh_grp);
1949
1950         if (newg->hash_threshold != oldg->hash_threshold) {
1951                 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with one of a different type.");
1952                 return -EINVAL;
1953         }
1954
1955         if (newg->hash_threshold) {
1956                 err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new,
1957                                              extack);
1958                 if (err)
1959                         return err;
1960         } else if (newg->resilient) {
1961                 new_res_table = rtnl_dereference(newg->res_table);
1962                 old_res_table = rtnl_dereference(oldg->res_table);
1963
1964                 /* Accept if num_nh_buckets was not given, but if it was
1965                  * given, demand that the value be correct.
1966                  */
1967                 if (cfg->nh_grp_res_has_num_buckets &&
1968                     cfg->nh_grp_res_num_buckets !=
1969                     old_res_table->num_nh_buckets) {
1970                         NL_SET_ERR_MSG(extack, "Can not change number of buckets of a resilient nexthop group.");
1971                         return -EINVAL;
1972                 }
1973
1974                 /* Emit a pre-replace notification so that listeners could veto
1975                  * a potentially unsupported configuration. Otherwise,
1976                  * individual bucket replacement notifications would need to be
1977                  * vetoed, which is something that should only happen if the
1978                  * bucket is currently active.
1979                  */
1980                 err = call_nexthop_res_table_notifiers(net, new, extack);
1981                 if (err)
1982                         return err;
1983
1984                 if (cfg->nh_grp_res_has_idle_timer)
1985                         old_res_table->idle_timer = cfg->nh_grp_res_idle_timer;
1986                 if (cfg->nh_grp_res_has_unbalanced_timer)
1987                         old_res_table->unbalanced_timer =
1988                                 cfg->nh_grp_res_unbalanced_timer;
1989
1990                 replace_nexthop_grp_res(oldg, newg);
1991
1992                 tmp_table = new_res_table;
1993                 rcu_assign_pointer(newg->res_table, old_res_table);
1994                 rcu_assign_pointer(newg->spare->res_table, old_res_table);
1995         }
1996
1997         /* update parents - used by nexthop code for cleanup */
1998         for (i = 0; i < newg->num_nh; i++)
1999                 newg->nh_entries[i].nh_parent = old;
2000
2001         rcu_assign_pointer(old->nh_grp, newg);
2002
2003         /* Make sure concurrent readers are not using 'oldg' anymore. */
2004         synchronize_net();
2005
2006         if (newg->resilient) {
2007                 rcu_assign_pointer(oldg->res_table, tmp_table);
2008                 rcu_assign_pointer(oldg->spare->res_table, tmp_table);
2009         }
2010
2011         for (i = 0; i < oldg->num_nh; i++)
2012                 oldg->nh_entries[i].nh_parent = new;
2013
2014         rcu_assign_pointer(new->nh_grp, oldg);
2015
2016         return 0;
2017 }
2018
2019 static void nh_group_v4_update(struct nh_group *nhg)
2020 {
2021         struct nh_grp_entry *nhges;
2022         bool has_v4 = false;
2023         int i;
2024
2025         nhges = nhg->nh_entries;
2026         for (i = 0; i < nhg->num_nh; i++) {
2027                 struct nh_info *nhi;
2028
2029                 nhi = rtnl_dereference(nhges[i].nh->nh_info);
2030                 if (nhi->family == AF_INET)
2031                         has_v4 = true;
2032         }
2033         nhg->has_v4 = has_v4;
2034 }
2035
2036 static int replace_nexthop_single_notify_res(struct net *net,
2037                                              struct nh_res_table *res_table,
2038                                              struct nexthop *old,
2039                                              struct nh_info *oldi,
2040                                              struct nh_info *newi,
2041                                              struct netlink_ext_ack *extack)
2042 {
2043         u32 nhg_id = res_table->nhg_id;
2044         int err;
2045         u16 i;
2046
2047         for (i = 0; i < res_table->num_nh_buckets; i++) {
2048                 struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
2049                 struct nh_grp_entry *nhge;
2050
2051                 nhge = rtnl_dereference(bucket->nh_entry);
2052                 if (nhge->nh == old) {
2053                         err = __call_nexthop_res_bucket_notifiers(net, nhg_id,
2054                                                                   i, true,
2055                                                                   oldi, newi,
2056                                                                   extack);
2057                         if (err)
2058                                 goto err_notify;
2059                 }
2060         }
2061
2062         return 0;
2063
2064 err_notify:
2065         while (i-- > 0) {
2066                 struct nh_res_bucket *bucket = &res_table->nh_buckets[i];
2067                 struct nh_grp_entry *nhge;
2068
2069                 nhge = rtnl_dereference(bucket->nh_entry);
2070                 if (nhge->nh == old)
2071                         __call_nexthop_res_bucket_notifiers(net, nhg_id, i,
2072                                                             true, newi, oldi,
2073                                                             extack);
2074         }
2075         return err;
2076 }
2077
2078 static int replace_nexthop_single_notify(struct net *net,
2079                                          struct nexthop *group_nh,
2080                                          struct nexthop *old,
2081                                          struct nh_info *oldi,
2082                                          struct nh_info *newi,
2083                                          struct netlink_ext_ack *extack)
2084 {
2085         struct nh_group *nhg = rtnl_dereference(group_nh->nh_grp);
2086         struct nh_res_table *res_table;
2087
2088         if (nhg->hash_threshold) {
2089                 return call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE,
2090                                               group_nh, extack);
2091         } else if (nhg->resilient) {
2092                 res_table = rtnl_dereference(nhg->res_table);
2093                 return replace_nexthop_single_notify_res(net, res_table,
2094                                                          old, oldi, newi,
2095                                                          extack);
2096         }
2097
2098         return -EINVAL;
2099 }
2100
2101 static int replace_nexthop_single(struct net *net, struct nexthop *old,
2102                                   struct nexthop *new,
2103                                   struct netlink_ext_ack *extack)
2104 {
2105         u8 old_protocol, old_nh_flags;
2106         struct nh_info *oldi, *newi;
2107         struct nh_grp_entry *nhge;
2108         int err;
2109
2110         if (new->is_group) {
2111                 NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
2112                 return -EINVAL;
2113         }
2114
2115         err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
2116         if (err)
2117                 return err;
2118
2119         /* Hardware flags were set on 'old' as 'new' is not in the red-black
2120          * tree. Therefore, inherit the flags from 'old' to 'new'.
2121          */
2122         new->nh_flags |= old->nh_flags & (RTNH_F_OFFLOAD | RTNH_F_TRAP);
2123
2124         oldi = rtnl_dereference(old->nh_info);
2125         newi = rtnl_dereference(new->nh_info);
2126
2127         newi->nh_parent = old;
2128         oldi->nh_parent = new;
2129
2130         old_protocol = old->protocol;
2131         old_nh_flags = old->nh_flags;
2132
2133         old->protocol = new->protocol;
2134         old->nh_flags = new->nh_flags;
2135
2136         rcu_assign_pointer(old->nh_info, newi);
2137         rcu_assign_pointer(new->nh_info, oldi);
2138
2139         /* Send a replace notification for all the groups using the nexthop. */
2140         list_for_each_entry(nhge, &old->grp_list, nh_list) {
2141                 struct nexthop *nhp = nhge->nh_parent;
2142
2143                 err = replace_nexthop_single_notify(net, nhp, old, oldi, newi,
2144                                                     extack);
2145                 if (err)
2146                         goto err_notify;
2147         }
2148
2149         /* When replacing an IPv4 nexthop with an IPv6 nexthop, potentially
2150          * update IPv4 indication in all the groups using the nexthop.
2151          */
2152         if (oldi->family == AF_INET && newi->family == AF_INET6) {
2153                 list_for_each_entry(nhge, &old->grp_list, nh_list) {
2154                         struct nexthop *nhp = nhge->nh_parent;
2155                         struct nh_group *nhg;
2156
2157                         nhg = rtnl_dereference(nhp->nh_grp);
2158                         nh_group_v4_update(nhg);
2159                 }
2160         }
2161
2162         return 0;
2163
2164 err_notify:
2165         rcu_assign_pointer(new->nh_info, newi);
2166         rcu_assign_pointer(old->nh_info, oldi);
2167         old->nh_flags = old_nh_flags;
2168         old->protocol = old_protocol;
2169         oldi->nh_parent = old;
2170         newi->nh_parent = new;
2171         list_for_each_entry_continue_reverse(nhge, &old->grp_list, nh_list) {
2172                 struct nexthop *nhp = nhge->nh_parent;
2173
2174                 replace_nexthop_single_notify(net, nhp, old, newi, oldi, NULL);
2175         }
2176         call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, old, extack);
2177         return err;
2178 }
2179
2180 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
2181                                      struct nl_info *info)
2182 {
2183         struct fib6_info *f6i;
2184
2185         if (!list_empty(&nh->fi_list)) {
2186                 struct fib_info *fi;
2187
2188                 /* expectation is a few fib_info per nexthop and then
2189                  * a lot of routes per fib_info. So mark the fib_info
2190                  * and then walk the fib tables once
2191                  */
2192                 list_for_each_entry(fi, &nh->fi_list, nh_list)
2193                         fi->nh_updated = true;
2194
2195                 fib_info_notify_update(net, info);
2196
2197                 list_for_each_entry(fi, &nh->fi_list, nh_list)
2198                         fi->nh_updated = false;
2199         }
2200
2201         list_for_each_entry(f6i, &nh->f6i_list, nh_list)
2202                 ipv6_stub->fib6_rt_update(net, f6i, info);
2203 }
2204
2205 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
2206  * linked to this nexthop and for all groups that the nexthop
2207  * is a member of
2208  */
2209 static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
2210                                    struct nl_info *info)
2211 {
2212         struct nh_grp_entry *nhge;
2213
2214         __nexthop_replace_notify(net, nh, info);
2215
2216         list_for_each_entry(nhge, &nh->grp_list, nh_list)
2217                 __nexthop_replace_notify(net, nhge->nh_parent, info);
2218 }
2219
2220 static int replace_nexthop(struct net *net, struct nexthop *old,
2221                            struct nexthop *new, const struct nh_config *cfg,
2222                            struct netlink_ext_ack *extack)
2223 {
2224         bool new_is_reject = false;
2225         struct nh_grp_entry *nhge;
2226         int err;
2227
2228         /* check that existing FIB entries are ok with the
2229          * new nexthop definition
2230          */
2231         err = fib_check_nh_list(old, new, extack);
2232         if (err)
2233                 return err;
2234
2235         err = fib6_check_nh_list(old, new, extack);
2236         if (err)
2237                 return err;
2238
2239         if (!new->is_group) {
2240                 struct nh_info *nhi = rtnl_dereference(new->nh_info);
2241
2242                 new_is_reject = nhi->reject_nh;
2243         }
2244
2245         list_for_each_entry(nhge, &old->grp_list, nh_list) {
2246                 /* if new nexthop is a blackhole, any groups using this
2247                  * nexthop cannot have more than 1 path
2248                  */
2249                 if (new_is_reject &&
2250                     nexthop_num_path(nhge->nh_parent) > 1) {
2251                         NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
2252                         return -EINVAL;
2253                 }
2254
2255                 err = fib_check_nh_list(nhge->nh_parent, new, extack);
2256                 if (err)
2257                         return err;
2258
2259                 err = fib6_check_nh_list(nhge->nh_parent, new, extack);
2260                 if (err)
2261                         return err;
2262         }
2263
2264         if (old->is_group)
2265                 err = replace_nexthop_grp(net, old, new, cfg, extack);
2266         else
2267                 err = replace_nexthop_single(net, old, new, extack);
2268
2269         if (!err) {
2270                 nh_rt_cache_flush(net, old, new);
2271
2272                 __remove_nexthop(net, new, NULL);
2273                 nexthop_put(new);
2274         }
2275
2276         return err;
2277 }
2278
2279 /* called with rtnl_lock held */
2280 static int insert_nexthop(struct net *net, struct nexthop *new_nh,
2281                           struct nh_config *cfg, struct netlink_ext_ack *extack)
2282 {
2283         struct rb_node **pp, *parent = NULL, *next;
2284         struct rb_root *root = &net->nexthop.rb_root;
2285         bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
2286         bool create = !!(cfg->nlflags & NLM_F_CREATE);
2287         u32 new_id = new_nh->id;
2288         int replace_notify = 0;
2289         int rc = -EEXIST;
2290
2291         pp = &root->rb_node;
2292         while (1) {
2293                 struct nexthop *nh;
2294
2295                 next = *pp;
2296                 if (!next)
2297                         break;
2298
2299                 parent = next;
2300
2301                 nh = rb_entry(parent, struct nexthop, rb_node);
2302                 if (new_id < nh->id) {
2303                         pp = &next->rb_left;
2304                 } else if (new_id > nh->id) {
2305                         pp = &next->rb_right;
2306                 } else if (replace) {
2307                         rc = replace_nexthop(net, nh, new_nh, cfg, extack);
2308                         if (!rc) {
2309                                 new_nh = nh; /* send notification with old nh */
2310                                 replace_notify = 1;
2311                         }
2312                         goto out;
2313                 } else {
2314                         /* id already exists and not a replace */
2315                         goto out;
2316                 }
2317         }
2318
2319         if (replace && !create) {
2320                 NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
2321                 rc = -ENOENT;
2322                 goto out;
2323         }
2324
2325         if (new_nh->is_group) {
2326                 struct nh_group *nhg = rtnl_dereference(new_nh->nh_grp);
2327                 struct nh_res_table *res_table;
2328
2329                 if (nhg->resilient) {
2330                         res_table = rtnl_dereference(nhg->res_table);
2331
2332                         /* Not passing the number of buckets is OK when
2333                          * replacing, but not when creating a new group.
2334                          */
2335                         if (!cfg->nh_grp_res_has_num_buckets) {
2336                                 NL_SET_ERR_MSG(extack, "Number of buckets not specified for nexthop group insertion");
2337                                 rc = -EINVAL;
2338                                 goto out;
2339                         }
2340
2341                         nh_res_group_rebalance(nhg, res_table);
2342
2343                         /* Do not send bucket notifications, we do full
2344                          * notification below.
2345                          */
2346                         nh_res_table_upkeep(res_table, false, false);
2347                 }
2348         }
2349
2350         rb_link_node_rcu(&new_nh->rb_node, parent, pp);
2351         rb_insert_color(&new_nh->rb_node, root);
2352
2353         /* The initial insertion is a full notification for hash-threshold as
2354          * well as resilient groups.
2355          */
2356         rc = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new_nh, extack);
2357         if (rc)
2358                 rb_erase(&new_nh->rb_node, &net->nexthop.rb_root);
2359
2360 out:
2361         if (!rc) {
2362                 nh_base_seq_inc(net);
2363                 nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
2364                 if (replace_notify &&
2365                     READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode))
2366                         nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
2367         }
2368
2369         return rc;
2370 }
2371
2372 /* rtnl */
2373 /* remove all nexthops tied to a device being deleted */
2374 static void nexthop_flush_dev(struct net_device *dev, unsigned long event)
2375 {
2376         unsigned int hash = nh_dev_hashfn(dev->ifindex);
2377         struct net *net = dev_net(dev);
2378         struct hlist_head *head = &net->nexthop.devhash[hash];
2379         struct hlist_node *n;
2380         struct nh_info *nhi;
2381
2382         hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
2383                 if (nhi->fib_nhc.nhc_dev != dev)
2384                         continue;
2385
2386                 if (nhi->reject_nh &&
2387                     (event == NETDEV_DOWN || event == NETDEV_CHANGE))
2388                         continue;
2389
2390                 remove_nexthop(net, nhi->nh_parent, NULL);
2391         }
2392 }
2393
2394 /* rtnl; called when net namespace is deleted */
2395 static void flush_all_nexthops(struct net *net)
2396 {
2397         struct rb_root *root = &net->nexthop.rb_root;
2398         struct rb_node *node;
2399         struct nexthop *nh;
2400
2401         while ((node = rb_first(root))) {
2402                 nh = rb_entry(node, struct nexthop, rb_node);
2403                 remove_nexthop(net, nh, NULL);
2404                 cond_resched();
2405         }
2406 }
2407
2408 static struct nexthop *nexthop_create_group(struct net *net,
2409                                             struct nh_config *cfg)
2410 {
2411         struct nlattr *grps_attr = cfg->nh_grp;
2412         struct nexthop_grp *entry = nla_data(grps_attr);
2413         u16 num_nh = nla_len(grps_attr) / sizeof(*entry);
2414         struct nh_group *nhg;
2415         struct nexthop *nh;
2416         int err;
2417         int i;
2418
2419         if (WARN_ON(!num_nh))
2420                 return ERR_PTR(-EINVAL);
2421
2422         nh = nexthop_alloc();
2423         if (!nh)
2424                 return ERR_PTR(-ENOMEM);
2425
2426         nh->is_group = 1;
2427
2428         nhg = nexthop_grp_alloc(num_nh);
2429         if (!nhg) {
2430                 kfree(nh);
2431                 return ERR_PTR(-ENOMEM);
2432         }
2433
2434         /* spare group used for removals */
2435         nhg->spare = nexthop_grp_alloc(num_nh);
2436         if (!nhg->spare) {
2437                 kfree(nhg);
2438                 kfree(nh);
2439                 return ERR_PTR(-ENOMEM);
2440         }
2441         nhg->spare->spare = nhg;
2442
2443         for (i = 0; i < nhg->num_nh; ++i) {
2444                 struct nexthop *nhe;
2445                 struct nh_info *nhi;
2446
2447                 nhe = nexthop_find_by_id(net, entry[i].id);
2448                 if (!nexthop_get(nhe)) {
2449                         err = -ENOENT;
2450                         goto out_no_nh;
2451                 }
2452
2453                 nhi = rtnl_dereference(nhe->nh_info);
2454                 if (nhi->family == AF_INET)
2455                         nhg->has_v4 = true;
2456
2457                 nhg->nh_entries[i].nh = nhe;
2458                 nhg->nh_entries[i].weight = entry[i].weight + 1;
2459                 list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
2460                 nhg->nh_entries[i].nh_parent = nh;
2461         }
2462
2463         if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) {
2464                 nhg->hash_threshold = 1;
2465                 nhg->is_multipath = true;
2466         } else if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_RES) {
2467                 struct nh_res_table *res_table;
2468
2469                 res_table = nexthop_res_table_alloc(net, cfg->nh_id, cfg);
2470                 if (!res_table) {
2471                         err = -ENOMEM;
2472                         goto out_no_nh;
2473                 }
2474
2475                 rcu_assign_pointer(nhg->spare->res_table, res_table);
2476                 rcu_assign_pointer(nhg->res_table, res_table);
2477                 nhg->resilient = true;
2478                 nhg->is_multipath = true;
2479         }
2480
2481         WARN_ON_ONCE(nhg->hash_threshold + nhg->resilient != 1);
2482
2483         if (nhg->hash_threshold)
2484                 nh_hthr_group_rebalance(nhg);
2485
2486         if (cfg->nh_fdb)
2487                 nhg->fdb_nh = 1;
2488
2489         rcu_assign_pointer(nh->nh_grp, nhg);
2490
2491         return nh;
2492
2493 out_no_nh:
2494         for (i--; i >= 0; --i) {
2495                 list_del(&nhg->nh_entries[i].nh_list);
2496                 nexthop_put(nhg->nh_entries[i].nh);
2497         }
2498
2499         kfree(nhg->spare);
2500         kfree(nhg);
2501         kfree(nh);
2502
2503         return ERR_PTR(err);
2504 }
2505
2506 static int nh_create_ipv4(struct net *net, struct nexthop *nh,
2507                           struct nh_info *nhi, struct nh_config *cfg,
2508                           struct netlink_ext_ack *extack)
2509 {
2510         struct fib_nh *fib_nh = &nhi->fib_nh;
2511         struct fib_config fib_cfg = {
2512                 .fc_oif   = cfg->nh_ifindex,
2513                 .fc_gw4   = cfg->gw.ipv4,
2514                 .fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
2515                 .fc_flags = cfg->nh_flags,
2516                 .fc_nlinfo = cfg->nlinfo,
2517                 .fc_encap = cfg->nh_encap,
2518                 .fc_encap_type = cfg->nh_encap_type,
2519         };
2520         u32 tb_id = (cfg->dev ? l3mdev_fib_table(cfg->dev) : RT_TABLE_MAIN);
2521         int err;
2522
2523         err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
2524         if (err) {
2525                 fib_nh_release(net, fib_nh);
2526                 goto out;
2527         }
2528
2529         if (nhi->fdb_nh)
2530                 goto out;
2531
2532         /* sets nh_dev if successful */
2533         err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
2534         if (!err) {
2535                 nh->nh_flags = fib_nh->fib_nh_flags;
2536                 fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
2537                                           !fib_nh->fib_nh_scope ? 0 : fib_nh->fib_nh_scope - 1);
2538         } else {
2539                 fib_nh_release(net, fib_nh);
2540         }
2541 out:
2542         return err;
2543 }
2544
2545 static int nh_create_ipv6(struct net *net,  struct nexthop *nh,
2546                           struct nh_info *nhi, struct nh_config *cfg,
2547                           struct netlink_ext_ack *extack)
2548 {
2549         struct fib6_nh *fib6_nh = &nhi->fib6_nh;
2550         struct fib6_config fib6_cfg = {
2551                 .fc_table = l3mdev_fib_table(cfg->dev),
2552                 .fc_ifindex = cfg->nh_ifindex,
2553                 .fc_gateway = cfg->gw.ipv6,
2554                 .fc_flags = cfg->nh_flags,
2555                 .fc_nlinfo = cfg->nlinfo,
2556                 .fc_encap = cfg->nh_encap,
2557                 .fc_encap_type = cfg->nh_encap_type,
2558                 .fc_is_fdb = cfg->nh_fdb,
2559         };
2560         int err;
2561
2562         if (!ipv6_addr_any(&cfg->gw.ipv6))
2563                 fib6_cfg.fc_flags |= RTF_GATEWAY;
2564
2565         /* sets nh_dev if successful */
2566         err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
2567                                       extack);
2568         if (err) {
2569                 /* IPv6 is not enabled, don't call fib6_nh_release */
2570                 if (err == -EAFNOSUPPORT)
2571                         goto out;
2572                 ipv6_stub->fib6_nh_release(fib6_nh);
2573         } else {
2574                 nh->nh_flags = fib6_nh->fib_nh_flags;
2575         }
2576 out:
2577         return err;
2578 }
2579
2580 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
2581                                       struct netlink_ext_ack *extack)
2582 {
2583         struct nh_info *nhi;
2584         struct nexthop *nh;
2585         int err = 0;
2586
2587         nh = nexthop_alloc();
2588         if (!nh)
2589                 return ERR_PTR(-ENOMEM);
2590
2591         nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
2592         if (!nhi) {
2593                 kfree(nh);
2594                 return ERR_PTR(-ENOMEM);
2595         }
2596
2597         nh->nh_flags = cfg->nh_flags;
2598         nh->net = net;
2599
2600         nhi->nh_parent = nh;
2601         nhi->family = cfg->nh_family;
2602         nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
2603
2604         if (cfg->nh_fdb)
2605                 nhi->fdb_nh = 1;
2606
2607         if (cfg->nh_blackhole) {
2608                 nhi->reject_nh = 1;
2609                 cfg->nh_ifindex = net->loopback_dev->ifindex;
2610         }
2611
2612         switch (cfg->nh_family) {
2613         case AF_INET:
2614                 err = nh_create_ipv4(net, nh, nhi, cfg, extack);
2615                 break;
2616         case AF_INET6:
2617                 err = nh_create_ipv6(net, nh, nhi, cfg, extack);
2618                 break;
2619         }
2620
2621         if (err) {
2622                 kfree(nhi);
2623                 kfree(nh);
2624                 return ERR_PTR(err);
2625         }
2626
2627         /* add the entry to the device based hash */
2628         if (!nhi->fdb_nh)
2629                 nexthop_devhash_add(net, nhi);
2630
2631         rcu_assign_pointer(nh->nh_info, nhi);
2632
2633         return nh;
2634 }
2635
2636 /* called with rtnl lock held */
2637 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
2638                                    struct netlink_ext_ack *extack)
2639 {
2640         struct nexthop *nh;
2641         int err;
2642
2643         if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
2644                 NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
2645                 return ERR_PTR(-EINVAL);
2646         }
2647
2648         if (!cfg->nh_id) {
2649                 cfg->nh_id = nh_find_unused_id(net);
2650                 if (!cfg->nh_id) {
2651                         NL_SET_ERR_MSG(extack, "No unused id");
2652                         return ERR_PTR(-EINVAL);
2653                 }
2654         }
2655
2656         if (cfg->nh_grp)
2657                 nh = nexthop_create_group(net, cfg);
2658         else
2659                 nh = nexthop_create(net, cfg, extack);
2660
2661         if (IS_ERR(nh))
2662                 return nh;
2663
2664         refcount_set(&nh->refcnt, 1);
2665         nh->id = cfg->nh_id;
2666         nh->protocol = cfg->nh_protocol;
2667         nh->net = net;
2668
2669         err = insert_nexthop(net, nh, cfg, extack);
2670         if (err) {
2671                 __remove_nexthop(net, nh, NULL);
2672                 nexthop_put(nh);
2673                 nh = ERR_PTR(err);
2674         }
2675
2676         return nh;
2677 }
2678
2679 static int rtm_nh_get_timer(struct nlattr *attr, unsigned long fallback,
2680                             unsigned long *timer_p, bool *has_p,
2681                             struct netlink_ext_ack *extack)
2682 {
2683         unsigned long timer;
2684         u32 value;
2685
2686         if (!attr) {
2687                 *timer_p = fallback;
2688                 *has_p = false;
2689                 return 0;
2690         }
2691
2692         value = nla_get_u32(attr);
2693         timer = clock_t_to_jiffies(value);
2694         if (timer == ~0UL) {
2695                 NL_SET_ERR_MSG(extack, "Timer value too large");
2696                 return -EINVAL;
2697         }
2698
2699         *timer_p = timer;
2700         *has_p = true;
2701         return 0;
2702 }
2703
2704 static int rtm_to_nh_config_grp_res(struct nlattr *res, struct nh_config *cfg,
2705                                     struct netlink_ext_ack *extack)
2706 {
2707         struct nlattr *tb[ARRAY_SIZE(rtm_nh_res_policy_new)] = {};
2708         int err;
2709
2710         if (res) {
2711                 err = nla_parse_nested(tb,
2712                                        ARRAY_SIZE(rtm_nh_res_policy_new) - 1,
2713                                        res, rtm_nh_res_policy_new, extack);
2714                 if (err < 0)
2715                         return err;
2716         }
2717
2718         if (tb[NHA_RES_GROUP_BUCKETS]) {
2719                 cfg->nh_grp_res_num_buckets =
2720                         nla_get_u16(tb[NHA_RES_GROUP_BUCKETS]);
2721                 cfg->nh_grp_res_has_num_buckets = true;
2722                 if (!cfg->nh_grp_res_num_buckets) {
2723                         NL_SET_ERR_MSG(extack, "Number of buckets needs to be non-0");
2724                         return -EINVAL;
2725                 }
2726         }
2727
2728         err = rtm_nh_get_timer(tb[NHA_RES_GROUP_IDLE_TIMER],
2729                                NH_RES_DEFAULT_IDLE_TIMER,
2730                                &cfg->nh_grp_res_idle_timer,
2731                                &cfg->nh_grp_res_has_idle_timer,
2732                                extack);
2733         if (err)
2734                 return err;
2735
2736         return rtm_nh_get_timer(tb[NHA_RES_GROUP_UNBALANCED_TIMER],
2737                                 NH_RES_DEFAULT_UNBALANCED_TIMER,
2738                                 &cfg->nh_grp_res_unbalanced_timer,
2739                                 &cfg->nh_grp_res_has_unbalanced_timer,
2740                                 extack);
2741 }
2742
2743 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
2744                             struct nlmsghdr *nlh, struct nh_config *cfg,
2745                             struct netlink_ext_ack *extack)
2746 {
2747         struct nhmsg *nhm = nlmsg_data(nlh);
2748         struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_new)];
2749         int err;
2750
2751         err = nlmsg_parse(nlh, sizeof(*nhm), tb,
2752                           ARRAY_SIZE(rtm_nh_policy_new) - 1,
2753                           rtm_nh_policy_new, extack);
2754         if (err < 0)
2755                 return err;
2756
2757         err = -EINVAL;
2758         if (nhm->resvd || nhm->nh_scope) {
2759                 NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
2760                 goto out;
2761         }
2762         if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
2763                 NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
2764                 goto out;
2765         }
2766
2767         switch (nhm->nh_family) {
2768         case AF_INET:
2769         case AF_INET6:
2770                 break;
2771         case AF_UNSPEC:
2772                 if (tb[NHA_GROUP])
2773                         break;
2774                 fallthrough;
2775         default:
2776                 NL_SET_ERR_MSG(extack, "Invalid address family");
2777                 goto out;
2778         }
2779
2780         memset(cfg, 0, sizeof(*cfg));
2781         cfg->nlflags = nlh->nlmsg_flags;
2782         cfg->nlinfo.portid = NETLINK_CB(skb).portid;
2783         cfg->nlinfo.nlh = nlh;
2784         cfg->nlinfo.nl_net = net;
2785
2786         cfg->nh_family = nhm->nh_family;
2787         cfg->nh_protocol = nhm->nh_protocol;
2788         cfg->nh_flags = nhm->nh_flags;
2789
2790         if (tb[NHA_ID])
2791                 cfg->nh_id = nla_get_u32(tb[NHA_ID]);
2792
2793         if (tb[NHA_FDB]) {
2794                 if (tb[NHA_OIF] || tb[NHA_BLACKHOLE] ||
2795                     tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE]) {
2796                         NL_SET_ERR_MSG(extack, "Fdb attribute can not be used with encap, oif or blackhole");
2797                         goto out;
2798                 }
2799                 if (nhm->nh_flags) {
2800                         NL_SET_ERR_MSG(extack, "Unsupported nexthop flags in ancillary header");
2801                         goto out;
2802                 }
2803                 cfg->nh_fdb = nla_get_flag(tb[NHA_FDB]);
2804         }
2805
2806         if (tb[NHA_GROUP]) {
2807                 if (nhm->nh_family != AF_UNSPEC) {
2808                         NL_SET_ERR_MSG(extack, "Invalid family for group");
2809                         goto out;
2810                 }
2811                 cfg->nh_grp = tb[NHA_GROUP];
2812
2813                 cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
2814                 if (tb[NHA_GROUP_TYPE])
2815                         cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
2816
2817                 if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
2818                         NL_SET_ERR_MSG(extack, "Invalid group type");
2819                         goto out;
2820                 }
2821                 err = nh_check_attr_group(net, tb, ARRAY_SIZE(tb),
2822                                           cfg->nh_grp_type, extack);
2823                 if (err)
2824                         goto out;
2825
2826                 if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_RES)
2827                         err = rtm_to_nh_config_grp_res(tb[NHA_RES_GROUP],
2828                                                        cfg, extack);
2829
2830                 /* no other attributes should be set */
2831                 goto out;
2832         }
2833
2834         if (tb[NHA_BLACKHOLE]) {
2835                 if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
2836                     tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE] || tb[NHA_FDB]) {
2837                         NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway, oif, encap or fdb");
2838                         goto out;
2839                 }
2840
2841                 cfg->nh_blackhole = 1;
2842                 err = 0;
2843                 goto out;
2844         }
2845
2846         if (!cfg->nh_fdb && !tb[NHA_OIF]) {
2847                 NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole and non-fdb nexthops");
2848                 goto out;
2849         }
2850
2851         if (!cfg->nh_fdb && tb[NHA_OIF]) {
2852                 cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
2853                 if (cfg->nh_ifindex)
2854                         cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
2855
2856                 if (!cfg->dev) {
2857                         NL_SET_ERR_MSG(extack, "Invalid device index");
2858                         goto out;
2859                 } else if (!(cfg->dev->flags & IFF_UP)) {
2860                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
2861                         err = -ENETDOWN;
2862                         goto out;
2863                 } else if (!netif_carrier_ok(cfg->dev)) {
2864                         NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
2865                         err = -ENETDOWN;
2866                         goto out;
2867                 }
2868         }
2869
2870         err = -EINVAL;
2871         if (tb[NHA_GATEWAY]) {
2872                 struct nlattr *gwa = tb[NHA_GATEWAY];
2873
2874                 switch (cfg->nh_family) {
2875                 case AF_INET:
2876                         if (nla_len(gwa) != sizeof(u32)) {
2877                                 NL_SET_ERR_MSG(extack, "Invalid gateway");
2878                                 goto out;
2879                         }
2880                         cfg->gw.ipv4 = nla_get_be32(gwa);
2881                         break;
2882                 case AF_INET6:
2883                         if (nla_len(gwa) != sizeof(struct in6_addr)) {
2884                                 NL_SET_ERR_MSG(extack, "Invalid gateway");
2885                                 goto out;
2886                         }
2887                         cfg->gw.ipv6 = nla_get_in6_addr(gwa);
2888                         break;
2889                 default:
2890                         NL_SET_ERR_MSG(extack,
2891                                        "Unknown address family for gateway");
2892                         goto out;
2893                 }
2894         } else {
2895                 /* device only nexthop (no gateway) */
2896                 if (cfg->nh_flags & RTNH_F_ONLINK) {
2897                         NL_SET_ERR_MSG(extack,
2898                                        "ONLINK flag can not be set for nexthop without a gateway");
2899                         goto out;
2900                 }
2901         }
2902
2903         if (tb[NHA_ENCAP]) {
2904                 cfg->nh_encap = tb[NHA_ENCAP];
2905
2906                 if (!tb[NHA_ENCAP_TYPE]) {
2907                         NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
2908                         goto out;
2909                 }
2910
2911                 cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
2912                 err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
2913                 if (err < 0)
2914                         goto out;
2915
2916         } else if (tb[NHA_ENCAP_TYPE]) {
2917                 NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
2918                 goto out;
2919         }
2920
2921
2922         err = 0;
2923 out:
2924         return err;
2925 }
2926
2927 /* rtnl */
2928 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
2929                            struct netlink_ext_ack *extack)
2930 {
2931         struct net *net = sock_net(skb->sk);
2932         struct nh_config cfg;
2933         struct nexthop *nh;
2934         int err;
2935
2936         err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
2937         if (!err) {
2938                 nh = nexthop_add(net, &cfg, extack);
2939                 if (IS_ERR(nh))
2940                         err = PTR_ERR(nh);
2941         }
2942
2943         return err;
2944 }
2945
2946 static int __nh_valid_get_del_req(const struct nlmsghdr *nlh,
2947                                   struct nlattr **tb, u32 *id,
2948                                   struct netlink_ext_ack *extack)
2949 {
2950         struct nhmsg *nhm = nlmsg_data(nlh);
2951
2952         if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
2953                 NL_SET_ERR_MSG(extack, "Invalid values in header");
2954                 return -EINVAL;
2955         }
2956
2957         if (!tb[NHA_ID]) {
2958                 NL_SET_ERR_MSG(extack, "Nexthop id is missing");
2959                 return -EINVAL;
2960         }
2961
2962         *id = nla_get_u32(tb[NHA_ID]);
2963         if (!(*id)) {
2964                 NL_SET_ERR_MSG(extack, "Invalid nexthop id");
2965                 return -EINVAL;
2966         }
2967
2968         return 0;
2969 }
2970
2971 static int nh_valid_get_del_req(const struct nlmsghdr *nlh, u32 *id,
2972                                 struct netlink_ext_ack *extack)
2973 {
2974         struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get)];
2975         int err;
2976
2977         err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
2978                           ARRAY_SIZE(rtm_nh_policy_get) - 1,
2979                           rtm_nh_policy_get, extack);
2980         if (err < 0)
2981                 return err;
2982
2983         return __nh_valid_get_del_req(nlh, tb, id, extack);
2984 }
2985
2986 /* rtnl */
2987 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
2988                            struct netlink_ext_ack *extack)
2989 {
2990         struct net *net = sock_net(skb->sk);
2991         struct nl_info nlinfo = {
2992                 .nlh = nlh,
2993                 .nl_net = net,
2994                 .portid = NETLINK_CB(skb).portid,
2995         };
2996         struct nexthop *nh;
2997         int err;
2998         u32 id;
2999
3000         err = nh_valid_get_del_req(nlh, &id, extack);
3001         if (err)
3002                 return err;
3003
3004         nh = nexthop_find_by_id(net, id);
3005         if (!nh)
3006                 return -ENOENT;
3007
3008         remove_nexthop(net, nh, &nlinfo);
3009
3010         return 0;
3011 }
3012
3013 /* rtnl */
3014 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3015                            struct netlink_ext_ack *extack)
3016 {
3017         struct net *net = sock_net(in_skb->sk);
3018         struct sk_buff *skb = NULL;
3019         struct nexthop *nh;
3020         int err;
3021         u32 id;
3022
3023         err = nh_valid_get_del_req(nlh, &id, extack);
3024         if (err)
3025                 return err;
3026
3027         err = -ENOBUFS;
3028         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3029         if (!skb)
3030                 goto out;
3031
3032         err = -ENOENT;
3033         nh = nexthop_find_by_id(net, id);
3034         if (!nh)
3035                 goto errout_free;
3036
3037         err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
3038                            nlh->nlmsg_seq, 0);
3039         if (err < 0) {
3040                 WARN_ON(err == -EMSGSIZE);
3041                 goto errout_free;
3042         }
3043
3044         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3045 out:
3046         return err;
3047 errout_free:
3048         kfree_skb(skb);
3049         goto out;
3050 }
3051
3052 struct nh_dump_filter {
3053         u32 nh_id;
3054         int dev_idx;
3055         int master_idx;
3056         bool group_filter;
3057         bool fdb_filter;
3058         u32 res_bucket_nh_id;
3059 };
3060
3061 static bool nh_dump_filtered(struct nexthop *nh,
3062                              struct nh_dump_filter *filter, u8 family)
3063 {
3064         const struct net_device *dev;
3065         const struct nh_info *nhi;
3066
3067         if (filter->group_filter && !nh->is_group)
3068                 return true;
3069
3070         if (!filter->dev_idx && !filter->master_idx && !family)
3071                 return false;
3072
3073         if (nh->is_group)
3074                 return true;
3075
3076         nhi = rtnl_dereference(nh->nh_info);
3077         if (family && nhi->family != family)
3078                 return true;
3079
3080         dev = nhi->fib_nhc.nhc_dev;
3081         if (filter->dev_idx && (!dev || dev->ifindex != filter->dev_idx))
3082                 return true;
3083
3084         if (filter->master_idx) {
3085                 struct net_device *master;
3086
3087                 if (!dev)
3088                         return true;
3089
3090                 master = netdev_master_upper_dev_get((struct net_device *)dev);
3091                 if (!master || master->ifindex != filter->master_idx)
3092                         return true;
3093         }
3094
3095         return false;
3096 }
3097
3098 static int __nh_valid_dump_req(const struct nlmsghdr *nlh, struct nlattr **tb,
3099                                struct nh_dump_filter *filter,
3100                                struct netlink_ext_ack *extack)
3101 {
3102         struct nhmsg *nhm;
3103         u32 idx;
3104
3105         if (tb[NHA_OIF]) {
3106                 idx = nla_get_u32(tb[NHA_OIF]);
3107                 if (idx > INT_MAX) {
3108                         NL_SET_ERR_MSG(extack, "Invalid device index");
3109                         return -EINVAL;
3110                 }
3111                 filter->dev_idx = idx;
3112         }
3113         if (tb[NHA_MASTER]) {
3114                 idx = nla_get_u32(tb[NHA_MASTER]);
3115                 if (idx > INT_MAX) {
3116                         NL_SET_ERR_MSG(extack, "Invalid master device index");
3117                         return -EINVAL;
3118                 }
3119                 filter->master_idx = idx;
3120         }
3121         filter->group_filter = nla_get_flag(tb[NHA_GROUPS]);
3122         filter->fdb_filter = nla_get_flag(tb[NHA_FDB]);
3123
3124         nhm = nlmsg_data(nlh);
3125         if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
3126                 NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
3127                 return -EINVAL;
3128         }
3129
3130         return 0;
3131 }
3132
3133 static int nh_valid_dump_req(const struct nlmsghdr *nlh,
3134                              struct nh_dump_filter *filter,
3135                              struct netlink_callback *cb)
3136 {
3137         struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump)];
3138         int err;
3139
3140         err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
3141                           ARRAY_SIZE(rtm_nh_policy_dump) - 1,
3142                           rtm_nh_policy_dump, cb->extack);
3143         if (err < 0)
3144                 return err;
3145
3146         return __nh_valid_dump_req(nlh, tb, filter, cb->extack);
3147 }
3148
3149 struct rtm_dump_nh_ctx {
3150         u32 idx;
3151 };
3152
3153 static struct rtm_dump_nh_ctx *
3154 rtm_dump_nh_ctx(struct netlink_callback *cb)
3155 {
3156         struct rtm_dump_nh_ctx *ctx = (void *)cb->ctx;
3157
3158         BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx));
3159         return ctx;
3160 }
3161
3162 static int rtm_dump_walk_nexthops(struct sk_buff *skb,
3163                                   struct netlink_callback *cb,
3164                                   struct rb_root *root,
3165                                   struct rtm_dump_nh_ctx *ctx,
3166                                   int (*nh_cb)(struct sk_buff *skb,
3167                                                struct netlink_callback *cb,
3168                                                struct nexthop *nh, void *data),
3169                                   void *data)
3170 {
3171         struct rb_node *node;
3172         int s_idx;
3173         int err;
3174
3175         s_idx = ctx->idx;
3176         for (node = rb_first(root); node; node = rb_next(node)) {
3177                 struct nexthop *nh;
3178
3179                 nh = rb_entry(node, struct nexthop, rb_node);
3180                 if (nh->id < s_idx)
3181                         continue;
3182
3183                 ctx->idx = nh->id;
3184                 err = nh_cb(skb, cb, nh, data);
3185                 if (err)
3186                         return err;
3187         }
3188
3189         ctx->idx++;
3190         return 0;
3191 }
3192
3193 static int rtm_dump_nexthop_cb(struct sk_buff *skb, struct netlink_callback *cb,
3194                                struct nexthop *nh, void *data)
3195 {
3196         struct nhmsg *nhm = nlmsg_data(cb->nlh);
3197         struct nh_dump_filter *filter = data;
3198
3199         if (nh_dump_filtered(nh, filter, nhm->nh_family))
3200                 return 0;
3201
3202         return nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
3203                             NETLINK_CB(cb->skb).portid,
3204                             cb->nlh->nlmsg_seq, NLM_F_MULTI);
3205 }
3206
3207 /* rtnl */
3208 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
3209 {
3210         struct rtm_dump_nh_ctx *ctx = rtm_dump_nh_ctx(cb);
3211         struct net *net = sock_net(skb->sk);
3212         struct rb_root *root = &net->nexthop.rb_root;
3213         struct nh_dump_filter filter = {};
3214         int err;
3215
3216         err = nh_valid_dump_req(cb->nlh, &filter, cb);
3217         if (err < 0)
3218                 return err;
3219
3220         err = rtm_dump_walk_nexthops(skb, cb, root, ctx,
3221                                      &rtm_dump_nexthop_cb, &filter);
3222         if (err < 0) {
3223                 if (likely(skb->len))
3224                         goto out;
3225                 goto out_err;
3226         }
3227
3228 out:
3229         err = skb->len;
3230 out_err:
3231         cb->seq = net->nexthop.seq;
3232         nl_dump_check_consistent(cb, nlmsg_hdr(skb));
3233         return err;
3234 }
3235
3236 static struct nexthop *
3237 nexthop_find_group_resilient(struct net *net, u32 id,
3238                              struct netlink_ext_ack *extack)
3239 {
3240         struct nh_group *nhg;
3241         struct nexthop *nh;
3242
3243         nh = nexthop_find_by_id(net, id);
3244         if (!nh)
3245                 return ERR_PTR(-ENOENT);
3246
3247         if (!nh->is_group) {
3248                 NL_SET_ERR_MSG(extack, "Not a nexthop group");
3249                 return ERR_PTR(-EINVAL);
3250         }
3251
3252         nhg = rtnl_dereference(nh->nh_grp);
3253         if (!nhg->resilient) {
3254                 NL_SET_ERR_MSG(extack, "Nexthop group not of type resilient");
3255                 return ERR_PTR(-EINVAL);
3256         }
3257
3258         return nh;
3259 }
3260
3261 static int nh_valid_dump_nhid(struct nlattr *attr, u32 *nh_id_p,
3262                               struct netlink_ext_ack *extack)
3263 {
3264         u32 idx;
3265
3266         if (attr) {
3267                 idx = nla_get_u32(attr);
3268                 if (!idx) {
3269                         NL_SET_ERR_MSG(extack, "Invalid nexthop id");
3270                         return -EINVAL;
3271                 }
3272                 *nh_id_p = idx;
3273         } else {
3274                 *nh_id_p = 0;
3275         }
3276
3277         return 0;
3278 }
3279
3280 static int nh_valid_dump_bucket_req(const struct nlmsghdr *nlh,
3281                                     struct nh_dump_filter *filter,
3282                                     struct netlink_callback *cb)
3283 {
3284         struct nlattr *res_tb[ARRAY_SIZE(rtm_nh_res_bucket_policy_dump)];
3285         struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump_bucket)];
3286         int err;
3287
3288         err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
3289                           ARRAY_SIZE(rtm_nh_policy_dump_bucket) - 1,
3290                           rtm_nh_policy_dump_bucket, NULL);
3291         if (err < 0)
3292                 return err;
3293
3294         err = nh_valid_dump_nhid(tb[NHA_ID], &filter->nh_id, cb->extack);
3295         if (err)
3296                 return err;
3297
3298         if (tb[NHA_RES_BUCKET]) {
3299                 size_t max = ARRAY_SIZE(rtm_nh_res_bucket_policy_dump) - 1;
3300
3301                 err = nla_parse_nested(res_tb, max,
3302                                        tb[NHA_RES_BUCKET],
3303                                        rtm_nh_res_bucket_policy_dump,
3304                                        cb->extack);
3305                 if (err < 0)
3306                         return err;
3307
3308                 err = nh_valid_dump_nhid(res_tb[NHA_RES_BUCKET_NH_ID],
3309                                          &filter->res_bucket_nh_id,
3310                                          cb->extack);
3311                 if (err)
3312                         return err;
3313         }
3314
3315         return __nh_valid_dump_req(nlh, tb, filter, cb->extack);
3316 }
3317
3318 struct rtm_dump_res_bucket_ctx {
3319         struct rtm_dump_nh_ctx nh;
3320         u16 bucket_index;
3321         u32 done_nh_idx; /* 1 + the index of the last fully processed NH. */
3322 };
3323
3324 static struct rtm_dump_res_bucket_ctx *
3325 rtm_dump_res_bucket_ctx(struct netlink_callback *cb)
3326 {
3327         struct rtm_dump_res_bucket_ctx *ctx = (void *)cb->ctx;
3328
3329         BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx));
3330         return ctx;
3331 }
3332
3333 struct rtm_dump_nexthop_bucket_data {
3334         struct rtm_dump_res_bucket_ctx *ctx;
3335         struct nh_dump_filter filter;
3336 };
3337
3338 static int rtm_dump_nexthop_bucket_nh(struct sk_buff *skb,
3339                                       struct netlink_callback *cb,
3340                                       struct nexthop *nh,
3341                                       struct rtm_dump_nexthop_bucket_data *dd)
3342 {
3343         u32 portid = NETLINK_CB(cb->skb).portid;
3344         struct nhmsg *nhm = nlmsg_data(cb->nlh);
3345         struct nh_res_table *res_table;
3346         struct nh_group *nhg;
3347         u16 bucket_index;
3348         int err;
3349
3350         if (dd->ctx->nh.idx < dd->ctx->done_nh_idx)
3351                 return 0;
3352
3353         nhg = rtnl_dereference(nh->nh_grp);
3354         res_table = rtnl_dereference(nhg->res_table);
3355         for (bucket_index = dd->ctx->bucket_index;
3356              bucket_index < res_table->num_nh_buckets;
3357              bucket_index++) {
3358                 struct nh_res_bucket *bucket;
3359                 struct nh_grp_entry *nhge;
3360
3361                 bucket = &res_table->nh_buckets[bucket_index];
3362                 nhge = rtnl_dereference(bucket->nh_entry);
3363                 if (nh_dump_filtered(nhge->nh, &dd->filter, nhm->nh_family))
3364                         continue;
3365
3366                 if (dd->filter.res_bucket_nh_id &&
3367                     dd->filter.res_bucket_nh_id != nhge->nh->id)
3368                         continue;
3369
3370                 err = nh_fill_res_bucket(skb, nh, bucket, bucket_index,
3371                                          RTM_NEWNEXTHOPBUCKET, portid,
3372                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
3373                                          cb->extack);
3374                 if (err < 0) {
3375                         if (likely(skb->len))
3376                                 goto out;
3377                         goto out_err;
3378                 }
3379         }
3380
3381         dd->ctx->done_nh_idx = dd->ctx->nh.idx + 1;
3382         bucket_index = 0;
3383
3384 out:
3385         err = skb->len;
3386 out_err:
3387         dd->ctx->bucket_index = bucket_index;
3388         return err;
3389 }
3390
3391 static int rtm_dump_nexthop_bucket_cb(struct sk_buff *skb,
3392                                       struct netlink_callback *cb,
3393                                       struct nexthop *nh, void *data)
3394 {
3395         struct rtm_dump_nexthop_bucket_data *dd = data;
3396         struct nh_group *nhg;
3397
3398         if (!nh->is_group)
3399                 return 0;
3400
3401         nhg = rtnl_dereference(nh->nh_grp);
3402         if (!nhg->resilient)
3403                 return 0;
3404
3405         return rtm_dump_nexthop_bucket_nh(skb, cb, nh, dd);
3406 }
3407
3408 /* rtnl */
3409 static int rtm_dump_nexthop_bucket(struct sk_buff *skb,
3410                                    struct netlink_callback *cb)
3411 {
3412         struct rtm_dump_res_bucket_ctx *ctx = rtm_dump_res_bucket_ctx(cb);
3413         struct rtm_dump_nexthop_bucket_data dd = { .ctx = ctx };
3414         struct net *net = sock_net(skb->sk);
3415         struct nexthop *nh;
3416         int err;
3417
3418         err = nh_valid_dump_bucket_req(cb->nlh, &dd.filter, cb);
3419         if (err)
3420                 return err;
3421
3422         if (dd.filter.nh_id) {
3423                 nh = nexthop_find_group_resilient(net, dd.filter.nh_id,
3424                                                   cb->extack);
3425                 if (IS_ERR(nh))
3426                         return PTR_ERR(nh);
3427                 err = rtm_dump_nexthop_bucket_nh(skb, cb, nh, &dd);
3428         } else {
3429                 struct rb_root *root = &net->nexthop.rb_root;
3430
3431                 err = rtm_dump_walk_nexthops(skb, cb, root, &ctx->nh,
3432                                              &rtm_dump_nexthop_bucket_cb, &dd);
3433         }
3434
3435         if (err < 0) {
3436                 if (likely(skb->len))
3437                         goto out;
3438                 goto out_err;
3439         }
3440
3441 out:
3442         err = skb->len;
3443 out_err:
3444         cb->seq = net->nexthop.seq;
3445         nl_dump_check_consistent(cb, nlmsg_hdr(skb));
3446         return err;
3447 }
3448
3449 static int nh_valid_get_bucket_req_res_bucket(struct nlattr *res,
3450                                               u16 *bucket_index,
3451                                               struct netlink_ext_ack *extack)
3452 {
3453         struct nlattr *tb[ARRAY_SIZE(rtm_nh_res_bucket_policy_get)];
3454         int err;
3455
3456         err = nla_parse_nested(tb, ARRAY_SIZE(rtm_nh_res_bucket_policy_get) - 1,
3457                                res, rtm_nh_res_bucket_policy_get, extack);
3458         if (err < 0)
3459                 return err;
3460
3461         if (!tb[NHA_RES_BUCKET_INDEX]) {
3462                 NL_SET_ERR_MSG(extack, "Bucket index is missing");
3463                 return -EINVAL;
3464         }
3465
3466         *bucket_index = nla_get_u16(tb[NHA_RES_BUCKET_INDEX]);
3467         return 0;
3468 }
3469
3470 static int nh_valid_get_bucket_req(const struct nlmsghdr *nlh,
3471                                    u32 *id, u16 *bucket_index,
3472                                    struct netlink_ext_ack *extack)
3473 {
3474         struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get_bucket)];
3475         int err;
3476
3477         err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb,
3478                           ARRAY_SIZE(rtm_nh_policy_get_bucket) - 1,
3479                           rtm_nh_policy_get_bucket, extack);
3480         if (err < 0)
3481                 return err;
3482
3483         err = __nh_valid_get_del_req(nlh, tb, id, extack);
3484         if (err)
3485                 return err;
3486
3487         if (!tb[NHA_RES_BUCKET]) {
3488                 NL_SET_ERR_MSG(extack, "Bucket information is missing");
3489                 return -EINVAL;
3490         }
3491
3492         err = nh_valid_get_bucket_req_res_bucket(tb[NHA_RES_BUCKET],
3493                                                  bucket_index, extack);
3494         if (err)
3495                 return err;
3496
3497         return 0;
3498 }
3499
3500 /* rtnl */
3501 static int rtm_get_nexthop_bucket(struct sk_buff *in_skb, struct nlmsghdr *nlh,
3502                                   struct netlink_ext_ack *extack)
3503 {
3504         struct net *net = sock_net(in_skb->sk);
3505         struct nh_res_table *res_table;
3506         struct sk_buff *skb = NULL;
3507         struct nh_group *nhg;
3508         struct nexthop *nh;
3509         u16 bucket_index;
3510         int err;
3511         u32 id;
3512
3513         err = nh_valid_get_bucket_req(nlh, &id, &bucket_index, extack);
3514         if (err)
3515                 return err;
3516
3517         nh = nexthop_find_group_resilient(net, id, extack);
3518         if (IS_ERR(nh))
3519                 return PTR_ERR(nh);
3520
3521         nhg = rtnl_dereference(nh->nh_grp);
3522         res_table = rtnl_dereference(nhg->res_table);
3523         if (bucket_index >= res_table->num_nh_buckets) {
3524                 NL_SET_ERR_MSG(extack, "Bucket index out of bounds");
3525                 return -ENOENT;
3526         }
3527
3528         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
3529         if (!skb)
3530                 return -ENOBUFS;
3531
3532         err = nh_fill_res_bucket(skb, nh, &res_table->nh_buckets[bucket_index],
3533                                  bucket_index, RTM_NEWNEXTHOPBUCKET,
3534                                  NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
3535                                  0, extack);
3536         if (err < 0) {
3537                 WARN_ON(err == -EMSGSIZE);
3538                 goto errout_free;
3539         }
3540
3541         return rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
3542
3543 errout_free:
3544         kfree_skb(skb);
3545         return err;
3546 }
3547
3548 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
3549 {
3550         unsigned int hash = nh_dev_hashfn(dev->ifindex);
3551         struct net *net = dev_net(dev);
3552         struct hlist_head *head = &net->nexthop.devhash[hash];
3553         struct hlist_node *n;
3554         struct nh_info *nhi;
3555
3556         hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
3557                 if (nhi->fib_nhc.nhc_dev == dev) {
3558                         if (nhi->family == AF_INET)
3559                                 fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
3560                                                    orig_mtu);
3561                 }
3562         }
3563 }
3564
3565 /* rtnl */
3566 static int nh_netdev_event(struct notifier_block *this,
3567                            unsigned long event, void *ptr)
3568 {
3569         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3570         struct netdev_notifier_info_ext *info_ext;
3571
3572         switch (event) {
3573         case NETDEV_DOWN:
3574         case NETDEV_UNREGISTER:
3575                 nexthop_flush_dev(dev, event);
3576                 break;
3577         case NETDEV_CHANGE:
3578                 if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
3579                         nexthop_flush_dev(dev, event);
3580                 break;
3581         case NETDEV_CHANGEMTU:
3582                 info_ext = ptr;
3583                 nexthop_sync_mtu(dev, info_ext->ext.mtu);
3584                 rt_cache_flush(dev_net(dev));
3585                 break;
3586         }
3587         return NOTIFY_DONE;
3588 }
3589
3590 static struct notifier_block nh_netdev_notifier = {
3591         .notifier_call = nh_netdev_event,
3592 };
3593
3594 static int nexthops_dump(struct net *net, struct notifier_block *nb,
3595                          enum nexthop_event_type event_type,
3596                          struct netlink_ext_ack *extack)
3597 {
3598         struct rb_root *root = &net->nexthop.rb_root;
3599         struct rb_node *node;
3600         int err = 0;
3601
3602         for (node = rb_first(root); node; node = rb_next(node)) {
3603                 struct nexthop *nh;
3604
3605                 nh = rb_entry(node, struct nexthop, rb_node);
3606                 err = call_nexthop_notifier(nb, net, event_type, nh, extack);
3607                 if (err)
3608                         break;
3609         }
3610
3611         return err;
3612 }
3613
3614 int register_nexthop_notifier(struct net *net, struct notifier_block *nb,
3615                               struct netlink_ext_ack *extack)
3616 {
3617         int err;
3618
3619         rtnl_lock();
3620         err = nexthops_dump(net, nb, NEXTHOP_EVENT_REPLACE, extack);
3621         if (err)
3622                 goto unlock;
3623         err = blocking_notifier_chain_register(&net->nexthop.notifier_chain,
3624                                                nb);
3625 unlock:
3626         rtnl_unlock();
3627         return err;
3628 }
3629 EXPORT_SYMBOL(register_nexthop_notifier);
3630
3631 int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb)
3632 {
3633         int err;
3634
3635         rtnl_lock();
3636         err = blocking_notifier_chain_unregister(&net->nexthop.notifier_chain,
3637                                                  nb);
3638         if (err)
3639                 goto unlock;
3640         nexthops_dump(net, nb, NEXTHOP_EVENT_DEL, NULL);
3641 unlock:
3642         rtnl_unlock();
3643         return err;
3644 }
3645 EXPORT_SYMBOL(unregister_nexthop_notifier);
3646
3647 void nexthop_set_hw_flags(struct net *net, u32 id, bool offload, bool trap)
3648 {
3649         struct nexthop *nexthop;
3650
3651         rcu_read_lock();
3652
3653         nexthop = nexthop_find_by_id(net, id);
3654         if (!nexthop)
3655                 goto out;
3656
3657         nexthop->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
3658         if (offload)
3659                 nexthop->nh_flags |= RTNH_F_OFFLOAD;
3660         if (trap)
3661                 nexthop->nh_flags |= RTNH_F_TRAP;
3662
3663 out:
3664         rcu_read_unlock();
3665 }
3666 EXPORT_SYMBOL(nexthop_set_hw_flags);
3667
3668 void nexthop_bucket_set_hw_flags(struct net *net, u32 id, u16 bucket_index,
3669                                  bool offload, bool trap)
3670 {
3671         struct nh_res_table *res_table;
3672         struct nh_res_bucket *bucket;
3673         struct nexthop *nexthop;
3674         struct nh_group *nhg;
3675
3676         rcu_read_lock();
3677
3678         nexthop = nexthop_find_by_id(net, id);
3679         if (!nexthop || !nexthop->is_group)
3680                 goto out;
3681
3682         nhg = rcu_dereference(nexthop->nh_grp);
3683         if (!nhg->resilient)
3684                 goto out;
3685
3686         if (bucket_index >= nhg->res_table->num_nh_buckets)
3687                 goto out;
3688
3689         res_table = rcu_dereference(nhg->res_table);
3690         bucket = &res_table->nh_buckets[bucket_index];
3691         bucket->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP);
3692         if (offload)
3693                 bucket->nh_flags |= RTNH_F_OFFLOAD;
3694         if (trap)
3695                 bucket->nh_flags |= RTNH_F_TRAP;
3696
3697 out:
3698         rcu_read_unlock();
3699 }
3700 EXPORT_SYMBOL(nexthop_bucket_set_hw_flags);
3701
3702 void nexthop_res_grp_activity_update(struct net *net, u32 id, u16 num_buckets,
3703                                      unsigned long *activity)
3704 {
3705         struct nh_res_table *res_table;
3706         struct nexthop *nexthop;
3707         struct nh_group *nhg;
3708         u16 i;
3709
3710         rcu_read_lock();
3711
3712         nexthop = nexthop_find_by_id(net, id);
3713         if (!nexthop || !nexthop->is_group)
3714                 goto out;
3715
3716         nhg = rcu_dereference(nexthop->nh_grp);
3717         if (!nhg->resilient)
3718                 goto out;
3719
3720         /* Instead of silently ignoring some buckets, demand that the sizes
3721          * be the same.
3722          */
3723         res_table = rcu_dereference(nhg->res_table);
3724         if (num_buckets != res_table->num_nh_buckets)
3725                 goto out;
3726
3727         for (i = 0; i < num_buckets; i++) {
3728                 if (test_bit(i, activity))
3729                         nh_res_bucket_set_busy(&res_table->nh_buckets[i]);
3730         }
3731
3732 out:
3733         rcu_read_unlock();
3734 }
3735 EXPORT_SYMBOL(nexthop_res_grp_activity_update);
3736
3737 static void __net_exit nexthop_net_exit_batch(struct list_head *net_list)
3738 {
3739         struct net *net;
3740
3741         rtnl_lock();
3742         list_for_each_entry(net, net_list, exit_list) {
3743                 flush_all_nexthops(net);
3744                 kfree(net->nexthop.devhash);
3745         }
3746         rtnl_unlock();
3747 }
3748
3749 static int __net_init nexthop_net_init(struct net *net)
3750 {
3751         size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
3752
3753         net->nexthop.rb_root = RB_ROOT;
3754         net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
3755         if (!net->nexthop.devhash)
3756                 return -ENOMEM;
3757         BLOCKING_INIT_NOTIFIER_HEAD(&net->nexthop.notifier_chain);
3758
3759         return 0;
3760 }
3761
3762 static struct pernet_operations nexthop_net_ops = {
3763         .init = nexthop_net_init,
3764         .exit_batch = nexthop_net_exit_batch,
3765 };
3766
3767 static int __init nexthop_init(void)
3768 {
3769         register_pernet_subsys(&nexthop_net_ops);
3770
3771         register_netdevice_notifier(&nh_netdev_notifier);
3772
3773         rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
3774         rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
3775         rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
3776                       rtm_dump_nexthop, 0);
3777
3778         rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
3779         rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
3780
3781         rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
3782         rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
3783
3784         rtnl_register(PF_UNSPEC, RTM_GETNEXTHOPBUCKET, rtm_get_nexthop_bucket,
3785                       rtm_dump_nexthop_bucket, 0);
3786
3787         return 0;
3788 }
3789 subsys_initcall(nexthop_init);