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 <net/arp.h>
  12 #include <net/ipv6_stubs.h>
  13 #include <net/lwtunnel.h>
  14 #include <net/ndisc.h>
  15 #include <net/nexthop.h>
  16 #include <net/route.h>
  17 #include <net/sock.h>
  18
  19 static void remove_nexthop(struct net *net, struct nexthop *nh,
  20                            struct nl_info *nlinfo);
  21
  22 #define NH_DEV_HASHBITS  8
  23 #define NH_DEV_HASHSIZE (1U << NH_DEV_HASHBITS)
  24
  25 static const struct nla_policy rtm_nh_policy[NHA_MAX + 1] = {
  26         [NHA_ID]                = { .type = NLA_U32 },
  27         [NHA_GROUP]             = { .type = NLA_BINARY },
  28         [NHA_GROUP_TYPE]        = { .type = NLA_U16 },
  29         [NHA_BLACKHOLE]         = { .type = NLA_FLAG },
  30         [NHA_OIF]               = { .type = NLA_U32 },
  31         [NHA_GATEWAY]           = { .type = NLA_BINARY },
  32         [NHA_ENCAP_TYPE]        = { .type = NLA_U16 },
  33         [NHA_ENCAP]             = { .type = NLA_NESTED },
  34         [NHA_GROUPS]            = { .type = NLA_FLAG },
  35         [NHA_MASTER]            = { .type = NLA_U32 },
  36         [NHA_FDB]               = { .type = NLA_FLAG },
  37 };
  38
  39 static int call_nexthop_notifiers(struct net *net,
  40                                   enum nexthop_event_type event_type,
  41                                   struct nexthop *nh)
  42 {
  43         int err;
  44
  45         err = atomic_notifier_call_chain(&net->nexthop.notifier_chain,
  46                                          event_type, nh);
  47         return notifier_to_errno(err);
  48 }
  49
  50 static unsigned int nh_dev_hashfn(unsigned int val)
  51 {
  52         unsigned int mask = NH_DEV_HASHSIZE - 1;
  53
  54         return (val ^
  55                 (val >> NH_DEV_HASHBITS) ^
  56                 (val >> (NH_DEV_HASHBITS * 2))) & mask;
  57 }
  58
  59 static void nexthop_devhash_add(struct net *net, struct nh_info *nhi)
  60 {
  61         struct net_device *dev = nhi->fib_nhc.nhc_dev;
  62         struct hlist_head *head;
  63         unsigned int hash;
  64
  65         WARN_ON(!dev);
  66
  67         hash = nh_dev_hashfn(dev->ifindex);
  68         head = &net->nexthop.devhash[hash];
  69         hlist_add_head(&nhi->dev_hash, head);
  70 }
  71
  72 static void nexthop_free_mpath(struct nexthop *nh)
  73 {
  74         struct nh_group *nhg;
  75         int i;
  76
  77         nhg = rcu_dereference_raw(nh->nh_grp);
  78         for (i = 0; i < nhg->num_nh; ++i) {
  79                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
  80
  81                 WARN_ON(!list_empty(&nhge->nh_list));
  82                 nexthop_put(nhge->nh);
  83         }
  84
  85         WARN_ON(nhg->spare == nhg);
  86
  87         kfree(nhg->spare);
  88         kfree(nhg);
  89 }
  90
  91 static void nexthop_free_single(struct nexthop *nh)
  92 {
  93         struct nh_info *nhi;
  94
  95         nhi = rcu_dereference_raw(nh->nh_info);
  96         switch (nhi->family) {
  97         case AF_INET:
  98                 fib_nh_release(nh->net, &nhi->fib_nh);
  99                 break;
 100         case AF_INET6:
 101                 ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
 102                 break;
 103         }
 104         kfree(nhi);
 105 }
 106
 107 void nexthop_free_rcu(struct rcu_head *head)
 108 {
 109         struct nexthop *nh = container_of(head, struct nexthop, rcu);
 110
 111         if (nh->is_group)
 112                 nexthop_free_mpath(nh);
 113         else
 114                 nexthop_free_single(nh);
 115
 116         kfree(nh);
 117 }
 118 EXPORT_SYMBOL_GPL(nexthop_free_rcu);
 119
 120 static struct nexthop *nexthop_alloc(void)
 121 {
 122         struct nexthop *nh;
 123
 124         nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
 125         if (nh) {
 126                 INIT_LIST_HEAD(&nh->fi_list);
 127                 INIT_LIST_HEAD(&nh->f6i_list);
 128                 INIT_LIST_HEAD(&nh->grp_list);
 129                 INIT_LIST_HEAD(&nh->fdb_list);
 130         }
 131         return nh;
 132 }
 133
 134 static struct nh_group *nexthop_grp_alloc(u16 num_nh)
 135 {
 136         size_t sz = offsetof(struct nexthop, nh_grp)
 137                     + sizeof(struct nh_group)
 138                     + sizeof(struct nh_grp_entry) * num_nh;
 139         struct nh_group *nhg;
 140
 141         nhg = kzalloc(sz, GFP_KERNEL);
 142         if (nhg)
 143                 nhg->num_nh = num_nh;
 144
 145         return nhg;
 146 }
 147
 148 static void nh_base_seq_inc(struct net *net)
 149 {
 150         while (++net->nexthop.seq == 0)
 151                 ;
 152 }
 153
 154 /* no reference taken; rcu lock or rtnl must be held */
 155 struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
 156 {
 157         struct rb_node **pp, *parent = NULL, *next;
 158
 159         pp = &net->nexthop.rb_root.rb_node;
 160         while (1) {
 161                 struct nexthop *nh;
 162
 163                 next = rcu_dereference_raw(*pp);
 164                 if (!next)
 165                         break;
 166                 parent = next;
 167
 168                 nh = rb_entry(parent, struct nexthop, rb_node);
 169                 if (id < nh->id)
 170                         pp = &next->rb_left;
 171                 else if (id > nh->id)
 172                         pp = &next->rb_right;
 173                 else
 174                         return nh;
 175         }
 176         return NULL;
 177 }
 178 EXPORT_SYMBOL_GPL(nexthop_find_by_id);
 179
 180 /* used for auto id allocation; called with rtnl held */
 181 static u32 nh_find_unused_id(struct net *net)
 182 {
 183         u32 id_start = net->nexthop.last_id_allocated;
 184
 185         while (1) {
 186                 net->nexthop.last_id_allocated++;
 187                 if (net->nexthop.last_id_allocated == id_start)
 188                         break;
 189
 190                 if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
 191                         return net->nexthop.last_id_allocated;
 192         }
 193         return 0;
 194 }
 195
 196 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
 197 {
 198         struct nexthop_grp *p;
 199         size_t len = nhg->num_nh * sizeof(*p);
 200         struct nlattr *nla;
 201         u16 group_type = 0;
 202         int i;
 203
 204         if (nhg->mpath)
 205                 group_type = NEXTHOP_GRP_TYPE_MPATH;
 206
 207         if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
 208                 goto nla_put_failure;
 209
 210         nla = nla_reserve(skb, NHA_GROUP, len);
 211         if (!nla)
 212                 goto nla_put_failure;
 213
 214         p = nla_data(nla);
 215         for (i = 0; i < nhg->num_nh; ++i) {
 216                 p->id = nhg->nh_entries[i].nh->id;
 217                 p->weight = nhg->nh_entries[i].weight - 1;
 218                 p += 1;
 219         }
 220
 221         return 0;
 222
 223 nla_put_failure:
 224         return -EMSGSIZE;
 225 }
 226
 227 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
 228                         int event, u32 portid, u32 seq, unsigned int nlflags)
 229 {
 230         struct fib6_nh *fib6_nh;
 231         struct fib_nh *fib_nh;
 232         struct nlmsghdr *nlh;
 233         struct nh_info *nhi;
 234         struct nhmsg *nhm;
 235
 236         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
 237         if (!nlh)
 238                 return -EMSGSIZE;
 239
 240         nhm = nlmsg_data(nlh);
 241         nhm->nh_family = AF_UNSPEC;
 242         nhm->nh_flags = nh->nh_flags;
 243         nhm->nh_protocol = nh->protocol;
 244         nhm->nh_scope = 0;
 245         nhm->resvd = 0;
 246
 247         if (nla_put_u32(skb, NHA_ID, nh->id))
 248                 goto nla_put_failure;
 249
 250         if (nh->is_group) {
 251                 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 252
 253                 if (nhg->fdb_nh && nla_put_flag(skb, NHA_FDB))
 254                         goto nla_put_failure;
 255                 if (nla_put_nh_group(skb, nhg))
 256                         goto nla_put_failure;
 257                 goto out;
 258         }
 259
 260         nhi = rtnl_dereference(nh->nh_info);
 261         nhm->nh_family = nhi->family;
 262         if (nhi->reject_nh) {
 263                 if (nla_put_flag(skb, NHA_BLACKHOLE))
 264                         goto nla_put_failure;
 265                 goto out;
 266         } else if (nhi->fdb_nh) {
 267                 if (nla_put_flag(skb, NHA_FDB))
 268                         goto nla_put_failure;
 269         } else {
 270                 const struct net_device *dev;
 271
 272                 dev = nhi->fib_nhc.nhc_dev;
 273                 if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
 274                         goto nla_put_failure;
 275         }
 276
 277         nhm->nh_scope = nhi->fib_nhc.nhc_scope;
 278         switch (nhi->family) {
 279         case AF_INET:
 280                 fib_nh = &nhi->fib_nh;
 281                 if (fib_nh->fib_nh_gw_family &&
 282                     nla_put_u32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
 283                         goto nla_put_failure;
 284                 break;
 285
 286         case AF_INET6:
 287                 fib6_nh = &nhi->fib6_nh;
 288                 if (fib6_nh->fib_nh_gw_family &&
 289                     nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
 290                         goto nla_put_failure;
 291                 break;
 292         }
 293
 294         if (nhi->fib_nhc.nhc_lwtstate &&
 295             lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
 296                                 NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
 297                 goto nla_put_failure;
 298
 299 out:
 300         nlmsg_end(skb, nlh);
 301         return 0;
 302
 303 nla_put_failure:
 304         nlmsg_cancel(skb, nlh);
 305         return -EMSGSIZE;
 306 }
 307
 308 static size_t nh_nlmsg_size_grp(struct nexthop *nh)
 309 {
 310         struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 311         size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
 312
 313         return nla_total_size(sz) +
 314                nla_total_size(2);  /* NHA_GROUP_TYPE */
 315 }
 316
 317 static size_t nh_nlmsg_size_single(struct nexthop *nh)
 318 {
 319         struct nh_info *nhi = rtnl_dereference(nh->nh_info);
 320         size_t sz;
 321
 322         /* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
 323          * are mutually exclusive
 324          */
 325         sz = nla_total_size(4);  /* NHA_OIF */
 326
 327         switch (nhi->family) {
 328         case AF_INET:
 329                 if (nhi->fib_nh.fib_nh_gw_family)
 330                         sz += nla_total_size(4);  /* NHA_GATEWAY */
 331                 break;
 332
 333         case AF_INET6:
 334                 /* NHA_GATEWAY */
 335                 if (nhi->fib6_nh.fib_nh_gw_family)
 336                         sz += nla_total_size(sizeof(const struct in6_addr));
 337                 break;
 338         }
 339
 340         if (nhi->fib_nhc.nhc_lwtstate) {
 341                 sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
 342                 sz += nla_total_size(2);  /* NHA_ENCAP_TYPE */
 343         }
 344
 345         return sz;
 346 }
 347
 348 static size_t nh_nlmsg_size(struct nexthop *nh)
 349 {
 350         size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg));
 351
 352         sz += nla_total_size(4); /* NHA_ID */
 353
 354         if (nh->is_group)
 355                 sz += nh_nlmsg_size_grp(nh);
 356         else
 357                 sz += nh_nlmsg_size_single(nh);
 358
 359         return sz;
 360 }
 361
 362 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
 363 {
 364         unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
 365         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 366         struct sk_buff *skb;
 367         int err = -ENOBUFS;
 368
 369         skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
 370         if (!skb)
 371                 goto errout;
 372
 373         err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
 374         if (err < 0) {
 375                 /* -EMSGSIZE implies BUG in nh_nlmsg_size() */
 376                 WARN_ON(err == -EMSGSIZE);
 377                 kfree_skb(skb);
 378                 goto errout;
 379         }
 380
 381         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
 382                     info->nlh, gfp_any());
 383         return;
 384 errout:
 385         if (err < 0)
 386                 rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
 387 }
 388
 389 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
 390                            bool *is_fdb, struct netlink_ext_ack *extack)
 391 {
 392         if (nh->is_group) {
 393                 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 394
 395                 /* nested multipath (group within a group) is not
 396                  * supported
 397                  */
 398                 if (nhg->mpath) {
 399                         NL_SET_ERR_MSG(extack,
 400                                        "Multipath group can not be a nexthop within a group");
 401                         return false;
 402                 }
 403                 *is_fdb = nhg->fdb_nh;
 404         } else {
 405                 struct nh_info *nhi = rtnl_dereference(nh->nh_info);
 406
 407                 if (nhi->reject_nh && npaths > 1) {
 408                         NL_SET_ERR_MSG(extack,
 409                                        "Blackhole nexthop can not be used in a group with more than 1 path");
 410                         return false;
 411                 }
 412                 *is_fdb = nhi->fdb_nh;
 413         }
 414
 415         return true;
 416 }
 417
 418 static int nh_check_attr_fdb_group(struct nexthop *nh, u8 *nh_family,
 419                                    struct netlink_ext_ack *extack)
 420 {
 421         struct nh_info *nhi;
 422
 423         nhi = rtnl_dereference(nh->nh_info);
 424
 425         if (!nhi->fdb_nh) {
 426                 NL_SET_ERR_MSG(extack, "FDB nexthop group can only have fdb nexthops");
 427                 return -EINVAL;
 428         }
 429
 430         if (*nh_family == AF_UNSPEC) {
 431                 *nh_family = nhi->family;
 432         } else if (*nh_family != nhi->family) {
 433                 NL_SET_ERR_MSG(extack, "FDB nexthop group cannot have mixed family nexthops");
 434                 return -EINVAL;
 435         }
 436
 437         return 0;
 438 }
 439
 440 static int nh_check_attr_group(struct net *net, struct nlattr *tb[],
 441                                struct netlink_ext_ack *extack)
 442 {
 443         unsigned int len = nla_len(tb[NHA_GROUP]);
 444         u8 nh_family = AF_UNSPEC;
 445         struct nexthop_grp *nhg;
 446         unsigned int i, j;
 447         u8 nhg_fdb = 0;
 448
 449         if (len & (sizeof(struct nexthop_grp) - 1)) {
 450                 NL_SET_ERR_MSG(extack,
 451                                "Invalid length for nexthop group attribute");
 452                 return -EINVAL;
 453         }
 454
 455         /* convert len to number of nexthop ids */
 456         len /= sizeof(*nhg);
 457
 458         nhg = nla_data(tb[NHA_GROUP]);
 459         for (i = 0; i < len; ++i) {
 460                 if (nhg[i].resvd1 || nhg[i].resvd2) {
 461                         NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
 462                         return -EINVAL;
 463                 }
 464                 if (nhg[i].weight > 254) {
 465                         NL_SET_ERR_MSG(extack, "Invalid value for weight");
 466                         return -EINVAL;
 467                 }
 468                 for (j = i + 1; j < len; ++j) {
 469                         if (nhg[i].id == nhg[j].id) {
 470                                 NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
 471                                 return -EINVAL;
 472                         }
 473                 }
 474         }
 475
 476         if (tb[NHA_FDB])
 477                 nhg_fdb = 1;
 478         nhg = nla_data(tb[NHA_GROUP]);
 479         for (i = 0; i < len; ++i) {
 480                 struct nexthop *nh;
 481                 bool is_fdb_nh;
 482
 483                 nh = nexthop_find_by_id(net, nhg[i].id);
 484                 if (!nh) {
 485                         NL_SET_ERR_MSG(extack, "Invalid nexthop id");
 486                         return -EINVAL;
 487                 }
 488                 if (!valid_group_nh(nh, len, &is_fdb_nh, extack))
 489                         return -EINVAL;
 490
 491                 if (nhg_fdb && nh_check_attr_fdb_group(nh, &nh_family, extack))
 492                         return -EINVAL;
 493
 494                 if (!nhg_fdb && is_fdb_nh) {
 495                         NL_SET_ERR_MSG(extack, "Non FDB nexthop group cannot have fdb nexthops");
 496                         return -EINVAL;
 497                 }
 498         }
 499         for (i = NHA_GROUP_TYPE + 1; i < __NHA_MAX; ++i) {
 500                 if (!tb[i])
 501                         continue;
 502                 if (tb[NHA_FDB])
 503                         continue;
 504                 NL_SET_ERR_MSG(extack,
 505                                "No other attributes can be set in nexthop groups");
 506                 return -EINVAL;
 507         }
 508
 509         return 0;
 510 }
 511
 512 static bool ipv6_good_nh(const struct fib6_nh *nh)
 513 {
 514         int state = NUD_REACHABLE;
 515         struct neighbour *n;
 516
 517         rcu_read_lock_bh();
 518
 519         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
 520         if (n)
 521                 state = n->nud_state;
 522
 523         rcu_read_unlock_bh();
 524
 525         return !!(state & NUD_VALID);
 526 }
 527
 528 static bool ipv4_good_nh(const struct fib_nh *nh)
 529 {
 530         int state = NUD_REACHABLE;
 531         struct neighbour *n;
 532
 533         rcu_read_lock_bh();
 534
 535         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
 536                                       (__force u32)nh->fib_nh_gw4);
 537         if (n)
 538                 state = n->nud_state;
 539
 540         rcu_read_unlock_bh();
 541
 542         return !!(state & NUD_VALID);
 543 }
 544
 545 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
 546 {
 547         struct nexthop *rc = NULL;
 548         struct nh_group *nhg;
 549         int i;
 550
 551         if (!nh->is_group)
 552                 return nh;
 553
 554         nhg = rcu_dereference(nh->nh_grp);
 555         for (i = 0; i < nhg->num_nh; ++i) {
 556                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 557                 struct nh_info *nhi;
 558
 559                 if (hash > atomic_read(&nhge->upper_bound))
 560                         continue;
 561
 562                 nhi = rcu_dereference(nhge->nh->nh_info);
 563                 if (nhi->fdb_nh)
 564                         return nhge->nh;
 565
 566                 /* nexthops always check if it is good and does
 567                  * not rely on a sysctl for this behavior
 568                  */
 569                 switch (nhi->family) {
 570                 case AF_INET:
 571                         if (ipv4_good_nh(&nhi->fib_nh))
 572                                 return nhge->nh;
 573                         break;
 574                 case AF_INET6:
 575                         if (ipv6_good_nh(&nhi->fib6_nh))
 576                                 return nhge->nh;
 577                         break;
 578                 }
 579
 580                 if (!rc)
 581                         rc = nhge->nh;
 582         }
 583
 584         return rc;
 585 }
 586 EXPORT_SYMBOL_GPL(nexthop_select_path);
 587
 588 int nexthop_for_each_fib6_nh(struct nexthop *nh,
 589                              int (*cb)(struct fib6_nh *nh, void *arg),
 590                              void *arg)
 591 {
 592         struct nh_info *nhi;
 593         int err;
 594
 595         if (nh->is_group) {
 596                 struct nh_group *nhg;
 597                 int i;
 598
 599                 nhg = rcu_dereference_rtnl(nh->nh_grp);
 600                 for (i = 0; i < nhg->num_nh; i++) {
 601                         struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 602
 603                         nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
 604                         err = cb(&nhi->fib6_nh, arg);
 605                         if (err)
 606                                 return err;
 607                 }
 608         } else {
 609                 nhi = rcu_dereference_rtnl(nh->nh_info);
 610                 err = cb(&nhi->fib6_nh, arg);
 611                 if (err)
 612                         return err;
 613         }
 614
 615         return 0;
 616 }
 617 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
 618
 619 static int check_src_addr(const struct in6_addr *saddr,
 620                           struct netlink_ext_ack *extack)
 621 {
 622         if (!ipv6_addr_any(saddr)) {
 623                 NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
 624                 return -EINVAL;
 625         }
 626         return 0;
 627 }
 628
 629 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
 630                        struct netlink_ext_ack *extack)
 631 {
 632         struct nh_info *nhi;
 633         bool is_fdb_nh;
 634
 635         /* fib6_src is unique to a fib6_info and limits the ability to cache
 636          * routes in fib6_nh within a nexthop that is potentially shared
 637          * across multiple fib entries. If the config wants to use source
 638          * routing it can not use nexthop objects. mlxsw also does not allow
 639          * fib6_src on routes.
 640          */
 641         if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
 642                 return -EINVAL;
 643
 644         if (nh->is_group) {
 645                 struct nh_group *nhg;
 646
 647                 nhg = rtnl_dereference(nh->nh_grp);
 648                 if (nhg->has_v4)
 649                         goto no_v4_nh;
 650                 is_fdb_nh = nhg->fdb_nh;
 651         } else {
 652                 nhi = rtnl_dereference(nh->nh_info);
 653                 if (nhi->family == AF_INET)
 654                         goto no_v4_nh;
 655                 is_fdb_nh = nhi->fdb_nh;
 656         }
 657
 658         if (is_fdb_nh) {
 659                 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
 660                 return -EINVAL;
 661         }
 662
 663         return 0;
 664 no_v4_nh:
 665         NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
 666         return -EINVAL;
 667 }
 668 EXPORT_SYMBOL_GPL(fib6_check_nexthop);
 669
 670 /* if existing nexthop has ipv6 routes linked to it, need
 671  * to verify this new spec works with ipv6
 672  */
 673 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
 674                               struct netlink_ext_ack *extack)
 675 {
 676         struct fib6_info *f6i;
 677
 678         if (list_empty(&old->f6i_list))
 679                 return 0;
 680
 681         list_for_each_entry(f6i, &old->f6i_list, nh_list) {
 682                 if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
 683                         return -EINVAL;
 684         }
 685
 686         return fib6_check_nexthop(new, NULL, extack);
 687 }
 688
 689 static int nexthop_check_scope(struct nh_info *nhi, u8 scope,
 690                                struct netlink_ext_ack *extack)
 691 {
 692         if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
 693                 NL_SET_ERR_MSG(extack,
 694                                "Route with host scope can not have a gateway");
 695                 return -EINVAL;
 696         }
 697
 698         if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
 699                 NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
 700                 return -EINVAL;
 701         }
 702
 703         return 0;
 704 }
 705
 706 /* Invoked by fib add code to verify nexthop by id is ok with
 707  * config for prefix; parts of fib_check_nh not done when nexthop
 708  * object is used.
 709  */
 710 int fib_check_nexthop(struct nexthop *nh, u8 scope,
 711                       struct netlink_ext_ack *extack)
 712 {
 713         struct nh_info *nhi;
 714         int err = 0;
 715
 716         if (nh->is_group) {
 717                 struct nh_group *nhg;
 718
 719                 nhg = rtnl_dereference(nh->nh_grp);
 720                 if (nhg->fdb_nh) {
 721                         NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
 722                         err = -EINVAL;
 723                         goto out;
 724                 }
 725
 726                 if (scope == RT_SCOPE_HOST) {
 727                         NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
 728                         err = -EINVAL;
 729                         goto out;
 730                 }
 731
 732                 /* all nexthops in a group have the same scope */
 733                 nhi = rtnl_dereference(nhg->nh_entries[0].nh->nh_info);
 734                 err = nexthop_check_scope(nhi, scope, extack);
 735         } else {
 736                 nhi = rtnl_dereference(nh->nh_info);
 737                 if (nhi->fdb_nh) {
 738                         NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop");
 739                         err = -EINVAL;
 740                         goto out;
 741                 }
 742                 err = nexthop_check_scope(nhi, scope, extack);
 743         }
 744
 745 out:
 746         return err;
 747 }
 748
 749 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
 750                              struct netlink_ext_ack *extack)
 751 {
 752         struct fib_info *fi;
 753
 754         list_for_each_entry(fi, &old->fi_list, nh_list) {
 755                 int err;
 756
 757                 err = fib_check_nexthop(new, fi->fib_scope, extack);
 758                 if (err)
 759                         return err;
 760         }
 761         return 0;
 762 }
 763
 764 static void nh_group_rebalance(struct nh_group *nhg)
 765 {
 766         int total = 0;
 767         int w = 0;
 768         int i;
 769
 770         for (i = 0; i < nhg->num_nh; ++i)
 771                 total += nhg->nh_entries[i].weight;
 772
 773         for (i = 0; i < nhg->num_nh; ++i) {
 774                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 775                 int upper_bound;
 776
 777                 w += nhge->weight;
 778                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
 779                 atomic_set(&nhge->upper_bound, upper_bound);
 780         }
 781 }
 782
 783 static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
 784                                 struct nl_info *nlinfo)
 785 {
 786         struct nh_grp_entry *nhges, *new_nhges;
 787         struct nexthop *nhp = nhge->nh_parent;
 788         struct nexthop *nh = nhge->nh;
 789         struct nh_group *nhg, *newg;
 790         int i, j;
 791
 792         WARN_ON(!nh);
 793
 794         nhg = rtnl_dereference(nhp->nh_grp);
 795         newg = nhg->spare;
 796
 797         /* last entry, keep it visible and remove the parent */
 798         if (nhg->num_nh == 1) {
 799                 remove_nexthop(net, nhp, nlinfo);
 800                 return;
 801         }
 802
 803         newg->has_v4 = nhg->has_v4;
 804         newg->mpath = nhg->mpath;
 805         newg->fdb_nh = nhg->fdb_nh;
 806         newg->num_nh = nhg->num_nh;
 807
 808         /* copy old entries to new except the one getting removed */
 809         nhges = nhg->nh_entries;
 810         new_nhges = newg->nh_entries;
 811         for (i = 0, j = 0; i < nhg->num_nh; ++i) {
 812                 /* current nexthop getting removed */
 813                 if (nhg->nh_entries[i].nh == nh) {
 814                         newg->num_nh--;
 815                         continue;
 816                 }
 817
 818                 list_del(&nhges[i].nh_list);
 819                 new_nhges[j].nh_parent = nhges[i].nh_parent;
 820                 new_nhges[j].nh = nhges[i].nh;
 821                 new_nhges[j].weight = nhges[i].weight;
 822                 list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list);
 823                 j++;
 824         }
 825
 826         nh_group_rebalance(newg);
 827         rcu_assign_pointer(nhp->nh_grp, newg);
 828
 829         list_del(&nhge->nh_list);
 830         nexthop_put(nhge->nh);
 831
 832         if (nlinfo)
 833                 nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo);
 834 }
 835
 836 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
 837                                        struct nl_info *nlinfo)
 838 {
 839         struct nh_grp_entry *nhge, *tmp;
 840
 841         list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list)
 842                 remove_nh_grp_entry(net, nhge, nlinfo);
 843
 844         /* make sure all see the newly published array before releasing rtnl */
 845         synchronize_rcu();
 846 }
 847
 848 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
 849 {
 850         struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
 851         int i, num_nh = nhg->num_nh;
 852
 853         for (i = 0; i < num_nh; ++i) {
 854                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 855
 856                 if (WARN_ON(!nhge->nh))
 857                         continue;
 858
 859                 list_del_init(&nhge->nh_list);
 860         }
 861 }
 862
 863 /* not called for nexthop replace */
 864 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
 865 {
 866         struct fib6_info *f6i, *tmp;
 867         bool do_flush = false;
 868         struct fib_info *fi;
 869
 870         call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh);
 871
 872         list_for_each_entry(fi, &nh->fi_list, nh_list) {
 873                 fi->fib_flags |= RTNH_F_DEAD;
 874                 do_flush = true;
 875         }
 876         if (do_flush)
 877                 fib_flush(net);
 878
 879         /* ip6_del_rt removes the entry from this list hence the _safe */
 880         list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
 881                 /* __ip6_del_rt does a release, so do a hold here */
 882                 fib6_info_hold(f6i);
 883                 ipv6_stub->ip6_del_rt(net, f6i,
 884                                       !net->ipv4.sysctl_nexthop_compat_mode);
 885         }
 886 }
 887
 888 static void __remove_nexthop(struct net *net, struct nexthop *nh,
 889                              struct nl_info *nlinfo)
 890 {
 891         __remove_nexthop_fib(net, nh);
 892
 893         if (nh->is_group) {
 894                 remove_nexthop_group(nh, nlinfo);
 895         } else {
 896                 struct nh_info *nhi;
 897
 898                 nhi = rtnl_dereference(nh->nh_info);
 899                 if (nhi->fib_nhc.nhc_dev)
 900                         hlist_del(&nhi->dev_hash);
 901
 902                 remove_nexthop_from_groups(net, nh, nlinfo);
 903         }
 904 }
 905
 906 static void remove_nexthop(struct net *net, struct nexthop *nh,
 907                            struct nl_info *nlinfo)
 908 {
 909         /* remove from the tree */
 910         rb_erase(&nh->rb_node, &net->nexthop.rb_root);
 911
 912         if (nlinfo)
 913                 nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
 914
 915         __remove_nexthop(net, nh, nlinfo);
 916         nh_base_seq_inc(net);
 917
 918         nexthop_put(nh);
 919 }
 920
 921 /* if any FIB entries reference this nexthop, any dst entries
 922  * need to be regenerated
 923  */
 924 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh)
 925 {
 926         struct fib6_info *f6i;
 927
 928         if (!list_empty(&nh->fi_list))
 929                 rt_cache_flush(net);
 930
 931         list_for_each_entry(f6i, &nh->f6i_list, nh_list)
 932                 ipv6_stub->fib6_update_sernum(net, f6i);
 933 }
 934
 935 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
 936                                struct nexthop *new,
 937                                struct netlink_ext_ack *extack)
 938 {
 939         struct nh_group *oldg, *newg;
 940         int i;
 941
 942         if (!new->is_group) {
 943                 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
 944                 return -EINVAL;
 945         }
 946
 947         oldg = rtnl_dereference(old->nh_grp);
 948         newg = rtnl_dereference(new->nh_grp);
 949
 950         /* update parents - used by nexthop code for cleanup */
 951         for (i = 0; i < newg->num_nh; i++)
 952                 newg->nh_entries[i].nh_parent = old;
 953
 954         rcu_assign_pointer(old->nh_grp, newg);
 955
 956         for (i = 0; i < oldg->num_nh; i++)
 957                 oldg->nh_entries[i].nh_parent = new;
 958
 959         rcu_assign_pointer(new->nh_grp, oldg);
 960
 961         return 0;
 962 }
 963
 964 static int replace_nexthop_single(struct net *net, struct nexthop *old,
 965                                   struct nexthop *new,
 966                                   struct netlink_ext_ack *extack)
 967 {
 968         struct nh_info *oldi, *newi;
 969
 970         if (new->is_group) {
 971                 NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
 972                 return -EINVAL;
 973         }
 974
 975         oldi = rtnl_dereference(old->nh_info);
 976         newi = rtnl_dereference(new->nh_info);
 977
 978         newi->nh_parent = old;
 979         oldi->nh_parent = new;
 980
 981         old->protocol = new->protocol;
 982         old->nh_flags = new->nh_flags;
 983
 984         rcu_assign_pointer(old->nh_info, newi);
 985         rcu_assign_pointer(new->nh_info, oldi);
 986
 987         return 0;
 988 }
 989
 990 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
 991                                      struct nl_info *info)
 992 {
 993         struct fib6_info *f6i;
 994
 995         if (!list_empty(&nh->fi_list)) {
 996                 struct fib_info *fi;
 997
 998                 /* expectation is a few fib_info per nexthop and then
 999                  * a lot of routes per fib_info. So mark the fib_info
1000                  * and then walk the fib tables once
1001                  */
1002                 list_for_each_entry(fi, &nh->fi_list, nh_list)
1003                         fi->nh_updated = true;
1004
1005                 fib_info_notify_update(net, info);
1006
1007                 list_for_each_entry(fi, &nh->fi_list, nh_list)
1008                         fi->nh_updated = false;
1009         }
1010
1011         list_for_each_entry(f6i, &nh->f6i_list, nh_list)
1012                 ipv6_stub->fib6_rt_update(net, f6i, info);
1013 }
1014
1015 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
1016  * linked to this nexthop and for all groups that the nexthop
1017  * is a member of
1018  */
1019 static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
1020                                    struct nl_info *info)
1021 {
1022         struct nh_grp_entry *nhge;
1023
1024         __nexthop_replace_notify(net, nh, info);
1025
1026         list_for_each_entry(nhge, &nh->grp_list, nh_list)
1027                 __nexthop_replace_notify(net, nhge->nh_parent, info);
1028 }
1029
1030 static int replace_nexthop(struct net *net, struct nexthop *old,
1031                            struct nexthop *new, struct netlink_ext_ack *extack)
1032 {
1033         bool new_is_reject = false;
1034         struct nh_grp_entry *nhge;
1035         int err;
1036
1037         /* check that existing FIB entries are ok with the
1038          * new nexthop definition
1039          */
1040         err = fib_check_nh_list(old, new, extack);
1041         if (err)
1042                 return err;
1043
1044         err = fib6_check_nh_list(old, new, extack);
1045         if (err)
1046                 return err;
1047
1048         if (!new->is_group) {
1049                 struct nh_info *nhi = rtnl_dereference(new->nh_info);
1050
1051                 new_is_reject = nhi->reject_nh;
1052         }
1053
1054         list_for_each_entry(nhge, &old->grp_list, nh_list) {
1055                 /* if new nexthop is a blackhole, any groups using this
1056                  * nexthop cannot have more than 1 path
1057                  */
1058                 if (new_is_reject &&
1059                     nexthop_num_path(nhge->nh_parent) > 1) {
1060                         NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
1061                         return -EINVAL;
1062                 }
1063
1064                 err = fib_check_nh_list(nhge->nh_parent, new, extack);
1065                 if (err)
1066                         return err;
1067
1068                 err = fib6_check_nh_list(nhge->nh_parent, new, extack);
1069                 if (err)
1070                         return err;
1071         }
1072
1073         if (old->is_group)
1074                 err = replace_nexthop_grp(net, old, new, extack);
1075         else
1076                 err = replace_nexthop_single(net, old, new, extack);
1077
1078         if (!err) {
1079                 nh_rt_cache_flush(net, old);
1080
1081                 __remove_nexthop(net, new, NULL);
1082                 nexthop_put(new);
1083         }
1084
1085         return err;
1086 }
1087
1088 /* called with rtnl_lock held */
1089 static int insert_nexthop(struct net *net, struct nexthop *new_nh,
1090                           struct nh_config *cfg, struct netlink_ext_ack *extack)
1091 {
1092         struct rb_node **pp, *parent = NULL, *next;
1093         struct rb_root *root = &net->nexthop.rb_root;
1094         bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
1095         bool create = !!(cfg->nlflags & NLM_F_CREATE);
1096         u32 new_id = new_nh->id;
1097         int replace_notify = 0;
1098         int rc = -EEXIST;
1099
1100         pp = &root->rb_node;
1101         while (1) {
1102                 struct nexthop *nh;
1103
1104                 next = rtnl_dereference(*pp);
1105                 if (!next)
1106                         break;
1107
1108                 parent = next;
1109
1110                 nh = rb_entry(parent, struct nexthop, rb_node);
1111                 if (new_id < nh->id) {
1112                         pp = &next->rb_left;
1113                 } else if (new_id > nh->id) {
1114                         pp = &next->rb_right;
1115                 } else if (replace) {
1116                         rc = replace_nexthop(net, nh, new_nh, extack);
1117                         if (!rc) {
1118                                 new_nh = nh; /* send notification with old nh */
1119                                 replace_notify = 1;
1120                         }
1121                         goto out;
1122                 } else {
1123                         /* id already exists and not a replace */
1124                         goto out;
1125                 }
1126         }
1127
1128         if (replace && !create) {
1129                 NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
1130                 rc = -ENOENT;
1131                 goto out;
1132         }
1133
1134         rb_link_node_rcu(&new_nh->rb_node, parent, pp);
1135         rb_insert_color(&new_nh->rb_node, root);
1136         rc = 0;
1137 out:
1138         if (!rc) {
1139                 nh_base_seq_inc(net);
1140                 nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
1141                 if (replace_notify && net->ipv4.sysctl_nexthop_compat_mode)
1142                         nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
1143         }
1144
1145         return rc;
1146 }
1147
1148 /* rtnl */
1149 /* remove all nexthops tied to a device being deleted */
1150 static void nexthop_flush_dev(struct net_device *dev)
1151 {
1152         unsigned int hash = nh_dev_hashfn(dev->ifindex);
1153         struct net *net = dev_net(dev);
1154         struct hlist_head *head = &net->nexthop.devhash[hash];
1155         struct hlist_node *n;
1156         struct nh_info *nhi;
1157
1158         hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1159                 if (nhi->fib_nhc.nhc_dev != dev)
1160                         continue;
1161
1162                 remove_nexthop(net, nhi->nh_parent, NULL);
1163         }
1164 }
1165
1166 /* rtnl; called when net namespace is deleted */
1167 static void flush_all_nexthops(struct net *net)
1168 {
1169         struct rb_root *root = &net->nexthop.rb_root;
1170         struct rb_node *node;
1171         struct nexthop *nh;
1172
1173         while ((node = rb_first(root))) {
1174                 nh = rb_entry(node, struct nexthop, rb_node);
1175                 remove_nexthop(net, nh, NULL);
1176                 cond_resched();
1177         }
1178 }
1179
1180 static struct nexthop *nexthop_create_group(struct net *net,
1181                                             struct nh_config *cfg)
1182 {
1183         struct nlattr *grps_attr = cfg->nh_grp;
1184         struct nexthop_grp *entry = nla_data(grps_attr);
1185         u16 num_nh = nla_len(grps_attr) / sizeof(*entry);
1186         struct nh_group *nhg;
1187         struct nexthop *nh;
1188         int i;
1189
1190         nh = nexthop_alloc();
1191         if (!nh)
1192                 return ERR_PTR(-ENOMEM);
1193
1194         nh->is_group = 1;
1195
1196         nhg = nexthop_grp_alloc(num_nh);
1197         if (!nhg) {
1198                 kfree(nh);
1199                 return ERR_PTR(-ENOMEM);
1200         }
1201
1202         /* spare group used for removals */
1203         nhg->spare = nexthop_grp_alloc(num_nh);
1204         if (!nhg->spare) {
1205                 kfree(nhg);
1206                 kfree(nh);
1207                 return ERR_PTR(-ENOMEM);
1208         }
1209         nhg->spare->spare = nhg;
1210
1211         for (i = 0; i < nhg->num_nh; ++i) {
1212                 struct nexthop *nhe;
1213                 struct nh_info *nhi;
1214
1215                 nhe = nexthop_find_by_id(net, entry[i].id);
1216                 if (!nexthop_get(nhe))
1217                         goto out_no_nh;
1218
1219                 nhi = rtnl_dereference(nhe->nh_info);
1220                 if (nhi->family == AF_INET)
1221                         nhg->has_v4 = true;
1222
1223                 nhg->nh_entries[i].nh = nhe;
1224                 nhg->nh_entries[i].weight = entry[i].weight + 1;
1225                 list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
1226                 nhg->nh_entries[i].nh_parent = nh;
1227         }
1228
1229         if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) {
1230                 nhg->mpath = 1;
1231                 nh_group_rebalance(nhg);
1232         }
1233
1234         if (cfg->nh_fdb)
1235                 nhg->fdb_nh = 1;
1236
1237         rcu_assign_pointer(nh->nh_grp, nhg);
1238
1239         return nh;
1240
1241 out_no_nh:
1242         for (; i >= 0; --i)
1243                 nexthop_put(nhg->nh_entries[i].nh);
1244
1245         kfree(nhg->spare);
1246         kfree(nhg);
1247         kfree(nh);
1248
1249         return ERR_PTR(-ENOENT);
1250 }
1251
1252 static int nh_create_ipv4(struct net *net, struct nexthop *nh,
1253                           struct nh_info *nhi, struct nh_config *cfg,
1254                           struct netlink_ext_ack *extack)
1255 {
1256         struct fib_nh *fib_nh = &nhi->fib_nh;
1257         struct fib_config fib_cfg = {
1258                 .fc_oif   = cfg->nh_ifindex,
1259                 .fc_gw4   = cfg->gw.ipv4,
1260                 .fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
1261                 .fc_flags = cfg->nh_flags,
1262                 .fc_encap = cfg->nh_encap,
1263                 .fc_encap_type = cfg->nh_encap_type,
1264         };
1265         u32 tb_id = (cfg->dev ? l3mdev_fib_table(cfg->dev) : RT_TABLE_MAIN);
1266         int err;
1267
1268         err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
1269         if (err) {
1270                 fib_nh_release(net, fib_nh);
1271                 goto out;
1272         }
1273
1274         if (nhi->fdb_nh)
1275                 goto out;
1276
1277         /* sets nh_dev if successful */
1278         err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
1279         if (!err) {
1280                 nh->nh_flags = fib_nh->fib_nh_flags;
1281                 fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
1282                                           fib_nh->fib_nh_scope);
1283         } else {
1284                 fib_nh_release(net, fib_nh);
1285         }
1286 out:
1287         return err;
1288 }
1289
1290 static int nh_create_ipv6(struct net *net,  struct nexthop *nh,
1291                           struct nh_info *nhi, struct nh_config *cfg,
1292                           struct netlink_ext_ack *extack)
1293 {
1294         struct fib6_nh *fib6_nh = &nhi->fib6_nh;
1295         struct fib6_config fib6_cfg = {
1296                 .fc_table = l3mdev_fib_table(cfg->dev),
1297                 .fc_ifindex = cfg->nh_ifindex,
1298                 .fc_gateway = cfg->gw.ipv6,
1299                 .fc_flags = cfg->nh_flags,
1300                 .fc_encap = cfg->nh_encap,
1301                 .fc_encap_type = cfg->nh_encap_type,
1302                 .fc_is_fdb = cfg->nh_fdb,
1303         };
1304         int err;
1305
1306         if (!ipv6_addr_any(&cfg->gw.ipv6))
1307                 fib6_cfg.fc_flags |= RTF_GATEWAY;
1308
1309         /* sets nh_dev if successful */
1310         err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
1311                                       extack);
1312         if (err)
1313                 ipv6_stub->fib6_nh_release(fib6_nh);
1314         else
1315                 nh->nh_flags = fib6_nh->fib_nh_flags;
1316
1317         return err;
1318 }
1319
1320 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
1321                                       struct netlink_ext_ack *extack)
1322 {
1323         struct nh_info *nhi;
1324         struct nexthop *nh;
1325         int err = 0;
1326
1327         nh = nexthop_alloc();
1328         if (!nh)
1329                 return ERR_PTR(-ENOMEM);
1330
1331         nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
1332         if (!nhi) {
1333                 kfree(nh);
1334                 return ERR_PTR(-ENOMEM);
1335         }
1336
1337         nh->nh_flags = cfg->nh_flags;
1338         nh->net = net;
1339
1340         nhi->nh_parent = nh;
1341         nhi->family = cfg->nh_family;
1342         nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
1343
1344         if (cfg->nh_fdb)
1345                 nhi->fdb_nh = 1;
1346
1347         if (cfg->nh_blackhole) {
1348                 nhi->reject_nh = 1;
1349                 cfg->nh_ifindex = net->loopback_dev->ifindex;
1350         }
1351
1352         switch (cfg->nh_family) {
1353         case AF_INET:
1354                 err = nh_create_ipv4(net, nh, nhi, cfg, extack);
1355                 break;
1356         case AF_INET6:
1357                 err = nh_create_ipv6(net, nh, nhi, cfg, extack);
1358                 break;
1359         }
1360
1361         if (err) {
1362                 kfree(nhi);
1363                 kfree(nh);
1364                 return ERR_PTR(err);
1365         }
1366
1367         /* add the entry to the device based hash */
1368         if (!nhi->fdb_nh)
1369                 nexthop_devhash_add(net, nhi);
1370
1371         rcu_assign_pointer(nh->nh_info, nhi);
1372
1373         return nh;
1374 }
1375
1376 /* called with rtnl lock held */
1377 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
1378                                    struct netlink_ext_ack *extack)
1379 {
1380         struct nexthop *nh;
1381         int err;
1382
1383         if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
1384                 NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
1385                 return ERR_PTR(-EINVAL);
1386         }
1387
1388         if (!cfg->nh_id) {
1389                 cfg->nh_id = nh_find_unused_id(net);
1390                 if (!cfg->nh_id) {
1391                         NL_SET_ERR_MSG(extack, "No unused id");
1392                         return ERR_PTR(-EINVAL);
1393                 }
1394         }
1395
1396         if (cfg->nh_grp)
1397                 nh = nexthop_create_group(net, cfg);
1398         else
1399                 nh = nexthop_create(net, cfg, extack);
1400
1401         if (IS_ERR(nh))
1402                 return nh;
1403
1404         refcount_set(&nh->refcnt, 1);
1405         nh->id = cfg->nh_id;
1406         nh->protocol = cfg->nh_protocol;
1407         nh->net = net;
1408
1409         err = insert_nexthop(net, nh, cfg, extack);
1410         if (err) {
1411                 __remove_nexthop(net, nh, NULL);
1412                 nexthop_put(nh);
1413                 nh = ERR_PTR(err);
1414         }
1415
1416         return nh;
1417 }
1418
1419 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
1420                             struct nlmsghdr *nlh, struct nh_config *cfg,
1421                             struct netlink_ext_ack *extack)
1422 {
1423         struct nhmsg *nhm = nlmsg_data(nlh);
1424         struct nlattr *tb[NHA_MAX + 1];
1425         int err;
1426
1427         err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1428                           extack);
1429         if (err < 0)
1430                 return err;
1431
1432         err = -EINVAL;
1433         if (nhm->resvd || nhm->nh_scope) {
1434                 NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
1435                 goto out;
1436         }
1437         if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
1438                 NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
1439                 goto out;
1440         }
1441
1442         switch (nhm->nh_family) {
1443         case AF_INET:
1444         case AF_INET6:
1445                 break;
1446         case AF_UNSPEC:
1447                 if (tb[NHA_GROUP])
1448                         break;
1449                 fallthrough;
1450         default:
1451                 NL_SET_ERR_MSG(extack, "Invalid address family");
1452                 goto out;
1453         }
1454
1455         if (tb[NHA_GROUPS] || tb[NHA_MASTER]) {
1456                 NL_SET_ERR_MSG(extack, "Invalid attributes in request");
1457                 goto out;
1458         }
1459
1460         memset(cfg, 0, sizeof(*cfg));
1461         cfg->nlflags = nlh->nlmsg_flags;
1462         cfg->nlinfo.portid = NETLINK_CB(skb).portid;
1463         cfg->nlinfo.nlh = nlh;
1464         cfg->nlinfo.nl_net = net;
1465
1466         cfg->nh_family = nhm->nh_family;
1467         cfg->nh_protocol = nhm->nh_protocol;
1468         cfg->nh_flags = nhm->nh_flags;
1469
1470         if (tb[NHA_ID])
1471                 cfg->nh_id = nla_get_u32(tb[NHA_ID]);
1472
1473         if (tb[NHA_FDB]) {
1474                 if (tb[NHA_OIF] || tb[NHA_BLACKHOLE] ||
1475                     tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE]) {
1476                         NL_SET_ERR_MSG(extack, "Fdb attribute can not be used with encap, oif or blackhole");
1477                         goto out;
1478                 }
1479                 if (nhm->nh_flags) {
1480                         NL_SET_ERR_MSG(extack, "Unsupported nexthop flags in ancillary header");
1481                         goto out;
1482                 }
1483                 cfg->nh_fdb = nla_get_flag(tb[NHA_FDB]);
1484         }
1485
1486         if (tb[NHA_GROUP]) {
1487                 if (nhm->nh_family != AF_UNSPEC) {
1488                         NL_SET_ERR_MSG(extack, "Invalid family for group");
1489                         goto out;
1490                 }
1491                 cfg->nh_grp = tb[NHA_GROUP];
1492
1493                 cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
1494                 if (tb[NHA_GROUP_TYPE])
1495                         cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
1496
1497                 if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
1498                         NL_SET_ERR_MSG(extack, "Invalid group type");
1499                         goto out;
1500                 }
1501                 err = nh_check_attr_group(net, tb, extack);
1502
1503                 /* no other attributes should be set */
1504                 goto out;
1505         }
1506
1507         if (tb[NHA_BLACKHOLE]) {
1508                 if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
1509                     tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE] || tb[NHA_FDB]) {
1510                         NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway, oif, encap or fdb");
1511                         goto out;
1512                 }
1513
1514                 cfg->nh_blackhole = 1;
1515                 err = 0;
1516                 goto out;
1517         }
1518
1519         if (!cfg->nh_fdb && !tb[NHA_OIF]) {
1520                 NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole and non-fdb nexthops");
1521                 goto out;
1522         }
1523
1524         if (!cfg->nh_fdb && tb[NHA_OIF]) {
1525                 cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
1526                 if (cfg->nh_ifindex)
1527                         cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
1528
1529                 if (!cfg->dev) {
1530                         NL_SET_ERR_MSG(extack, "Invalid device index");
1531                         goto out;
1532                 } else if (!(cfg->dev->flags & IFF_UP)) {
1533                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1534                         err = -ENETDOWN;
1535                         goto out;
1536                 } else if (!netif_carrier_ok(cfg->dev)) {
1537                         NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
1538                         err = -ENETDOWN;
1539                         goto out;
1540                 }
1541         }
1542
1543         err = -EINVAL;
1544         if (tb[NHA_GATEWAY]) {
1545                 struct nlattr *gwa = tb[NHA_GATEWAY];
1546
1547                 switch (cfg->nh_family) {
1548                 case AF_INET:
1549                         if (nla_len(gwa) != sizeof(u32)) {
1550                                 NL_SET_ERR_MSG(extack, "Invalid gateway");
1551                                 goto out;
1552                         }
1553                         cfg->gw.ipv4 = nla_get_be32(gwa);
1554                         break;
1555                 case AF_INET6:
1556                         if (nla_len(gwa) != sizeof(struct in6_addr)) {
1557                                 NL_SET_ERR_MSG(extack, "Invalid gateway");
1558                                 goto out;
1559                         }
1560                         cfg->gw.ipv6 = nla_get_in6_addr(gwa);
1561                         break;
1562                 default:
1563                         NL_SET_ERR_MSG(extack,
1564                                        "Unknown address family for gateway");
1565                         goto out;
1566                 }
1567         } else {
1568                 /* device only nexthop (no gateway) */
1569                 if (cfg->nh_flags & RTNH_F_ONLINK) {
1570                         NL_SET_ERR_MSG(extack,
1571                                        "ONLINK flag can not be set for nexthop without a gateway");
1572                         goto out;
1573                 }
1574         }
1575
1576         if (tb[NHA_ENCAP]) {
1577                 cfg->nh_encap = tb[NHA_ENCAP];
1578
1579                 if (!tb[NHA_ENCAP_TYPE]) {
1580                         NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
1581                         goto out;
1582                 }
1583
1584                 cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
1585                 err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
1586                 if (err < 0)
1587                         goto out;
1588
1589         } else if (tb[NHA_ENCAP_TYPE]) {
1590                 NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
1591                 goto out;
1592         }
1593
1594
1595         err = 0;
1596 out:
1597         return err;
1598 }
1599
1600 /* rtnl */
1601 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1602                            struct netlink_ext_ack *extack)
1603 {
1604         struct net *net = sock_net(skb->sk);
1605         struct nh_config cfg;
1606         struct nexthop *nh;
1607         int err;
1608
1609         err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
1610         if (!err) {
1611                 nh = nexthop_add(net, &cfg, extack);
1612                 if (IS_ERR(nh))
1613                         err = PTR_ERR(nh);
1614         }
1615
1616         return err;
1617 }
1618
1619 static int nh_valid_get_del_req(struct nlmsghdr *nlh, u32 *id,
1620                                 struct netlink_ext_ack *extack)
1621 {
1622         struct nhmsg *nhm = nlmsg_data(nlh);
1623         struct nlattr *tb[NHA_MAX + 1];
1624         int err, i;
1625
1626         err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1627                           extack);
1628         if (err < 0)
1629                 return err;
1630
1631         err = -EINVAL;
1632         for (i = 0; i < __NHA_MAX; ++i) {
1633                 if (!tb[i])
1634                         continue;
1635
1636                 switch (i) {
1637                 case NHA_ID:
1638                         break;
1639                 default:
1640                         NL_SET_ERR_MSG_ATTR(extack, tb[i],
1641                                             "Unexpected attribute in request");
1642                         goto out;
1643                 }
1644         }
1645         if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1646                 NL_SET_ERR_MSG(extack, "Invalid values in header");
1647                 goto out;
1648         }
1649
1650         if (!tb[NHA_ID]) {
1651                 NL_SET_ERR_MSG(extack, "Nexthop id is missing");
1652                 goto out;
1653         }
1654
1655         *id = nla_get_u32(tb[NHA_ID]);
1656         if (!(*id))
1657                 NL_SET_ERR_MSG(extack, "Invalid nexthop id");
1658         else
1659                 err = 0;
1660 out:
1661         return err;
1662 }
1663
1664 /* rtnl */
1665 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1666                            struct netlink_ext_ack *extack)
1667 {
1668         struct net *net = sock_net(skb->sk);
1669         struct nl_info nlinfo = {
1670                 .nlh = nlh,
1671                 .nl_net = net,
1672                 .portid = NETLINK_CB(skb).portid,
1673         };
1674         struct nexthop *nh;
1675         int err;
1676         u32 id;
1677
1678         err = nh_valid_get_del_req(nlh, &id, extack);
1679         if (err)
1680                 return err;
1681
1682         nh = nexthop_find_by_id(net, id);
1683         if (!nh)
1684                 return -ENOENT;
1685
1686         remove_nexthop(net, nh, &nlinfo);
1687
1688         return 0;
1689 }
1690
1691 /* rtnl */
1692 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1693                            struct netlink_ext_ack *extack)
1694 {
1695         struct net *net = sock_net(in_skb->sk);
1696         struct sk_buff *skb = NULL;
1697         struct nexthop *nh;
1698         int err;
1699         u32 id;
1700
1701         err = nh_valid_get_del_req(nlh, &id, extack);
1702         if (err)
1703                 return err;
1704
1705         err = -ENOBUFS;
1706         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1707         if (!skb)
1708                 goto out;
1709
1710         err = -ENOENT;
1711         nh = nexthop_find_by_id(net, id);
1712         if (!nh)
1713                 goto errout_free;
1714
1715         err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
1716                            nlh->nlmsg_seq, 0);
1717         if (err < 0) {
1718                 WARN_ON(err == -EMSGSIZE);
1719                 goto errout_free;
1720         }
1721
1722         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1723 out:
1724         return err;
1725 errout_free:
1726         kfree_skb(skb);
1727         goto out;
1728 }
1729
1730 static bool nh_dump_filtered(struct nexthop *nh, int dev_idx, int master_idx,
1731                              bool group_filter, u8 family)
1732 {
1733         const struct net_device *dev;
1734         const struct nh_info *nhi;
1735
1736         if (group_filter && !nh->is_group)
1737                 return true;
1738
1739         if (!dev_idx && !master_idx && !family)
1740                 return false;
1741
1742         if (nh->is_group)
1743                 return true;
1744
1745         nhi = rtnl_dereference(nh->nh_info);
1746         if (family && nhi->family != family)
1747                 return true;
1748
1749         dev = nhi->fib_nhc.nhc_dev;
1750         if (dev_idx && (!dev || dev->ifindex != dev_idx))
1751                 return true;
1752
1753         if (master_idx) {
1754                 struct net_device *master;
1755
1756                 if (!dev)
1757                         return true;
1758
1759                 master = netdev_master_upper_dev_get((struct net_device *)dev);
1760                 if (!master || master->ifindex != master_idx)
1761                         return true;
1762         }
1763
1764         return false;
1765 }
1766
1767 static int nh_valid_dump_req(const struct nlmsghdr *nlh, int *dev_idx,
1768                              int *master_idx, bool *group_filter,
1769                              bool *fdb_filter, struct netlink_callback *cb)
1770 {
1771         struct netlink_ext_ack *extack = cb->extack;
1772         struct nlattr *tb[NHA_MAX + 1];
1773         struct nhmsg *nhm;
1774         int err, i;
1775         u32 idx;
1776
1777         err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1778                           NULL);
1779         if (err < 0)
1780                 return err;
1781
1782         for (i = 0; i <= NHA_MAX; ++i) {
1783                 if (!tb[i])
1784                         continue;
1785
1786                 switch (i) {
1787                 case NHA_OIF:
1788                         idx = nla_get_u32(tb[i]);
1789                         if (idx > INT_MAX) {
1790                                 NL_SET_ERR_MSG(extack, "Invalid device index");
1791                                 return -EINVAL;
1792                         }
1793                         *dev_idx = idx;
1794                         break;
1795                 case NHA_MASTER:
1796                         idx = nla_get_u32(tb[i]);
1797                         if (idx > INT_MAX) {
1798                                 NL_SET_ERR_MSG(extack, "Invalid master device index");
1799                                 return -EINVAL;
1800                         }
1801                         *master_idx = idx;
1802                         break;
1803                 case NHA_GROUPS:
1804                         *group_filter = true;
1805                         break;
1806                 case NHA_FDB:
1807                         *fdb_filter = true;
1808                         break;
1809                 default:
1810                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
1811                         return -EINVAL;
1812                 }
1813         }
1814
1815         nhm = nlmsg_data(nlh);
1816         if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1817                 NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
1818                 return -EINVAL;
1819         }
1820
1821         return 0;
1822 }
1823
1824 /* rtnl */
1825 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
1826 {
1827         bool group_filter = false, fdb_filter = false;
1828         struct nhmsg *nhm = nlmsg_data(cb->nlh);
1829         int dev_filter_idx = 0, master_idx = 0;
1830         struct net *net = sock_net(skb->sk);
1831         struct rb_root *root = &net->nexthop.rb_root;
1832         struct rb_node *node;
1833         int idx = 0, s_idx;
1834         int err;
1835
1836         err = nh_valid_dump_req(cb->nlh, &dev_filter_idx, &master_idx,
1837                                 &group_filter, &fdb_filter, cb);
1838         if (err < 0)
1839                 return err;
1840
1841         s_idx = cb->args[0];
1842         for (node = rb_first(root); node; node = rb_next(node)) {
1843                 struct nexthop *nh;
1844
1845                 if (idx < s_idx)
1846                         goto cont;
1847
1848                 nh = rb_entry(node, struct nexthop, rb_node);
1849                 if (nh_dump_filtered(nh, dev_filter_idx, master_idx,
1850                                      group_filter, nhm->nh_family))
1851                         goto cont;
1852
1853                 err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
1854                                    NETLINK_CB(cb->skb).portid,
1855                                    cb->nlh->nlmsg_seq, NLM_F_MULTI);
1856                 if (err < 0) {
1857                         if (likely(skb->len))
1858                                 goto out;
1859
1860                         goto out_err;
1861                 }
1862 cont:
1863                 idx++;
1864         }
1865
1866 out:
1867         err = skb->len;
1868 out_err:
1869         cb->args[0] = idx;
1870         cb->seq = net->nexthop.seq;
1871         nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1872
1873         return err;
1874 }
1875
1876 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
1877 {
1878         unsigned int hash = nh_dev_hashfn(dev->ifindex);
1879         struct net *net = dev_net(dev);
1880         struct hlist_head *head = &net->nexthop.devhash[hash];
1881         struct hlist_node *n;
1882         struct nh_info *nhi;
1883
1884         hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1885                 if (nhi->fib_nhc.nhc_dev == dev) {
1886                         if (nhi->family == AF_INET)
1887                                 fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
1888                                                    orig_mtu);
1889                 }
1890         }
1891 }
1892
1893 /* rtnl */
1894 static int nh_netdev_event(struct notifier_block *this,
1895                            unsigned long event, void *ptr)
1896 {
1897         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1898         struct netdev_notifier_info_ext *info_ext;
1899
1900         switch (event) {
1901         case NETDEV_DOWN:
1902         case NETDEV_UNREGISTER:
1903                 nexthop_flush_dev(dev);
1904                 break;
1905         case NETDEV_CHANGE:
1906                 if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
1907                         nexthop_flush_dev(dev);
1908                 break;
1909         case NETDEV_CHANGEMTU:
1910                 info_ext = ptr;
1911                 nexthop_sync_mtu(dev, info_ext->ext.mtu);
1912                 rt_cache_flush(dev_net(dev));
1913                 break;
1914         }
1915         return NOTIFY_DONE;
1916 }
1917
1918 static struct notifier_block nh_netdev_notifier = {
1919         .notifier_call = nh_netdev_event,
1920 };
1921
1922 int register_nexthop_notifier(struct net *net, struct notifier_block *nb)
1923 {
1924         return atomic_notifier_chain_register(&net->nexthop.notifier_chain, nb);
1925 }
1926 EXPORT_SYMBOL(register_nexthop_notifier);
1927
1928 int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb)
1929 {
1930         return atomic_notifier_chain_unregister(&net->nexthop.notifier_chain,
1931                                                 nb);
1932 }
1933 EXPORT_SYMBOL(unregister_nexthop_notifier);
1934
1935 static void __net_exit nexthop_net_exit(struct net *net)
1936 {
1937         rtnl_lock();
1938         flush_all_nexthops(net);
1939         rtnl_unlock();
1940         kfree(net->nexthop.devhash);
1941 }
1942
1943 static int __net_init nexthop_net_init(struct net *net)
1944 {
1945         size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
1946
1947         net->nexthop.rb_root = RB_ROOT;
1948         net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
1949         if (!net->nexthop.devhash)
1950                 return -ENOMEM;
1951         ATOMIC_INIT_NOTIFIER_HEAD(&net->nexthop.notifier_chain);
1952
1953         return 0;
1954 }
1955
1956 static struct pernet_operations nexthop_net_ops = {
1957         .init = nexthop_net_init,
1958         .exit = nexthop_net_exit,
1959 };
1960
1961 static int __init nexthop_init(void)
1962 {
1963         register_pernet_subsys(&nexthop_net_ops);
1964
1965         register_netdevice_notifier(&nh_netdev_notifier);
1966
1967         rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1968         rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
1969         rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
1970                       rtm_dump_nexthop, 0);
1971
1972         rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1973         rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1974
1975         rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1976         rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1977
1978         return 0;
1979 }
1980 subsys_initcall(nexthop_init);