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