net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <net/net_namespace.h>
  45 #include <net/snmp.h>
  46 #include <net/ipv6.h>
  47 #include <net/ip6_fib.h>
  48 #include <net/ip6_route.h>
  49 #include <net/ndisc.h>
  50 #include <net/addrconf.h>
  51 #include <net/tcp.h>
  52 #include <linux/rtnetlink.h>
  53 #include <net/dst.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/xfrm.h>
  56 #include <net/netevent.h>
  57 #include <net/netlink.h>
  58 #include <net/rtnh.h>
  59 #include <net/lwtunnel.h>
  60 #include <net/ip_tunnels.h>
  61 #include <net/l3mdev.h>
  62 #include <net/ip.h>
  63 #include <linux/uaccess.h>
  64
  65 #ifdef CONFIG_SYSCTL
  66 #include <linux/sysctl.h>
  67 #endif
  68
  69 static int ip6_rt_type_to_error(u8 fib6_type);
  70
  71 #define CREATE_TRACE_POINTS
  72 #include <trace/events/fib6.h>
  73 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  74 #undef CREATE_TRACE_POINTS
  75
  76 enum rt6_nud_state {
  77         RT6_NUD_FAIL_HARD = -3,
  78         RT6_NUD_FAIL_PROBE = -2,
  79         RT6_NUD_FAIL_DO_RR = -1,
  80         RT6_NUD_SUCCEED = 1
  81 };
  82
  83 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  84 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  85 static unsigned int      ip6_mtu(const struct dst_entry *dst);
  86 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  87 static void             ip6_dst_destroy(struct dst_entry *);
  88 static void             ip6_dst_ifdown(struct dst_entry *,
  89                                        struct net_device *dev, int how);
  90 static int               ip6_dst_gc(struct dst_ops *ops);
  91
  92 static int              ip6_pkt_discard(struct sk_buff *skb);
  93 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  94 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  95 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  96 static void             ip6_link_failure(struct sk_buff *skb);
  97 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
  98                                            struct sk_buff *skb, u32 mtu);
  99 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 100                                         struct sk_buff *skb);
 101 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 102                            int strict);
 103 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 104 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 105                          struct fib6_info *rt, struct dst_entry *dst,
 106                          struct in6_addr *dest, struct in6_addr *src,
 107                          int iif, int type, u32 portid, u32 seq,
 108                          unsigned int flags);
 109 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 110                                            const struct in6_addr *daddr,
 111                                            const struct in6_addr *saddr);
 112
 113 #ifdef CONFIG_IPV6_ROUTE_INFO
 114 static struct fib6_info *rt6_add_route_info(struct net *net,
 115                                            const struct in6_addr *prefix, int prefixlen,
 116                                            const struct in6_addr *gwaddr,
 117                                            struct net_device *dev,
 118                                            unsigned int pref);
 119 static struct fib6_info *rt6_get_route_info(struct net *net,
 120                                            const struct in6_addr *prefix, int prefixlen,
 121                                            const struct in6_addr *gwaddr,
 122                                            struct net_device *dev);
 123 #endif
 124
 125 struct uncached_list {
 126         spinlock_t              lock;
 127         struct list_head        head;
 128 };
 129
 130 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 131
 132 void rt6_uncached_list_add(struct rt6_info *rt)
 133 {
 134         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 135
 136         rt->rt6i_uncached_list = ul;
 137
 138         spin_lock_bh(&ul->lock);
 139         list_add_tail(&rt->rt6i_uncached, &ul->head);
 140         spin_unlock_bh(&ul->lock);
 141 }
 142
 143 void rt6_uncached_list_del(struct rt6_info *rt)
 144 {
 145         if (!list_empty(&rt->rt6i_uncached)) {
 146                 struct uncached_list *ul = rt->rt6i_uncached_list;
 147                 struct net *net = dev_net(rt->dst.dev);
 148
 149                 spin_lock_bh(&ul->lock);
 150                 list_del(&rt->rt6i_uncached);
 151                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 152                 spin_unlock_bh(&ul->lock);
 153         }
 154 }
 155
 156 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 157 {
 158         struct net_device *loopback_dev = net->loopback_dev;
 159         int cpu;
 160
 161         if (dev == loopback_dev)
 162                 return;
 163
 164         for_each_possible_cpu(cpu) {
 165                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 166                 struct rt6_info *rt;
 167
 168                 spin_lock_bh(&ul->lock);
 169                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 170                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 171                         struct net_device *rt_dev = rt->dst.dev;
 172
 173                         if (rt_idev->dev == dev) {
 174                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 175                                 in6_dev_put(rt_idev);
 176                         }
 177
 178                         if (rt_dev == dev) {
 179                                 rt->dst.dev = blackhole_netdev;
 180                                 dev_hold(rt->dst.dev);
 181                                 dev_put(rt_dev);
 182                         }
 183                 }
 184                 spin_unlock_bh(&ul->lock);
 185         }
 186 }
 187
 188 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 189                                              struct sk_buff *skb,
 190                                              const void *daddr)
 191 {
 192         if (!ipv6_addr_any(p))
 193                 return (const void *) p;
 194         else if (skb)
 195                 return &ipv6_hdr(skb)->daddr;
 196         return daddr;
 197 }
 198
 199 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 200                                    struct net_device *dev,
 201                                    struct sk_buff *skb,
 202                                    const void *daddr)
 203 {
 204         struct neighbour *n;
 205
 206         daddr = choose_neigh_daddr(gw, skb, daddr);
 207         n = __ipv6_neigh_lookup(dev, daddr);
 208         if (n)
 209                 return n;
 210
 211         n = neigh_create(&nd_tbl, daddr, dev);
 212         return IS_ERR(n) ? NULL : n;
 213 }
 214
 215 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 216                                               struct sk_buff *skb,
 217                                               const void *daddr)
 218 {
 219         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 220
 221         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 222                                 dst->dev, skb, daddr);
 223 }
 224
 225 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 226 {
 227         struct net_device *dev = dst->dev;
 228         struct rt6_info *rt = (struct rt6_info *)dst;
 229
 230         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 231         if (!daddr)
 232                 return;
 233         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 234                 return;
 235         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 236                 return;
 237         __ipv6_confirm_neigh(dev, daddr);
 238 }
 239
 240 static struct dst_ops ip6_dst_ops_template = {
 241         .family                 =       AF_INET6,
 242         .gc                     =       ip6_dst_gc,
 243         .gc_thresh              =       1024,
 244         .check                  =       ip6_dst_check,
 245         .default_advmss         =       ip6_default_advmss,
 246         .mtu                    =       ip6_mtu,
 247         .cow_metrics            =       dst_cow_metrics_generic,
 248         .destroy                =       ip6_dst_destroy,
 249         .ifdown                 =       ip6_dst_ifdown,
 250         .negative_advice        =       ip6_negative_advice,
 251         .link_failure           =       ip6_link_failure,
 252         .update_pmtu            =       ip6_rt_update_pmtu,
 253         .redirect               =       rt6_do_redirect,
 254         .local_out              =       __ip6_local_out,
 255         .neigh_lookup           =       ip6_dst_neigh_lookup,
 256         .confirm_neigh          =       ip6_confirm_neigh,
 257 };
 258
 259 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 260 {
 261         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 262
 263         return mtu ? : dst->dev->mtu;
 264 }
 265
 266 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 267                                          struct sk_buff *skb, u32 mtu)
 268 {
 269 }
 270
 271 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 272                                       struct sk_buff *skb)
 273 {
 274 }
 275
 276 static struct dst_ops ip6_dst_blackhole_ops = {
 277         .family                 =       AF_INET6,
 278         .destroy                =       ip6_dst_destroy,
 279         .check                  =       ip6_dst_check,
 280         .mtu                    =       ip6_blackhole_mtu,
 281         .default_advmss         =       ip6_default_advmss,
 282         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 283         .redirect               =       ip6_rt_blackhole_redirect,
 284         .cow_metrics            =       dst_cow_metrics_generic,
 285         .neigh_lookup           =       ip6_dst_neigh_lookup,
 286 };
 287
 288 static const u32 ip6_template_metrics[RTAX_MAX] = {
 289         [RTAX_HOPLIMIT - 1] = 0,
 290 };
 291
 292 static const struct fib6_info fib6_null_entry_template = {
 293         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 294         .fib6_protocol  = RTPROT_KERNEL,
 295         .fib6_metric    = ~(u32)0,
 296         .fib6_ref       = REFCOUNT_INIT(1),
 297         .fib6_type      = RTN_UNREACHABLE,
 298         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 299 };
 300
 301 static const struct rt6_info ip6_null_entry_template = {
 302         .dst = {
 303                 .__refcnt       = ATOMIC_INIT(1),
 304                 .__use          = 1,
 305                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 306                 .error          = -ENETUNREACH,
 307                 .input          = ip6_pkt_discard,
 308                 .output         = ip6_pkt_discard_out,
 309         },
 310         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 311 };
 312
 313 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 314
 315 static const struct rt6_info ip6_prohibit_entry_template = {
 316         .dst = {
 317                 .__refcnt       = ATOMIC_INIT(1),
 318                 .__use          = 1,
 319                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 320                 .error          = -EACCES,
 321                 .input          = ip6_pkt_prohibit,
 322                 .output         = ip6_pkt_prohibit_out,
 323         },
 324         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 325 };
 326
 327 static const struct rt6_info ip6_blk_hole_entry_template = {
 328         .dst = {
 329                 .__refcnt       = ATOMIC_INIT(1),
 330                 .__use          = 1,
 331                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 332                 .error          = -EINVAL,
 333                 .input          = dst_discard,
 334                 .output         = dst_discard_out,
 335         },
 336         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 337 };
 338
 339 #endif
 340
 341 static void rt6_info_init(struct rt6_info *rt)
 342 {
 343         struct dst_entry *dst = &rt->dst;
 344
 345         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 346         INIT_LIST_HEAD(&rt->rt6i_uncached);
 347 }
 348
 349 /* allocate dst with ip6_dst_ops */
 350 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 351                                int flags)
 352 {
 353         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 354                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 355
 356         if (rt) {
 357                 rt6_info_init(rt);
 358                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 359         }
 360
 361         return rt;
 362 }
 363 EXPORT_SYMBOL(ip6_dst_alloc);
 364
 365 static void ip6_dst_destroy(struct dst_entry *dst)
 366 {
 367         struct rt6_info *rt = (struct rt6_info *)dst;
 368         struct fib6_info *from;
 369         struct inet6_dev *idev;
 370
 371         ip_dst_metrics_put(dst);
 372         rt6_uncached_list_del(rt);
 373
 374         idev = rt->rt6i_idev;
 375         if (idev) {
 376                 rt->rt6i_idev = NULL;
 377                 in6_dev_put(idev);
 378         }
 379
 380         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 381         fib6_info_release(from);
 382 }
 383
 384 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 385                            int how)
 386 {
 387         struct rt6_info *rt = (struct rt6_info *)dst;
 388         struct inet6_dev *idev = rt->rt6i_idev;
 389         struct net_device *loopback_dev =
 390                 dev_net(dev)->loopback_dev;
 391
 392         if (idev && idev->dev != loopback_dev) {
 393                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 394                 if (loopback_idev) {
 395                         rt->rt6i_idev = loopback_idev;
 396                         in6_dev_put(idev);
 397                 }
 398         }
 399 }
 400
 401 static bool __rt6_check_expired(const struct rt6_info *rt)
 402 {
 403         if (rt->rt6i_flags & RTF_EXPIRES)
 404                 return time_after(jiffies, rt->dst.expires);
 405         else
 406                 return false;
 407 }
 408
 409 static bool rt6_check_expired(const struct rt6_info *rt)
 410 {
 411         struct fib6_info *from;
 412
 413         from = rcu_dereference(rt->from);
 414
 415         if (rt->rt6i_flags & RTF_EXPIRES) {
 416                 if (time_after(jiffies, rt->dst.expires))
 417                         return true;
 418         } else if (from) {
 419                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 420                         fib6_check_expired(from);
 421         }
 422         return false;
 423 }
 424
 425 void fib6_select_path(const struct net *net, struct fib6_result *res,
 426                       struct flowi6 *fl6, int oif, bool have_oif_match,
 427                       const struct sk_buff *skb, int strict)
 428 {
 429         struct fib6_info *sibling, *next_sibling;
 430         struct fib6_info *match = res->f6i;
 431
 432         if ((!match->fib6_nsiblings && !match->nh) || have_oif_match)
 433                 goto out;
 434
 435         /* We might have already computed the hash for ICMPv6 errors. In such
 436          * case it will always be non-zero. Otherwise now is the time to do it.
 437          */
 438         if (!fl6->mp_hash &&
 439             (!match->nh || nexthop_is_multipath(match->nh)))
 440                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 441
 442         if (unlikely(match->nh)) {
 443                 nexthop_path_fib6_result(res, fl6->mp_hash);
 444                 return;
 445         }
 446
 447         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 448                 goto out;
 449
 450         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 451                                  fib6_siblings) {
 452                 const struct fib6_nh *nh = sibling->fib6_nh;
 453                 int nh_upper_bound;
 454
 455                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 456                 if (fl6->mp_hash > nh_upper_bound)
 457                         continue;
 458                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 459                         break;
 460                 match = sibling;
 461                 break;
 462         }
 463
 464 out:
 465         res->f6i = match;
 466         res->nh = match->fib6_nh;
 467 }
 468
 469 /*
 470  *      Route lookup. rcu_read_lock() should be held.
 471  */
 472
 473 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 474                                const struct in6_addr *saddr, int oif, int flags)
 475 {
 476         const struct net_device *dev;
 477
 478         if (nh->fib_nh_flags & RTNH_F_DEAD)
 479                 return false;
 480
 481         dev = nh->fib_nh_dev;
 482         if (oif) {
 483                 if (dev->ifindex == oif)
 484                         return true;
 485         } else {
 486                 if (ipv6_chk_addr(net, saddr, dev,
 487                                   flags & RT6_LOOKUP_F_IFACE))
 488                         return true;
 489         }
 490
 491         return false;
 492 }
 493
 494 struct fib6_nh_dm_arg {
 495         struct net              *net;
 496         const struct in6_addr   *saddr;
 497         int                     oif;
 498         int                     flags;
 499         struct fib6_nh          *nh;
 500 };
 501
 502 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 503 {
 504         struct fib6_nh_dm_arg *arg = _arg;
 505
 506         arg->nh = nh;
 507         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 508                                   arg->flags);
 509 }
 510
 511 /* returns fib6_nh from nexthop or NULL */
 512 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 513                                         struct fib6_result *res,
 514                                         const struct in6_addr *saddr,
 515                                         int oif, int flags)
 516 {
 517         struct fib6_nh_dm_arg arg = {
 518                 .net   = net,
 519                 .saddr = saddr,
 520                 .oif   = oif,
 521                 .flags = flags,
 522         };
 523
 524         if (nexthop_is_blackhole(nh))
 525                 return NULL;
 526
 527         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 528                 return arg.nh;
 529
 530         return NULL;
 531 }
 532
 533 static void rt6_device_match(struct net *net, struct fib6_result *res,
 534                              const struct in6_addr *saddr, int oif, int flags)
 535 {
 536         struct fib6_info *f6i = res->f6i;
 537         struct fib6_info *spf6i;
 538         struct fib6_nh *nh;
 539
 540         if (!oif && ipv6_addr_any(saddr)) {
 541                 if (unlikely(f6i->nh)) {
 542                         nh = nexthop_fib6_nh(f6i->nh);
 543                         if (nexthop_is_blackhole(f6i->nh))
 544                                 goto out_blackhole;
 545                 } else {
 546                         nh = f6i->fib6_nh;
 547                 }
 548                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 549                         goto out;
 550         }
 551
 552         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 553                 bool matched = false;
 554
 555                 if (unlikely(spf6i->nh)) {
 556                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 557                                               oif, flags);
 558                         if (nh)
 559                                 matched = true;
 560                 } else {
 561                         nh = spf6i->fib6_nh;
 562                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 563                                 matched = true;
 564                 }
 565                 if (matched) {
 566                         res->f6i = spf6i;
 567                         goto out;
 568                 }
 569         }
 570
 571         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 572                 res->f6i = net->ipv6.fib6_null_entry;
 573                 nh = res->f6i->fib6_nh;
 574                 goto out;
 575         }
 576
 577         if (unlikely(f6i->nh)) {
 578                 nh = nexthop_fib6_nh(f6i->nh);
 579                 if (nexthop_is_blackhole(f6i->nh))
 580                         goto out_blackhole;
 581         } else {
 582                 nh = f6i->fib6_nh;
 583         }
 584
 585         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 586                 res->f6i = net->ipv6.fib6_null_entry;
 587                 nh = res->f6i->fib6_nh;
 588         }
 589 out:
 590         res->nh = nh;
 591         res->fib6_type = res->f6i->fib6_type;
 592         res->fib6_flags = res->f6i->fib6_flags;
 593         return;
 594
 595 out_blackhole:
 596         res->fib6_flags |= RTF_REJECT;
 597         res->fib6_type = RTN_BLACKHOLE;
 598         res->nh = nh;
 599 }
 600
 601 #ifdef CONFIG_IPV6_ROUTER_PREF
 602 struct __rt6_probe_work {
 603         struct work_struct work;
 604         struct in6_addr target;
 605         struct net_device *dev;
 606 };
 607
 608 static void rt6_probe_deferred(struct work_struct *w)
 609 {
 610         struct in6_addr mcaddr;
 611         struct __rt6_probe_work *work =
 612                 container_of(w, struct __rt6_probe_work, work);
 613
 614         addrconf_addr_solict_mult(&work->target, &mcaddr);
 615         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 616         dev_put(work->dev);
 617         kfree(work);
 618 }
 619
 620 static void rt6_probe(struct fib6_nh *fib6_nh)
 621 {
 622         struct __rt6_probe_work *work = NULL;
 623         const struct in6_addr *nh_gw;
 624         unsigned long last_probe;
 625         struct neighbour *neigh;
 626         struct net_device *dev;
 627         struct inet6_dev *idev;
 628
 629         /*
 630          * Okay, this does not seem to be appropriate
 631          * for now, however, we need to check if it
 632          * is really so; aka Router Reachability Probing.
 633          *
 634          * Router Reachability Probe MUST be rate-limited
 635          * to no more than one per minute.
 636          */
 637         if (fib6_nh->fib_nh_gw_family)
 638                 return;
 639
 640         nh_gw = &fib6_nh->fib_nh_gw6;
 641         dev = fib6_nh->fib_nh_dev;
 642         rcu_read_lock_bh();
 643         last_probe = READ_ONCE(fib6_nh->last_probe);
 644         idev = __in6_dev_get(dev);
 645         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 646         if (neigh) {
 647                 if (neigh->nud_state & NUD_VALID)
 648                         goto out;
 649
 650                 write_lock(&neigh->lock);
 651                 if (!(neigh->nud_state & NUD_VALID) &&
 652                     time_after(jiffies,
 653                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 654                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 655                         if (work)
 656                                 __neigh_set_probe_once(neigh);
 657                 }
 658                 write_unlock(&neigh->lock);
 659         } else if (time_after(jiffies, last_probe +
 660                                        idev->cnf.rtr_probe_interval)) {
 661                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 662         }
 663
 664         if (!work || cmpxchg(&fib6_nh->last_probe,
 665                              last_probe, jiffies) != last_probe) {
 666                 kfree(work);
 667         } else {
 668                 INIT_WORK(&work->work, rt6_probe_deferred);
 669                 work->target = *nh_gw;
 670                 dev_hold(dev);
 671                 work->dev = dev;
 672                 schedule_work(&work->work);
 673         }
 674
 675 out:
 676         rcu_read_unlock_bh();
 677 }
 678 #else
 679 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 680 {
 681 }
 682 #endif
 683
 684 /*
 685  * Default Router Selection (RFC 2461 6.3.6)
 686  */
 687 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 688 {
 689         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 690         struct neighbour *neigh;
 691
 692         rcu_read_lock_bh();
 693         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 694                                           &fib6_nh->fib_nh_gw6);
 695         if (neigh) {
 696                 read_lock(&neigh->lock);
 697                 if (neigh->nud_state & NUD_VALID)
 698                         ret = RT6_NUD_SUCCEED;
 699 #ifdef CONFIG_IPV6_ROUTER_PREF
 700                 else if (!(neigh->nud_state & NUD_FAILED))
 701                         ret = RT6_NUD_SUCCEED;
 702                 else
 703                         ret = RT6_NUD_FAIL_PROBE;
 704 #endif
 705                 read_unlock(&neigh->lock);
 706         } else {
 707                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 708                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 709         }
 710         rcu_read_unlock_bh();
 711
 712         return ret;
 713 }
 714
 715 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 716                            int strict)
 717 {
 718         int m = 0;
 719
 720         if (!oif || nh->fib_nh_dev->ifindex == oif)
 721                 m = 2;
 722
 723         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 724                 return RT6_NUD_FAIL_HARD;
 725 #ifdef CONFIG_IPV6_ROUTER_PREF
 726         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 727 #endif
 728         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 729             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 730                 int n = rt6_check_neigh(nh);
 731                 if (n < 0)
 732                         return n;
 733         }
 734         return m;
 735 }
 736
 737 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 738                        int oif, int strict, int *mpri, bool *do_rr)
 739 {
 740         bool match_do_rr = false;
 741         bool rc = false;
 742         int m;
 743
 744         if (nh->fib_nh_flags & RTNH_F_DEAD)
 745                 goto out;
 746
 747         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 748             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 749             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 750                 goto out;
 751
 752         m = rt6_score_route(nh, fib6_flags, oif, strict);
 753         if (m == RT6_NUD_FAIL_DO_RR) {
 754                 match_do_rr = true;
 755                 m = 0; /* lowest valid score */
 756         } else if (m == RT6_NUD_FAIL_HARD) {
 757                 goto out;
 758         }
 759
 760         if (strict & RT6_LOOKUP_F_REACHABLE)
 761                 rt6_probe(nh);
 762
 763         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 764         if (m > *mpri) {
 765                 *do_rr = match_do_rr;
 766                 *mpri = m;
 767                 rc = true;
 768         }
 769 out:
 770         return rc;
 771 }
 772
 773 struct fib6_nh_frl_arg {
 774         u32             flags;
 775         int             oif;
 776         int             strict;
 777         int             *mpri;
 778         bool            *do_rr;
 779         struct fib6_nh  *nh;
 780 };
 781
 782 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 783 {
 784         struct fib6_nh_frl_arg *arg = _arg;
 785
 786         arg->nh = nh;
 787         return find_match(nh, arg->flags, arg->oif, arg->strict,
 788                           arg->mpri, arg->do_rr);
 789 }
 790
 791 static void __find_rr_leaf(struct fib6_info *f6i_start,
 792                            struct fib6_info *nomatch, u32 metric,
 793                            struct fib6_result *res, struct fib6_info **cont,
 794                            int oif, int strict, bool *do_rr, int *mpri)
 795 {
 796         struct fib6_info *f6i;
 797
 798         for (f6i = f6i_start;
 799              f6i && f6i != nomatch;
 800              f6i = rcu_dereference(f6i->fib6_next)) {
 801                 bool matched = false;
 802                 struct fib6_nh *nh;
 803
 804                 if (cont && f6i->fib6_metric != metric) {
 805                         *cont = f6i;
 806                         return;
 807                 }
 808
 809                 if (fib6_check_expired(f6i))
 810                         continue;
 811
 812                 if (unlikely(f6i->nh)) {
 813                         struct fib6_nh_frl_arg arg = {
 814                                 .flags  = f6i->fib6_flags,
 815                                 .oif    = oif,
 816                                 .strict = strict,
 817                                 .mpri   = mpri,
 818                                 .do_rr  = do_rr
 819                         };
 820
 821                         if (nexthop_is_blackhole(f6i->nh)) {
 822                                 res->fib6_flags = RTF_REJECT;
 823                                 res->fib6_type = RTN_BLACKHOLE;
 824                                 res->f6i = f6i;
 825                                 res->nh = nexthop_fib6_nh(f6i->nh);
 826                                 return;
 827                         }
 828                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 829                                                      &arg)) {
 830                                 matched = true;
 831                                 nh = arg.nh;
 832                         }
 833                 } else {
 834                         nh = f6i->fib6_nh;
 835                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 836                                        mpri, do_rr))
 837                                 matched = true;
 838                 }
 839                 if (matched) {
 840                         res->f6i = f6i;
 841                         res->nh = nh;
 842                         res->fib6_flags = f6i->fib6_flags;
 843                         res->fib6_type = f6i->fib6_type;
 844                 }
 845         }
 846 }
 847
 848 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 849                          struct fib6_info *rr_head, int oif, int strict,
 850                          bool *do_rr, struct fib6_result *res)
 851 {
 852         u32 metric = rr_head->fib6_metric;
 853         struct fib6_info *cont = NULL;
 854         int mpri = -1;
 855
 856         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 857                        oif, strict, do_rr, &mpri);
 858
 859         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 860                        oif, strict, do_rr, &mpri);
 861
 862         if (res->f6i || !cont)
 863                 return;
 864
 865         __find_rr_leaf(cont, NULL, metric, res, NULL,
 866                        oif, strict, do_rr, &mpri);
 867 }
 868
 869 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 870                        struct fib6_result *res, int strict)
 871 {
 872         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 873         struct fib6_info *rt0;
 874         bool do_rr = false;
 875         int key_plen;
 876
 877         /* make sure this function or its helpers sets f6i */
 878         res->f6i = NULL;
 879
 880         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 881                 goto out;
 882
 883         rt0 = rcu_dereference(fn->rr_ptr);
 884         if (!rt0)
 885                 rt0 = leaf;
 886
 887         /* Double check to make sure fn is not an intermediate node
 888          * and fn->leaf does not points to its child's leaf
 889          * (This might happen if all routes under fn are deleted from
 890          * the tree and fib6_repair_tree() is called on the node.)
 891          */
 892         key_plen = rt0->fib6_dst.plen;
 893 #ifdef CONFIG_IPV6_SUBTREES
 894         if (rt0->fib6_src.plen)
 895                 key_plen = rt0->fib6_src.plen;
 896 #endif
 897         if (fn->fn_bit != key_plen)
 898                 goto out;
 899
 900         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 901         if (do_rr) {
 902                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 903
 904                 /* no entries matched; do round-robin */
 905                 if (!next || next->fib6_metric != rt0->fib6_metric)
 906                         next = leaf;
 907
 908                 if (next != rt0) {
 909                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 910                         /* make sure next is not being deleted from the tree */
 911                         if (next->fib6_node)
 912                                 rcu_assign_pointer(fn->rr_ptr, next);
 913                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 914                 }
 915         }
 916
 917 out:
 918         if (!res->f6i) {
 919                 res->f6i = net->ipv6.fib6_null_entry;
 920                 res->nh = res->f6i->fib6_nh;
 921                 res->fib6_flags = res->f6i->fib6_flags;
 922                 res->fib6_type = res->f6i->fib6_type;
 923         }
 924 }
 925
 926 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 927 {
 928         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 929                res->nh->fib_nh_gw_family;
 930 }
 931
 932 #ifdef CONFIG_IPV6_ROUTE_INFO
 933 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 934                   const struct in6_addr *gwaddr)
 935 {
 936         struct net *net = dev_net(dev);
 937         struct route_info *rinfo = (struct route_info *) opt;
 938         struct in6_addr prefix_buf, *prefix;
 939         unsigned int pref;
 940         unsigned long lifetime;
 941         struct fib6_info *rt;
 942
 943         if (len < sizeof(struct route_info)) {
 944                 return -EINVAL;
 945         }
 946
 947         /* Sanity check for prefix_len and length */
 948         if (rinfo->length > 3) {
 949                 return -EINVAL;
 950         } else if (rinfo->prefix_len > 128) {
 951                 return -EINVAL;
 952         } else if (rinfo->prefix_len > 64) {
 953                 if (rinfo->length < 2) {
 954                         return -EINVAL;
 955                 }
 956         } else if (rinfo->prefix_len > 0) {
 957                 if (rinfo->length < 1) {
 958                         return -EINVAL;
 959                 }
 960         }
 961
 962         pref = rinfo->route_pref;
 963         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 964                 return -EINVAL;
 965
 966         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 967
 968         if (rinfo->length == 3)
 969                 prefix = (struct in6_addr *)rinfo->prefix;
 970         else {
 971                 /* this function is safe */
 972                 ipv6_addr_prefix(&prefix_buf,
 973                                  (struct in6_addr *)rinfo->prefix,
 974                                  rinfo->prefix_len);
 975                 prefix = &prefix_buf;
 976         }
 977
 978         if (rinfo->prefix_len == 0)
 979                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 980         else
 981                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 982                                         gwaddr, dev);
 983
 984         if (rt && !lifetime) {
 985                 ip6_del_rt(net, rt);
 986                 rt = NULL;
 987         }
 988
 989         if (!rt && lifetime)
 990                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 991                                         dev, pref);
 992         else if (rt)
 993                 rt->fib6_flags = RTF_ROUTEINFO |
 994                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 995
 996         if (rt) {
 997                 if (!addrconf_finite_timeout(lifetime))
 998                         fib6_clean_expires(rt);
 999                 else
1000                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1001
1002                 fib6_info_release(rt);
1003         }
1004         return 0;
1005 }
1006 #endif
1007
1008 /*
1009  *      Misc support functions
1010  */
1011
1012 /* called with rcu_lock held */
1013 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1014 {
1015         struct net_device *dev = res->nh->fib_nh_dev;
1016
1017         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1018                 /* for copies of local routes, dst->dev needs to be the
1019                  * device if it is a master device, the master device if
1020                  * device is enslaved, and the loopback as the default
1021                  */
1022                 if (netif_is_l3_slave(dev) &&
1023                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1024                         dev = l3mdev_master_dev_rcu(dev);
1025                 else if (!netif_is_l3_master(dev))
1026                         dev = dev_net(dev)->loopback_dev;
1027                 /* last case is netif_is_l3_master(dev) is true in which
1028                  * case we want dev returned to be dev
1029                  */
1030         }
1031
1032         return dev;
1033 }
1034
1035 static const int fib6_prop[RTN_MAX + 1] = {
1036         [RTN_UNSPEC]    = 0,
1037         [RTN_UNICAST]   = 0,
1038         [RTN_LOCAL]     = 0,
1039         [RTN_BROADCAST] = 0,
1040         [RTN_ANYCAST]   = 0,
1041         [RTN_MULTICAST] = 0,
1042         [RTN_BLACKHOLE] = -EINVAL,
1043         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1044         [RTN_PROHIBIT]  = -EACCES,
1045         [RTN_THROW]     = -EAGAIN,
1046         [RTN_NAT]       = -EINVAL,
1047         [RTN_XRESOLVE]  = -EINVAL,
1048 };
1049
1050 static int ip6_rt_type_to_error(u8 fib6_type)
1051 {
1052         return fib6_prop[fib6_type];
1053 }
1054
1055 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1056 {
1057         unsigned short flags = 0;
1058
1059         if (rt->dst_nocount)
1060                 flags |= DST_NOCOUNT;
1061         if (rt->dst_nopolicy)
1062                 flags |= DST_NOPOLICY;
1063         if (rt->dst_host)
1064                 flags |= DST_HOST;
1065
1066         return flags;
1067 }
1068
1069 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1070 {
1071         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1072
1073         switch (fib6_type) {
1074         case RTN_BLACKHOLE:
1075                 rt->dst.output = dst_discard_out;
1076                 rt->dst.input = dst_discard;
1077                 break;
1078         case RTN_PROHIBIT:
1079                 rt->dst.output = ip6_pkt_prohibit_out;
1080                 rt->dst.input = ip6_pkt_prohibit;
1081                 break;
1082         case RTN_THROW:
1083         case RTN_UNREACHABLE:
1084         default:
1085                 rt->dst.output = ip6_pkt_discard_out;
1086                 rt->dst.input = ip6_pkt_discard;
1087                 break;
1088         }
1089 }
1090
1091 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1092 {
1093         struct fib6_info *f6i = res->f6i;
1094
1095         if (res->fib6_flags & RTF_REJECT) {
1096                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1097                 return;
1098         }
1099
1100         rt->dst.error = 0;
1101         rt->dst.output = ip6_output;
1102
1103         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1104                 rt->dst.input = ip6_input;
1105         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1106                 rt->dst.input = ip6_mc_input;
1107         } else {
1108                 rt->dst.input = ip6_forward;
1109         }
1110
1111         if (res->nh->fib_nh_lws) {
1112                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1113                 lwtunnel_set_redirect(&rt->dst);
1114         }
1115
1116         rt->dst.lastuse = jiffies;
1117 }
1118
1119 /* Caller must already hold reference to @from */
1120 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1121 {
1122         rt->rt6i_flags &= ~RTF_EXPIRES;
1123         rcu_assign_pointer(rt->from, from);
1124         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1125 }
1126
1127 /* Caller must already hold reference to f6i in result */
1128 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1129 {
1130         const struct fib6_nh *nh = res->nh;
1131         const struct net_device *dev = nh->fib_nh_dev;
1132         struct fib6_info *f6i = res->f6i;
1133
1134         ip6_rt_init_dst(rt, res);
1135
1136         rt->rt6i_dst = f6i->fib6_dst;
1137         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1138         rt->rt6i_flags = res->fib6_flags;
1139         if (nh->fib_nh_gw_family) {
1140                 rt->rt6i_gateway = nh->fib_nh_gw6;
1141                 rt->rt6i_flags |= RTF_GATEWAY;
1142         }
1143         rt6_set_from(rt, f6i);
1144 #ifdef CONFIG_IPV6_SUBTREES
1145         rt->rt6i_src = f6i->fib6_src;
1146 #endif
1147 }
1148
1149 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1150                                         struct in6_addr *saddr)
1151 {
1152         struct fib6_node *pn, *sn;
1153         while (1) {
1154                 if (fn->fn_flags & RTN_TL_ROOT)
1155                         return NULL;
1156                 pn = rcu_dereference(fn->parent);
1157                 sn = FIB6_SUBTREE(pn);
1158                 if (sn && sn != fn)
1159                         fn = fib6_node_lookup(sn, NULL, saddr);
1160                 else
1161                         fn = pn;
1162                 if (fn->fn_flags & RTN_RTINFO)
1163                         return fn;
1164         }
1165 }
1166
1167 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1168 {
1169         struct rt6_info *rt = *prt;
1170
1171         if (dst_hold_safe(&rt->dst))
1172                 return true;
1173         if (net) {
1174                 rt = net->ipv6.ip6_null_entry;
1175                 dst_hold(&rt->dst);
1176         } else {
1177                 rt = NULL;
1178         }
1179         *prt = rt;
1180         return false;
1181 }
1182
1183 /* called with rcu_lock held */
1184 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1185 {
1186         struct net_device *dev = res->nh->fib_nh_dev;
1187         struct fib6_info *f6i = res->f6i;
1188         unsigned short flags;
1189         struct rt6_info *nrt;
1190
1191         if (!fib6_info_hold_safe(f6i))
1192                 goto fallback;
1193
1194         flags = fib6_info_dst_flags(f6i);
1195         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1196         if (!nrt) {
1197                 fib6_info_release(f6i);
1198                 goto fallback;
1199         }
1200
1201         ip6_rt_copy_init(nrt, res);
1202         return nrt;
1203
1204 fallback:
1205         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1206         dst_hold(&nrt->dst);
1207         return nrt;
1208 }
1209
1210 static struct rt6_info *ip6_pol_route_lookup(struct net *net,
1211                                              struct fib6_table *table,
1212                                              struct flowi6 *fl6,
1213                                              const struct sk_buff *skb,
1214                                              int flags)
1215 {
1216         struct fib6_result res = {};
1217         struct fib6_node *fn;
1218         struct rt6_info *rt;
1219
1220         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1221                 flags &= ~RT6_LOOKUP_F_IFACE;
1222
1223         rcu_read_lock();
1224         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1225 restart:
1226         res.f6i = rcu_dereference(fn->leaf);
1227         if (!res.f6i)
1228                 res.f6i = net->ipv6.fib6_null_entry;
1229         else
1230                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1231                                  flags);
1232
1233         if (res.f6i == net->ipv6.fib6_null_entry) {
1234                 fn = fib6_backtrack(fn, &fl6->saddr);
1235                 if (fn)
1236                         goto restart;
1237
1238                 rt = net->ipv6.ip6_null_entry;
1239                 dst_hold(&rt->dst);
1240                 goto out;
1241         } else if (res.fib6_flags & RTF_REJECT) {
1242                 goto do_create;
1243         }
1244
1245         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1246                          fl6->flowi6_oif != 0, skb, flags);
1247
1248         /* Search through exception table */
1249         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1250         if (rt) {
1251                 if (ip6_hold_safe(net, &rt))
1252                         dst_use_noref(&rt->dst, jiffies);
1253         } else {
1254 do_create:
1255                 rt = ip6_create_rt_rcu(&res);
1256         }
1257
1258 out:
1259         trace_fib6_table_lookup(net, &res, table, fl6);
1260
1261         rcu_read_unlock();
1262
1263         return rt;
1264 }
1265
1266 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1267                                    const struct sk_buff *skb, int flags)
1268 {
1269         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1270 }
1271 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1272
1273 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1274                             const struct in6_addr *saddr, int oif,
1275                             const struct sk_buff *skb, int strict)
1276 {
1277         struct flowi6 fl6 = {
1278                 .flowi6_oif = oif,
1279                 .daddr = *daddr,
1280         };
1281         struct dst_entry *dst;
1282         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1283
1284         if (saddr) {
1285                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1286                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1287         }
1288
1289         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1290         if (dst->error == 0)
1291                 return (struct rt6_info *) dst;
1292
1293         dst_release(dst);
1294
1295         return NULL;
1296 }
1297 EXPORT_SYMBOL(rt6_lookup);
1298
1299 /* ip6_ins_rt is called with FREE table->tb6_lock.
1300  * It takes new route entry, the addition fails by any reason the
1301  * route is released.
1302  * Caller must hold dst before calling it.
1303  */
1304
1305 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1306                         struct netlink_ext_ack *extack)
1307 {
1308         int err;
1309         struct fib6_table *table;
1310
1311         table = rt->fib6_table;
1312         spin_lock_bh(&table->tb6_lock);
1313         err = fib6_add(&table->tb6_root, rt, info, extack);
1314         spin_unlock_bh(&table->tb6_lock);
1315
1316         return err;
1317 }
1318
1319 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1320 {
1321         struct nl_info info = { .nl_net = net, };
1322
1323         return __ip6_ins_rt(rt, &info, NULL);
1324 }
1325
1326 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1327                                            const struct in6_addr *daddr,
1328                                            const struct in6_addr *saddr)
1329 {
1330         struct fib6_info *f6i = res->f6i;
1331         struct net_device *dev;
1332         struct rt6_info *rt;
1333
1334         /*
1335          *      Clone the route.
1336          */
1337
1338         if (!fib6_info_hold_safe(f6i))
1339                 return NULL;
1340
1341         dev = ip6_rt_get_dev_rcu(res);
1342         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1343         if (!rt) {
1344                 fib6_info_release(f6i);
1345                 return NULL;
1346         }
1347
1348         ip6_rt_copy_init(rt, res);
1349         rt->rt6i_flags |= RTF_CACHE;
1350         rt->dst.flags |= DST_HOST;
1351         rt->rt6i_dst.addr = *daddr;
1352         rt->rt6i_dst.plen = 128;
1353
1354         if (!rt6_is_gw_or_nonexthop(res)) {
1355                 if (f6i->fib6_dst.plen != 128 &&
1356                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1357                         rt->rt6i_flags |= RTF_ANYCAST;
1358 #ifdef CONFIG_IPV6_SUBTREES
1359                 if (rt->rt6i_src.plen && saddr) {
1360                         rt->rt6i_src.addr = *saddr;
1361                         rt->rt6i_src.plen = 128;
1362                 }
1363 #endif
1364         }
1365
1366         return rt;
1367 }
1368
1369 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1370 {
1371         struct fib6_info *f6i = res->f6i;
1372         unsigned short flags = fib6_info_dst_flags(f6i);
1373         struct net_device *dev;
1374         struct rt6_info *pcpu_rt;
1375
1376         if (!fib6_info_hold_safe(f6i))
1377                 return NULL;
1378
1379         rcu_read_lock();
1380         dev = ip6_rt_get_dev_rcu(res);
1381         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags);
1382         rcu_read_unlock();
1383         if (!pcpu_rt) {
1384                 fib6_info_release(f6i);
1385                 return NULL;
1386         }
1387         ip6_rt_copy_init(pcpu_rt, res);
1388         pcpu_rt->rt6i_flags |= RTF_PCPU;
1389         return pcpu_rt;
1390 }
1391
1392 /* It should be called with rcu_read_lock() acquired */
1393 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1394 {
1395         struct rt6_info *pcpu_rt;
1396
1397         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1398
1399         return pcpu_rt;
1400 }
1401
1402 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1403                                             const struct fib6_result *res)
1404 {
1405         struct rt6_info *pcpu_rt, *prev, **p;
1406
1407         pcpu_rt = ip6_rt_pcpu_alloc(res);
1408         if (!pcpu_rt)
1409                 return NULL;
1410
1411         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1412         prev = cmpxchg(p, NULL, pcpu_rt);
1413         BUG_ON(prev);
1414
1415         if (res->f6i->fib6_destroying) {
1416                 struct fib6_info *from;
1417
1418                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1419                 fib6_info_release(from);
1420         }
1421
1422         return pcpu_rt;
1423 }
1424
1425 /* exception hash table implementation
1426  */
1427 static DEFINE_SPINLOCK(rt6_exception_lock);
1428
1429 /* Remove rt6_ex from hash table and free the memory
1430  * Caller must hold rt6_exception_lock
1431  */
1432 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1433                                  struct rt6_exception *rt6_ex)
1434 {
1435         struct fib6_info *from;
1436         struct net *net;
1437
1438         if (!bucket || !rt6_ex)
1439                 return;
1440
1441         net = dev_net(rt6_ex->rt6i->dst.dev);
1442         net->ipv6.rt6_stats->fib_rt_cache--;
1443
1444         /* purge completely the exception to allow releasing the held resources:
1445          * some [sk] cache may keep the dst around for unlimited time
1446          */
1447         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1448         fib6_info_release(from);
1449         dst_dev_put(&rt6_ex->rt6i->dst);
1450
1451         hlist_del_rcu(&rt6_ex->hlist);
1452         dst_release(&rt6_ex->rt6i->dst);
1453         kfree_rcu(rt6_ex, rcu);
1454         WARN_ON_ONCE(!bucket->depth);
1455         bucket->depth--;
1456 }
1457
1458 /* Remove oldest rt6_ex in bucket and free the memory
1459  * Caller must hold rt6_exception_lock
1460  */
1461 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1462 {
1463         struct rt6_exception *rt6_ex, *oldest = NULL;
1464
1465         if (!bucket)
1466                 return;
1467
1468         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1469                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1470                         oldest = rt6_ex;
1471         }
1472         rt6_remove_exception(bucket, oldest);
1473 }
1474
1475 static u32 rt6_exception_hash(const struct in6_addr *dst,
1476                               const struct in6_addr *src)
1477 {
1478         static u32 seed __read_mostly;
1479         u32 val;
1480
1481         net_get_random_once(&seed, sizeof(seed));
1482         val = jhash(dst, sizeof(*dst), seed);
1483
1484 #ifdef CONFIG_IPV6_SUBTREES
1485         if (src)
1486                 val = jhash(src, sizeof(*src), val);
1487 #endif
1488         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1489 }
1490
1491 /* Helper function to find the cached rt in the hash table
1492  * and update bucket pointer to point to the bucket for this
1493  * (daddr, saddr) pair
1494  * Caller must hold rt6_exception_lock
1495  */
1496 static struct rt6_exception *
1497 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1498                               const struct in6_addr *daddr,
1499                               const struct in6_addr *saddr)
1500 {
1501         struct rt6_exception *rt6_ex;
1502         u32 hval;
1503
1504         if (!(*bucket) || !daddr)
1505                 return NULL;
1506
1507         hval = rt6_exception_hash(daddr, saddr);
1508         *bucket += hval;
1509
1510         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1511                 struct rt6_info *rt6 = rt6_ex->rt6i;
1512                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1513
1514 #ifdef CONFIG_IPV6_SUBTREES
1515                 if (matched && saddr)
1516                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1517 #endif
1518                 if (matched)
1519                         return rt6_ex;
1520         }
1521         return NULL;
1522 }
1523
1524 /* Helper function to find the cached rt in the hash table
1525  * and update bucket pointer to point to the bucket for this
1526  * (daddr, saddr) pair
1527  * Caller must hold rcu_read_lock()
1528  */
1529 static struct rt6_exception *
1530 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1531                          const struct in6_addr *daddr,
1532                          const struct in6_addr *saddr)
1533 {
1534         struct rt6_exception *rt6_ex;
1535         u32 hval;
1536
1537         WARN_ON_ONCE(!rcu_read_lock_held());
1538
1539         if (!(*bucket) || !daddr)
1540                 return NULL;
1541
1542         hval = rt6_exception_hash(daddr, saddr);
1543         *bucket += hval;
1544
1545         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1546                 struct rt6_info *rt6 = rt6_ex->rt6i;
1547                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1548
1549 #ifdef CONFIG_IPV6_SUBTREES
1550                 if (matched && saddr)
1551                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1552 #endif
1553                 if (matched)
1554                         return rt6_ex;
1555         }
1556         return NULL;
1557 }
1558
1559 static unsigned int fib6_mtu(const struct fib6_result *res)
1560 {
1561         const struct fib6_nh *nh = res->nh;
1562         unsigned int mtu;
1563
1564         if (res->f6i->fib6_pmtu) {
1565                 mtu = res->f6i->fib6_pmtu;
1566         } else {
1567                 struct net_device *dev = nh->fib_nh_dev;
1568                 struct inet6_dev *idev;
1569
1570                 rcu_read_lock();
1571                 idev = __in6_dev_get(dev);
1572                 mtu = idev->cnf.mtu6;
1573                 rcu_read_unlock();
1574         }
1575
1576         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1577
1578         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1579 }
1580
1581 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1582
1583 /* used when the flushed bit is not relevant, only access to the bucket
1584  * (ie., all bucket users except rt6_insert_exception);
1585  *
1586  * called under rcu lock; sometimes called with rt6_exception_lock held
1587  */
1588 static
1589 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1590                                                        spinlock_t *lock)
1591 {
1592         struct rt6_exception_bucket *bucket;
1593
1594         if (lock)
1595                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1596                                                    lockdep_is_held(lock));
1597         else
1598                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1599
1600         /* remove bucket flushed bit if set */
1601         if (bucket) {
1602                 unsigned long p = (unsigned long)bucket;
1603
1604                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1605                 bucket = (struct rt6_exception_bucket *)p;
1606         }
1607
1608         return bucket;
1609 }
1610
1611 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1612 {
1613         unsigned long p = (unsigned long)bucket;
1614
1615         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1616 }
1617
1618 /* called with rt6_exception_lock held */
1619 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1620                                               spinlock_t *lock)
1621 {
1622         struct rt6_exception_bucket *bucket;
1623         unsigned long p;
1624
1625         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1626                                            lockdep_is_held(lock));
1627
1628         p = (unsigned long)bucket;
1629         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1630         bucket = (struct rt6_exception_bucket *)p;
1631         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1632 }
1633
1634 static int rt6_insert_exception(struct rt6_info *nrt,
1635                                 const struct fib6_result *res)
1636 {
1637         struct net *net = dev_net(nrt->dst.dev);
1638         struct rt6_exception_bucket *bucket;
1639         struct fib6_info *f6i = res->f6i;
1640         struct in6_addr *src_key = NULL;
1641         struct rt6_exception *rt6_ex;
1642         struct fib6_nh *nh = res->nh;
1643         int err = 0;
1644
1645         spin_lock_bh(&rt6_exception_lock);
1646
1647         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1648                                           lockdep_is_held(&rt6_exception_lock));
1649         if (!bucket) {
1650                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1651                                  GFP_ATOMIC);
1652                 if (!bucket) {
1653                         err = -ENOMEM;
1654                         goto out;
1655                 }
1656                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1657         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1658                 err = -EINVAL;
1659                 goto out;
1660         }
1661
1662 #ifdef CONFIG_IPV6_SUBTREES
1663         /* fib6_src.plen != 0 indicates f6i is in subtree
1664          * and exception table is indexed by a hash of
1665          * both fib6_dst and fib6_src.
1666          * Otherwise, the exception table is indexed by
1667          * a hash of only fib6_dst.
1668          */
1669         if (f6i->fib6_src.plen)
1670                 src_key = &nrt->rt6i_src.addr;
1671 #endif
1672         /* rt6_mtu_change() might lower mtu on f6i.
1673          * Only insert this exception route if its mtu
1674          * is less than f6i's mtu value.
1675          */
1676         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1677                 err = -EINVAL;
1678                 goto out;
1679         }
1680
1681         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1682                                                src_key);
1683         if (rt6_ex)
1684                 rt6_remove_exception(bucket, rt6_ex);
1685
1686         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1687         if (!rt6_ex) {
1688                 err = -ENOMEM;
1689                 goto out;
1690         }
1691         rt6_ex->rt6i = nrt;
1692         rt6_ex->stamp = jiffies;
1693         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1694         bucket->depth++;
1695         net->ipv6.rt6_stats->fib_rt_cache++;
1696
1697         if (bucket->depth > FIB6_MAX_DEPTH)
1698                 rt6_exception_remove_oldest(bucket);
1699
1700 out:
1701         spin_unlock_bh(&rt6_exception_lock);
1702
1703         /* Update fn->fn_sernum to invalidate all cached dst */
1704         if (!err) {
1705                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1706                 fib6_update_sernum(net, f6i);
1707                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1708                 fib6_force_start_gc(net);
1709         }
1710
1711         return err;
1712 }
1713
1714 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1715 {
1716         struct rt6_exception_bucket *bucket;
1717         struct rt6_exception *rt6_ex;
1718         struct hlist_node *tmp;
1719         int i;
1720
1721         spin_lock_bh(&rt6_exception_lock);
1722
1723         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1724         if (!bucket)
1725                 goto out;
1726
1727         /* Prevent rt6_insert_exception() to recreate the bucket list */
1728         if (!from)
1729                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1730
1731         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1732                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1733                         if (!from ||
1734                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1735                                 rt6_remove_exception(bucket, rt6_ex);
1736                 }
1737                 WARN_ON_ONCE(!from && bucket->depth);
1738                 bucket++;
1739         }
1740 out:
1741         spin_unlock_bh(&rt6_exception_lock);
1742 }
1743
1744 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1745 {
1746         struct fib6_info *f6i = arg;
1747
1748         fib6_nh_flush_exceptions(nh, f6i);
1749
1750         return 0;
1751 }
1752
1753 void rt6_flush_exceptions(struct fib6_info *f6i)
1754 {
1755         if (f6i->nh)
1756                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1757                                          f6i);
1758         else
1759                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1760 }
1761
1762 /* Find cached rt in the hash table inside passed in rt
1763  * Caller has to hold rcu_read_lock()
1764  */
1765 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1766                                            const struct in6_addr *daddr,
1767                                            const struct in6_addr *saddr)
1768 {
1769         const struct in6_addr *src_key = NULL;
1770         struct rt6_exception_bucket *bucket;
1771         struct rt6_exception *rt6_ex;
1772         struct rt6_info *ret = NULL;
1773
1774 #ifdef CONFIG_IPV6_SUBTREES
1775         /* fib6i_src.plen != 0 indicates f6i is in subtree
1776          * and exception table is indexed by a hash of
1777          * both fib6_dst and fib6_src.
1778          * However, the src addr used to create the hash
1779          * might not be exactly the passed in saddr which
1780          * is a /128 addr from the flow.
1781          * So we need to use f6i->fib6_src to redo lookup
1782          * if the passed in saddr does not find anything.
1783          * (See the logic in ip6_rt_cache_alloc() on how
1784          * rt->rt6i_src is updated.)
1785          */
1786         if (res->f6i->fib6_src.plen)
1787                 src_key = saddr;
1788 find_ex:
1789 #endif
1790         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1791         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1792
1793         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1794                 ret = rt6_ex->rt6i;
1795
1796 #ifdef CONFIG_IPV6_SUBTREES
1797         /* Use fib6_src as src_key and redo lookup */
1798         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1799                 src_key = &res->f6i->fib6_src.addr;
1800                 goto find_ex;
1801         }
1802 #endif
1803
1804         return ret;
1805 }
1806
1807 /* Remove the passed in cached rt from the hash table that contains it */
1808 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1809                                     const struct rt6_info *rt)
1810 {
1811         const struct in6_addr *src_key = NULL;
1812         struct rt6_exception_bucket *bucket;
1813         struct rt6_exception *rt6_ex;
1814         int err;
1815
1816         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1817                 return -ENOENT;
1818
1819         spin_lock_bh(&rt6_exception_lock);
1820         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1821
1822 #ifdef CONFIG_IPV6_SUBTREES
1823         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1824          * and exception table is indexed by a hash of
1825          * both rt6i_dst and rt6i_src.
1826          * Otherwise, the exception table is indexed by
1827          * a hash of only rt6i_dst.
1828          */
1829         if (plen)
1830                 src_key = &rt->rt6i_src.addr;
1831 #endif
1832         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1833                                                &rt->rt6i_dst.addr,
1834                                                src_key);
1835         if (rt6_ex) {
1836                 rt6_remove_exception(bucket, rt6_ex);
1837                 err = 0;
1838         } else {
1839                 err = -ENOENT;
1840         }
1841
1842         spin_unlock_bh(&rt6_exception_lock);
1843         return err;
1844 }
1845
1846 struct fib6_nh_excptn_arg {
1847         struct rt6_info *rt;
1848         int             plen;
1849 };
1850
1851 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1852 {
1853         struct fib6_nh_excptn_arg *arg = _arg;
1854         int err;
1855
1856         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1857         if (err == 0)
1858                 return 1;
1859
1860         return 0;
1861 }
1862
1863 static int rt6_remove_exception_rt(struct rt6_info *rt)
1864 {
1865         struct fib6_info *from;
1866
1867         from = rcu_dereference(rt->from);
1868         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1869                 return -EINVAL;
1870
1871         if (from->nh) {
1872                 struct fib6_nh_excptn_arg arg = {
1873                         .rt = rt,
1874                         .plen = from->fib6_src.plen
1875                 };
1876                 int rc;
1877
1878                 /* rc = 1 means an entry was found */
1879                 rc = nexthop_for_each_fib6_nh(from->nh,
1880                                               rt6_nh_remove_exception_rt,
1881                                               &arg);
1882                 return rc ? 0 : -ENOENT;
1883         }
1884
1885         return fib6_nh_remove_exception(from->fib6_nh,
1886                                         from->fib6_src.plen, rt);
1887 }
1888
1889 /* Find rt6_ex which contains the passed in rt cache and
1890  * refresh its stamp
1891  */
1892 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1893                                      const struct rt6_info *rt)
1894 {
1895         const struct in6_addr *src_key = NULL;
1896         struct rt6_exception_bucket *bucket;
1897         struct rt6_exception *rt6_ex;
1898
1899         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1900 #ifdef CONFIG_IPV6_SUBTREES
1901         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1902          * and exception table is indexed by a hash of
1903          * both rt6i_dst and rt6i_src.
1904          * Otherwise, the exception table is indexed by
1905          * a hash of only rt6i_dst.
1906          */
1907         if (plen)
1908                 src_key = &rt->rt6i_src.addr;
1909 #endif
1910         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1911         if (rt6_ex)
1912                 rt6_ex->stamp = jiffies;
1913 }
1914
1915 struct fib6_nh_match_arg {
1916         const struct net_device *dev;
1917         const struct in6_addr   *gw;
1918         struct fib6_nh          *match;
1919 };
1920
1921 /* determine if fib6_nh has given device and gateway */
1922 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1923 {
1924         struct fib6_nh_match_arg *arg = _arg;
1925
1926         if (arg->dev != nh->fib_nh_dev ||
1927             (arg->gw && !nh->fib_nh_gw_family) ||
1928             (!arg->gw && nh->fib_nh_gw_family) ||
1929             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1930                 return 0;
1931
1932         arg->match = nh;
1933
1934         /* found a match, break the loop */
1935         return 1;
1936 }
1937
1938 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1939 {
1940         struct fib6_info *from;
1941         struct fib6_nh *fib6_nh;
1942
1943         rcu_read_lock();
1944
1945         from = rcu_dereference(rt->from);
1946         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1947                 goto unlock;
1948
1949         if (from->nh) {
1950                 struct fib6_nh_match_arg arg = {
1951                         .dev = rt->dst.dev,
1952                         .gw = &rt->rt6i_gateway,
1953                 };
1954
1955                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1956
1957                 if (!arg.match)
1958                         goto unlock;
1959                 fib6_nh = arg.match;
1960         } else {
1961                 fib6_nh = from->fib6_nh;
1962         }
1963         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1964 unlock:
1965         rcu_read_unlock();
1966 }
1967
1968 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1969                                          struct rt6_info *rt, int mtu)
1970 {
1971         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1972          * lowest MTU in the path: always allow updating the route PMTU to
1973          * reflect PMTU decreases.
1974          *
1975          * If the new MTU is higher, and the route PMTU is equal to the local
1976          * MTU, this means the old MTU is the lowest in the path, so allow
1977          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1978          * handle this.
1979          */
1980
1981         if (dst_mtu(&rt->dst) >= mtu)
1982                 return true;
1983
1984         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1985                 return true;
1986
1987         return false;
1988 }
1989
1990 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
1991                                        const struct fib6_nh *nh, int mtu)
1992 {
1993         struct rt6_exception_bucket *bucket;
1994         struct rt6_exception *rt6_ex;
1995         int i;
1996
1997         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1998         if (!bucket)
1999                 return;
2000
2001         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2002                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2003                         struct rt6_info *entry = rt6_ex->rt6i;
2004
2005                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2006                          * route), the metrics of its rt->from have already
2007                          * been updated.
2008                          */
2009                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2010                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2011                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2012                 }
2013                 bucket++;
2014         }
2015 }
2016
2017 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2018
2019 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2020                                             const struct in6_addr *gateway)
2021 {
2022         struct rt6_exception_bucket *bucket;
2023         struct rt6_exception *rt6_ex;
2024         struct hlist_node *tmp;
2025         int i;
2026
2027         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2028                 return;
2029
2030         spin_lock_bh(&rt6_exception_lock);
2031         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2032         if (bucket) {
2033                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2034                         hlist_for_each_entry_safe(rt6_ex, tmp,
2035                                                   &bucket->chain, hlist) {
2036                                 struct rt6_info *entry = rt6_ex->rt6i;
2037
2038                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2039                                     RTF_CACHE_GATEWAY &&
2040                                     ipv6_addr_equal(gateway,
2041                                                     &entry->rt6i_gateway)) {
2042                                         rt6_remove_exception(bucket, rt6_ex);
2043                                 }
2044                         }
2045                         bucket++;
2046                 }
2047         }
2048
2049         spin_unlock_bh(&rt6_exception_lock);
2050 }
2051
2052 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2053                                       struct rt6_exception *rt6_ex,
2054                                       struct fib6_gc_args *gc_args,
2055                                       unsigned long now)
2056 {
2057         struct rt6_info *rt = rt6_ex->rt6i;
2058
2059         /* we are pruning and obsoleting aged-out and non gateway exceptions
2060          * even if others have still references to them, so that on next
2061          * dst_check() such references can be dropped.
2062          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2063          * expired, independently from their aging, as per RFC 8201 section 4
2064          */
2065         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2066                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2067                         RT6_TRACE("aging clone %p\n", rt);
2068                         rt6_remove_exception(bucket, rt6_ex);
2069                         return;
2070                 }
2071         } else if (time_after(jiffies, rt->dst.expires)) {
2072                 RT6_TRACE("purging expired route %p\n", rt);
2073                 rt6_remove_exception(bucket, rt6_ex);
2074                 return;
2075         }
2076
2077         if (rt->rt6i_flags & RTF_GATEWAY) {
2078                 struct neighbour *neigh;
2079                 __u8 neigh_flags = 0;
2080
2081                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2082                 if (neigh)
2083                         neigh_flags = neigh->flags;
2084
2085                 if (!(neigh_flags & NTF_ROUTER)) {
2086                         RT6_TRACE("purging route %p via non-router but gateway\n",
2087                                   rt);
2088                         rt6_remove_exception(bucket, rt6_ex);
2089                         return;
2090                 }
2091         }
2092
2093         gc_args->more++;
2094 }
2095
2096 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2097                                    struct fib6_gc_args *gc_args,
2098                                    unsigned long now)
2099 {
2100         struct rt6_exception_bucket *bucket;
2101         struct rt6_exception *rt6_ex;
2102         struct hlist_node *tmp;
2103         int i;
2104
2105         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2106                 return;
2107
2108         rcu_read_lock_bh();
2109         spin_lock(&rt6_exception_lock);
2110         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2111         if (bucket) {
2112                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2113                         hlist_for_each_entry_safe(rt6_ex, tmp,
2114                                                   &bucket->chain, hlist) {
2115                                 rt6_age_examine_exception(bucket, rt6_ex,
2116                                                           gc_args, now);
2117                         }
2118                         bucket++;
2119                 }
2120         }
2121         spin_unlock(&rt6_exception_lock);
2122         rcu_read_unlock_bh();
2123 }
2124
2125 struct fib6_nh_age_excptn_arg {
2126         struct fib6_gc_args     *gc_args;
2127         unsigned long           now;
2128 };
2129
2130 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2131 {
2132         struct fib6_nh_age_excptn_arg *arg = _arg;
2133
2134         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2135         return 0;
2136 }
2137
2138 void rt6_age_exceptions(struct fib6_info *f6i,
2139                         struct fib6_gc_args *gc_args,
2140                         unsigned long now)
2141 {
2142         if (f6i->nh) {
2143                 struct fib6_nh_age_excptn_arg arg = {
2144                         .gc_args = gc_args,
2145                         .now = now
2146                 };
2147
2148                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2149                                          &arg);
2150         } else {
2151                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2152         }
2153 }
2154
2155 /* must be called with rcu lock held */
2156 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2157                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2158 {
2159         struct fib6_node *fn, *saved_fn;
2160
2161         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2162         saved_fn = fn;
2163
2164         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2165                 oif = 0;
2166
2167 redo_rt6_select:
2168         rt6_select(net, fn, oif, res, strict);
2169         if (res->f6i == net->ipv6.fib6_null_entry) {
2170                 fn = fib6_backtrack(fn, &fl6->saddr);
2171                 if (fn)
2172                         goto redo_rt6_select;
2173                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2174                         /* also consider unreachable route */
2175                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2176                         fn = saved_fn;
2177                         goto redo_rt6_select;
2178                 }
2179         }
2180
2181         trace_fib6_table_lookup(net, res, table, fl6);
2182
2183         return 0;
2184 }
2185
2186 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2187                                int oif, struct flowi6 *fl6,
2188                                const struct sk_buff *skb, int flags)
2189 {
2190         struct fib6_result res = {};
2191         struct rt6_info *rt = NULL;
2192         int strict = 0;
2193
2194         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2195                      !rcu_read_lock_held());
2196
2197         strict |= flags & RT6_LOOKUP_F_IFACE;
2198         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2199         if (net->ipv6.devconf_all->forwarding == 0)
2200                 strict |= RT6_LOOKUP_F_REACHABLE;
2201
2202         rcu_read_lock();
2203
2204         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2205         if (res.f6i == net->ipv6.fib6_null_entry)
2206                 goto out;
2207
2208         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2209
2210         /*Search through exception table */
2211         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2212         if (rt) {
2213                 goto out;
2214         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2215                             !res.nh->fib_nh_gw_family)) {
2216                 /* Create a RTF_CACHE clone which will not be
2217                  * owned by the fib6 tree.  It is for the special case where
2218                  * the daddr in the skb during the neighbor look-up is different
2219                  * from the fl6->daddr used to look-up route here.
2220                  */
2221                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2222
2223                 if (rt) {
2224                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2225                          * As rt6_uncached_list_add() does not consume refcnt,
2226                          * this refcnt is always returned to the caller even
2227                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2228                          */
2229                         rt6_uncached_list_add(rt);
2230                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2231                         rcu_read_unlock();
2232
2233                         return rt;
2234                 }
2235         } else {
2236                 /* Get a percpu copy */
2237                 local_bh_disable();
2238                 rt = rt6_get_pcpu_route(&res);
2239
2240                 if (!rt)
2241                         rt = rt6_make_pcpu_route(net, &res);
2242
2243                 local_bh_enable();
2244         }
2245 out:
2246         if (!rt)
2247                 rt = net->ipv6.ip6_null_entry;
2248         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2249                 ip6_hold_safe(net, &rt);
2250         rcu_read_unlock();
2251
2252         return rt;
2253 }
2254 EXPORT_SYMBOL_GPL(ip6_pol_route);
2255
2256 static struct rt6_info *ip6_pol_route_input(struct net *net,
2257                                             struct fib6_table *table,
2258                                             struct flowi6 *fl6,
2259                                             const struct sk_buff *skb,
2260                                             int flags)
2261 {
2262         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2263 }
2264
2265 struct dst_entry *ip6_route_input_lookup(struct net *net,
2266                                          struct net_device *dev,
2267                                          struct flowi6 *fl6,
2268                                          const struct sk_buff *skb,
2269                                          int flags)
2270 {
2271         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2272                 flags |= RT6_LOOKUP_F_IFACE;
2273
2274         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2275 }
2276 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2277
2278 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2279                                   struct flow_keys *keys,
2280                                   struct flow_keys *flkeys)
2281 {
2282         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2283         const struct ipv6hdr *key_iph = outer_iph;
2284         struct flow_keys *_flkeys = flkeys;
2285         const struct ipv6hdr *inner_iph;
2286         const struct icmp6hdr *icmph;
2287         struct ipv6hdr _inner_iph;
2288         struct icmp6hdr _icmph;
2289
2290         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2291                 goto out;
2292
2293         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2294                                    sizeof(_icmph), &_icmph);
2295         if (!icmph)
2296                 goto out;
2297
2298         if (icmph->icmp6_type != ICMPV6_DEST_UNREACH &&
2299             icmph->icmp6_type != ICMPV6_PKT_TOOBIG &&
2300             icmph->icmp6_type != ICMPV6_TIME_EXCEED &&
2301             icmph->icmp6_type != ICMPV6_PARAMPROB)
2302                 goto out;
2303
2304         inner_iph = skb_header_pointer(skb,
2305                                        skb_transport_offset(skb) + sizeof(*icmph),
2306                                        sizeof(_inner_iph), &_inner_iph);
2307         if (!inner_iph)
2308                 goto out;
2309
2310         key_iph = inner_iph;
2311         _flkeys = NULL;
2312 out:
2313         if (_flkeys) {
2314                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2315                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2316                 keys->tags.flow_label = _flkeys->tags.flow_label;
2317                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2318         } else {
2319                 keys->addrs.v6addrs.src = key_iph->saddr;
2320                 keys->addrs.v6addrs.dst = key_iph->daddr;
2321                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2322                 keys->basic.ip_proto = key_iph->nexthdr;
2323         }
2324 }
2325
2326 /* if skb is set it will be used and fl6 can be NULL */
2327 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2328                        const struct sk_buff *skb, struct flow_keys *flkeys)
2329 {
2330         struct flow_keys hash_keys;
2331         u32 mhash;
2332
2333         switch (ip6_multipath_hash_policy(net)) {
2334         case 0:
2335                 memset(&hash_keys, 0, sizeof(hash_keys));
2336                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2337                 if (skb) {
2338                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2339                 } else {
2340                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2341                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2342                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2343                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2344                 }
2345                 break;
2346         case 1:
2347                 if (skb) {
2348                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2349                         struct flow_keys keys;
2350
2351                         /* short-circuit if we already have L4 hash present */
2352                         if (skb->l4_hash)
2353                                 return skb_get_hash_raw(skb) >> 1;
2354
2355                         memset(&hash_keys, 0, sizeof(hash_keys));
2356
2357                         if (!flkeys) {
2358                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2359                                 flkeys = &keys;
2360                         }
2361                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2362                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2363                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2364                         hash_keys.ports.src = flkeys->ports.src;
2365                         hash_keys.ports.dst = flkeys->ports.dst;
2366                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2367                 } else {
2368                         memset(&hash_keys, 0, sizeof(hash_keys));
2369                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2370                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2371                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2372                         hash_keys.ports.src = fl6->fl6_sport;
2373                         hash_keys.ports.dst = fl6->fl6_dport;
2374                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2375                 }
2376                 break;
2377         case 2:
2378                 memset(&hash_keys, 0, sizeof(hash_keys));
2379                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2380                 if (skb) {
2381                         struct flow_keys keys;
2382
2383                         if (!flkeys) {
2384                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2385                                 flkeys = &keys;
2386                         }
2387
2388                         /* Inner can be v4 or v6 */
2389                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2390                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2391                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2392                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2393                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2394                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2395                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2396                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2397                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2398                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2399                         } else {
2400                                 /* Same as case 0 */
2401                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2402                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2403                         }
2404                 } else {
2405                         /* Same as case 0 */
2406                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2407                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2408                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2409                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2410                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2411                 }
2412                 break;
2413         }
2414         mhash = flow_hash_from_keys(&hash_keys);
2415
2416         return mhash >> 1;
2417 }
2418
2419 /* Called with rcu held */
2420 void ip6_route_input(struct sk_buff *skb)
2421 {
2422         const struct ipv6hdr *iph = ipv6_hdr(skb);
2423         struct net *net = dev_net(skb->dev);
2424         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2425         struct ip_tunnel_info *tun_info;
2426         struct flowi6 fl6 = {
2427                 .flowi6_iif = skb->dev->ifindex,
2428                 .daddr = iph->daddr,
2429                 .saddr = iph->saddr,
2430                 .flowlabel = ip6_flowinfo(iph),
2431                 .flowi6_mark = skb->mark,
2432                 .flowi6_proto = iph->nexthdr,
2433         };
2434         struct flow_keys *flkeys = NULL, _flkeys;
2435
2436         tun_info = skb_tunnel_info(skb);
2437         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2438                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2439
2440         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2441                 flkeys = &_flkeys;
2442
2443         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2444                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2445         skb_dst_drop(skb);
2446         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2447                                                       &fl6, skb, flags));
2448 }
2449
2450 static struct rt6_info *ip6_pol_route_output(struct net *net,
2451                                              struct fib6_table *table,
2452                                              struct flowi6 *fl6,
2453                                              const struct sk_buff *skb,
2454                                              int flags)
2455 {
2456         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2457 }
2458
2459 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2460                                                const struct sock *sk,
2461                                                struct flowi6 *fl6, int flags)
2462 {
2463         bool any_src;
2464
2465         if (ipv6_addr_type(&fl6->daddr) &
2466             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2467                 struct dst_entry *dst;
2468
2469                 /* This function does not take refcnt on the dst */
2470                 dst = l3mdev_link_scope_lookup(net, fl6);
2471                 if (dst)
2472                         return dst;
2473         }
2474
2475         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2476
2477         flags |= RT6_LOOKUP_F_DST_NOREF;
2478         any_src = ipv6_addr_any(&fl6->saddr);
2479         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2480             (fl6->flowi6_oif && any_src))
2481                 flags |= RT6_LOOKUP_F_IFACE;
2482
2483         if (!any_src)
2484                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2485         else if (sk)
2486                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2487
2488         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2489 }
2490 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2491
2492 struct dst_entry *ip6_route_output_flags(struct net *net,
2493                                          const struct sock *sk,
2494                                          struct flowi6 *fl6,
2495                                          int flags)
2496 {
2497         struct dst_entry *dst;
2498         struct rt6_info *rt6;
2499
2500         rcu_read_lock();
2501         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2502         rt6 = (struct rt6_info *)dst;
2503         /* For dst cached in uncached_list, refcnt is already taken. */
2504         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2505                 dst = &net->ipv6.ip6_null_entry->dst;
2506                 dst_hold(dst);
2507         }
2508         rcu_read_unlock();
2509
2510         return dst;
2511 }
2512 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2513
2514 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2515 {
2516         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2517         struct net_device *loopback_dev = net->loopback_dev;
2518         struct dst_entry *new = NULL;
2519
2520         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2521                        DST_OBSOLETE_DEAD, 0);
2522         if (rt) {
2523                 rt6_info_init(rt);
2524                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2525
2526                 new = &rt->dst;
2527                 new->__use = 1;
2528                 new->input = dst_discard;
2529                 new->output = dst_discard_out;
2530
2531                 dst_copy_metrics(new, &ort->dst);
2532
2533                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2534                 rt->rt6i_gateway = ort->rt6i_gateway;
2535                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2536
2537                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2538 #ifdef CONFIG_IPV6_SUBTREES
2539                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2540 #endif
2541         }
2542
2543         dst_release(dst_orig);
2544         return new ? new : ERR_PTR(-ENOMEM);
2545 }
2546
2547 /*
2548  *      Destination cache support functions
2549  */
2550
2551 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2552 {
2553         u32 rt_cookie = 0;
2554
2555         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2556                 return false;
2557
2558         if (fib6_check_expired(f6i))
2559                 return false;
2560
2561         return true;
2562 }
2563
2564 static struct dst_entry *rt6_check(struct rt6_info *rt,
2565                                    struct fib6_info *from,
2566                                    u32 cookie)
2567 {
2568         u32 rt_cookie = 0;
2569
2570         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2571             rt_cookie != cookie)
2572                 return NULL;
2573
2574         if (rt6_check_expired(rt))
2575                 return NULL;
2576
2577         return &rt->dst;
2578 }
2579
2580 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2581                                             struct fib6_info *from,
2582                                             u32 cookie)
2583 {
2584         if (!__rt6_check_expired(rt) &&
2585             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2586             fib6_check(from, cookie))
2587                 return &rt->dst;
2588         else
2589                 return NULL;
2590 }
2591
2592 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2593 {
2594         struct dst_entry *dst_ret;
2595         struct fib6_info *from;
2596         struct rt6_info *rt;
2597
2598         rt = container_of(dst, struct rt6_info, dst);
2599
2600         rcu_read_lock();
2601
2602         /* All IPV6 dsts are created with ->obsolete set to the value
2603          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2604          * into this function always.
2605          */
2606
2607         from = rcu_dereference(rt->from);
2608
2609         if (from && (rt->rt6i_flags & RTF_PCPU ||
2610             unlikely(!list_empty(&rt->rt6i_uncached))))
2611                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2612         else
2613                 dst_ret = rt6_check(rt, from, cookie);
2614
2615         rcu_read_unlock();
2616
2617         return dst_ret;
2618 }
2619
2620 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2621 {
2622         struct rt6_info *rt = (struct rt6_info *) dst;
2623
2624         if (rt) {
2625                 if (rt->rt6i_flags & RTF_CACHE) {
2626                         rcu_read_lock();
2627                         if (rt6_check_expired(rt)) {
2628                                 rt6_remove_exception_rt(rt);
2629                                 dst = NULL;
2630                         }
2631                         rcu_read_unlock();
2632                 } else {
2633                         dst_release(dst);
2634                         dst = NULL;
2635                 }
2636         }
2637         return dst;
2638 }
2639
2640 static void ip6_link_failure(struct sk_buff *skb)
2641 {
2642         struct rt6_info *rt;
2643
2644         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2645
2646         rt = (struct rt6_info *) skb_dst(skb);
2647         if (rt) {
2648                 rcu_read_lock();
2649                 if (rt->rt6i_flags & RTF_CACHE) {
2650                         rt6_remove_exception_rt(rt);
2651                 } else {
2652                         struct fib6_info *from;
2653                         struct fib6_node *fn;
2654
2655                         from = rcu_dereference(rt->from);
2656                         if (from) {
2657                                 fn = rcu_dereference(from->fib6_node);
2658                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2659                                         fn->fn_sernum = -1;
2660                         }
2661                 }
2662                 rcu_read_unlock();
2663         }
2664 }
2665
2666 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2667 {
2668         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2669                 struct fib6_info *from;
2670
2671                 rcu_read_lock();
2672                 from = rcu_dereference(rt0->from);
2673                 if (from)
2674                         rt0->dst.expires = from->expires;
2675                 rcu_read_unlock();
2676         }
2677
2678         dst_set_expires(&rt0->dst, timeout);
2679         rt0->rt6i_flags |= RTF_EXPIRES;
2680 }
2681
2682 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2683 {
2684         struct net *net = dev_net(rt->dst.dev);
2685
2686         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2687         rt->rt6i_flags |= RTF_MODIFIED;
2688         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2689 }
2690
2691 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2692 {
2693         return !(rt->rt6i_flags & RTF_CACHE) &&
2694                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2695 }
2696
2697 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2698                                  const struct ipv6hdr *iph, u32 mtu)
2699 {
2700         const struct in6_addr *daddr, *saddr;
2701         struct rt6_info *rt6 = (struct rt6_info *)dst;
2702
2703         if (dst_metric_locked(dst, RTAX_MTU))
2704                 return;
2705
2706         if (iph) {
2707                 daddr = &iph->daddr;
2708                 saddr = &iph->saddr;
2709         } else if (sk) {
2710                 daddr = &sk->sk_v6_daddr;
2711                 saddr = &inet6_sk(sk)->saddr;
2712         } else {
2713                 daddr = NULL;
2714                 saddr = NULL;
2715         }
2716         dst_confirm_neigh(dst, daddr);
2717         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2718         if (mtu >= dst_mtu(dst))
2719                 return;
2720
2721         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2722                 rt6_do_update_pmtu(rt6, mtu);
2723                 /* update rt6_ex->stamp for cache */
2724                 if (rt6->rt6i_flags & RTF_CACHE)
2725                         rt6_update_exception_stamp_rt(rt6);
2726         } else if (daddr) {
2727                 struct fib6_result res = {};
2728                 struct rt6_info *nrt6;
2729
2730                 rcu_read_lock();
2731                 res.f6i = rcu_dereference(rt6->from);
2732                 if (!res.f6i)
2733                         goto out_unlock;
2734
2735                 res.fib6_flags = res.f6i->fib6_flags;
2736                 res.fib6_type = res.f6i->fib6_type;
2737
2738                 if (res.f6i->nh) {
2739                         struct fib6_nh_match_arg arg = {
2740                                 .dev = dst->dev,
2741                                 .gw = &rt6->rt6i_gateway,
2742                         };
2743
2744                         nexthop_for_each_fib6_nh(res.f6i->nh,
2745                                                  fib6_nh_find_match, &arg);
2746
2747                         /* fib6_info uses a nexthop that does not have fib6_nh
2748                          * using the dst->dev + gw. Should be impossible.
2749                          */
2750                         if (!arg.match)
2751                                 goto out_unlock;
2752
2753                         res.nh = arg.match;
2754                 } else {
2755                         res.nh = res.f6i->fib6_nh;
2756                 }
2757
2758                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2759                 if (nrt6) {
2760                         rt6_do_update_pmtu(nrt6, mtu);
2761                         if (rt6_insert_exception(nrt6, &res))
2762                                 dst_release_immediate(&nrt6->dst);
2763                 }
2764 out_unlock:
2765                 rcu_read_unlock();
2766         }
2767 }
2768
2769 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2770                                struct sk_buff *skb, u32 mtu)
2771 {
2772         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu);
2773 }
2774
2775 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2776                      int oif, u32 mark, kuid_t uid)
2777 {
2778         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2779         struct dst_entry *dst;
2780         struct flowi6 fl6 = {
2781                 .flowi6_oif = oif,
2782                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2783                 .daddr = iph->daddr,
2784                 .saddr = iph->saddr,
2785                 .flowlabel = ip6_flowinfo(iph),
2786                 .flowi6_uid = uid,
2787         };
2788
2789         dst = ip6_route_output(net, NULL, &fl6);
2790         if (!dst->error)
2791                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu));
2792         dst_release(dst);
2793 }
2794 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2795
2796 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2797 {
2798         int oif = sk->sk_bound_dev_if;
2799         struct dst_entry *dst;
2800
2801         if (!oif && skb->dev)
2802                 oif = l3mdev_master_ifindex(skb->dev);
2803
2804         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2805
2806         dst = __sk_dst_get(sk);
2807         if (!dst || !dst->obsolete ||
2808             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2809                 return;
2810
2811         bh_lock_sock(sk);
2812         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2813                 ip6_datagram_dst_update(sk, false);
2814         bh_unlock_sock(sk);
2815 }
2816 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2817
2818 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2819                            const struct flowi6 *fl6)
2820 {
2821 #ifdef CONFIG_IPV6_SUBTREES
2822         struct ipv6_pinfo *np = inet6_sk(sk);
2823 #endif
2824
2825         ip6_dst_store(sk, dst,
2826                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2827                       &sk->sk_v6_daddr : NULL,
2828 #ifdef CONFIG_IPV6_SUBTREES
2829                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2830                       &np->saddr :
2831 #endif
2832                       NULL);
2833 }
2834
2835 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2836                                   struct flowi6 *fl6,
2837                                   const struct in6_addr *gw,
2838                                   struct rt6_info **ret)
2839 {
2840         const struct fib6_nh *nh = res->nh;
2841
2842         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2843             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2844                 return false;
2845
2846         /* rt_cache's gateway might be different from its 'parent'
2847          * in the case of an ip redirect.
2848          * So we keep searching in the exception table if the gateway
2849          * is different.
2850          */
2851         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2852                 struct rt6_info *rt_cache;
2853
2854                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2855                 if (rt_cache &&
2856                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2857                         *ret = rt_cache;
2858                         return true;
2859                 }
2860                 return false;
2861         }
2862         return true;
2863 }
2864
2865 struct fib6_nh_rd_arg {
2866         struct fib6_result      *res;
2867         struct flowi6           *fl6;
2868         const struct in6_addr   *gw;
2869         struct rt6_info         **ret;
2870 };
2871
2872 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2873 {
2874         struct fib6_nh_rd_arg *arg = _arg;
2875
2876         arg->res->nh = nh;
2877         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2878 }
2879
2880 /* Handle redirects */
2881 struct ip6rd_flowi {
2882         struct flowi6 fl6;
2883         struct in6_addr gateway;
2884 };
2885
2886 static struct rt6_info *__ip6_route_redirect(struct net *net,
2887                                              struct fib6_table *table,
2888                                              struct flowi6 *fl6,
2889                                              const struct sk_buff *skb,
2890                                              int flags)
2891 {
2892         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2893         struct rt6_info *ret = NULL;
2894         struct fib6_result res = {};
2895         struct fib6_nh_rd_arg arg = {
2896                 .res = &res,
2897                 .fl6 = fl6,
2898                 .gw  = &rdfl->gateway,
2899                 .ret = &ret
2900         };
2901         struct fib6_info *rt;
2902         struct fib6_node *fn;
2903
2904         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2905          * this case we must match on the real ingress device, so reset it
2906          */
2907         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2908                 fl6->flowi6_oif = skb->dev->ifindex;
2909
2910         /* Get the "current" route for this destination and
2911          * check if the redirect has come from appropriate router.
2912          *
2913          * RFC 4861 specifies that redirects should only be
2914          * accepted if they come from the nexthop to the target.
2915          * Due to the way the routes are chosen, this notion
2916          * is a bit fuzzy and one might need to check all possible
2917          * routes.
2918          */
2919
2920         rcu_read_lock();
2921         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2922 restart:
2923         for_each_fib6_node_rt_rcu(fn) {
2924                 res.f6i = rt;
2925                 if (fib6_check_expired(rt))
2926                         continue;
2927                 if (rt->fib6_flags & RTF_REJECT)
2928                         break;
2929                 if (unlikely(rt->nh)) {
2930                         if (nexthop_is_blackhole(rt->nh))
2931                                 continue;
2932                         /* on match, res->nh is filled in and potentially ret */
2933                         if (nexthop_for_each_fib6_nh(rt->nh,
2934                                                      fib6_nh_redirect_match,
2935                                                      &arg))
2936                                 goto out;
2937                 } else {
2938                         res.nh = rt->fib6_nh;
2939                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2940                                                   &ret))
2941                                 goto out;
2942                 }
2943         }
2944
2945         if (!rt)
2946                 rt = net->ipv6.fib6_null_entry;
2947         else if (rt->fib6_flags & RTF_REJECT) {
2948                 ret = net->ipv6.ip6_null_entry;
2949                 goto out;
2950         }
2951
2952         if (rt == net->ipv6.fib6_null_entry) {
2953                 fn = fib6_backtrack(fn, &fl6->saddr);
2954                 if (fn)
2955                         goto restart;
2956         }
2957
2958         res.f6i = rt;
2959         res.nh = rt->fib6_nh;
2960 out:
2961         if (ret) {
2962                 ip6_hold_safe(net, &ret);
2963         } else {
2964                 res.fib6_flags = res.f6i->fib6_flags;
2965                 res.fib6_type = res.f6i->fib6_type;
2966                 ret = ip6_create_rt_rcu(&res);
2967         }
2968
2969         rcu_read_unlock();
2970
2971         trace_fib6_table_lookup(net, &res, table, fl6);
2972         return ret;
2973 };
2974
2975 static struct dst_entry *ip6_route_redirect(struct net *net,
2976                                             const struct flowi6 *fl6,
2977                                             const struct sk_buff *skb,
2978                                             const struct in6_addr *gateway)
2979 {
2980         int flags = RT6_LOOKUP_F_HAS_SADDR;
2981         struct ip6rd_flowi rdfl;
2982
2983         rdfl.fl6 = *fl6;
2984         rdfl.gateway = *gateway;
2985
2986         return fib6_rule_lookup(net, &rdfl.fl6, skb,
2987                                 flags, __ip6_route_redirect);
2988 }
2989
2990 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
2991                   kuid_t uid)
2992 {
2993         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2994         struct dst_entry *dst;
2995         struct flowi6 fl6 = {
2996                 .flowi6_iif = LOOPBACK_IFINDEX,
2997                 .flowi6_oif = oif,
2998                 .flowi6_mark = mark,
2999                 .daddr = iph->daddr,
3000                 .saddr = iph->saddr,
3001                 .flowlabel = ip6_flowinfo(iph),
3002                 .flowi6_uid = uid,
3003         };
3004
3005         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3006         rt6_do_redirect(dst, NULL, skb);
3007         dst_release(dst);
3008 }
3009 EXPORT_SYMBOL_GPL(ip6_redirect);
3010
3011 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3012 {
3013         const struct ipv6hdr *iph = ipv6_hdr(skb);
3014         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3015         struct dst_entry *dst;
3016         struct flowi6 fl6 = {
3017                 .flowi6_iif = LOOPBACK_IFINDEX,
3018                 .flowi6_oif = oif,
3019                 .daddr = msg->dest,
3020                 .saddr = iph->daddr,
3021                 .flowi6_uid = sock_net_uid(net, NULL),
3022         };
3023
3024         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3025         rt6_do_redirect(dst, NULL, skb);
3026         dst_release(dst);
3027 }
3028
3029 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3030 {
3031         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3032                      sk->sk_uid);
3033 }
3034 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3035
3036 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3037 {
3038         struct net_device *dev = dst->dev;
3039         unsigned int mtu = dst_mtu(dst);
3040         struct net *net = dev_net(dev);
3041
3042         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3043
3044         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3045                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3046
3047         /*
3048          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3049          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3050          * IPV6_MAXPLEN is also valid and means: "any MSS,
3051          * rely only on pmtu discovery"
3052          */
3053         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3054                 mtu = IPV6_MAXPLEN;
3055         return mtu;
3056 }
3057
3058 static unsigned int ip6_mtu(const struct dst_entry *dst)
3059 {
3060         struct inet6_dev *idev;
3061         unsigned int mtu;
3062
3063         mtu = dst_metric_raw(dst, RTAX_MTU);
3064         if (mtu)
3065                 goto out;
3066
3067         mtu = IPV6_MIN_MTU;
3068
3069         rcu_read_lock();
3070         idev = __in6_dev_get(dst->dev);
3071         if (idev)
3072                 mtu = idev->cnf.mtu6;
3073         rcu_read_unlock();
3074
3075 out:
3076         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3077
3078         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3079 }
3080
3081 /* MTU selection:
3082  * 1. mtu on route is locked - use it
3083  * 2. mtu from nexthop exception
3084  * 3. mtu from egress device
3085  *
3086  * based on ip6_dst_mtu_forward and exception logic of
3087  * rt6_find_cached_rt; called with rcu_read_lock
3088  */
3089 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3090                       const struct in6_addr *daddr,
3091                       const struct in6_addr *saddr)
3092 {
3093         const struct fib6_nh *nh = res->nh;
3094         struct fib6_info *f6i = res->f6i;
3095         struct inet6_dev *idev;
3096         struct rt6_info *rt;
3097         u32 mtu = 0;
3098
3099         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3100                 mtu = f6i->fib6_pmtu;
3101                 if (mtu)
3102                         goto out;
3103         }
3104
3105         rt = rt6_find_cached_rt(res, daddr, saddr);
3106         if (unlikely(rt)) {
3107                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3108         } else {
3109                 struct net_device *dev = nh->fib_nh_dev;
3110
3111                 mtu = IPV6_MIN_MTU;
3112                 idev = __in6_dev_get(dev);
3113                 if (idev && idev->cnf.mtu6 > mtu)
3114                         mtu = idev->cnf.mtu6;
3115         }
3116
3117         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3118 out:
3119         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3120 }
3121
3122 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3123                                   struct flowi6 *fl6)
3124 {
3125         struct dst_entry *dst;
3126         struct rt6_info *rt;
3127         struct inet6_dev *idev = in6_dev_get(dev);
3128         struct net *net = dev_net(dev);
3129
3130         if (unlikely(!idev))
3131                 return ERR_PTR(-ENODEV);
3132
3133         rt = ip6_dst_alloc(net, dev, 0);
3134         if (unlikely(!rt)) {
3135                 in6_dev_put(idev);
3136                 dst = ERR_PTR(-ENOMEM);
3137                 goto out;
3138         }
3139
3140         rt->dst.flags |= DST_HOST;
3141         rt->dst.input = ip6_input;
3142         rt->dst.output  = ip6_output;
3143         rt->rt6i_gateway  = fl6->daddr;
3144         rt->rt6i_dst.addr = fl6->daddr;
3145         rt->rt6i_dst.plen = 128;
3146         rt->rt6i_idev     = idev;
3147         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3148
3149         /* Add this dst into uncached_list so that rt6_disable_ip() can
3150          * do proper release of the net_device
3151          */
3152         rt6_uncached_list_add(rt);
3153         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3154
3155         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3156
3157 out:
3158         return dst;
3159 }
3160
3161 static int ip6_dst_gc(struct dst_ops *ops)
3162 {
3163         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3164         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3165         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3166         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3167         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3168         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3169         int entries;
3170
3171         entries = dst_entries_get_fast(ops);
3172         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3173             entries <= rt_max_size)
3174                 goto out;
3175
3176         net->ipv6.ip6_rt_gc_expire++;
3177         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3178         entries = dst_entries_get_slow(ops);
3179         if (entries < ops->gc_thresh)
3180                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3181 out:
3182         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3183         return entries > rt_max_size;
3184 }
3185
3186 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3187                                const struct in6_addr *gw_addr, u32 tbid,
3188                                int flags, struct fib6_result *res)
3189 {
3190         struct flowi6 fl6 = {
3191                 .flowi6_oif = cfg->fc_ifindex,
3192                 .daddr = *gw_addr,
3193                 .saddr = cfg->fc_prefsrc,
3194         };
3195         struct fib6_table *table;
3196         int err;
3197
3198         table = fib6_get_table(net, tbid);
3199         if (!table)
3200                 return -EINVAL;
3201
3202         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3203                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3204
3205         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3206
3207         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3208         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3209                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3210                                  cfg->fc_ifindex != 0, NULL, flags);
3211
3212         return err;
3213 }
3214
3215 static int ip6_route_check_nh_onlink(struct net *net,
3216                                      struct fib6_config *cfg,
3217                                      const struct net_device *dev,
3218                                      struct netlink_ext_ack *extack)
3219 {
3220         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3221         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3222         struct fib6_result res = {};
3223         int err;
3224
3225         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3226         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3227             /* ignore match if it is the default route */
3228             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3229             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3230                 NL_SET_ERR_MSG(extack,
3231                                "Nexthop has invalid gateway or device mismatch");
3232                 err = -EINVAL;
3233         }
3234
3235         return err;
3236 }
3237
3238 static int ip6_route_check_nh(struct net *net,
3239                               struct fib6_config *cfg,
3240                               struct net_device **_dev,
3241                               struct inet6_dev **idev)
3242 {
3243         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3244         struct net_device *dev = _dev ? *_dev : NULL;
3245         int flags = RT6_LOOKUP_F_IFACE;
3246         struct fib6_result res = {};
3247         int err = -EHOSTUNREACH;
3248
3249         if (cfg->fc_table) {
3250                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3251                                           cfg->fc_table, flags, &res);
3252                 /* gw_addr can not require a gateway or resolve to a reject
3253                  * route. If a device is given, it must match the result.
3254                  */
3255                 if (err || res.fib6_flags & RTF_REJECT ||
3256                     res.nh->fib_nh_gw_family ||
3257                     (dev && dev != res.nh->fib_nh_dev))
3258                         err = -EHOSTUNREACH;
3259         }
3260
3261         if (err < 0) {
3262                 struct flowi6 fl6 = {
3263                         .flowi6_oif = cfg->fc_ifindex,
3264                         .daddr = *gw_addr,
3265                 };
3266
3267                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3268                 if (err || res.fib6_flags & RTF_REJECT ||
3269                     res.nh->fib_nh_gw_family)
3270                         err = -EHOSTUNREACH;
3271
3272                 if (err)
3273                         return err;
3274
3275                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3276                                  cfg->fc_ifindex != 0, NULL, flags);
3277         }
3278
3279         err = 0;
3280         if (dev) {
3281                 if (dev != res.nh->fib_nh_dev)
3282                         err = -EHOSTUNREACH;
3283         } else {
3284                 *_dev = dev = res.nh->fib_nh_dev;
3285                 dev_hold(dev);
3286                 *idev = in6_dev_get(dev);
3287         }
3288
3289         return err;
3290 }
3291
3292 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3293                            struct net_device **_dev, struct inet6_dev **idev,
3294                            struct netlink_ext_ack *extack)
3295 {
3296         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3297         int gwa_type = ipv6_addr_type(gw_addr);
3298         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3299         const struct net_device *dev = *_dev;
3300         bool need_addr_check = !dev;
3301         int err = -EINVAL;
3302
3303         /* if gw_addr is local we will fail to detect this in case
3304          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3305          * will return already-added prefix route via interface that
3306          * prefix route was assigned to, which might be non-loopback.
3307          */
3308         if (dev &&
3309             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3310                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3311                 goto out;
3312         }
3313
3314         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3315                 /* IPv6 strictly inhibits using not link-local
3316                  * addresses as nexthop address.
3317                  * Otherwise, router will not able to send redirects.
3318                  * It is very good, but in some (rare!) circumstances
3319                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3320                  * some exceptions. --ANK
3321                  * We allow IPv4-mapped nexthops to support RFC4798-type
3322                  * addressing
3323                  */
3324                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3325                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3326                         goto out;
3327                 }
3328
3329                 rcu_read_lock();
3330
3331                 if (cfg->fc_flags & RTNH_F_ONLINK)
3332                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3333                 else
3334                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3335
3336                 rcu_read_unlock();
3337
3338                 if (err)
3339                         goto out;
3340         }
3341
3342         /* reload in case device was changed */
3343         dev = *_dev;
3344
3345         err = -EINVAL;
3346         if (!dev) {
3347                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3348                 goto out;
3349         } else if (dev->flags & IFF_LOOPBACK) {
3350                 NL_SET_ERR_MSG(extack,
3351                                "Egress device can not be loopback device for this route");
3352                 goto out;
3353         }
3354
3355         /* if we did not check gw_addr above, do so now that the
3356          * egress device has been resolved.
3357          */
3358         if (need_addr_check &&
3359             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3360                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3361                 goto out;
3362         }
3363
3364         err = 0;
3365 out:
3366         return err;
3367 }
3368
3369 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3370 {
3371         if ((flags & RTF_REJECT) ||
3372             (dev && (dev->flags & IFF_LOOPBACK) &&
3373              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3374              !(flags & RTF_LOCAL)))
3375                 return true;
3376
3377         return false;
3378 }
3379
3380 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3381                  struct fib6_config *cfg, gfp_t gfp_flags,
3382                  struct netlink_ext_ack *extack)
3383 {
3384         struct net_device *dev = NULL;
3385         struct inet6_dev *idev = NULL;
3386         int addr_type;
3387         int err;
3388
3389         fib6_nh->fib_nh_family = AF_INET6;
3390 #ifdef CONFIG_IPV6_ROUTER_PREF
3391         fib6_nh->last_probe = jiffies;
3392 #endif
3393
3394         err = -ENODEV;
3395         if (cfg->fc_ifindex) {
3396                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3397                 if (!dev)
3398                         goto out;
3399                 idev = in6_dev_get(dev);
3400                 if (!idev)
3401                         goto out;
3402         }
3403
3404         if (cfg->fc_flags & RTNH_F_ONLINK) {
3405                 if (!dev) {
3406                         NL_SET_ERR_MSG(extack,
3407                                        "Nexthop device required for onlink");
3408                         goto out;
3409                 }
3410
3411                 if (!(dev->flags & IFF_UP)) {
3412                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3413                         err = -ENETDOWN;
3414                         goto out;
3415                 }
3416
3417                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3418         }
3419
3420         fib6_nh->fib_nh_weight = 1;
3421
3422         /* We cannot add true routes via loopback here,
3423          * they would result in kernel looping; promote them to reject routes
3424          */
3425         addr_type = ipv6_addr_type(&cfg->fc_dst);
3426         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3427                 /* hold loopback dev/idev if we haven't done so. */
3428                 if (dev != net->loopback_dev) {
3429                         if (dev) {
3430                                 dev_put(dev);
3431                                 in6_dev_put(idev);
3432                         }
3433                         dev = net->loopback_dev;
3434                         dev_hold(dev);
3435                         idev = in6_dev_get(dev);
3436                         if (!idev) {
3437                                 err = -ENODEV;
3438                                 goto out;
3439                         }
3440                 }
3441                 goto pcpu_alloc;
3442         }
3443
3444         if (cfg->fc_flags & RTF_GATEWAY) {
3445                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3446                 if (err)
3447                         goto out;
3448
3449                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3450                 fib6_nh->fib_nh_gw_family = AF_INET6;
3451         }
3452
3453         err = -ENODEV;
3454         if (!dev)
3455                 goto out;
3456
3457         if (idev->cnf.disable_ipv6) {
3458                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3459                 err = -EACCES;
3460                 goto out;
3461         }
3462
3463         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3464                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3465                 err = -ENETDOWN;
3466                 goto out;
3467         }
3468
3469         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3470             !netif_carrier_ok(dev))
3471                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3472
3473         err = fib_nh_common_init(&fib6_nh->nh_common, cfg->fc_encap,
3474                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3475         if (err)
3476                 goto out;
3477
3478 pcpu_alloc:
3479         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3480         if (!fib6_nh->rt6i_pcpu) {
3481                 err = -ENOMEM;
3482                 goto out;
3483         }
3484
3485         fib6_nh->fib_nh_dev = dev;
3486         fib6_nh->fib_nh_oif = dev->ifindex;
3487         err = 0;
3488 out:
3489         if (idev)
3490                 in6_dev_put(idev);
3491
3492         if (err) {
3493                 lwtstate_put(fib6_nh->fib_nh_lws);
3494                 fib6_nh->fib_nh_lws = NULL;
3495                 if (dev)
3496                         dev_put(dev);
3497         }
3498
3499         return err;
3500 }
3501
3502 void fib6_nh_release(struct fib6_nh *fib6_nh)
3503 {
3504         struct rt6_exception_bucket *bucket;
3505
3506         rcu_read_lock();
3507
3508         fib6_nh_flush_exceptions(fib6_nh, NULL);
3509         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3510         if (bucket) {
3511                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3512                 kfree(bucket);
3513         }
3514
3515         rcu_read_unlock();
3516
3517         if (fib6_nh->rt6i_pcpu) {
3518                 int cpu;
3519
3520                 for_each_possible_cpu(cpu) {
3521                         struct rt6_info **ppcpu_rt;
3522                         struct rt6_info *pcpu_rt;
3523
3524                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3525                         pcpu_rt = *ppcpu_rt;
3526                         if (pcpu_rt) {
3527                                 dst_dev_put(&pcpu_rt->dst);
3528                                 dst_release(&pcpu_rt->dst);
3529                                 *ppcpu_rt = NULL;
3530                         }
3531                 }
3532
3533                 free_percpu(fib6_nh->rt6i_pcpu);
3534         }
3535
3536         fib_nh_common_release(&fib6_nh->nh_common);
3537 }
3538
3539 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3540                                               gfp_t gfp_flags,
3541                                               struct netlink_ext_ack *extack)
3542 {
3543         struct net *net = cfg->fc_nlinfo.nl_net;
3544         struct fib6_info *rt = NULL;
3545         struct nexthop *nh = NULL;
3546         struct fib6_table *table;
3547         struct fib6_nh *fib6_nh;
3548         int err = -EINVAL;
3549         int addr_type;
3550
3551         /* RTF_PCPU is an internal flag; can not be set by userspace */
3552         if (cfg->fc_flags & RTF_PCPU) {
3553                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3554                 goto out;
3555         }
3556
3557         /* RTF_CACHE is an internal flag; can not be set by userspace */
3558         if (cfg->fc_flags & RTF_CACHE) {
3559                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3560                 goto out;
3561         }
3562
3563         if (cfg->fc_type > RTN_MAX) {
3564                 NL_SET_ERR_MSG(extack, "Invalid route type");
3565                 goto out;
3566         }
3567
3568         if (cfg->fc_dst_len > 128) {
3569                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3570                 goto out;
3571         }
3572         if (cfg->fc_src_len > 128) {
3573                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3574                 goto out;
3575         }
3576 #ifndef CONFIG_IPV6_SUBTREES
3577         if (cfg->fc_src_len) {
3578                 NL_SET_ERR_MSG(extack,
3579                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3580                 goto out;
3581         }
3582 #endif
3583         if (cfg->fc_nh_id) {
3584                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3585                 if (!nh) {
3586                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3587                         goto out;
3588                 }
3589                 err = fib6_check_nexthop(nh, cfg, extack);
3590                 if (err)
3591                         goto out;
3592         }
3593
3594         err = -ENOBUFS;
3595         if (cfg->fc_nlinfo.nlh &&
3596             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3597                 table = fib6_get_table(net, cfg->fc_table);
3598                 if (!table) {
3599                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3600                         table = fib6_new_table(net, cfg->fc_table);
3601                 }
3602         } else {
3603                 table = fib6_new_table(net, cfg->fc_table);
3604         }
3605
3606         if (!table)
3607                 goto out;
3608
3609         err = -ENOMEM;
3610         rt = fib6_info_alloc(gfp_flags, !nh);
3611         if (!rt)
3612                 goto out;
3613
3614         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3615                                                extack);
3616         if (IS_ERR(rt->fib6_metrics)) {
3617                 err = PTR_ERR(rt->fib6_metrics);
3618                 /* Do not leave garbage there. */
3619                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3620                 goto out;
3621         }
3622
3623         if (cfg->fc_flags & RTF_ADDRCONF)
3624                 rt->dst_nocount = true;
3625
3626         if (cfg->fc_flags & RTF_EXPIRES)
3627                 fib6_set_expires(rt, jiffies +
3628                                 clock_t_to_jiffies(cfg->fc_expires));
3629         else
3630                 fib6_clean_expires(rt);
3631
3632         if (cfg->fc_protocol == RTPROT_UNSPEC)
3633                 cfg->fc_protocol = RTPROT_BOOT;
3634         rt->fib6_protocol = cfg->fc_protocol;
3635
3636         rt->fib6_table = table;
3637         rt->fib6_metric = cfg->fc_metric;
3638         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3639         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3640
3641         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3642         rt->fib6_dst.plen = cfg->fc_dst_len;
3643         if (rt->fib6_dst.plen == 128)
3644                 rt->dst_host = true;
3645
3646 #ifdef CONFIG_IPV6_SUBTREES
3647         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3648         rt->fib6_src.plen = cfg->fc_src_len;
3649 #endif
3650         if (nh) {
3651                 if (!nexthop_get(nh)) {
3652                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3653                         goto out;
3654                 }
3655                 if (rt->fib6_src.plen) {
3656                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3657                         goto out;
3658                 }
3659                 rt->nh = nh;
3660                 fib6_nh = nexthop_fib6_nh(rt->nh);
3661         } else {
3662                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3663                 if (err)
3664                         goto out;
3665
3666                 fib6_nh = rt->fib6_nh;
3667
3668                 /* We cannot add true routes via loopback here, they would
3669                  * result in kernel looping; promote them to reject routes
3670                  */
3671                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3672                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3673                                    addr_type))
3674                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3675         }
3676
3677         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3678                 struct net_device *dev = fib6_nh->fib_nh_dev;
3679
3680                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3681                         NL_SET_ERR_MSG(extack, "Invalid source address");
3682                         err = -EINVAL;
3683                         goto out;
3684                 }
3685                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3686                 rt->fib6_prefsrc.plen = 128;
3687         } else
3688                 rt->fib6_prefsrc.plen = 0;
3689
3690         return rt;
3691 out:
3692         fib6_info_release(rt);
3693         return ERR_PTR(err);
3694 }
3695
3696 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3697                   struct netlink_ext_ack *extack)
3698 {
3699         struct fib6_info *rt;
3700         int err;
3701
3702         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3703         if (IS_ERR(rt))
3704                 return PTR_ERR(rt);
3705
3706         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3707         fib6_info_release(rt);
3708
3709         return err;
3710 }
3711
3712 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3713 {
3714         struct net *net = info->nl_net;
3715         struct fib6_table *table;
3716         int err;
3717
3718         if (rt == net->ipv6.fib6_null_entry) {
3719                 err = -ENOENT;
3720                 goto out;
3721         }
3722
3723         table = rt->fib6_table;
3724         spin_lock_bh(&table->tb6_lock);
3725         err = fib6_del(rt, info);
3726         spin_unlock_bh(&table->tb6_lock);
3727
3728 out:
3729         fib6_info_release(rt);
3730         return err;
3731 }
3732
3733 int ip6_del_rt(struct net *net, struct fib6_info *rt)
3734 {
3735         struct nl_info info = { .nl_net = net };
3736
3737         return __ip6_del_rt(rt, &info);
3738 }
3739
3740 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3741 {
3742         struct nl_info *info = &cfg->fc_nlinfo;
3743         struct net *net = info->nl_net;
3744         struct sk_buff *skb = NULL;
3745         struct fib6_table *table;
3746         int err = -ENOENT;
3747
3748         if (rt == net->ipv6.fib6_null_entry)
3749                 goto out_put;
3750         table = rt->fib6_table;
3751         spin_lock_bh(&table->tb6_lock);
3752
3753         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3754                 struct fib6_info *sibling, *next_sibling;
3755
3756                 /* prefer to send a single notification with all hops */
3757                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3758                 if (skb) {
3759                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3760
3761                         if (rt6_fill_node(net, skb, rt, NULL,
3762                                           NULL, NULL, 0, RTM_DELROUTE,
3763                                           info->portid, seq, 0) < 0) {
3764                                 kfree_skb(skb);
3765                                 skb = NULL;
3766                         } else
3767                                 info->skip_notify = 1;
3768                 }
3769
3770                 info->skip_notify_kernel = 1;
3771                 call_fib6_multipath_entry_notifiers(net,
3772                                                     FIB_EVENT_ENTRY_DEL,
3773                                                     rt,
3774                                                     rt->fib6_nsiblings,
3775                                                     NULL);
3776                 list_for_each_entry_safe(sibling, next_sibling,
3777                                          &rt->fib6_siblings,
3778                                          fib6_siblings) {
3779                         err = fib6_del(sibling, info);
3780                         if (err)
3781                                 goto out_unlock;
3782                 }
3783         }
3784
3785         err = fib6_del(rt, info);
3786 out_unlock:
3787         spin_unlock_bh(&table->tb6_lock);
3788 out_put:
3789         fib6_info_release(rt);
3790
3791         if (skb) {
3792                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3793                             info->nlh, gfp_any());
3794         }
3795         return err;
3796 }
3797
3798 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3799 {
3800         int rc = -ESRCH;
3801
3802         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3803                 goto out;
3804
3805         if (cfg->fc_flags & RTF_GATEWAY &&
3806             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3807                 goto out;
3808
3809         rc = rt6_remove_exception_rt(rt);
3810 out:
3811         return rc;
3812 }
3813
3814 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3815                              struct fib6_nh *nh)
3816 {
3817         struct fib6_result res = {
3818                 .f6i = rt,
3819                 .nh = nh,
3820         };
3821         struct rt6_info *rt_cache;
3822
3823         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3824         if (rt_cache)
3825                 return __ip6_del_cached_rt(rt_cache, cfg);
3826
3827         return 0;
3828 }
3829
3830 struct fib6_nh_del_cached_rt_arg {
3831         struct fib6_config *cfg;
3832         struct fib6_info *f6i;
3833 };
3834
3835 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3836 {
3837         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3838         int rc;
3839
3840         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3841         return rc != -ESRCH ? rc : 0;
3842 }
3843
3844 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3845 {
3846         struct fib6_nh_del_cached_rt_arg arg = {
3847                 .cfg = cfg,
3848                 .f6i = f6i
3849         };
3850
3851         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3852 }
3853
3854 static int ip6_route_del(struct fib6_config *cfg,
3855                          struct netlink_ext_ack *extack)
3856 {
3857         struct fib6_table *table;
3858         struct fib6_info *rt;
3859         struct fib6_node *fn;
3860         int err = -ESRCH;
3861
3862         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3863         if (!table) {
3864                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3865                 return err;
3866         }
3867
3868         rcu_read_lock();
3869
3870         fn = fib6_locate(&table->tb6_root,
3871                          &cfg->fc_dst, cfg->fc_dst_len,
3872                          &cfg->fc_src, cfg->fc_src_len,
3873                          !(cfg->fc_flags & RTF_CACHE));
3874
3875         if (fn) {
3876                 for_each_fib6_node_rt_rcu(fn) {
3877                         struct fib6_nh *nh;
3878
3879                         if (rt->nh && cfg->fc_nh_id &&
3880                             rt->nh->id != cfg->fc_nh_id)
3881                                 continue;
3882
3883                         if (cfg->fc_flags & RTF_CACHE) {
3884                                 int rc = 0;
3885
3886                                 if (rt->nh) {
3887                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3888                                 } else if (cfg->fc_nh_id) {
3889                                         continue;
3890                                 } else {
3891                                         nh = rt->fib6_nh;
3892                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3893                                 }
3894                                 if (rc != -ESRCH) {
3895                                         rcu_read_unlock();
3896                                         return rc;
3897                                 }
3898                                 continue;
3899                         }
3900
3901                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3902                                 continue;
3903                         if (cfg->fc_protocol &&
3904                             cfg->fc_protocol != rt->fib6_protocol)
3905                                 continue;
3906
3907                         if (rt->nh) {
3908                                 if (!fib6_info_hold_safe(rt))
3909                                         continue;
3910                                 rcu_read_unlock();
3911
3912                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3913                         }
3914                         if (cfg->fc_nh_id)
3915                                 continue;
3916
3917                         nh = rt->fib6_nh;
3918                         if (cfg->fc_ifindex &&
3919                             (!nh->fib_nh_dev ||
3920                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3921                                 continue;
3922                         if (cfg->fc_flags & RTF_GATEWAY &&
3923                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3924                                 continue;
3925                         if (!fib6_info_hold_safe(rt))
3926                                 continue;
3927                         rcu_read_unlock();
3928
3929                         /* if gateway was specified only delete the one hop */
3930                         if (cfg->fc_flags & RTF_GATEWAY)
3931                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3932
3933                         return __ip6_del_rt_siblings(rt, cfg);
3934                 }
3935         }
3936         rcu_read_unlock();
3937
3938         return err;
3939 }
3940
3941 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3942 {
3943         struct netevent_redirect netevent;
3944         struct rt6_info *rt, *nrt = NULL;
3945         struct fib6_result res = {};
3946         struct ndisc_options ndopts;
3947         struct inet6_dev *in6_dev;
3948         struct neighbour *neigh;
3949         struct rd_msg *msg;
3950         int optlen, on_link;
3951         u8 *lladdr;
3952
3953         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
3954         optlen -= sizeof(*msg);
3955
3956         if (optlen < 0) {
3957                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
3958                 return;
3959         }
3960
3961         msg = (struct rd_msg *)icmp6_hdr(skb);
3962
3963         if (ipv6_addr_is_multicast(&msg->dest)) {
3964                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
3965                 return;
3966         }
3967
3968         on_link = 0;
3969         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
3970                 on_link = 1;
3971         } else if (ipv6_addr_type(&msg->target) !=
3972                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
3973                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
3974                 return;
3975         }
3976
3977         in6_dev = __in6_dev_get(skb->dev);
3978         if (!in6_dev)
3979                 return;
3980         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
3981                 return;
3982
3983         /* RFC2461 8.1:
3984          *      The IP source address of the Redirect MUST be the same as the current
3985          *      first-hop router for the specified ICMP Destination Address.
3986          */
3987
3988         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
3989                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
3990                 return;
3991         }
3992
3993         lladdr = NULL;
3994         if (ndopts.nd_opts_tgt_lladdr) {
3995                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
3996                                              skb->dev);
3997                 if (!lladdr) {
3998                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
3999                         return;
4000                 }
4001         }
4002
4003         rt = (struct rt6_info *) dst;
4004         if (rt->rt6i_flags & RTF_REJECT) {
4005                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4006                 return;
4007         }
4008
4009         /* Redirect received -> path was valid.
4010          * Look, redirects are sent only in response to data packets,
4011          * so that this nexthop apparently is reachable. --ANK
4012          */
4013         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4014
4015         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4016         if (!neigh)
4017                 return;
4018
4019         /*
4020          *      We have finally decided to accept it.
4021          */
4022
4023         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4024                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4025                      NEIGH_UPDATE_F_OVERRIDE|
4026                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4027                                      NEIGH_UPDATE_F_ISROUTER)),
4028                      NDISC_REDIRECT, &ndopts);
4029
4030         rcu_read_lock();
4031         res.f6i = rcu_dereference(rt->from);
4032         if (!res.f6i)
4033                 goto out;
4034
4035         if (res.f6i->nh) {
4036                 struct fib6_nh_match_arg arg = {
4037                         .dev = dst->dev,
4038                         .gw = &rt->rt6i_gateway,
4039                 };
4040
4041                 nexthop_for_each_fib6_nh(res.f6i->nh,
4042                                          fib6_nh_find_match, &arg);
4043
4044                 /* fib6_info uses a nexthop that does not have fib6_nh
4045                  * using the dst->dev. Should be impossible
4046                  */
4047                 if (!arg.match)
4048                         goto out;
4049                 res.nh = arg.match;
4050         } else {
4051                 res.nh = res.f6i->fib6_nh;
4052         }
4053
4054         res.fib6_flags = res.f6i->fib6_flags;
4055         res.fib6_type = res.f6i->fib6_type;
4056         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4057         if (!nrt)
4058                 goto out;
4059
4060         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4061         if (on_link)
4062                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4063
4064         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4065
4066         /* rt6_insert_exception() will take care of duplicated exceptions */
4067         if (rt6_insert_exception(nrt, &res)) {
4068                 dst_release_immediate(&nrt->dst);
4069                 goto out;
4070         }
4071
4072         netevent.old = &rt->dst;
4073         netevent.new = &nrt->dst;
4074         netevent.daddr = &msg->dest;
4075         netevent.neigh = neigh;
4076         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4077
4078 out:
4079         rcu_read_unlock();
4080         neigh_release(neigh);
4081 }
4082
4083 #ifdef CONFIG_IPV6_ROUTE_INFO
4084 static struct fib6_info *rt6_get_route_info(struct net *net,
4085                                            const struct in6_addr *prefix, int prefixlen,
4086                                            const struct in6_addr *gwaddr,
4087                                            struct net_device *dev)
4088 {
4089         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4090         int ifindex = dev->ifindex;
4091         struct fib6_node *fn;
4092         struct fib6_info *rt = NULL;
4093         struct fib6_table *table;
4094
4095         table = fib6_get_table(net, tb_id);
4096         if (!table)
4097                 return NULL;
4098
4099         rcu_read_lock();
4100         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4101         if (!fn)
4102                 goto out;
4103
4104         for_each_fib6_node_rt_rcu(fn) {
4105                 /* these routes do not use nexthops */
4106                 if (rt->nh)
4107                         continue;
4108                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4109                         continue;
4110                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4111                     !rt->fib6_nh->fib_nh_gw_family)
4112                         continue;
4113                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4114                         continue;
4115                 if (!fib6_info_hold_safe(rt))
4116                         continue;
4117                 break;
4118         }
4119 out:
4120         rcu_read_unlock();
4121         return rt;
4122 }
4123
4124 static struct fib6_info *rt6_add_route_info(struct net *net,
4125                                            const struct in6_addr *prefix, int prefixlen,
4126                                            const struct in6_addr *gwaddr,
4127                                            struct net_device *dev,
4128                                            unsigned int pref)
4129 {
4130         struct fib6_config cfg = {
4131                 .fc_metric      = IP6_RT_PRIO_USER,
4132                 .fc_ifindex     = dev->ifindex,
4133                 .fc_dst_len     = prefixlen,
4134                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4135                                   RTF_UP | RTF_PREF(pref),
4136                 .fc_protocol = RTPROT_RA,
4137                 .fc_type = RTN_UNICAST,
4138                 .fc_nlinfo.portid = 0,
4139                 .fc_nlinfo.nlh = NULL,
4140                 .fc_nlinfo.nl_net = net,
4141         };
4142
4143         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
4144         cfg.fc_dst = *prefix;
4145         cfg.fc_gateway = *gwaddr;
4146
4147         /* We should treat it as a default route if prefix length is 0. */
4148         if (!prefixlen)
4149                 cfg.fc_flags |= RTF_DEFAULT;
4150
4151         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4152
4153         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4154 }
4155 #endif
4156
4157 struct fib6_info *rt6_get_dflt_router(struct net *net,
4158                                      const struct in6_addr *addr,
4159                                      struct net_device *dev)
4160 {
4161         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4162         struct fib6_info *rt;
4163         struct fib6_table *table;
4164
4165         table = fib6_get_table(net, tb_id);
4166         if (!table)
4167                 return NULL;
4168
4169         rcu_read_lock();
4170         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4171                 struct fib6_nh *nh;
4172
4173                 /* RA routes do not use nexthops */
4174                 if (rt->nh)
4175                         continue;
4176
4177                 nh = rt->fib6_nh;
4178                 if (dev == nh->fib_nh_dev &&
4179                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4180                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4181                         break;
4182         }
4183         if (rt && !fib6_info_hold_safe(rt))
4184                 rt = NULL;
4185         rcu_read_unlock();
4186         return rt;
4187 }
4188
4189 struct fib6_info *rt6_add_dflt_router(struct net *net,
4190                                      const struct in6_addr *gwaddr,
4191                                      struct net_device *dev,
4192                                      unsigned int pref)
4193 {
4194         struct fib6_config cfg = {
4195                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4196                 .fc_metric      = IP6_RT_PRIO_USER,
4197                 .fc_ifindex     = dev->ifindex,
4198                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4199                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4200                 .fc_protocol = RTPROT_RA,
4201                 .fc_type = RTN_UNICAST,
4202                 .fc_nlinfo.portid = 0,
4203                 .fc_nlinfo.nlh = NULL,
4204                 .fc_nlinfo.nl_net = net,
4205         };
4206
4207         cfg.fc_gateway = *gwaddr;
4208
4209         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4210                 struct fib6_table *table;
4211
4212                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4213                 if (table)
4214                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4215         }
4216
4217         return rt6_get_dflt_router(net, gwaddr, dev);
4218 }
4219
4220 static void __rt6_purge_dflt_routers(struct net *net,
4221                                      struct fib6_table *table)
4222 {
4223         struct fib6_info *rt;
4224
4225 restart:
4226         rcu_read_lock();
4227         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4228                 struct net_device *dev = fib6_info_nh_dev(rt);
4229                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4230
4231                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4232                     (!idev || idev->cnf.accept_ra != 2) &&
4233                     fib6_info_hold_safe(rt)) {
4234                         rcu_read_unlock();
4235                         ip6_del_rt(net, rt);
4236                         goto restart;
4237                 }
4238         }
4239         rcu_read_unlock();
4240
4241         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4242 }
4243
4244 void rt6_purge_dflt_routers(struct net *net)
4245 {
4246         struct fib6_table *table;
4247         struct hlist_head *head;
4248         unsigned int h;
4249
4250         rcu_read_lock();
4251
4252         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4253                 head = &net->ipv6.fib_table_hash[h];
4254                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4255                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4256                                 __rt6_purge_dflt_routers(net, table);
4257                 }
4258         }
4259
4260         rcu_read_unlock();
4261 }
4262
4263 static void rtmsg_to_fib6_config(struct net *net,
4264                                  struct in6_rtmsg *rtmsg,
4265                                  struct fib6_config *cfg)
4266 {
4267         *cfg = (struct fib6_config){
4268                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4269                          : RT6_TABLE_MAIN,
4270                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4271                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4272                 .fc_expires = rtmsg->rtmsg_info,
4273                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4274                 .fc_src_len = rtmsg->rtmsg_src_len,
4275                 .fc_flags = rtmsg->rtmsg_flags,
4276                 .fc_type = rtmsg->rtmsg_type,
4277
4278                 .fc_nlinfo.nl_net = net,
4279
4280                 .fc_dst = rtmsg->rtmsg_dst,
4281                 .fc_src = rtmsg->rtmsg_src,
4282                 .fc_gateway = rtmsg->rtmsg_gateway,
4283         };
4284 }
4285
4286 int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4287 {
4288         struct fib6_config cfg;
4289         struct in6_rtmsg rtmsg;
4290         int err;
4291
4292         switch (cmd) {
4293         case SIOCADDRT:         /* Add a route */
4294         case SIOCDELRT:         /* Delete a route */
4295                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4296                         return -EPERM;
4297                 err = copy_from_user(&rtmsg, arg,
4298                                      sizeof(struct in6_rtmsg));
4299                 if (err)
4300                         return -EFAULT;
4301
4302                 rtmsg_to_fib6_config(net, &rtmsg, &cfg);
4303
4304                 rtnl_lock();
4305                 switch (cmd) {
4306                 case SIOCADDRT:
4307                         err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4308                         break;
4309                 case SIOCDELRT:
4310                         err = ip6_route_del(&cfg, NULL);
4311                         break;
4312                 default:
4313                         err = -EINVAL;
4314                 }
4315                 rtnl_unlock();
4316
4317                 return err;
4318         }
4319
4320         return -EINVAL;
4321 }
4322
4323 /*
4324  *      Drop the packet on the floor
4325  */
4326
4327 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4328 {
4329         struct dst_entry *dst = skb_dst(skb);
4330         struct net *net = dev_net(dst->dev);
4331         struct inet6_dev *idev;
4332         int type;
4333
4334         if (netif_is_l3_master(skb->dev) &&
4335             dst->dev == net->loopback_dev)
4336                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4337         else
4338                 idev = ip6_dst_idev(dst);
4339
4340         switch (ipstats_mib_noroutes) {
4341         case IPSTATS_MIB_INNOROUTES:
4342                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4343                 if (type == IPV6_ADDR_ANY) {
4344                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4345                         break;
4346                 }
4347                 /* FALLTHROUGH */
4348         case IPSTATS_MIB_OUTNOROUTES:
4349                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4350                 break;
4351         }
4352
4353         /* Start over by dropping the dst for l3mdev case */
4354         if (netif_is_l3_master(skb->dev))
4355                 skb_dst_drop(skb);
4356
4357         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4358         kfree_skb(skb);
4359         return 0;
4360 }
4361
4362 static int ip6_pkt_discard(struct sk_buff *skb)
4363 {
4364         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4365 }
4366
4367 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4368 {
4369         skb->dev = skb_dst(skb)->dev;
4370         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4371 }
4372
4373 static int ip6_pkt_prohibit(struct sk_buff *skb)
4374 {
4375         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4376 }
4377
4378 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4379 {
4380         skb->dev = skb_dst(skb)->dev;
4381         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4382 }
4383
4384 /*
4385  *      Allocate a dst for local (unicast / anycast) address.
4386  */
4387
4388 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4389                                      struct inet6_dev *idev,
4390                                      const struct in6_addr *addr,
4391                                      bool anycast, gfp_t gfp_flags)
4392 {
4393         struct fib6_config cfg = {
4394                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4395                 .fc_ifindex = idev->dev->ifindex,
4396                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4397                 .fc_dst = *addr,
4398                 .fc_dst_len = 128,
4399                 .fc_protocol = RTPROT_KERNEL,
4400                 .fc_nlinfo.nl_net = net,
4401                 .fc_ignore_dev_down = true,
4402         };
4403         struct fib6_info *f6i;
4404
4405         if (anycast) {
4406                 cfg.fc_type = RTN_ANYCAST;
4407                 cfg.fc_flags |= RTF_ANYCAST;
4408         } else {
4409                 cfg.fc_type = RTN_LOCAL;
4410                 cfg.fc_flags |= RTF_LOCAL;
4411         }
4412
4413         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4414         if (!IS_ERR(f6i))
4415                 f6i->dst_nocount = true;
4416         return f6i;
4417 }
4418
4419 /* remove deleted ip from prefsrc entries */
4420 struct arg_dev_net_ip {
4421         struct net_device *dev;
4422         struct net *net;
4423         struct in6_addr *addr;
4424 };
4425
4426 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4427 {
4428         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4429         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4430         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4431
4432         if (!rt->nh &&
4433             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4434             rt != net->ipv6.fib6_null_entry &&
4435             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4436                 spin_lock_bh(&rt6_exception_lock);
4437                 /* remove prefsrc entry */
4438                 rt->fib6_prefsrc.plen = 0;
4439                 spin_unlock_bh(&rt6_exception_lock);
4440         }
4441         return 0;
4442 }
4443
4444 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4445 {
4446         struct net *net = dev_net(ifp->idev->dev);
4447         struct arg_dev_net_ip adni = {
4448                 .dev = ifp->idev->dev,
4449                 .net = net,
4450                 .addr = &ifp->addr,
4451         };
4452         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4453 }
4454
4455 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4456
4457 /* Remove routers and update dst entries when gateway turn into host. */
4458 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4459 {
4460         struct in6_addr *gateway = (struct in6_addr *)arg;
4461         struct fib6_nh *nh;
4462
4463         /* RA routes do not use nexthops */
4464         if (rt->nh)
4465                 return 0;
4466
4467         nh = rt->fib6_nh;
4468         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4469             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4470                 return -1;
4471
4472         /* Further clean up cached routes in exception table.
4473          * This is needed because cached route may have a different
4474          * gateway than its 'parent' in the case of an ip redirect.
4475          */
4476         fib6_nh_exceptions_clean_tohost(nh, gateway);
4477
4478         return 0;
4479 }
4480
4481 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4482 {
4483         fib6_clean_all(net, fib6_clean_tohost, gateway);
4484 }
4485
4486 struct arg_netdev_event {
4487         const struct net_device *dev;
4488         union {
4489                 unsigned char nh_flags;
4490                 unsigned long event;
4491         };
4492 };
4493
4494 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4495 {
4496         struct fib6_info *iter;
4497         struct fib6_node *fn;
4498
4499         fn = rcu_dereference_protected(rt->fib6_node,
4500                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4501         iter = rcu_dereference_protected(fn->leaf,
4502                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4503         while (iter) {
4504                 if (iter->fib6_metric == rt->fib6_metric &&
4505                     rt6_qualify_for_ecmp(iter))
4506                         return iter;
4507                 iter = rcu_dereference_protected(iter->fib6_next,
4508                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4509         }
4510
4511         return NULL;
4512 }
4513
4514 /* only called for fib entries with builtin fib6_nh */
4515 static bool rt6_is_dead(const struct fib6_info *rt)
4516 {
4517         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4518             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4519              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4520                 return true;
4521
4522         return false;
4523 }
4524
4525 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4526 {
4527         struct fib6_info *iter;
4528         int total = 0;
4529
4530         if (!rt6_is_dead(rt))
4531                 total += rt->fib6_nh->fib_nh_weight;
4532
4533         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4534                 if (!rt6_is_dead(iter))
4535                         total += iter->fib6_nh->fib_nh_weight;
4536         }
4537
4538         return total;
4539 }
4540
4541 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4542 {
4543         int upper_bound = -1;
4544
4545         if (!rt6_is_dead(rt)) {
4546                 *weight += rt->fib6_nh->fib_nh_weight;
4547                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4548                                                     total) - 1;
4549         }
4550         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4551 }
4552
4553 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4554 {
4555         struct fib6_info *iter;
4556         int weight = 0;
4557
4558         rt6_upper_bound_set(rt, &weight, total);
4559
4560         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4561                 rt6_upper_bound_set(iter, &weight, total);
4562 }
4563
4564 void rt6_multipath_rebalance(struct fib6_info *rt)
4565 {
4566         struct fib6_info *first;
4567         int total;
4568
4569         /* In case the entire multipath route was marked for flushing,
4570          * then there is no need to rebalance upon the removal of every
4571          * sibling route.
4572          */
4573         if (!rt->fib6_nsiblings || rt->should_flush)
4574                 return;
4575
4576         /* During lookup routes are evaluated in order, so we need to
4577          * make sure upper bounds are assigned from the first sibling
4578          * onwards.
4579          */
4580         first = rt6_multipath_first_sibling(rt);
4581         if (WARN_ON_ONCE(!first))
4582                 return;
4583
4584         total = rt6_multipath_total_weight(first);
4585         rt6_multipath_upper_bound_set(first, total);
4586 }
4587
4588 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4589 {
4590         const struct arg_netdev_event *arg = p_arg;
4591         struct net *net = dev_net(arg->dev);
4592
4593         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4594             rt->fib6_nh->fib_nh_dev == arg->dev) {
4595                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4596                 fib6_update_sernum_upto_root(net, rt);
4597                 rt6_multipath_rebalance(rt);
4598         }
4599
4600         return 0;
4601 }
4602
4603 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4604 {
4605         struct arg_netdev_event arg = {
4606                 .dev = dev,
4607                 {
4608                         .nh_flags = nh_flags,
4609                 },
4610         };
4611
4612         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4613                 arg.nh_flags |= RTNH_F_LINKDOWN;
4614
4615         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4616 }
4617
4618 /* only called for fib entries with inline fib6_nh */
4619 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4620                                    const struct net_device *dev)
4621 {
4622         struct fib6_info *iter;
4623
4624         if (rt->fib6_nh->fib_nh_dev == dev)
4625                 return true;
4626         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4627                 if (iter->fib6_nh->fib_nh_dev == dev)
4628                         return true;
4629
4630         return false;
4631 }
4632
4633 static void rt6_multipath_flush(struct fib6_info *rt)
4634 {
4635         struct fib6_info *iter;
4636
4637         rt->should_flush = 1;
4638         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4639                 iter->should_flush = 1;
4640 }
4641
4642 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4643                                              const struct net_device *down_dev)
4644 {
4645         struct fib6_info *iter;
4646         unsigned int dead = 0;
4647
4648         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4649             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4650                 dead++;
4651         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4652                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4653                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4654                         dead++;
4655
4656         return dead;
4657 }
4658
4659 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4660                                        const struct net_device *dev,
4661                                        unsigned char nh_flags)
4662 {
4663         struct fib6_info *iter;
4664
4665         if (rt->fib6_nh->fib_nh_dev == dev)
4666                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4667         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4668                 if (iter->fib6_nh->fib_nh_dev == dev)
4669                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4670 }
4671
4672 /* called with write lock held for table with rt */
4673 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4674 {
4675         const struct arg_netdev_event *arg = p_arg;
4676         const struct net_device *dev = arg->dev;
4677         struct net *net = dev_net(dev);
4678
4679         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4680                 return 0;
4681
4682         switch (arg->event) {
4683         case NETDEV_UNREGISTER:
4684                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4685         case NETDEV_DOWN:
4686                 if (rt->should_flush)
4687                         return -1;
4688                 if (!rt->fib6_nsiblings)
4689                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4690                 if (rt6_multipath_uses_dev(rt, dev)) {
4691                         unsigned int count;
4692
4693                         count = rt6_multipath_dead_count(rt, dev);
4694                         if (rt->fib6_nsiblings + 1 == count) {
4695                                 rt6_multipath_flush(rt);
4696                                 return -1;
4697                         }
4698                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4699                                                    RTNH_F_LINKDOWN);
4700                         fib6_update_sernum(net, rt);
4701                         rt6_multipath_rebalance(rt);
4702                 }
4703                 return -2;
4704         case NETDEV_CHANGE:
4705                 if (rt->fib6_nh->fib_nh_dev != dev ||
4706                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4707                         break;
4708                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4709                 rt6_multipath_rebalance(rt);
4710                 break;
4711         }
4712
4713         return 0;
4714 }
4715
4716 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4717 {
4718         struct arg_netdev_event arg = {
4719                 .dev = dev,
4720                 {
4721                         .event = event,
4722                 },
4723         };
4724         struct net *net = dev_net(dev);
4725
4726         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4727                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4728         else
4729                 fib6_clean_all(net, fib6_ifdown, &arg);
4730 }
4731
4732 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4733 {
4734         rt6_sync_down_dev(dev, event);
4735         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4736         neigh_ifdown(&nd_tbl, dev);
4737 }
4738
4739 struct rt6_mtu_change_arg {
4740         struct net_device *dev;
4741         unsigned int mtu;
4742         struct fib6_info *f6i;
4743 };
4744
4745 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4746 {
4747         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4748         struct fib6_info *f6i = arg->f6i;
4749
4750         /* For administrative MTU increase, there is no way to discover
4751          * IPv6 PMTU increase, so PMTU increase should be updated here.
4752          * Since RFC 1981 doesn't include administrative MTU increase
4753          * update PMTU increase is a MUST. (i.e. jumbo frame)
4754          */
4755         if (nh->fib_nh_dev == arg->dev) {
4756                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4757                 u32 mtu = f6i->fib6_pmtu;
4758
4759                 if (mtu >= arg->mtu ||
4760                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4761                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4762
4763                 spin_lock_bh(&rt6_exception_lock);
4764                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4765                 spin_unlock_bh(&rt6_exception_lock);
4766         }
4767
4768         return 0;
4769 }
4770
4771 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4772 {
4773         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4774         struct inet6_dev *idev;
4775
4776         /* In IPv6 pmtu discovery is not optional,
4777            so that RTAX_MTU lock cannot disable it.
4778            We still use this lock to block changes
4779            caused by addrconf/ndisc.
4780         */
4781
4782         idev = __in6_dev_get(arg->dev);
4783         if (!idev)
4784                 return 0;
4785
4786         if (fib6_metric_locked(f6i, RTAX_MTU))
4787                 return 0;
4788
4789         arg->f6i = f6i;
4790         if (f6i->nh) {
4791                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4792                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4793                                                 arg);
4794         }
4795
4796         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4797 }
4798
4799 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4800 {
4801         struct rt6_mtu_change_arg arg = {
4802                 .dev = dev,
4803                 .mtu = mtu,
4804         };
4805
4806         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4807 }
4808
4809 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4810         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4811         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4812         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4813         [RTA_OIF]               = { .type = NLA_U32 },
4814         [RTA_IIF]               = { .type = NLA_U32 },
4815         [RTA_PRIORITY]          = { .type = NLA_U32 },
4816         [RTA_METRICS]           = { .type = NLA_NESTED },
4817         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4818         [RTA_PREF]              = { .type = NLA_U8 },
4819         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4820         [RTA_ENCAP]             = { .type = NLA_NESTED },
4821         [RTA_EXPIRES]           = { .type = NLA_U32 },
4822         [RTA_UID]               = { .type = NLA_U32 },
4823         [RTA_MARK]              = { .type = NLA_U32 },
4824         [RTA_TABLE]             = { .type = NLA_U32 },
4825         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4826         [RTA_SPORT]             = { .type = NLA_U16 },
4827         [RTA_DPORT]             = { .type = NLA_U16 },
4828         [RTA_NH_ID]             = { .type = NLA_U32 },
4829 };
4830
4831 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4832                               struct fib6_config *cfg,
4833                               struct netlink_ext_ack *extack)
4834 {
4835         struct rtmsg *rtm;
4836         struct nlattr *tb[RTA_MAX+1];
4837         unsigned int pref;
4838         int err;
4839
4840         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4841                                      rtm_ipv6_policy, extack);
4842         if (err < 0)
4843                 goto errout;
4844
4845         err = -EINVAL;
4846         rtm = nlmsg_data(nlh);
4847
4848         *cfg = (struct fib6_config){
4849                 .fc_table = rtm->rtm_table,
4850                 .fc_dst_len = rtm->rtm_dst_len,
4851                 .fc_src_len = rtm->rtm_src_len,
4852                 .fc_flags = RTF_UP,
4853                 .fc_protocol = rtm->rtm_protocol,
4854                 .fc_type = rtm->rtm_type,
4855
4856                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4857                 .fc_nlinfo.nlh = nlh,
4858                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4859         };
4860
4861         if (rtm->rtm_type == RTN_UNREACHABLE ||
4862             rtm->rtm_type == RTN_BLACKHOLE ||
4863             rtm->rtm_type == RTN_PROHIBIT ||
4864             rtm->rtm_type == RTN_THROW)
4865                 cfg->fc_flags |= RTF_REJECT;
4866
4867         if (rtm->rtm_type == RTN_LOCAL)
4868                 cfg->fc_flags |= RTF_LOCAL;
4869
4870         if (rtm->rtm_flags & RTM_F_CLONED)
4871                 cfg->fc_flags |= RTF_CACHE;
4872
4873         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4874
4875         if (tb[RTA_NH_ID]) {
4876                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4877                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4878                         NL_SET_ERR_MSG(extack,
4879                                        "Nexthop specification and nexthop id are mutually exclusive");
4880                         goto errout;
4881                 }
4882                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4883         }
4884
4885         if (tb[RTA_GATEWAY]) {
4886                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4887                 cfg->fc_flags |= RTF_GATEWAY;
4888         }
4889         if (tb[RTA_VIA]) {
4890                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4891                 goto errout;
4892         }
4893
4894         if (tb[RTA_DST]) {
4895                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4896
4897                 if (nla_len(tb[RTA_DST]) < plen)
4898                         goto errout;
4899
4900                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4901         }
4902
4903         if (tb[RTA_SRC]) {
4904                 int plen = (rtm->rtm_src_len + 7) >> 3;
4905
4906                 if (nla_len(tb[RTA_SRC]) < plen)
4907                         goto errout;
4908
4909                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4910         }
4911
4912         if (tb[RTA_PREFSRC])
4913                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4914
4915         if (tb[RTA_OIF])
4916                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4917
4918         if (tb[RTA_PRIORITY])
4919                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4920
4921         if (tb[RTA_METRICS]) {
4922                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4923                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4924         }
4925
4926         if (tb[RTA_TABLE])
4927                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4928
4929         if (tb[RTA_MULTIPATH]) {
4930                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4931                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4932
4933                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4934                                                      cfg->fc_mp_len, extack);
4935                 if (err < 0)
4936                         goto errout;
4937         }
4938
4939         if (tb[RTA_PREF]) {
4940                 pref = nla_get_u8(tb[RTA_PREF]);
4941                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4942                     pref != ICMPV6_ROUTER_PREF_HIGH)
4943                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4944                 cfg->fc_flags |= RTF_PREF(pref);
4945         }
4946
4947         if (tb[RTA_ENCAP])
4948                 cfg->fc_encap = tb[RTA_ENCAP];
4949
4950         if (tb[RTA_ENCAP_TYPE]) {
4951                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4952
4953                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4954                 if (err < 0)
4955                         goto errout;
4956         }
4957
4958         if (tb[RTA_EXPIRES]) {
4959                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4960
4961                 if (addrconf_finite_timeout(timeout)) {
4962                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
4963                         cfg->fc_flags |= RTF_EXPIRES;
4964                 }
4965         }
4966
4967         err = 0;
4968 errout:
4969         return err;
4970 }
4971
4972 struct rt6_nh {
4973         struct fib6_info *fib6_info;
4974         struct fib6_config r_cfg;
4975         struct list_head next;
4976 };
4977
4978 static int ip6_route_info_append(struct net *net,
4979                                  struct list_head *rt6_nh_list,
4980                                  struct fib6_info *rt,
4981                                  struct fib6_config *r_cfg)
4982 {
4983         struct rt6_nh *nh;
4984         int err = -EEXIST;
4985
4986         list_for_each_entry(nh, rt6_nh_list, next) {
4987                 /* check if fib6_info already exists */
4988                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
4989                         return err;
4990         }
4991
4992         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
4993         if (!nh)
4994                 return -ENOMEM;
4995         nh->fib6_info = rt;
4996         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
4997         list_add_tail(&nh->next, rt6_nh_list);
4998
4999         return 0;
5000 }
5001
5002 static void ip6_route_mpath_notify(struct fib6_info *rt,
5003                                    struct fib6_info *rt_last,
5004                                    struct nl_info *info,
5005                                    __u16 nlflags)
5006 {
5007         /* if this is an APPEND route, then rt points to the first route
5008          * inserted and rt_last points to last route inserted. Userspace
5009          * wants a consistent dump of the route which starts at the first
5010          * nexthop. Since sibling routes are always added at the end of
5011          * the list, find the first sibling of the last route appended
5012          */
5013         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5014                 rt = list_first_entry(&rt_last->fib6_siblings,
5015                                       struct fib6_info,
5016                                       fib6_siblings);
5017         }
5018
5019         if (rt)
5020                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5021 }
5022
5023 static int ip6_route_multipath_add(struct fib6_config *cfg,
5024                                    struct netlink_ext_ack *extack)
5025 {
5026         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5027         struct nl_info *info = &cfg->fc_nlinfo;
5028         enum fib_event_type event_type;
5029         struct fib6_config r_cfg;
5030         struct rtnexthop *rtnh;
5031         struct fib6_info *rt;
5032         struct rt6_nh *err_nh;
5033         struct rt6_nh *nh, *nh_safe;
5034         __u16 nlflags;
5035         int remaining;
5036         int attrlen;
5037         int err = 1;
5038         int nhn = 0;
5039         int replace = (cfg->fc_nlinfo.nlh &&
5040                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5041         LIST_HEAD(rt6_nh_list);
5042
5043         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5044         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5045                 nlflags |= NLM_F_APPEND;
5046
5047         remaining = cfg->fc_mp_len;
5048         rtnh = (struct rtnexthop *)cfg->fc_mp;
5049
5050         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5051          * fib6_info structs per nexthop
5052          */
5053         while (rtnh_ok(rtnh, remaining)) {
5054                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5055                 if (rtnh->rtnh_ifindex)
5056                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5057
5058                 attrlen = rtnh_attrlen(rtnh);
5059                 if (attrlen > 0) {
5060                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5061
5062                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5063                         if (nla) {
5064                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5065                                 r_cfg.fc_flags |= RTF_GATEWAY;
5066                         }
5067                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5068                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5069                         if (nla)
5070                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5071                 }
5072
5073                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5074                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5075                 if (IS_ERR(rt)) {
5076                         err = PTR_ERR(rt);
5077                         rt = NULL;
5078                         goto cleanup;
5079                 }
5080                 if (!rt6_qualify_for_ecmp(rt)) {
5081                         err = -EINVAL;
5082                         NL_SET_ERR_MSG(extack,
5083                                        "Device only routes can not be added for IPv6 using the multipath API.");
5084                         fib6_info_release(rt);
5085                         goto cleanup;
5086                 }
5087
5088                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5089
5090                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5091                                             rt, &r_cfg);
5092                 if (err) {
5093                         fib6_info_release(rt);
5094                         goto cleanup;
5095                 }
5096
5097                 rtnh = rtnh_next(rtnh, &remaining);
5098         }
5099
5100         if (list_empty(&rt6_nh_list)) {
5101                 NL_SET_ERR_MSG(extack,
5102                                "Invalid nexthop configuration - no valid nexthops");
5103                 return -EINVAL;
5104         }
5105
5106         /* for add and replace send one notification with all nexthops.
5107          * Skip the notification in fib6_add_rt2node and send one with
5108          * the full route when done
5109          */
5110         info->skip_notify = 1;
5111
5112         /* For add and replace, send one notification with all nexthops. For
5113          * append, send one notification with all appended nexthops.
5114          */
5115         info->skip_notify_kernel = 1;
5116
5117         err_nh = NULL;
5118         list_for_each_entry(nh, &rt6_nh_list, next) {
5119                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5120                 fib6_info_release(nh->fib6_info);
5121
5122                 if (!err) {
5123                         /* save reference to last route successfully inserted */
5124                         rt_last = nh->fib6_info;
5125
5126                         /* save reference to first route for notification */
5127                         if (!rt_notif)
5128                                 rt_notif = nh->fib6_info;
5129                 }
5130
5131                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5132                 nh->fib6_info = NULL;
5133                 if (err) {
5134                         if (replace && nhn)
5135                                 NL_SET_ERR_MSG_MOD(extack,
5136                                                    "multipath route replace failed (check consistency of installed routes)");
5137                         err_nh = nh;
5138                         goto add_errout;
5139                 }
5140
5141                 /* Because each route is added like a single route we remove
5142                  * these flags after the first nexthop: if there is a collision,
5143                  * we have already failed to add the first nexthop:
5144                  * fib6_add_rt2node() has rejected it; when replacing, old
5145                  * nexthops have been replaced by first new, the rest should
5146                  * be added to it.
5147                  */
5148                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5149                                                      NLM_F_REPLACE);
5150                 nhn++;
5151         }
5152
5153         event_type = replace ? FIB_EVENT_ENTRY_REPLACE : FIB_EVENT_ENTRY_ADD;
5154         err = call_fib6_multipath_entry_notifiers(info->nl_net, event_type,
5155                                                   rt_notif, nhn - 1, extack);
5156         if (err) {
5157                 /* Delete all the siblings that were just added */
5158                 err_nh = NULL;
5159                 goto add_errout;
5160         }
5161
5162         /* success ... tell user about new route */
5163         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5164         goto cleanup;
5165
5166 add_errout:
5167         /* send notification for routes that were added so that
5168          * the delete notifications sent by ip6_route_del are
5169          * coherent
5170          */
5171         if (rt_notif)
5172                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5173
5174         /* Delete routes that were already added */
5175         list_for_each_entry(nh, &rt6_nh_list, next) {
5176                 if (err_nh == nh)
5177                         break;
5178                 ip6_route_del(&nh->r_cfg, extack);
5179         }
5180
5181 cleanup:
5182         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5183                 if (nh->fib6_info)
5184                         fib6_info_release(nh->fib6_info);
5185                 list_del(&nh->next);
5186                 kfree(nh);
5187         }
5188
5189         return err;
5190 }
5191
5192 static int ip6_route_multipath_del(struct fib6_config *cfg,
5193                                    struct netlink_ext_ack *extack)
5194 {
5195         struct fib6_config r_cfg;
5196         struct rtnexthop *rtnh;
5197         int remaining;
5198         int attrlen;
5199         int err = 1, last_err = 0;
5200
5201         remaining = cfg->fc_mp_len;
5202         rtnh = (struct rtnexthop *)cfg->fc_mp;
5203
5204         /* Parse a Multipath Entry */
5205         while (rtnh_ok(rtnh, remaining)) {
5206                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5207                 if (rtnh->rtnh_ifindex)
5208                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5209
5210                 attrlen = rtnh_attrlen(rtnh);
5211                 if (attrlen > 0) {
5212                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5213
5214                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5215                         if (nla) {
5216                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5217                                 r_cfg.fc_flags |= RTF_GATEWAY;
5218                         }
5219                 }
5220                 err = ip6_route_del(&r_cfg, extack);
5221                 if (err)
5222                         last_err = err;
5223
5224                 rtnh = rtnh_next(rtnh, &remaining);
5225         }
5226
5227         return last_err;
5228 }
5229
5230 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5231                               struct netlink_ext_ack *extack)
5232 {
5233         struct fib6_config cfg;
5234         int err;
5235
5236         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5237         if (err < 0)
5238                 return err;
5239
5240         if (cfg.fc_nh_id &&
5241             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5242                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5243                 return -EINVAL;
5244         }
5245
5246         if (cfg.fc_mp)
5247                 return ip6_route_multipath_del(&cfg, extack);
5248         else {
5249                 cfg.fc_delete_all_nh = 1;
5250                 return ip6_route_del(&cfg, extack);
5251         }
5252 }
5253
5254 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5255                               struct netlink_ext_ack *extack)
5256 {
5257         struct fib6_config cfg;
5258         int err;
5259
5260         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5261         if (err < 0)
5262                 return err;
5263
5264         if (cfg.fc_metric == 0)
5265                 cfg.fc_metric = IP6_RT_PRIO_USER;
5266
5267         if (cfg.fc_mp)
5268                 return ip6_route_multipath_add(&cfg, extack);
5269         else
5270                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5271 }
5272
5273 /* add the overhead of this fib6_nh to nexthop_len */
5274 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5275 {
5276         int *nexthop_len = arg;
5277
5278         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5279                      + NLA_ALIGN(sizeof(struct rtnexthop))
5280                      + nla_total_size(16); /* RTA_GATEWAY */
5281
5282         if (nh->fib_nh_lws) {
5283                 /* RTA_ENCAP_TYPE */
5284                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5285                 /* RTA_ENCAP */
5286                 *nexthop_len += nla_total_size(2);
5287         }
5288
5289         return 0;
5290 }
5291
5292 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5293 {
5294         int nexthop_len;
5295
5296         if (f6i->nh) {
5297                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5298                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5299                                          &nexthop_len);
5300         } else {
5301                 struct fib6_nh *nh = f6i->fib6_nh;
5302
5303                 nexthop_len = 0;
5304                 if (f6i->fib6_nsiblings) {
5305                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5306                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5307                                     + nla_total_size(16) /* RTA_GATEWAY */
5308                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5309
5310                         nexthop_len *= f6i->fib6_nsiblings;
5311                 }
5312                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5313         }
5314
5315         return NLMSG_ALIGN(sizeof(struct rtmsg))
5316                + nla_total_size(16) /* RTA_SRC */
5317                + nla_total_size(16) /* RTA_DST */
5318                + nla_total_size(16) /* RTA_GATEWAY */
5319                + nla_total_size(16) /* RTA_PREFSRC */
5320                + nla_total_size(4) /* RTA_TABLE */
5321                + nla_total_size(4) /* RTA_IIF */
5322                + nla_total_size(4) /* RTA_OIF */
5323                + nla_total_size(4) /* RTA_PRIORITY */
5324                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5325                + nla_total_size(sizeof(struct rta_cacheinfo))
5326                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5327                + nla_total_size(1) /* RTA_PREF */
5328                + nexthop_len;
5329 }
5330
5331 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5332                                  unsigned char *flags)
5333 {
5334         if (nexthop_is_multipath(nh)) {
5335                 struct nlattr *mp;
5336
5337                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5338                 if (!mp)
5339                         goto nla_put_failure;
5340
5341                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5342                         goto nla_put_failure;
5343
5344                 nla_nest_end(skb, mp);
5345         } else {
5346                 struct fib6_nh *fib6_nh;
5347
5348                 fib6_nh = nexthop_fib6_nh(nh);
5349                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5350                                      flags, false) < 0)
5351                         goto nla_put_failure;
5352         }
5353
5354         return 0;
5355
5356 nla_put_failure:
5357         return -EMSGSIZE;
5358 }
5359
5360 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5361                          struct fib6_info *rt, struct dst_entry *dst,
5362                          struct in6_addr *dest, struct in6_addr *src,
5363                          int iif, int type, u32 portid, u32 seq,
5364                          unsigned int flags)
5365 {
5366         struct rt6_info *rt6 = (struct rt6_info *)dst;
5367         struct rt6key *rt6_dst, *rt6_src;
5368         u32 *pmetrics, table, rt6_flags;
5369         unsigned char nh_flags = 0;
5370         struct nlmsghdr *nlh;
5371         struct rtmsg *rtm;
5372         long expires = 0;
5373
5374         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5375         if (!nlh)
5376                 return -EMSGSIZE;
5377
5378         if (rt6) {
5379                 rt6_dst = &rt6->rt6i_dst;
5380                 rt6_src = &rt6->rt6i_src;
5381                 rt6_flags = rt6->rt6i_flags;
5382         } else {
5383                 rt6_dst = &rt->fib6_dst;
5384                 rt6_src = &rt->fib6_src;
5385                 rt6_flags = rt->fib6_flags;
5386         }
5387
5388         rtm = nlmsg_data(nlh);
5389         rtm->rtm_family = AF_INET6;
5390         rtm->rtm_dst_len = rt6_dst->plen;
5391         rtm->rtm_src_len = rt6_src->plen;
5392         rtm->rtm_tos = 0;
5393         if (rt->fib6_table)
5394                 table = rt->fib6_table->tb6_id;
5395         else
5396                 table = RT6_TABLE_UNSPEC;
5397         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5398         if (nla_put_u32(skb, RTA_TABLE, table))
5399                 goto nla_put_failure;
5400
5401         rtm->rtm_type = rt->fib6_type;
5402         rtm->rtm_flags = 0;
5403         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5404         rtm->rtm_protocol = rt->fib6_protocol;
5405
5406         if (rt6_flags & RTF_CACHE)
5407                 rtm->rtm_flags |= RTM_F_CLONED;
5408
5409         if (dest) {
5410                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5411                         goto nla_put_failure;
5412                 rtm->rtm_dst_len = 128;
5413         } else if (rtm->rtm_dst_len)
5414                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5415                         goto nla_put_failure;
5416 #ifdef CONFIG_IPV6_SUBTREES
5417         if (src) {
5418                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5419                         goto nla_put_failure;
5420                 rtm->rtm_src_len = 128;
5421         } else if (rtm->rtm_src_len &&
5422                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5423                 goto nla_put_failure;
5424 #endif
5425         if (iif) {
5426 #ifdef CONFIG_IPV6_MROUTE
5427                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5428                         int err = ip6mr_get_route(net, skb, rtm, portid);
5429
5430                         if (err == 0)
5431                                 return 0;
5432                         if (err < 0)
5433                                 goto nla_put_failure;
5434                 } else
5435 #endif
5436                         if (nla_put_u32(skb, RTA_IIF, iif))
5437                                 goto nla_put_failure;
5438         } else if (dest) {
5439                 struct in6_addr saddr_buf;
5440                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5441                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5442                         goto nla_put_failure;
5443         }
5444
5445         if (rt->fib6_prefsrc.plen) {
5446                 struct in6_addr saddr_buf;
5447                 saddr_buf = rt->fib6_prefsrc.addr;
5448                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5449                         goto nla_put_failure;
5450         }
5451
5452         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5453         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5454                 goto nla_put_failure;
5455
5456         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5457                 goto nla_put_failure;
5458
5459         /* For multipath routes, walk the siblings list and add
5460          * each as a nexthop within RTA_MULTIPATH.
5461          */
5462         if (rt6) {
5463                 if (rt6_flags & RTF_GATEWAY &&
5464                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5465                         goto nla_put_failure;
5466
5467                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5468                         goto nla_put_failure;
5469         } else if (rt->fib6_nsiblings) {
5470                 struct fib6_info *sibling, *next_sibling;
5471                 struct nlattr *mp;
5472
5473                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5474                 if (!mp)
5475                         goto nla_put_failure;
5476
5477                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5478                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5479                         goto nla_put_failure;
5480
5481                 list_for_each_entry_safe(sibling, next_sibling,
5482                                          &rt->fib6_siblings, fib6_siblings) {
5483                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5484                                             sibling->fib6_nh->fib_nh_weight,
5485                                             AF_INET6) < 0)
5486                                 goto nla_put_failure;
5487                 }
5488
5489                 nla_nest_end(skb, mp);
5490         } else if (rt->nh) {
5491                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5492                         goto nla_put_failure;
5493
5494                 if (nexthop_is_blackhole(rt->nh))
5495                         rtm->rtm_type = RTN_BLACKHOLE;
5496
5497                 if (rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5498                         goto nla_put_failure;
5499
5500                 rtm->rtm_flags |= nh_flags;
5501         } else {
5502                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5503                                      &nh_flags, false) < 0)
5504                         goto nla_put_failure;
5505
5506                 rtm->rtm_flags |= nh_flags;
5507         }
5508
5509         if (rt6_flags & RTF_EXPIRES) {
5510                 expires = dst ? dst->expires : rt->expires;
5511                 expires -= jiffies;
5512         }
5513
5514         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5515                 goto nla_put_failure;
5516
5517         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5518                 goto nla_put_failure;
5519
5520
5521         nlmsg_end(skb, nlh);
5522         return 0;
5523
5524 nla_put_failure:
5525         nlmsg_cancel(skb, nlh);
5526         return -EMSGSIZE;
5527 }
5528
5529 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5530 {
5531         const struct net_device *dev = arg;
5532
5533         if (nh->fib_nh_dev == dev)
5534                 return 1;
5535
5536         return 0;
5537 }
5538
5539 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5540                                const struct net_device *dev)
5541 {
5542         if (f6i->nh) {
5543                 struct net_device *_dev = (struct net_device *)dev;
5544
5545                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5546                                                   fib6_info_nh_uses_dev,
5547                                                   _dev);
5548         }
5549
5550         if (f6i->fib6_nh->fib_nh_dev == dev)
5551                 return true;
5552
5553         if (f6i->fib6_nsiblings) {
5554                 struct fib6_info *sibling, *next_sibling;
5555
5556                 list_for_each_entry_safe(sibling, next_sibling,
5557                                          &f6i->fib6_siblings, fib6_siblings) {
5558                         if (sibling->fib6_nh->fib_nh_dev == dev)
5559                                 return true;
5560                 }
5561         }
5562
5563         return false;
5564 }
5565
5566 struct fib6_nh_exception_dump_walker {
5567         struct rt6_rtnl_dump_arg *dump;
5568         struct fib6_info *rt;
5569         unsigned int flags;
5570         unsigned int skip;
5571         unsigned int count;
5572 };
5573
5574 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5575 {
5576         struct fib6_nh_exception_dump_walker *w = arg;
5577         struct rt6_rtnl_dump_arg *dump = w->dump;
5578         struct rt6_exception_bucket *bucket;
5579         struct rt6_exception *rt6_ex;
5580         int i, err;
5581
5582         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5583         if (!bucket)
5584                 return 0;
5585
5586         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5587                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5588                         if (w->skip) {
5589                                 w->skip--;
5590                                 continue;
5591                         }
5592
5593                         /* Expiration of entries doesn't bump sernum, insertion
5594                          * does. Removal is triggered by insertion, so we can
5595                          * rely on the fact that if entries change between two
5596                          * partial dumps, this node is scanned again completely,
5597                          * see rt6_insert_exception() and fib6_dump_table().
5598                          *
5599                          * Count expired entries we go through as handled
5600                          * entries that we'll skip next time, in case of partial
5601                          * node dump. Otherwise, if entries expire meanwhile,
5602                          * we'll skip the wrong amount.
5603                          */
5604                         if (rt6_check_expired(rt6_ex->rt6i)) {
5605                                 w->count++;
5606                                 continue;
5607                         }
5608
5609                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5610                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5611                                             RTM_NEWROUTE,
5612                                             NETLINK_CB(dump->cb->skb).portid,
5613                                             dump->cb->nlh->nlmsg_seq, w->flags);
5614                         if (err)
5615                                 return err;
5616
5617                         w->count++;
5618                 }
5619                 bucket++;
5620         }
5621
5622         return 0;
5623 }
5624
5625 /* Return -1 if done with node, number of handled routes on partial dump */
5626 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5627 {
5628         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5629         struct fib_dump_filter *filter = &arg->filter;
5630         unsigned int flags = NLM_F_MULTI;
5631         struct net *net = arg->net;
5632         int count = 0;
5633
5634         if (rt == net->ipv6.fib6_null_entry)
5635                 return -1;
5636
5637         if ((filter->flags & RTM_F_PREFIX) &&
5638             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5639                 /* success since this is not a prefix route */
5640                 return -1;
5641         }
5642         if (filter->filter_set &&
5643             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5644              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5645              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5646                 return -1;
5647         }
5648
5649         if (filter->filter_set ||
5650             !filter->dump_routes || !filter->dump_exceptions) {
5651                 flags |= NLM_F_DUMP_FILTERED;
5652         }
5653
5654         if (filter->dump_routes) {
5655                 if (skip) {
5656                         skip--;
5657                 } else {
5658                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5659                                           0, RTM_NEWROUTE,
5660                                           NETLINK_CB(arg->cb->skb).portid,
5661                                           arg->cb->nlh->nlmsg_seq, flags)) {
5662                                 return 0;
5663                         }
5664                         count++;
5665                 }
5666         }
5667
5668         if (filter->dump_exceptions) {
5669                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5670                                                            .rt = rt,
5671                                                            .flags = flags,
5672                                                            .skip = skip,
5673                                                            .count = 0 };
5674                 int err;
5675
5676                 rcu_read_lock();
5677                 if (rt->nh) {
5678                         err = nexthop_for_each_fib6_nh(rt->nh,
5679                                                        rt6_nh_dump_exceptions,
5680                                                        &w);
5681                 } else {
5682                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5683                 }
5684                 rcu_read_unlock();
5685
5686                 if (err)
5687                         return count += w.count;
5688         }
5689
5690         return -1;
5691 }
5692
5693 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5694                                         const struct nlmsghdr *nlh,
5695                                         struct nlattr **tb,
5696                                         struct netlink_ext_ack *extack)
5697 {
5698         struct rtmsg *rtm;
5699         int i, err;
5700
5701         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5702                 NL_SET_ERR_MSG_MOD(extack,
5703                                    "Invalid header for get route request");
5704                 return -EINVAL;
5705         }
5706
5707         if (!netlink_strict_get_check(skb))
5708                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5709                                               rtm_ipv6_policy, extack);
5710
5711         rtm = nlmsg_data(nlh);
5712         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5713             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5714             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5715             rtm->rtm_type) {
5716                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5717                 return -EINVAL;
5718         }
5719         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5720                 NL_SET_ERR_MSG_MOD(extack,
5721                                    "Invalid flags for get route request");
5722                 return -EINVAL;
5723         }
5724
5725         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5726                                             rtm_ipv6_policy, extack);
5727         if (err)
5728                 return err;
5729
5730         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5731             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5732                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5733                 return -EINVAL;
5734         }
5735
5736         for (i = 0; i <= RTA_MAX; i++) {
5737                 if (!tb[i])
5738                         continue;
5739
5740                 switch (i) {
5741                 case RTA_SRC:
5742                 case RTA_DST:
5743                 case RTA_IIF:
5744                 case RTA_OIF:
5745                 case RTA_MARK:
5746                 case RTA_UID:
5747                 case RTA_SPORT:
5748                 case RTA_DPORT:
5749                 case RTA_IP_PROTO:
5750                         break;
5751                 default:
5752                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5753                         return -EINVAL;
5754                 }
5755         }
5756
5757         return 0;
5758 }
5759
5760 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5761                               struct netlink_ext_ack *extack)
5762 {
5763         struct net *net = sock_net(in_skb->sk);
5764         struct nlattr *tb[RTA_MAX+1];
5765         int err, iif = 0, oif = 0;
5766         struct fib6_info *from;
5767         struct dst_entry *dst;
5768         struct rt6_info *rt;
5769         struct sk_buff *skb;
5770         struct rtmsg *rtm;
5771         struct flowi6 fl6 = {};
5772         bool fibmatch;
5773
5774         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5775         if (err < 0)
5776                 goto errout;
5777
5778         err = -EINVAL;
5779         rtm = nlmsg_data(nlh);
5780         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5781         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5782
5783         if (tb[RTA_SRC]) {
5784                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5785                         goto errout;
5786
5787                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5788         }
5789
5790         if (tb[RTA_DST]) {
5791                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5792                         goto errout;
5793
5794                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5795         }
5796
5797         if (tb[RTA_IIF])
5798                 iif = nla_get_u32(tb[RTA_IIF]);
5799
5800         if (tb[RTA_OIF])
5801                 oif = nla_get_u32(tb[RTA_OIF]);
5802
5803         if (tb[RTA_MARK])
5804                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5805
5806         if (tb[RTA_UID])
5807                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5808                                            nla_get_u32(tb[RTA_UID]));
5809         else
5810                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5811
5812         if (tb[RTA_SPORT])
5813                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5814
5815         if (tb[RTA_DPORT])
5816                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5817
5818         if (tb[RTA_IP_PROTO]) {
5819                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5820                                                   &fl6.flowi6_proto, AF_INET6,
5821                                                   extack);
5822                 if (err)
5823                         goto errout;
5824         }
5825
5826         if (iif) {
5827                 struct net_device *dev;
5828                 int flags = 0;
5829
5830                 rcu_read_lock();
5831
5832                 dev = dev_get_by_index_rcu(net, iif);
5833                 if (!dev) {
5834                         rcu_read_unlock();
5835                         err = -ENODEV;
5836                         goto errout;
5837                 }
5838
5839                 fl6.flowi6_iif = iif;
5840
5841                 if (!ipv6_addr_any(&fl6.saddr))
5842                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5843
5844                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5845
5846                 rcu_read_unlock();
5847         } else {
5848                 fl6.flowi6_oif = oif;
5849
5850                 dst = ip6_route_output(net, NULL, &fl6);
5851         }
5852
5853
5854         rt = container_of(dst, struct rt6_info, dst);
5855         if (rt->dst.error) {
5856                 err = rt->dst.error;
5857                 ip6_rt_put(rt);
5858                 goto errout;
5859         }
5860
5861         if (rt == net->ipv6.ip6_null_entry) {
5862                 err = rt->dst.error;
5863                 ip6_rt_put(rt);
5864                 goto errout;
5865         }
5866
5867         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5868         if (!skb) {
5869                 ip6_rt_put(rt);
5870                 err = -ENOBUFS;
5871                 goto errout;
5872         }
5873
5874         skb_dst_set(skb, &rt->dst);
5875
5876         rcu_read_lock();
5877         from = rcu_dereference(rt->from);
5878         if (from) {
5879                 if (fibmatch)
5880                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5881                                             iif, RTM_NEWROUTE,
5882                                             NETLINK_CB(in_skb).portid,
5883                                             nlh->nlmsg_seq, 0);
5884                 else
5885                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5886                                             &fl6.saddr, iif, RTM_NEWROUTE,
5887                                             NETLINK_CB(in_skb).portid,
5888                                             nlh->nlmsg_seq, 0);
5889         } else {
5890                 err = -ENETUNREACH;
5891         }
5892         rcu_read_unlock();
5893
5894         if (err < 0) {
5895                 kfree_skb(skb);
5896                 goto errout;
5897         }
5898
5899         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5900 errout:
5901         return err;
5902 }
5903
5904 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
5905                      unsigned int nlm_flags)
5906 {
5907         struct sk_buff *skb;
5908         struct net *net = info->nl_net;
5909         u32 seq;
5910         int err;
5911
5912         err = -ENOBUFS;
5913         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5914
5915         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5916         if (!skb)
5917                 goto errout;
5918
5919         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5920                             event, info->portid, seq, nlm_flags);
5921         if (err < 0) {
5922                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5923                 WARN_ON(err == -EMSGSIZE);
5924                 kfree_skb(skb);
5925                 goto errout;
5926         }
5927         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5928                     info->nlh, gfp_any());
5929         return;
5930 errout:
5931         if (err < 0)
5932                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5933 }
5934
5935 void fib6_rt_update(struct net *net, struct fib6_info *rt,
5936                     struct nl_info *info)
5937 {
5938         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5939         struct sk_buff *skb;
5940         int err = -ENOBUFS;
5941
5942         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
5943          * is implemented and supported for nexthop objects
5944          */
5945         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
5946
5947         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5948         if (!skb)
5949                 goto errout;
5950
5951         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5952                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
5953         if (err < 0) {
5954                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5955                 WARN_ON(err == -EMSGSIZE);
5956                 kfree_skb(skb);
5957                 goto errout;
5958         }
5959         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5960                     info->nlh, gfp_any());
5961         return;
5962 errout:
5963         if (err < 0)
5964                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5965 }
5966
5967 static int ip6_route_dev_notify(struct notifier_block *this,
5968                                 unsigned long event, void *ptr)
5969 {
5970         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
5971         struct net *net = dev_net(dev);
5972
5973         if (!(dev->flags & IFF_LOOPBACK))
5974                 return NOTIFY_OK;
5975
5976         if (event == NETDEV_REGISTER) {
5977                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
5978                 net->ipv6.ip6_null_entry->dst.dev = dev;
5979                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
5980 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5981                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
5982                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
5983                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
5984                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
5985 #endif
5986          } else if (event == NETDEV_UNREGISTER &&
5987                     dev->reg_state != NETREG_UNREGISTERED) {
5988                 /* NETDEV_UNREGISTER could be fired for multiple times by
5989                  * netdev_wait_allrefs(). Make sure we only call this once.
5990                  */
5991                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
5992 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5993                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
5994                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
5995 #endif
5996         }
5997
5998         return NOTIFY_OK;
5999 }
6000
6001 /*
6002  *      /proc
6003  */
6004
6005 #ifdef CONFIG_PROC_FS
6006 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6007 {
6008         struct net *net = (struct net *)seq->private;
6009         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6010                    net->ipv6.rt6_stats->fib_nodes,
6011                    net->ipv6.rt6_stats->fib_route_nodes,
6012                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6013                    net->ipv6.rt6_stats->fib_rt_entries,
6014                    net->ipv6.rt6_stats->fib_rt_cache,
6015                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6016                    net->ipv6.rt6_stats->fib_discarded_routes);
6017
6018         return 0;
6019 }
6020 #endif  /* CONFIG_PROC_FS */
6021
6022 #ifdef CONFIG_SYSCTL
6023
6024 static
6025 int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6026                               void __user *buffer, size_t *lenp, loff_t *ppos)
6027 {
6028         struct net *net;
6029         int delay;
6030         int ret;
6031         if (!write)
6032                 return -EINVAL;
6033
6034         net = (struct net *)ctl->extra1;
6035         delay = net->ipv6.sysctl.flush_delay;
6036         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6037         if (ret)
6038                 return ret;
6039
6040         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6041         return 0;
6042 }
6043
6044 static struct ctl_table ipv6_route_table_template[] = {
6045         {
6046                 .procname       =       "flush",
6047                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6048                 .maxlen         =       sizeof(int),
6049                 .mode           =       0200,
6050                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6051         },
6052         {
6053                 .procname       =       "gc_thresh",
6054                 .data           =       &ip6_dst_ops_template.gc_thresh,
6055                 .maxlen         =       sizeof(int),
6056                 .mode           =       0644,
6057                 .proc_handler   =       proc_dointvec,
6058         },
6059         {
6060                 .procname       =       "max_size",
6061                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6062                 .maxlen         =       sizeof(int),
6063                 .mode           =       0644,
6064                 .proc_handler   =       proc_dointvec,
6065         },
6066         {
6067                 .procname       =       "gc_min_interval",
6068                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6069                 .maxlen         =       sizeof(int),
6070                 .mode           =       0644,
6071                 .proc_handler   =       proc_dointvec_jiffies,
6072         },
6073         {
6074                 .procname       =       "gc_timeout",
6075                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6076                 .maxlen         =       sizeof(int),
6077                 .mode           =       0644,
6078                 .proc_handler   =       proc_dointvec_jiffies,
6079         },
6080         {
6081                 .procname       =       "gc_interval",
6082                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6083                 .maxlen         =       sizeof(int),
6084                 .mode           =       0644,
6085                 .proc_handler   =       proc_dointvec_jiffies,
6086         },
6087         {
6088                 .procname       =       "gc_elasticity",
6089                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6090                 .maxlen         =       sizeof(int),
6091                 .mode           =       0644,
6092                 .proc_handler   =       proc_dointvec,
6093         },
6094         {
6095                 .procname       =       "mtu_expires",
6096                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6097                 .maxlen         =       sizeof(int),
6098                 .mode           =       0644,
6099                 .proc_handler   =       proc_dointvec_jiffies,
6100         },
6101         {
6102                 .procname       =       "min_adv_mss",
6103                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6104                 .maxlen         =       sizeof(int),
6105                 .mode           =       0644,
6106                 .proc_handler   =       proc_dointvec,
6107         },
6108         {
6109                 .procname       =       "gc_min_interval_ms",
6110                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6111                 .maxlen         =       sizeof(int),
6112                 .mode           =       0644,
6113                 .proc_handler   =       proc_dointvec_ms_jiffies,
6114         },
6115         {
6116                 .procname       =       "skip_notify_on_dev_down",
6117                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6118                 .maxlen         =       sizeof(int),
6119                 .mode           =       0644,
6120                 .proc_handler   =       proc_dointvec_minmax,
6121                 .extra1         =       SYSCTL_ZERO,
6122                 .extra2         =       SYSCTL_ONE,
6123         },
6124         { }
6125 };
6126
6127 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6128 {
6129         struct ctl_table *table;
6130
6131         table = kmemdup(ipv6_route_table_template,
6132                         sizeof(ipv6_route_table_template),
6133                         GFP_KERNEL);
6134
6135         if (table) {
6136                 table[0].data = &net->ipv6.sysctl.flush_delay;
6137                 table[0].extra1 = net;
6138                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6139                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6140                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6141                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6142                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6143                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6144                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6145                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6146                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6147                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6148
6149                 /* Don't export sysctls to unprivileged users */
6150                 if (net->user_ns != &init_user_ns)
6151                         table[0].procname = NULL;
6152         }
6153
6154         return table;
6155 }
6156 #endif
6157
6158 static int __net_init ip6_route_net_init(struct net *net)
6159 {
6160         int ret = -ENOMEM;
6161
6162         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6163                sizeof(net->ipv6.ip6_dst_ops));
6164
6165         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6166                 goto out_ip6_dst_ops;
6167
6168         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6169         if (!net->ipv6.fib6_null_entry)
6170                 goto out_ip6_dst_entries;
6171         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6172                sizeof(*net->ipv6.fib6_null_entry));
6173
6174         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6175                                            sizeof(*net->ipv6.ip6_null_entry),
6176                                            GFP_KERNEL);
6177         if (!net->ipv6.ip6_null_entry)
6178                 goto out_fib6_null_entry;
6179         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6180         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6181                          ip6_template_metrics, true);
6182         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6183
6184 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6185         net->ipv6.fib6_has_custom_rules = false;
6186         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6187                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6188                                                GFP_KERNEL);
6189         if (!net->ipv6.ip6_prohibit_entry)
6190                 goto out_ip6_null_entry;
6191         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6192         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6193                          ip6_template_metrics, true);
6194         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6195
6196         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6197                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6198                                                GFP_KERNEL);
6199         if (!net->ipv6.ip6_blk_hole_entry)
6200                 goto out_ip6_prohibit_entry;
6201         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6202         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6203                          ip6_template_metrics, true);
6204         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6205 #endif
6206
6207         net->ipv6.sysctl.flush_delay = 0;
6208         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6209         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6210         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6211         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6212         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6213         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6214         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6215         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6216
6217         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6218
6219         ret = 0;
6220 out:
6221         return ret;
6222
6223 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6224 out_ip6_prohibit_entry:
6225         kfree(net->ipv6.ip6_prohibit_entry);
6226 out_ip6_null_entry:
6227         kfree(net->ipv6.ip6_null_entry);
6228 #endif
6229 out_fib6_null_entry:
6230         kfree(net->ipv6.fib6_null_entry);
6231 out_ip6_dst_entries:
6232         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6233 out_ip6_dst_ops:
6234         goto out;
6235 }
6236
6237 static void __net_exit ip6_route_net_exit(struct net *net)
6238 {
6239         kfree(net->ipv6.fib6_null_entry);
6240         kfree(net->ipv6.ip6_null_entry);
6241 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6242         kfree(net->ipv6.ip6_prohibit_entry);
6243         kfree(net->ipv6.ip6_blk_hole_entry);
6244 #endif
6245         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6246 }
6247
6248 static int __net_init ip6_route_net_init_late(struct net *net)
6249 {
6250 #ifdef CONFIG_PROC_FS
6251         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6252                         sizeof(struct ipv6_route_iter));
6253         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6254                         rt6_stats_seq_show, NULL);
6255 #endif
6256         return 0;
6257 }
6258
6259 static void __net_exit ip6_route_net_exit_late(struct net *net)
6260 {
6261 #ifdef CONFIG_PROC_FS
6262         remove_proc_entry("ipv6_route", net->proc_net);
6263         remove_proc_entry("rt6_stats", net->proc_net);
6264 #endif
6265 }
6266
6267 static struct pernet_operations ip6_route_net_ops = {
6268         .init = ip6_route_net_init,
6269         .exit = ip6_route_net_exit,
6270 };
6271
6272 static int __net_init ipv6_inetpeer_init(struct net *net)
6273 {
6274         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6275
6276         if (!bp)
6277                 return -ENOMEM;
6278         inet_peer_base_init(bp);
6279         net->ipv6.peers = bp;
6280         return 0;
6281 }
6282
6283 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6284 {
6285         struct inet_peer_base *bp = net->ipv6.peers;
6286
6287         net->ipv6.peers = NULL;
6288         inetpeer_invalidate_tree(bp);
6289         kfree(bp);
6290 }
6291
6292 static struct pernet_operations ipv6_inetpeer_ops = {
6293         .init   =       ipv6_inetpeer_init,
6294         .exit   =       ipv6_inetpeer_exit,
6295 };
6296
6297 static struct pernet_operations ip6_route_net_late_ops = {
6298         .init = ip6_route_net_init_late,
6299         .exit = ip6_route_net_exit_late,
6300 };
6301
6302 static struct notifier_block ip6_route_dev_notifier = {
6303         .notifier_call = ip6_route_dev_notify,
6304         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6305 };
6306
6307 void __init ip6_route_init_special_entries(void)
6308 {
6309         /* Registering of the loopback is done before this portion of code,
6310          * the loopback reference in rt6_info will not be taken, do it
6311          * manually for init_net */
6312         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6313         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6314         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6315   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6316         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6317         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6318         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6319         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6320   #endif
6321 }
6322
6323 int __init ip6_route_init(void)
6324 {
6325         int ret;
6326         int cpu;
6327
6328         ret = -ENOMEM;
6329         ip6_dst_ops_template.kmem_cachep =
6330                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6331                                   SLAB_HWCACHE_ALIGN, NULL);
6332         if (!ip6_dst_ops_template.kmem_cachep)
6333                 goto out;
6334
6335         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6336         if (ret)
6337                 goto out_kmem_cache;
6338
6339         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6340         if (ret)
6341                 goto out_dst_entries;
6342
6343         ret = register_pernet_subsys(&ip6_route_net_ops);
6344         if (ret)
6345                 goto out_register_inetpeer;
6346
6347         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6348
6349         ret = fib6_init();
6350         if (ret)
6351                 goto out_register_subsys;
6352
6353         ret = xfrm6_init();
6354         if (ret)
6355                 goto out_fib6_init;
6356
6357         ret = fib6_rules_init();
6358         if (ret)
6359                 goto xfrm6_init;
6360
6361         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6362         if (ret)
6363                 goto fib6_rules_init;
6364
6365         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6366                                    inet6_rtm_newroute, NULL, 0);
6367         if (ret < 0)
6368                 goto out_register_late_subsys;
6369
6370         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6371                                    inet6_rtm_delroute, NULL, 0);
6372         if (ret < 0)
6373                 goto out_register_late_subsys;
6374
6375         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6376                                    inet6_rtm_getroute, NULL,
6377                                    RTNL_FLAG_DOIT_UNLOCKED);
6378         if (ret < 0)
6379                 goto out_register_late_subsys;
6380
6381         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6382         if (ret)
6383                 goto out_register_late_subsys;
6384
6385         for_each_possible_cpu(cpu) {
6386                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6387
6388                 INIT_LIST_HEAD(&ul->head);
6389                 spin_lock_init(&ul->lock);
6390         }
6391
6392 out:
6393         return ret;
6394
6395 out_register_late_subsys:
6396         rtnl_unregister_all(PF_INET6);
6397         unregister_pernet_subsys(&ip6_route_net_late_ops);
6398 fib6_rules_init:
6399         fib6_rules_cleanup();
6400 xfrm6_init:
6401         xfrm6_fini();
6402 out_fib6_init:
6403         fib6_gc_cleanup();
6404 out_register_subsys:
6405         unregister_pernet_subsys(&ip6_route_net_ops);
6406 out_register_inetpeer:
6407         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6408 out_dst_entries:
6409         dst_entries_destroy(&ip6_dst_blackhole_ops);
6410 out_kmem_cache:
6411         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6412         goto out;
6413 }
6414
6415 void ip6_route_cleanup(void)
6416 {
6417         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6418         unregister_pernet_subsys(&ip6_route_net_late_ops);
6419         fib6_rules_cleanup();
6420         xfrm6_fini();
6421         fib6_gc_cleanup();
6422         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6423         unregister_pernet_subsys(&ip6_route_net_ops);
6424         dst_entries_destroy(&ip6_dst_blackhole_ops);
6425         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6426 }