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