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                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
  69
  70                 WARN_ON(!list_empty(&nhge->nh_list));
  71                 nexthop_put(nhge->nh);
  72         }
  73
  74         WARN_ON(nhg->spare == nhg);
  75
  76         kfree(nhg->spare);
  77         kfree(nhg);
  78 }
  79
  80 static void nexthop_free_single(struct nexthop *nh)
  81 {
  82         struct nh_info *nhi;
  83
  84         nhi = rcu_dereference_raw(nh->nh_info);
  85         switch (nhi->family) {
  86         case AF_INET:
  87                 fib_nh_release(nh->net, &nhi->fib_nh);
  88                 break;
  89         case AF_INET6:
  90                 ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
  91                 break;
  92         }
  93         kfree(nhi);
  94 }
  95
  96 void nexthop_free_rcu(struct rcu_head *head)
  97 {
  98         struct nexthop *nh = container_of(head, struct nexthop, rcu);
  99
 100         if (nh->is_group)
 101                 nexthop_free_mpath(nh);
 102         else
 103                 nexthop_free_single(nh);
 104
 105         kfree(nh);
 106 }
 107 EXPORT_SYMBOL_GPL(nexthop_free_rcu);
 108
 109 static struct nexthop *nexthop_alloc(void)
 110 {
 111         struct nexthop *nh;
 112
 113         nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
 114         if (nh) {
 115                 INIT_LIST_HEAD(&nh->fi_list);
 116                 INIT_LIST_HEAD(&nh->f6i_list);
 117                 INIT_LIST_HEAD(&nh->grp_list);
 118         }
 119         return nh;
 120 }
 121
 122 static struct nh_group *nexthop_grp_alloc(u16 num_nh)
 123 {
 124         size_t sz = offsetof(struct nexthop, nh_grp)
 125                     + sizeof(struct nh_group)
 126                     + sizeof(struct nh_grp_entry) * num_nh;
 127         struct nh_group *nhg;
 128
 129         nhg = kzalloc(sz, GFP_KERNEL);
 130         if (nhg)
 131                 nhg->num_nh = num_nh;
 132
 133         return nhg;
 134 }
 135
 136 static void nh_base_seq_inc(struct net *net)
 137 {
 138         while (++net->nexthop.seq == 0)
 139                 ;
 140 }
 141
 142 /* no reference taken; rcu lock or rtnl must be held */
 143 struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
 144 {
 145         struct rb_node **pp, *parent = NULL, *next;
 146
 147         pp = &net->nexthop.rb_root.rb_node;
 148         while (1) {
 149                 struct nexthop *nh;
 150
 151                 next = rcu_dereference_raw(*pp);
 152                 if (!next)
 153                         break;
 154                 parent = next;
 155
 156                 nh = rb_entry(parent, struct nexthop, rb_node);
 157                 if (id < nh->id)
 158                         pp = &next->rb_left;
 159                 else if (id > nh->id)
 160                         pp = &next->rb_right;
 161                 else
 162                         return nh;
 163         }
 164         return NULL;
 165 }
 166 EXPORT_SYMBOL_GPL(nexthop_find_by_id);
 167
 168 /* used for auto id allocation; called with rtnl held */
 169 static u32 nh_find_unused_id(struct net *net)
 170 {
 171         u32 id_start = net->nexthop.last_id_allocated;
 172
 173         while (1) {
 174                 net->nexthop.last_id_allocated++;
 175                 if (net->nexthop.last_id_allocated == id_start)
 176                         break;
 177
 178                 if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
 179                         return net->nexthop.last_id_allocated;
 180         }
 181         return 0;
 182 }
 183
 184 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
 185 {
 186         struct nexthop_grp *p;
 187         size_t len = nhg->num_nh * sizeof(*p);
 188         struct nlattr *nla;
 189         u16 group_type = 0;
 190         int i;
 191
 192         if (nhg->mpath)
 193                 group_type = NEXTHOP_GRP_TYPE_MPATH;
 194
 195         if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
 196                 goto nla_put_failure;
 197
 198         nla = nla_reserve(skb, NHA_GROUP, len);
 199         if (!nla)
 200                 goto nla_put_failure;
 201
 202         p = nla_data(nla);
 203         for (i = 0; i < nhg->num_nh; ++i) {
 204                 p->id = nhg->nh_entries[i].nh->id;
 205                 p->weight = nhg->nh_entries[i].weight - 1;
 206                 p += 1;
 207         }
 208
 209         return 0;
 210
 211 nla_put_failure:
 212         return -EMSGSIZE;
 213 }
 214
 215 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
 216                         int event, u32 portid, u32 seq, unsigned int nlflags)
 217 {
 218         struct fib6_nh *fib6_nh;
 219         struct fib_nh *fib_nh;
 220         struct nlmsghdr *nlh;
 221         struct nh_info *nhi;
 222         struct nhmsg *nhm;
 223
 224         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
 225         if (!nlh)
 226                 return -EMSGSIZE;
 227
 228         nhm = nlmsg_data(nlh);
 229         nhm->nh_family = AF_UNSPEC;
 230         nhm->nh_flags = nh->nh_flags;
 231         nhm->nh_protocol = nh->protocol;
 232         nhm->nh_scope = 0;
 233         nhm->resvd = 0;
 234
 235         if (nla_put_u32(skb, NHA_ID, nh->id))
 236                 goto nla_put_failure;
 237
 238         if (nh->is_group) {
 239                 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 240
 241                 if (nla_put_nh_group(skb, nhg))
 242                         goto nla_put_failure;
 243                 goto out;
 244         }
 245
 246         nhi = rtnl_dereference(nh->nh_info);
 247         nhm->nh_family = nhi->family;
 248         if (nhi->reject_nh) {
 249                 if (nla_put_flag(skb, NHA_BLACKHOLE))
 250                         goto nla_put_failure;
 251                 goto out;
 252         } else {
 253                 const struct net_device *dev;
 254
 255                 dev = nhi->fib_nhc.nhc_dev;
 256                 if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
 257                         goto nla_put_failure;
 258         }
 259
 260         nhm->nh_scope = nhi->fib_nhc.nhc_scope;
 261         switch (nhi->family) {
 262         case AF_INET:
 263                 fib_nh = &nhi->fib_nh;
 264                 if (fib_nh->fib_nh_gw_family &&
 265                     nla_put_u32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
 266                         goto nla_put_failure;
 267                 break;
 268
 269         case AF_INET6:
 270                 fib6_nh = &nhi->fib6_nh;
 271                 if (fib6_nh->fib_nh_gw_family &&
 272                     nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
 273                         goto nla_put_failure;
 274                 break;
 275         }
 276
 277         if (nhi->fib_nhc.nhc_lwtstate &&
 278             lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
 279                                 NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
 280                 goto nla_put_failure;
 281
 282 out:
 283         nlmsg_end(skb, nlh);
 284         return 0;
 285
 286 nla_put_failure:
 287         nlmsg_cancel(skb, nlh);
 288         return -EMSGSIZE;
 289 }
 290
 291 static size_t nh_nlmsg_size_grp(struct nexthop *nh)
 292 {
 293         struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 294         size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
 295
 296         return nla_total_size(sz) +
 297                nla_total_size(2);  /* NHA_GROUP_TYPE */
 298 }
 299
 300 static size_t nh_nlmsg_size_single(struct nexthop *nh)
 301 {
 302         struct nh_info *nhi = rtnl_dereference(nh->nh_info);
 303         size_t sz;
 304
 305         /* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
 306          * are mutually exclusive
 307          */
 308         sz = nla_total_size(4);  /* NHA_OIF */
 309
 310         switch (nhi->family) {
 311         case AF_INET:
 312                 if (nhi->fib_nh.fib_nh_gw_family)
 313                         sz += nla_total_size(4);  /* NHA_GATEWAY */
 314                 break;
 315
 316         case AF_INET6:
 317                 /* NHA_GATEWAY */
 318                 if (nhi->fib6_nh.fib_nh_gw_family)
 319                         sz += nla_total_size(sizeof(const struct in6_addr));
 320                 break;
 321         }
 322
 323         if (nhi->fib_nhc.nhc_lwtstate) {
 324                 sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
 325                 sz += nla_total_size(2);  /* NHA_ENCAP_TYPE */
 326         }
 327
 328         return sz;
 329 }
 330
 331 static size_t nh_nlmsg_size(struct nexthop *nh)
 332 {
 333         size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg));
 334
 335         sz += nla_total_size(4); /* NHA_ID */
 336
 337         if (nh->is_group)
 338                 sz += nh_nlmsg_size_grp(nh);
 339         else
 340                 sz += nh_nlmsg_size_single(nh);
 341
 342         return sz;
 343 }
 344
 345 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
 346 {
 347         unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
 348         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
 349         struct sk_buff *skb;
 350         int err = -ENOBUFS;
 351
 352         skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
 353         if (!skb)
 354                 goto errout;
 355
 356         err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
 357         if (err < 0) {
 358                 /* -EMSGSIZE implies BUG in nh_nlmsg_size() */
 359                 WARN_ON(err == -EMSGSIZE);
 360                 kfree_skb(skb);
 361                 goto errout;
 362         }
 363
 364         rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
 365                     info->nlh, gfp_any());
 366         return;
 367 errout:
 368         if (err < 0)
 369                 rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
 370 }
 371
 372 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
 373                            struct netlink_ext_ack *extack)
 374 {
 375         if (nh->is_group) {
 376                 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
 377
 378                 /* nested multipath (group within a group) is not
 379                  * supported
 380                  */
 381                 if (nhg->mpath) {
 382                         NL_SET_ERR_MSG(extack,
 383                                        "Multipath group can not be a nexthop within a group");
 384                         return false;
 385                 }
 386         } else {
 387                 struct nh_info *nhi = rtnl_dereference(nh->nh_info);
 388
 389                 if (nhi->reject_nh && npaths > 1) {
 390                         NL_SET_ERR_MSG(extack,
 391                                        "Blackhole nexthop can not be used in a group with more than 1 path");
 392                         return false;
 393                 }
 394         }
 395
 396         return true;
 397 }
 398
 399 static int nh_check_attr_group(struct net *net, struct nlattr *tb[],
 400                                struct netlink_ext_ack *extack)
 401 {
 402         unsigned int len = nla_len(tb[NHA_GROUP]);
 403         struct nexthop_grp *nhg;
 404         unsigned int i, j;
 405
 406         if (!len || len & (sizeof(struct nexthop_grp) - 1)) {
 407                 NL_SET_ERR_MSG(extack,
 408                                "Invalid length for nexthop group attribute");
 409                 return -EINVAL;
 410         }
 411
 412         /* convert len to number of nexthop ids */
 413         len /= sizeof(*nhg);
 414
 415         nhg = nla_data(tb[NHA_GROUP]);
 416         for (i = 0; i < len; ++i) {
 417                 if (nhg[i].resvd1 || nhg[i].resvd2) {
 418                         NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
 419                         return -EINVAL;
 420                 }
 421                 if (nhg[i].weight > 254) {
 422                         NL_SET_ERR_MSG(extack, "Invalid value for weight");
 423                         return -EINVAL;
 424                 }
 425                 for (j = i + 1; j < len; ++j) {
 426                         if (nhg[i].id == nhg[j].id) {
 427                                 NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
 428                                 return -EINVAL;
 429                         }
 430                 }
 431         }
 432
 433         nhg = nla_data(tb[NHA_GROUP]);
 434         for (i = 0; i < len; ++i) {
 435                 struct nexthop *nh;
 436
 437                 nh = nexthop_find_by_id(net, nhg[i].id);
 438                 if (!nh) {
 439                         NL_SET_ERR_MSG(extack, "Invalid nexthop id");
 440                         return -EINVAL;
 441                 }
 442                 if (!valid_group_nh(nh, len, extack))
 443                         return -EINVAL;
 444         }
 445         for (i = NHA_GROUP_TYPE + 1; i < __NHA_MAX; ++i) {
 446                 if (!tb[i])
 447                         continue;
 448
 449                 NL_SET_ERR_MSG(extack,
 450                                "No other attributes can be set in nexthop groups");
 451                 return -EINVAL;
 452         }
 453
 454         return 0;
 455 }
 456
 457 static bool ipv6_good_nh(const struct fib6_nh *nh)
 458 {
 459         int state = NUD_REACHABLE;
 460         struct neighbour *n;
 461
 462         rcu_read_lock_bh();
 463
 464         n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
 465         if (n)
 466                 state = n->nud_state;
 467
 468         rcu_read_unlock_bh();
 469
 470         return !!(state & NUD_VALID);
 471 }
 472
 473 static bool ipv4_good_nh(const struct fib_nh *nh)
 474 {
 475         int state = NUD_REACHABLE;
 476         struct neighbour *n;
 477
 478         rcu_read_lock_bh();
 479
 480         n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
 481                                       (__force u32)nh->fib_nh_gw4);
 482         if (n)
 483                 state = n->nud_state;
 484
 485         rcu_read_unlock_bh();
 486
 487         return !!(state & NUD_VALID);
 488 }
 489
 490 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
 491 {
 492         struct nexthop *rc = NULL;
 493         struct nh_group *nhg;
 494         int i;
 495
 496         if (!nh->is_group)
 497                 return nh;
 498
 499         nhg = rcu_dereference(nh->nh_grp);
 500         for (i = 0; i < nhg->num_nh; ++i) {
 501                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 502                 struct nh_info *nhi;
 503
 504                 if (hash > atomic_read(&nhge->upper_bound))
 505                         continue;
 506
 507                 /* nexthops always check if it is good and does
 508                  * not rely on a sysctl for this behavior
 509                  */
 510                 nhi = rcu_dereference(nhge->nh->nh_info);
 511                 switch (nhi->family) {
 512                 case AF_INET:
 513                         if (ipv4_good_nh(&nhi->fib_nh))
 514                                 return nhge->nh;
 515                         break;
 516                 case AF_INET6:
 517                         if (ipv6_good_nh(&nhi->fib6_nh))
 518                                 return nhge->nh;
 519                         break;
 520                 }
 521
 522                 if (!rc)
 523                         rc = nhge->nh;
 524         }
 525
 526         return rc;
 527 }
 528 EXPORT_SYMBOL_GPL(nexthop_select_path);
 529
 530 int nexthop_for_each_fib6_nh(struct nexthop *nh,
 531                              int (*cb)(struct fib6_nh *nh, void *arg),
 532                              void *arg)
 533 {
 534         struct nh_info *nhi;
 535         int err;
 536
 537         if (nh->is_group) {
 538                 struct nh_group *nhg;
 539                 int i;
 540
 541                 nhg = rcu_dereference_rtnl(nh->nh_grp);
 542                 for (i = 0; i < nhg->num_nh; i++) {
 543                         struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 544
 545                         nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
 546                         err = cb(&nhi->fib6_nh, arg);
 547                         if (err)
 548                                 return err;
 549                 }
 550         } else {
 551                 nhi = rcu_dereference_rtnl(nh->nh_info);
 552                 err = cb(&nhi->fib6_nh, arg);
 553                 if (err)
 554                         return err;
 555         }
 556
 557         return 0;
 558 }
 559 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
 560
 561 static int check_src_addr(const struct in6_addr *saddr,
 562                           struct netlink_ext_ack *extack)
 563 {
 564         if (!ipv6_addr_any(saddr)) {
 565                 NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
 566                 return -EINVAL;
 567         }
 568         return 0;
 569 }
 570
 571 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
 572                        struct netlink_ext_ack *extack)
 573 {
 574         struct nh_info *nhi;
 575
 576         /* fib6_src is unique to a fib6_info and limits the ability to cache
 577          * routes in fib6_nh within a nexthop that is potentially shared
 578          * across multiple fib entries. If the config wants to use source
 579          * routing it can not use nexthop objects. mlxsw also does not allow
 580          * fib6_src on routes.
 581          */
 582         if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
 583                 return -EINVAL;
 584
 585         if (nh->is_group) {
 586                 struct nh_group *nhg;
 587
 588                 nhg = rtnl_dereference(nh->nh_grp);
 589                 if (nhg->has_v4)
 590                         goto no_v4_nh;
 591         } else {
 592                 nhi = rtnl_dereference(nh->nh_info);
 593                 if (nhi->family == AF_INET)
 594                         goto no_v4_nh;
 595         }
 596
 597         return 0;
 598 no_v4_nh:
 599         NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
 600         return -EINVAL;
 601 }
 602 EXPORT_SYMBOL_GPL(fib6_check_nexthop);
 603
 604 /* if existing nexthop has ipv6 routes linked to it, need
 605  * to verify this new spec works with ipv6
 606  */
 607 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
 608                               struct netlink_ext_ack *extack)
 609 {
 610         struct fib6_info *f6i;
 611
 612         if (list_empty(&old->f6i_list))
 613                 return 0;
 614
 615         list_for_each_entry(f6i, &old->f6i_list, nh_list) {
 616                 if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
 617                         return -EINVAL;
 618         }
 619
 620         return fib6_check_nexthop(new, NULL, extack);
 621 }
 622
 623 static int nexthop_check_scope(struct nexthop *nh, u8 scope,
 624                                struct netlink_ext_ack *extack)
 625 {
 626         struct nh_info *nhi;
 627
 628         nhi = rtnl_dereference(nh->nh_info);
 629         if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
 630                 NL_SET_ERR_MSG(extack,
 631                                "Route with host scope can not have a gateway");
 632                 return -EINVAL;
 633         }
 634
 635         if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
 636                 NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
 637                 return -EINVAL;
 638         }
 639
 640         return 0;
 641 }
 642
 643 /* Invoked by fib add code to verify nexthop by id is ok with
 644  * config for prefix; parts of fib_check_nh not done when nexthop
 645  * object is used.
 646  */
 647 int fib_check_nexthop(struct nexthop *nh, u8 scope,
 648                       struct netlink_ext_ack *extack)
 649 {
 650         int err = 0;
 651
 652         if (nh->is_group) {
 653                 struct nh_group *nhg;
 654
 655                 if (scope == RT_SCOPE_HOST) {
 656                         NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
 657                         err = -EINVAL;
 658                         goto out;
 659                 }
 660
 661                 nhg = rtnl_dereference(nh->nh_grp);
 662                 /* all nexthops in a group have the same scope */
 663                 err = nexthop_check_scope(nhg->nh_entries[0].nh, scope, extack);
 664         } else {
 665                 err = nexthop_check_scope(nh, scope, extack);
 666         }
 667 out:
 668         return err;
 669 }
 670
 671 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
 672                              struct netlink_ext_ack *extack)
 673 {
 674         struct fib_info *fi;
 675
 676         list_for_each_entry(fi, &old->fi_list, nh_list) {
 677                 int err;
 678
 679                 err = fib_check_nexthop(new, fi->fib_scope, extack);
 680                 if (err)
 681                         return err;
 682         }
 683         return 0;
 684 }
 685
 686 static void nh_group_rebalance(struct nh_group *nhg)
 687 {
 688         int total = 0;
 689         int w = 0;
 690         int i;
 691
 692         for (i = 0; i < nhg->num_nh; ++i)
 693                 total += nhg->nh_entries[i].weight;
 694
 695         for (i = 0; i < nhg->num_nh; ++i) {
 696                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 697                 int upper_bound;
 698
 699                 w += nhge->weight;
 700                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
 701                 atomic_set(&nhge->upper_bound, upper_bound);
 702         }
 703 }
 704
 705 static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge,
 706                                 struct nl_info *nlinfo)
 707 {
 708         struct nh_grp_entry *nhges, *new_nhges;
 709         struct nexthop *nhp = nhge->nh_parent;
 710         struct nexthop *nh = nhge->nh;
 711         struct nh_group *nhg, *newg;
 712         int i, j;
 713
 714         WARN_ON(!nh);
 715
 716         nhg = rtnl_dereference(nhp->nh_grp);
 717         newg = nhg->spare;
 718
 719         /* last entry, keep it visible and remove the parent */
 720         if (nhg->num_nh == 1) {
 721                 remove_nexthop(net, nhp, nlinfo);
 722                 return;
 723         }
 724
 725         newg->has_v4 = nhg->has_v4;
 726         newg->mpath = nhg->mpath;
 727         newg->num_nh = nhg->num_nh;
 728
 729         /* copy old entries to new except the one getting removed */
 730         nhges = nhg->nh_entries;
 731         new_nhges = newg->nh_entries;
 732         for (i = 0, j = 0; i < nhg->num_nh; ++i) {
 733                 /* current nexthop getting removed */
 734                 if (nhg->nh_entries[i].nh == nh) {
 735                         newg->num_nh--;
 736                         continue;
 737                 }
 738
 739                 list_del(&nhges[i].nh_list);
 740                 new_nhges[j].nh_parent = nhges[i].nh_parent;
 741                 new_nhges[j].nh = nhges[i].nh;
 742                 new_nhges[j].weight = nhges[i].weight;
 743                 list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list);
 744                 j++;
 745         }
 746
 747         nh_group_rebalance(newg);
 748         rcu_assign_pointer(nhp->nh_grp, newg);
 749
 750         list_del(&nhge->nh_list);
 751         nexthop_put(nhge->nh);
 752
 753         if (nlinfo)
 754                 nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo);
 755 }
 756
 757 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
 758                                        struct nl_info *nlinfo)
 759 {
 760         struct nh_grp_entry *nhge, *tmp;
 761
 762         list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list)
 763                 remove_nh_grp_entry(net, nhge, nlinfo);
 764
 765         /* make sure all see the newly published array before releasing rtnl */
 766         synchronize_net();
 767 }
 768
 769 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
 770 {
 771         struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
 772         int i, num_nh = nhg->num_nh;
 773
 774         for (i = 0; i < num_nh; ++i) {
 775                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 776
 777                 if (WARN_ON(!nhge->nh))
 778                         continue;
 779
 780                 list_del_init(&nhge->nh_list);
 781         }
 782 }
 783
 784 /* not called for nexthop replace */
 785 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
 786 {
 787         struct fib6_info *f6i, *tmp;
 788         bool do_flush = false;
 789         struct fib_info *fi;
 790
 791         list_for_each_entry(fi, &nh->fi_list, nh_list) {
 792                 fi->fib_flags |= RTNH_F_DEAD;
 793                 do_flush = true;
 794         }
 795         if (do_flush)
 796                 fib_flush(net);
 797
 798         /* ip6_del_rt removes the entry from this list hence the _safe */
 799         list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
 800                 /* __ip6_del_rt does a release, so do a hold here */
 801                 fib6_info_hold(f6i);
 802                 ipv6_stub->ip6_del_rt(net, f6i);
 803         }
 804 }
 805
 806 static void __remove_nexthop(struct net *net, struct nexthop *nh,
 807                              struct nl_info *nlinfo)
 808 {
 809         __remove_nexthop_fib(net, nh);
 810
 811         if (nh->is_group) {
 812                 remove_nexthop_group(nh, nlinfo);
 813         } else {
 814                 struct nh_info *nhi;
 815
 816                 nhi = rtnl_dereference(nh->nh_info);
 817                 if (nhi->fib_nhc.nhc_dev)
 818                         hlist_del(&nhi->dev_hash);
 819
 820                 remove_nexthop_from_groups(net, nh, nlinfo);
 821         }
 822 }
 823
 824 static void remove_nexthop(struct net *net, struct nexthop *nh,
 825                            struct nl_info *nlinfo)
 826 {
 827         /* remove from the tree */
 828         rb_erase(&nh->rb_node, &net->nexthop.rb_root);
 829
 830         if (nlinfo)
 831                 nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
 832
 833         __remove_nexthop(net, nh, nlinfo);
 834         nh_base_seq_inc(net);
 835
 836         nexthop_put(nh);
 837 }
 838
 839 /* if any FIB entries reference this nexthop, any dst entries
 840  * need to be regenerated
 841  */
 842 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh,
 843                               struct nexthop *replaced_nh)
 844 {
 845         struct fib6_info *f6i;
 846         struct nh_group *nhg;
 847         int i;
 848
 849         if (!list_empty(&nh->fi_list))
 850                 rt_cache_flush(net);
 851
 852         list_for_each_entry(f6i, &nh->f6i_list, nh_list)
 853                 ipv6_stub->fib6_update_sernum(net, f6i);
 854
 855         /* if an IPv6 group was replaced, we have to release all old
 856          * dsts to make sure all refcounts are released
 857          */
 858         if (!replaced_nh->is_group)
 859                 return;
 860
 861         /* new dsts must use only the new nexthop group */
 862         synchronize_net();
 863
 864         nhg = rtnl_dereference(replaced_nh->nh_grp);
 865         for (i = 0; i < nhg->num_nh; i++) {
 866                 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
 867                 struct nh_info *nhi = rtnl_dereference(nhge->nh->nh_info);
 868
 869                 if (nhi->family == AF_INET6)
 870                         ipv6_stub->fib6_nh_release_dsts(&nhi->fib6_nh);
 871         }
 872 }
 873
 874 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
 875                                struct nexthop *new,
 876                                struct netlink_ext_ack *extack)
 877 {
 878         struct nh_group *oldg, *newg;
 879         int i;
 880
 881         if (!new->is_group) {
 882                 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
 883                 return -EINVAL;
 884         }
 885
 886         oldg = rtnl_dereference(old->nh_grp);
 887         newg = rtnl_dereference(new->nh_grp);
 888
 889         /* update parents - used by nexthop code for cleanup */
 890         for (i = 0; i < newg->num_nh; i++)
 891                 newg->nh_entries[i].nh_parent = old;
 892
 893         rcu_assign_pointer(old->nh_grp, newg);
 894
 895         for (i = 0; i < oldg->num_nh; i++)
 896                 oldg->nh_entries[i].nh_parent = new;
 897
 898         rcu_assign_pointer(new->nh_grp, oldg);
 899
 900         return 0;
 901 }
 902
 903 static int replace_nexthop_single(struct net *net, struct nexthop *old,
 904                                   struct nexthop *new,
 905                                   struct netlink_ext_ack *extack)
 906 {
 907         struct nh_info *oldi, *newi;
 908
 909         if (new->is_group) {
 910                 NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
 911                 return -EINVAL;
 912         }
 913
 914         oldi = rtnl_dereference(old->nh_info);
 915         newi = rtnl_dereference(new->nh_info);
 916
 917         newi->nh_parent = old;
 918         oldi->nh_parent = new;
 919
 920         old->protocol = new->protocol;
 921         old->nh_flags = new->nh_flags;
 922
 923         rcu_assign_pointer(old->nh_info, newi);
 924         rcu_assign_pointer(new->nh_info, oldi);
 925
 926         return 0;
 927 }
 928
 929 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
 930                                      struct nl_info *info)
 931 {
 932         struct fib6_info *f6i;
 933
 934         if (!list_empty(&nh->fi_list)) {
 935                 struct fib_info *fi;
 936
 937                 /* expectation is a few fib_info per nexthop and then
 938                  * a lot of routes per fib_info. So mark the fib_info
 939                  * and then walk the fib tables once
 940                  */
 941                 list_for_each_entry(fi, &nh->fi_list, nh_list)
 942                         fi->nh_updated = true;
 943
 944                 fib_info_notify_update(net, info);
 945
 946                 list_for_each_entry(fi, &nh->fi_list, nh_list)
 947                         fi->nh_updated = false;
 948         }
 949
 950         list_for_each_entry(f6i, &nh->f6i_list, nh_list)
 951                 ipv6_stub->fib6_rt_update(net, f6i, info);
 952 }
 953
 954 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
 955  * linked to this nexthop and for all groups that the nexthop
 956  * is a member of
 957  */
 958 static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
 959                                    struct nl_info *info)
 960 {
 961         struct nh_grp_entry *nhge;
 962
 963         __nexthop_replace_notify(net, nh, info);
 964
 965         list_for_each_entry(nhge, &nh->grp_list, nh_list)
 966                 __nexthop_replace_notify(net, nhge->nh_parent, info);
 967 }
 968
 969 static int replace_nexthop(struct net *net, struct nexthop *old,
 970                            struct nexthop *new, struct netlink_ext_ack *extack)
 971 {
 972         bool new_is_reject = false;
 973         struct nh_grp_entry *nhge;
 974         int err;
 975
 976         /* check that existing FIB entries are ok with the
 977          * new nexthop definition
 978          */
 979         err = fib_check_nh_list(old, new, extack);
 980         if (err)
 981                 return err;
 982
 983         err = fib6_check_nh_list(old, new, extack);
 984         if (err)
 985                 return err;
 986
 987         if (!new->is_group) {
 988                 struct nh_info *nhi = rtnl_dereference(new->nh_info);
 989
 990                 new_is_reject = nhi->reject_nh;
 991         }
 992
 993         list_for_each_entry(nhge, &old->grp_list, nh_list) {
 994                 /* if new nexthop is a blackhole, any groups using this
 995                  * nexthop cannot have more than 1 path
 996                  */
 997                 if (new_is_reject &&
 998                     nexthop_num_path(nhge->nh_parent) > 1) {
 999                         NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
1000                         return -EINVAL;
1001                 }
1002
1003                 err = fib_check_nh_list(nhge->nh_parent, new, extack);
1004                 if (err)
1005                         return err;
1006
1007                 err = fib6_check_nh_list(nhge->nh_parent, new, extack);
1008                 if (err)
1009                         return err;
1010         }
1011
1012         if (old->is_group)
1013                 err = replace_nexthop_grp(net, old, new, extack);
1014         else
1015                 err = replace_nexthop_single(net, old, new, extack);
1016
1017         if (!err) {
1018                 nh_rt_cache_flush(net, old, new);
1019
1020                 __remove_nexthop(net, new, NULL);
1021                 nexthop_put(new);
1022         }
1023
1024         return err;
1025 }
1026
1027 /* called with rtnl_lock held */
1028 static int insert_nexthop(struct net *net, struct nexthop *new_nh,
1029                           struct nh_config *cfg, struct netlink_ext_ack *extack)
1030 {
1031         struct rb_node **pp, *parent = NULL, *next;
1032         struct rb_root *root = &net->nexthop.rb_root;
1033         bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
1034         bool create = !!(cfg->nlflags & NLM_F_CREATE);
1035         u32 new_id = new_nh->id;
1036         int replace_notify = 0;
1037         int rc = -EEXIST;
1038
1039         pp = &root->rb_node;
1040         while (1) {
1041                 struct nexthop *nh;
1042
1043                 next = rtnl_dereference(*pp);
1044                 if (!next)
1045                         break;
1046
1047                 parent = next;
1048
1049                 nh = rb_entry(parent, struct nexthop, rb_node);
1050                 if (new_id < nh->id) {
1051                         pp = &next->rb_left;
1052                 } else if (new_id > nh->id) {
1053                         pp = &next->rb_right;
1054                 } else if (replace) {
1055                         rc = replace_nexthop(net, nh, new_nh, extack);
1056                         if (!rc) {
1057                                 new_nh = nh; /* send notification with old nh */
1058                                 replace_notify = 1;
1059                         }
1060                         goto out;
1061                 } else {
1062                         /* id already exists and not a replace */
1063                         goto out;
1064                 }
1065         }
1066
1067         if (replace && !create) {
1068                 NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
1069                 rc = -ENOENT;
1070                 goto out;
1071         }
1072
1073         rb_link_node_rcu(&new_nh->rb_node, parent, pp);
1074         rb_insert_color(&new_nh->rb_node, root);
1075         rc = 0;
1076 out:
1077         if (!rc) {
1078                 nh_base_seq_inc(net);
1079                 nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
1080                 if (replace_notify)
1081                         nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
1082         }
1083
1084         return rc;
1085 }
1086
1087 /* rtnl */
1088 /* remove all nexthops tied to a device being deleted */
1089 static void nexthop_flush_dev(struct net_device *dev, unsigned long event)
1090 {
1091         unsigned int hash = nh_dev_hashfn(dev->ifindex);
1092         struct net *net = dev_net(dev);
1093         struct hlist_head *head = &net->nexthop.devhash[hash];
1094         struct hlist_node *n;
1095         struct nh_info *nhi;
1096
1097         hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1098                 if (nhi->fib_nhc.nhc_dev != dev)
1099                         continue;
1100
1101                 if (nhi->reject_nh &&
1102                     (event == NETDEV_DOWN || event == NETDEV_CHANGE))
1103                         continue;
1104
1105                 remove_nexthop(net, nhi->nh_parent, NULL);
1106         }
1107 }
1108
1109 /* rtnl; called when net namespace is deleted */
1110 static void flush_all_nexthops(struct net *net)
1111 {
1112         struct rb_root *root = &net->nexthop.rb_root;
1113         struct rb_node *node;
1114         struct nexthop *nh;
1115
1116         while ((node = rb_first(root))) {
1117                 nh = rb_entry(node, struct nexthop, rb_node);
1118                 remove_nexthop(net, nh, NULL);
1119                 cond_resched();
1120         }
1121 }
1122
1123 static struct nexthop *nexthop_create_group(struct net *net,
1124                                             struct nh_config *cfg)
1125 {
1126         struct nlattr *grps_attr = cfg->nh_grp;
1127         struct nexthop_grp *entry = nla_data(grps_attr);
1128         u16 num_nh = nla_len(grps_attr) / sizeof(*entry);
1129         struct nh_group *nhg;
1130         struct nexthop *nh;
1131         int i;
1132
1133         if (WARN_ON(!num_nh))
1134                 return ERR_PTR(-EINVAL);
1135
1136         nh = nexthop_alloc();
1137         if (!nh)
1138                 return ERR_PTR(-ENOMEM);
1139
1140         nh->is_group = 1;
1141
1142         nhg = nexthop_grp_alloc(num_nh);
1143         if (!nhg) {
1144                 kfree(nh);
1145                 return ERR_PTR(-ENOMEM);
1146         }
1147
1148         /* spare group used for removals */
1149         nhg->spare = nexthop_grp_alloc(num_nh);
1150         if (!nhg) {
1151                 kfree(nhg);
1152                 kfree(nh);
1153                 return NULL;
1154         }
1155         nhg->spare->spare = nhg;
1156
1157         for (i = 0; i < nhg->num_nh; ++i) {
1158                 struct nexthop *nhe;
1159                 struct nh_info *nhi;
1160
1161                 nhe = nexthop_find_by_id(net, entry[i].id);
1162                 if (!nexthop_get(nhe))
1163                         goto out_no_nh;
1164
1165                 nhi = rtnl_dereference(nhe->nh_info);
1166                 if (nhi->family == AF_INET)
1167                         nhg->has_v4 = true;
1168
1169                 nhg->nh_entries[i].nh = nhe;
1170                 nhg->nh_entries[i].weight = entry[i].weight + 1;
1171                 list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
1172                 nhg->nh_entries[i].nh_parent = nh;
1173         }
1174
1175         if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) {
1176                 nhg->mpath = 1;
1177                 nh_group_rebalance(nhg);
1178         }
1179
1180         rcu_assign_pointer(nh->nh_grp, nhg);
1181
1182         return nh;
1183
1184 out_no_nh:
1185         for (i--; i >= 0; --i) {
1186                 list_del(&nhg->nh_entries[i].nh_list);
1187                 nexthop_put(nhg->nh_entries[i].nh);
1188         }
1189
1190         kfree(nhg->spare);
1191         kfree(nhg);
1192         kfree(nh);
1193
1194         return ERR_PTR(-ENOENT);
1195 }
1196
1197 static int nh_create_ipv4(struct net *net, struct nexthop *nh,
1198                           struct nh_info *nhi, struct nh_config *cfg,
1199                           struct netlink_ext_ack *extack)
1200 {
1201         struct fib_nh *fib_nh = &nhi->fib_nh;
1202         struct fib_config fib_cfg = {
1203                 .fc_oif   = cfg->nh_ifindex,
1204                 .fc_gw4   = cfg->gw.ipv4,
1205                 .fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
1206                 .fc_flags = cfg->nh_flags,
1207                 .fc_nlinfo = cfg->nlinfo,
1208                 .fc_encap = cfg->nh_encap,
1209                 .fc_encap_type = cfg->nh_encap_type,
1210         };
1211         u32 tb_id = l3mdev_fib_table(cfg->dev);
1212         int err;
1213
1214         err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
1215         if (err) {
1216                 fib_nh_release(net, fib_nh);
1217                 goto out;
1218         }
1219
1220         /* sets nh_dev if successful */
1221         err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
1222         if (!err) {
1223                 nh->nh_flags = fib_nh->fib_nh_flags;
1224                 fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
1225                                           !fib_nh->fib_nh_scope ? 0 : fib_nh->fib_nh_scope - 1);
1226         } else {
1227                 fib_nh_release(net, fib_nh);
1228         }
1229 out:
1230         return err;
1231 }
1232
1233 static int nh_create_ipv6(struct net *net,  struct nexthop *nh,
1234                           struct nh_info *nhi, struct nh_config *cfg,
1235                           struct netlink_ext_ack *extack)
1236 {
1237         struct fib6_nh *fib6_nh = &nhi->fib6_nh;
1238         struct fib6_config fib6_cfg = {
1239                 .fc_table = l3mdev_fib_table(cfg->dev),
1240                 .fc_ifindex = cfg->nh_ifindex,
1241                 .fc_gateway = cfg->gw.ipv6,
1242                 .fc_flags = cfg->nh_flags,
1243                 .fc_nlinfo = cfg->nlinfo,
1244                 .fc_encap = cfg->nh_encap,
1245                 .fc_encap_type = cfg->nh_encap_type,
1246         };
1247         int err;
1248
1249         if (!ipv6_addr_any(&cfg->gw.ipv6))
1250                 fib6_cfg.fc_flags |= RTF_GATEWAY;
1251
1252         /* sets nh_dev if successful */
1253         err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
1254                                       extack);
1255         if (err) {
1256                 /* IPv6 is not enabled, don't call fib6_nh_release */
1257                 if (err == -EAFNOSUPPORT)
1258                         goto out;
1259                 ipv6_stub->fib6_nh_release(fib6_nh);
1260         } else {
1261                 nh->nh_flags = fib6_nh->fib_nh_flags;
1262         }
1263 out:
1264         return err;
1265 }
1266
1267 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
1268                                       struct netlink_ext_ack *extack)
1269 {
1270         struct nh_info *nhi;
1271         struct nexthop *nh;
1272         int err = 0;
1273
1274         nh = nexthop_alloc();
1275         if (!nh)
1276                 return ERR_PTR(-ENOMEM);
1277
1278         nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
1279         if (!nhi) {
1280                 kfree(nh);
1281                 return ERR_PTR(-ENOMEM);
1282         }
1283
1284         nh->nh_flags = cfg->nh_flags;
1285         nh->net = net;
1286
1287         nhi->nh_parent = nh;
1288         nhi->family = cfg->nh_family;
1289         nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
1290
1291         if (cfg->nh_blackhole) {
1292                 nhi->reject_nh = 1;
1293                 cfg->nh_ifindex = net->loopback_dev->ifindex;
1294         }
1295
1296         switch (cfg->nh_family) {
1297         case AF_INET:
1298                 err = nh_create_ipv4(net, nh, nhi, cfg, extack);
1299                 break;
1300         case AF_INET6:
1301                 err = nh_create_ipv6(net, nh, nhi, cfg, extack);
1302                 break;
1303         }
1304
1305         if (err) {
1306                 kfree(nhi);
1307                 kfree(nh);
1308                 return ERR_PTR(err);
1309         }
1310
1311         /* add the entry to the device based hash */
1312         nexthop_devhash_add(net, nhi);
1313
1314         rcu_assign_pointer(nh->nh_info, nhi);
1315
1316         return nh;
1317 }
1318
1319 /* called with rtnl lock held */
1320 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
1321                                    struct netlink_ext_ack *extack)
1322 {
1323         struct nexthop *nh;
1324         int err;
1325
1326         if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
1327                 NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
1328                 return ERR_PTR(-EINVAL);
1329         }
1330
1331         if (!cfg->nh_id) {
1332                 cfg->nh_id = nh_find_unused_id(net);
1333                 if (!cfg->nh_id) {
1334                         NL_SET_ERR_MSG(extack, "No unused id");
1335                         return ERR_PTR(-EINVAL);
1336                 }
1337         }
1338
1339         if (cfg->nh_grp)
1340                 nh = nexthop_create_group(net, cfg);
1341         else
1342                 nh = nexthop_create(net, cfg, extack);
1343
1344         if (IS_ERR(nh))
1345                 return nh;
1346
1347         refcount_set(&nh->refcnt, 1);
1348         nh->id = cfg->nh_id;
1349         nh->protocol = cfg->nh_protocol;
1350         nh->net = net;
1351
1352         err = insert_nexthop(net, nh, cfg, extack);
1353         if (err) {
1354                 __remove_nexthop(net, nh, NULL);
1355                 nexthop_put(nh);
1356                 nh = ERR_PTR(err);
1357         }
1358
1359         return nh;
1360 }
1361
1362 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
1363                             struct nlmsghdr *nlh, struct nh_config *cfg,
1364                             struct netlink_ext_ack *extack)
1365 {
1366         struct nhmsg *nhm = nlmsg_data(nlh);
1367         struct nlattr *tb[NHA_MAX + 1];
1368         int err;
1369
1370         err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1371                           extack);
1372         if (err < 0)
1373                 return err;
1374
1375         err = -EINVAL;
1376         if (nhm->resvd || nhm->nh_scope) {
1377                 NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
1378                 goto out;
1379         }
1380         if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
1381                 NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
1382                 goto out;
1383         }
1384
1385         switch (nhm->nh_family) {
1386         case AF_INET:
1387         case AF_INET6:
1388                 break;
1389         case AF_UNSPEC:
1390                 if (tb[NHA_GROUP])
1391                         break;
1392                 /* fallthrough */
1393         default:
1394                 NL_SET_ERR_MSG(extack, "Invalid address family");
1395                 goto out;
1396         }
1397
1398         if (tb[NHA_GROUPS] || tb[NHA_MASTER]) {
1399                 NL_SET_ERR_MSG(extack, "Invalid attributes in request");
1400                 goto out;
1401         }
1402
1403         memset(cfg, 0, sizeof(*cfg));
1404         cfg->nlflags = nlh->nlmsg_flags;
1405         cfg->nlinfo.portid = NETLINK_CB(skb).portid;
1406         cfg->nlinfo.nlh = nlh;
1407         cfg->nlinfo.nl_net = net;
1408
1409         cfg->nh_family = nhm->nh_family;
1410         cfg->nh_protocol = nhm->nh_protocol;
1411         cfg->nh_flags = nhm->nh_flags;
1412
1413         if (tb[NHA_ID])
1414                 cfg->nh_id = nla_get_u32(tb[NHA_ID]);
1415
1416         if (tb[NHA_GROUP]) {
1417                 if (nhm->nh_family != AF_UNSPEC) {
1418                         NL_SET_ERR_MSG(extack, "Invalid family for group");
1419                         goto out;
1420                 }
1421                 cfg->nh_grp = tb[NHA_GROUP];
1422
1423                 cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
1424                 if (tb[NHA_GROUP_TYPE])
1425                         cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
1426
1427                 if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
1428                         NL_SET_ERR_MSG(extack, "Invalid group type");
1429                         goto out;
1430                 }
1431                 err = nh_check_attr_group(net, tb, extack);
1432
1433                 /* no other attributes should be set */
1434                 goto out;
1435         }
1436
1437         if (tb[NHA_BLACKHOLE]) {
1438                 if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
1439                     tb[NHA_ENCAP]   || tb[NHA_ENCAP_TYPE]) {
1440                         NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway or oif");
1441                         goto out;
1442                 }
1443
1444                 cfg->nh_blackhole = 1;
1445                 err = 0;
1446                 goto out;
1447         }
1448
1449         if (!tb[NHA_OIF]) {
1450                 NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole nexthops");
1451                 goto out;
1452         }
1453
1454         cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
1455         if (cfg->nh_ifindex)
1456                 cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
1457
1458         if (!cfg->dev) {
1459                 NL_SET_ERR_MSG(extack, "Invalid device index");
1460                 goto out;
1461         } else if (!(cfg->dev->flags & IFF_UP)) {
1462                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1463                 err = -ENETDOWN;
1464                 goto out;
1465         } else if (!netif_carrier_ok(cfg->dev)) {
1466                 NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
1467                 err = -ENETDOWN;
1468                 goto out;
1469         }
1470
1471         err = -EINVAL;
1472         if (tb[NHA_GATEWAY]) {
1473                 struct nlattr *gwa = tb[NHA_GATEWAY];
1474
1475                 switch (cfg->nh_family) {
1476                 case AF_INET:
1477                         if (nla_len(gwa) != sizeof(u32)) {
1478                                 NL_SET_ERR_MSG(extack, "Invalid gateway");
1479                                 goto out;
1480                         }
1481                         cfg->gw.ipv4 = nla_get_be32(gwa);
1482                         break;
1483                 case AF_INET6:
1484                         if (nla_len(gwa) != sizeof(struct in6_addr)) {
1485                                 NL_SET_ERR_MSG(extack, "Invalid gateway");
1486                                 goto out;
1487                         }
1488                         cfg->gw.ipv6 = nla_get_in6_addr(gwa);
1489                         break;
1490                 default:
1491                         NL_SET_ERR_MSG(extack,
1492                                        "Unknown address family for gateway");
1493                         goto out;
1494                 }
1495         } else {
1496                 /* device only nexthop (no gateway) */
1497                 if (cfg->nh_flags & RTNH_F_ONLINK) {
1498                         NL_SET_ERR_MSG(extack,
1499                                        "ONLINK flag can not be set for nexthop without a gateway");
1500                         goto out;
1501                 }
1502         }
1503
1504         if (tb[NHA_ENCAP]) {
1505                 cfg->nh_encap = tb[NHA_ENCAP];
1506
1507                 if (!tb[NHA_ENCAP_TYPE]) {
1508                         NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
1509                         goto out;
1510                 }
1511
1512                 cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
1513                 err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
1514                 if (err < 0)
1515                         goto out;
1516
1517         } else if (tb[NHA_ENCAP_TYPE]) {
1518                 NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
1519                 goto out;
1520         }
1521
1522
1523         err = 0;
1524 out:
1525         return err;
1526 }
1527
1528 /* rtnl */
1529 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1530                            struct netlink_ext_ack *extack)
1531 {
1532         struct net *net = sock_net(skb->sk);
1533         struct nh_config cfg;
1534         struct nexthop *nh;
1535         int err;
1536
1537         err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
1538         if (!err) {
1539                 nh = nexthop_add(net, &cfg, extack);
1540                 if (IS_ERR(nh))
1541                         err = PTR_ERR(nh);
1542         }
1543
1544         return err;
1545 }
1546
1547 static int nh_valid_get_del_req(struct nlmsghdr *nlh, u32 *id,
1548                                 struct netlink_ext_ack *extack)
1549 {
1550         struct nhmsg *nhm = nlmsg_data(nlh);
1551         struct nlattr *tb[NHA_MAX + 1];
1552         int err, i;
1553
1554         err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1555                           extack);
1556         if (err < 0)
1557                 return err;
1558
1559         err = -EINVAL;
1560         for (i = 0; i < __NHA_MAX; ++i) {
1561                 if (!tb[i])
1562                         continue;
1563
1564                 switch (i) {
1565                 case NHA_ID:
1566                         break;
1567                 default:
1568                         NL_SET_ERR_MSG_ATTR(extack, tb[i],
1569                                             "Unexpected attribute in request");
1570                         goto out;
1571                 }
1572         }
1573         if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1574                 NL_SET_ERR_MSG(extack, "Invalid values in header");
1575                 goto out;
1576         }
1577
1578         if (!tb[NHA_ID]) {
1579                 NL_SET_ERR_MSG(extack, "Nexthop id is missing");
1580                 goto out;
1581         }
1582
1583         *id = nla_get_u32(tb[NHA_ID]);
1584         if (!(*id))
1585                 NL_SET_ERR_MSG(extack, "Invalid nexthop id");
1586         else
1587                 err = 0;
1588 out:
1589         return err;
1590 }
1591
1592 /* rtnl */
1593 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1594                            struct netlink_ext_ack *extack)
1595 {
1596         struct net *net = sock_net(skb->sk);
1597         struct nl_info nlinfo = {
1598                 .nlh = nlh,
1599                 .nl_net = net,
1600                 .portid = NETLINK_CB(skb).portid,
1601         };
1602         struct nexthop *nh;
1603         int err;
1604         u32 id;
1605
1606         err = nh_valid_get_del_req(nlh, &id, extack);
1607         if (err)
1608                 return err;
1609
1610         nh = nexthop_find_by_id(net, id);
1611         if (!nh)
1612                 return -ENOENT;
1613
1614         remove_nexthop(net, nh, &nlinfo);
1615
1616         return 0;
1617 }
1618
1619 /* rtnl */
1620 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1621                            struct netlink_ext_ack *extack)
1622 {
1623         struct net *net = sock_net(in_skb->sk);
1624         struct sk_buff *skb = NULL;
1625         struct nexthop *nh;
1626         int err;
1627         u32 id;
1628
1629         err = nh_valid_get_del_req(nlh, &id, extack);
1630         if (err)
1631                 return err;
1632
1633         err = -ENOBUFS;
1634         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1635         if (!skb)
1636                 goto out;
1637
1638         err = -ENOENT;
1639         nh = nexthop_find_by_id(net, id);
1640         if (!nh)
1641                 goto errout_free;
1642
1643         err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
1644                            nlh->nlmsg_seq, 0);
1645         if (err < 0) {
1646                 WARN_ON(err == -EMSGSIZE);
1647                 goto errout_free;
1648         }
1649
1650         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1651 out:
1652         return err;
1653 errout_free:
1654         kfree_skb(skb);
1655         goto out;
1656 }
1657
1658 static bool nh_dump_filtered(struct nexthop *nh, int dev_idx, int master_idx,
1659                              bool group_filter, u8 family)
1660 {
1661         const struct net_device *dev;
1662         const struct nh_info *nhi;
1663
1664         if (group_filter && !nh->is_group)
1665                 return true;
1666
1667         if (!dev_idx && !master_idx && !family)
1668                 return false;
1669
1670         if (nh->is_group)
1671                 return true;
1672
1673         nhi = rtnl_dereference(nh->nh_info);
1674         if (family && nhi->family != family)
1675                 return true;
1676
1677         dev = nhi->fib_nhc.nhc_dev;
1678         if (dev_idx && (!dev || dev->ifindex != dev_idx))
1679                 return true;
1680
1681         if (master_idx) {
1682                 struct net_device *master;
1683
1684                 if (!dev)
1685                         return true;
1686
1687                 master = netdev_master_upper_dev_get((struct net_device *)dev);
1688                 if (!master || master->ifindex != master_idx)
1689                         return true;
1690         }
1691
1692         return false;
1693 }
1694
1695 static int nh_valid_dump_req(const struct nlmsghdr *nlh, int *dev_idx,
1696                              int *master_idx, bool *group_filter,
1697                              struct netlink_callback *cb)
1698 {
1699         struct netlink_ext_ack *extack = cb->extack;
1700         struct nlattr *tb[NHA_MAX + 1];
1701         struct nhmsg *nhm;
1702         int err, i;
1703         u32 idx;
1704
1705         err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1706                           NULL);
1707         if (err < 0)
1708                 return err;
1709
1710         for (i = 0; i <= NHA_MAX; ++i) {
1711                 if (!tb[i])
1712                         continue;
1713
1714                 switch (i) {
1715                 case NHA_OIF:
1716                         idx = nla_get_u32(tb[i]);
1717                         if (idx > INT_MAX) {
1718                                 NL_SET_ERR_MSG(extack, "Invalid device index");
1719                                 return -EINVAL;
1720                         }
1721                         *dev_idx = idx;
1722                         break;
1723                 case NHA_MASTER:
1724                         idx = nla_get_u32(tb[i]);
1725                         if (idx > INT_MAX) {
1726                                 NL_SET_ERR_MSG(extack, "Invalid master device index");
1727                                 return -EINVAL;
1728                         }
1729                         *master_idx = idx;
1730                         break;
1731                 case NHA_GROUPS:
1732                         *group_filter = true;
1733                         break;
1734                 default:
1735                         NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
1736                         return -EINVAL;
1737                 }
1738         }
1739
1740         nhm = nlmsg_data(nlh);
1741         if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1742                 NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
1743                 return -EINVAL;
1744         }
1745
1746         return 0;
1747 }
1748
1749 /* rtnl */
1750 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
1751 {
1752         struct nhmsg *nhm = nlmsg_data(cb->nlh);
1753         int dev_filter_idx = 0, master_idx = 0;
1754         struct net *net = sock_net(skb->sk);
1755         struct rb_root *root = &net->nexthop.rb_root;
1756         bool group_filter = false;
1757         struct rb_node *node;
1758         int idx = 0, s_idx;
1759         int err;
1760
1761         err = nh_valid_dump_req(cb->nlh, &dev_filter_idx, &master_idx,
1762                                 &group_filter, cb);
1763         if (err < 0)
1764                 return err;
1765
1766         s_idx = cb->args[0];
1767         for (node = rb_first(root); node; node = rb_next(node)) {
1768                 struct nexthop *nh;
1769
1770                 if (idx < s_idx)
1771                         goto cont;
1772
1773                 nh = rb_entry(node, struct nexthop, rb_node);
1774                 if (nh_dump_filtered(nh, dev_filter_idx, master_idx,
1775                                      group_filter, nhm->nh_family))
1776                         goto cont;
1777
1778                 err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
1779                                    NETLINK_CB(cb->skb).portid,
1780                                    cb->nlh->nlmsg_seq, NLM_F_MULTI);
1781                 if (err < 0) {
1782                         if (likely(skb->len))
1783                                 goto out;
1784
1785                         goto out_err;
1786                 }
1787 cont:
1788                 idx++;
1789         }
1790
1791 out:
1792         err = skb->len;
1793 out_err:
1794         cb->args[0] = idx;
1795         cb->seq = net->nexthop.seq;
1796         nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1797
1798         return err;
1799 }
1800
1801 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
1802 {
1803         unsigned int hash = nh_dev_hashfn(dev->ifindex);
1804         struct net *net = dev_net(dev);
1805         struct hlist_head *head = &net->nexthop.devhash[hash];
1806         struct hlist_node *n;
1807         struct nh_info *nhi;
1808
1809         hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1810                 if (nhi->fib_nhc.nhc_dev == dev) {
1811                         if (nhi->family == AF_INET)
1812                                 fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
1813                                                    orig_mtu);
1814                 }
1815         }
1816 }
1817
1818 /* rtnl */
1819 static int nh_netdev_event(struct notifier_block *this,
1820                            unsigned long event, void *ptr)
1821 {
1822         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1823         struct netdev_notifier_info_ext *info_ext;
1824
1825         switch (event) {
1826         case NETDEV_DOWN:
1827         case NETDEV_UNREGISTER:
1828                 nexthop_flush_dev(dev, event);
1829                 break;
1830         case NETDEV_CHANGE:
1831                 if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
1832                         nexthop_flush_dev(dev, event);
1833                 break;
1834         case NETDEV_CHANGEMTU:
1835                 info_ext = ptr;
1836                 nexthop_sync_mtu(dev, info_ext->ext.mtu);
1837                 rt_cache_flush(dev_net(dev));
1838                 break;
1839         }
1840         return NOTIFY_DONE;
1841 }
1842
1843 static struct notifier_block nh_netdev_notifier = {
1844         .notifier_call = nh_netdev_event,
1845 };
1846
1847 static void __net_exit nexthop_net_exit(struct net *net)
1848 {
1849         rtnl_lock();
1850         flush_all_nexthops(net);
1851         rtnl_unlock();
1852         kfree(net->nexthop.devhash);
1853 }
1854
1855 static int __net_init nexthop_net_init(struct net *net)
1856 {
1857         size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
1858
1859         net->nexthop.rb_root = RB_ROOT;
1860         net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
1861         if (!net->nexthop.devhash)
1862                 return -ENOMEM;
1863
1864         return 0;
1865 }
1866
1867 static struct pernet_operations nexthop_net_ops = {
1868         .init = nexthop_net_init,
1869         .exit = nexthop_net_exit,
1870 };
1871
1872 static int __init nexthop_init(void)
1873 {
1874         register_pernet_subsys(&nexthop_net_ops);
1875
1876         register_netdevice_notifier(&nh_netdev_notifier);
1877
1878         rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1879         rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
1880         rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
1881                       rtm_dump_nexthop, 0);
1882
1883         rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1884         rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1885
1886         rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1887         rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1888
1889         return 0;
1890 }
1891 subsys_initcall(nexthop_init);