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 = loopback_dev;
 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(&rt->rt6i_gateway, dst->dev, skb, daddr);
 222 }
 223
 224 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 225 {
 226         struct net_device *dev = dst->dev;
 227         struct rt6_info *rt = (struct rt6_info *)dst;
 228
 229         daddr = choose_neigh_daddr(&rt->rt6i_gateway, NULL, daddr);
 230         if (!daddr)
 231                 return;
 232         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 233                 return;
 234         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 235                 return;
 236         __ipv6_confirm_neigh(dev, daddr);
 237 }
 238
 239 static struct dst_ops ip6_dst_ops_template = {
 240         .family                 =       AF_INET6,
 241         .gc                     =       ip6_dst_gc,
 242         .gc_thresh              =       1024,
 243         .check                  =       ip6_dst_check,
 244         .default_advmss         =       ip6_default_advmss,
 245         .mtu                    =       ip6_mtu,
 246         .cow_metrics            =       dst_cow_metrics_generic,
 247         .destroy                =       ip6_dst_destroy,
 248         .ifdown                 =       ip6_dst_ifdown,
 249         .negative_advice        =       ip6_negative_advice,
 250         .link_failure           =       ip6_link_failure,
 251         .update_pmtu            =       ip6_rt_update_pmtu,
 252         .redirect               =       rt6_do_redirect,
 253         .local_out              =       __ip6_local_out,
 254         .neigh_lookup           =       ip6_dst_neigh_lookup,
 255         .confirm_neigh          =       ip6_confirm_neigh,
 256 };
 257
 258 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 259 {
 260         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 261
 262         return mtu ? : dst->dev->mtu;
 263 }
 264
 265 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 266                                          struct sk_buff *skb, u32 mtu)
 267 {
 268 }
 269
 270 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 271                                       struct sk_buff *skb)
 272 {
 273 }
 274
 275 static struct dst_ops ip6_dst_blackhole_ops = {
 276         .family                 =       AF_INET6,
 277         .destroy                =       ip6_dst_destroy,
 278         .check                  =       ip6_dst_check,
 279         .mtu                    =       ip6_blackhole_mtu,
 280         .default_advmss         =       ip6_default_advmss,
 281         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 282         .redirect               =       ip6_rt_blackhole_redirect,
 283         .cow_metrics            =       dst_cow_metrics_generic,
 284         .neigh_lookup           =       ip6_dst_neigh_lookup,
 285 };
 286
 287 static const u32 ip6_template_metrics[RTAX_MAX] = {
 288         [RTAX_HOPLIMIT - 1] = 0,
 289 };
 290
 291 static const struct fib6_info fib6_null_entry_template = {
 292         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 293         .fib6_protocol  = RTPROT_KERNEL,
 294         .fib6_metric    = ~(u32)0,
 295         .fib6_ref       = REFCOUNT_INIT(1),
 296         .fib6_type      = RTN_UNREACHABLE,
 297         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 298 };
 299
 300 static const struct rt6_info ip6_null_entry_template = {
 301         .dst = {
 302                 .__refcnt       = ATOMIC_INIT(1),
 303                 .__use          = 1,
 304                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 305                 .error          = -ENETUNREACH,
 306                 .input          = ip6_pkt_discard,
 307                 .output         = ip6_pkt_discard_out,
 308         },
 309         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 310 };
 311
 312 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 313
 314 static const struct rt6_info ip6_prohibit_entry_template = {
 315         .dst = {
 316                 .__refcnt       = ATOMIC_INIT(1),
 317                 .__use          = 1,
 318                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 319                 .error          = -EACCES,
 320                 .input          = ip6_pkt_prohibit,
 321                 .output         = ip6_pkt_prohibit_out,
 322         },
 323         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 324 };
 325
 326 static const struct rt6_info ip6_blk_hole_entry_template = {
 327         .dst = {
 328                 .__refcnt       = ATOMIC_INIT(1),
 329                 .__use          = 1,
 330                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 331                 .error          = -EINVAL,
 332                 .input          = dst_discard,
 333                 .output         = dst_discard_out,
 334         },
 335         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 336 };
 337
 338 #endif
 339
 340 static void rt6_info_init(struct rt6_info *rt)
 341 {
 342         struct dst_entry *dst = &rt->dst;
 343
 344         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 345         INIT_LIST_HEAD(&rt->rt6i_uncached);
 346 }
 347
 348 /* allocate dst with ip6_dst_ops */
 349 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 350                                int flags)
 351 {
 352         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 353                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 354
 355         if (rt) {
 356                 rt6_info_init(rt);
 357                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 358         }
 359
 360         return rt;
 361 }
 362 EXPORT_SYMBOL(ip6_dst_alloc);
 363
 364 static void ip6_dst_destroy(struct dst_entry *dst)
 365 {
 366         struct rt6_info *rt = (struct rt6_info *)dst;
 367         struct fib6_info *from;
 368         struct inet6_dev *idev;
 369
 370         ip_dst_metrics_put(dst);
 371         rt6_uncached_list_del(rt);
 372
 373         idev = rt->rt6i_idev;
 374         if (idev) {
 375                 rt->rt6i_idev = NULL;
 376                 in6_dev_put(idev);
 377         }
 378
 379         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 380         fib6_info_release(from);
 381 }
 382
 383 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 384                            int how)
 385 {
 386         struct rt6_info *rt = (struct rt6_info *)dst;
 387         struct inet6_dev *idev = rt->rt6i_idev;
 388         struct net_device *loopback_dev =
 389                 dev_net(dev)->loopback_dev;
 390
 391         if (idev && idev->dev != loopback_dev) {
 392                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 393                 if (loopback_idev) {
 394                         rt->rt6i_idev = loopback_idev;
 395                         in6_dev_put(idev);
 396                 }
 397         }
 398 }
 399
 400 static bool __rt6_check_expired(const struct rt6_info *rt)
 401 {
 402         if (rt->rt6i_flags & RTF_EXPIRES)
 403                 return time_after(jiffies, rt->dst.expires);
 404         else
 405                 return false;
 406 }
 407
 408 static bool rt6_check_expired(const struct rt6_info *rt)
 409 {
 410         struct fib6_info *from;
 411
 412         from = rcu_dereference(rt->from);
 413
 414         if (rt->rt6i_flags & RTF_EXPIRES) {
 415                 if (time_after(jiffies, rt->dst.expires))
 416                         return true;
 417         } else if (from) {
 418                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 419                         fib6_check_expired(from);
 420         }
 421         return false;
 422 }
 423
 424 void fib6_select_path(const struct net *net, struct fib6_result *res,
 425                       struct flowi6 *fl6, int oif, bool have_oif_match,
 426                       const struct sk_buff *skb, int strict)
 427 {
 428         struct fib6_info *sibling, *next_sibling;
 429         struct fib6_info *match = res->f6i;
 430
 431         if ((!match->fib6_nsiblings && !match->nh) || have_oif_match)
 432                 goto out;
 433
 434         /* We might have already computed the hash for ICMPv6 errors. In such
 435          * case it will always be non-zero. Otherwise now is the time to do it.
 436          */
 437         if (!fl6->mp_hash &&
 438             (!match->nh || nexthop_is_multipath(match->nh)))
 439                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 440
 441         if (unlikely(match->nh)) {
 442                 nexthop_path_fib6_result(res, fl6->mp_hash);
 443                 return;
 444         }
 445
 446         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 447                 goto out;
 448
 449         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 450                                  fib6_siblings) {
 451                 const struct fib6_nh *nh = sibling->fib6_nh;
 452                 int nh_upper_bound;
 453
 454                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 455                 if (fl6->mp_hash > nh_upper_bound)
 456                         continue;
 457                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 458                         break;
 459                 match = sibling;
 460                 break;
 461         }
 462
 463 out:
 464         res->f6i = match;
 465         res->nh = match->fib6_nh;
 466 }
 467
 468 /*
 469  *      Route lookup. rcu_read_lock() should be held.
 470  */
 471
 472 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 473                                const struct in6_addr *saddr, int oif, int flags)
 474 {
 475         const struct net_device *dev;
 476
 477         if (nh->fib_nh_flags & RTNH_F_DEAD)
 478                 return false;
 479
 480         dev = nh->fib_nh_dev;
 481         if (oif) {
 482                 if (dev->ifindex == oif)
 483                         return true;
 484         } else {
 485                 if (ipv6_chk_addr(net, saddr, dev,
 486                                   flags & RT6_LOOKUP_F_IFACE))
 487                         return true;
 488         }
 489
 490         return false;
 491 }
 492
 493 struct fib6_nh_dm_arg {
 494         struct net              *net;
 495         const struct in6_addr   *saddr;
 496         int                     oif;
 497         int                     flags;
 498         struct fib6_nh          *nh;
 499 };
 500
 501 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 502 {
 503         struct fib6_nh_dm_arg *arg = _arg;
 504
 505         arg->nh = nh;
 506         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 507                                   arg->flags);
 508 }
 509
 510 /* returns fib6_nh from nexthop or NULL */
 511 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 512                                         struct fib6_result *res,
 513                                         const struct in6_addr *saddr,
 514                                         int oif, int flags)
 515 {
 516         struct fib6_nh_dm_arg arg = {
 517                 .net   = net,
 518                 .saddr = saddr,
 519                 .oif   = oif,
 520                 .flags = flags,
 521         };
 522
 523         if (nexthop_is_blackhole(nh))
 524                 return NULL;
 525
 526         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 527                 return arg.nh;
 528
 529         return NULL;
 530 }
 531
 532 static void rt6_device_match(struct net *net, struct fib6_result *res,
 533                              const struct in6_addr *saddr, int oif, int flags)
 534 {
 535         struct fib6_info *f6i = res->f6i;
 536         struct fib6_info *spf6i;
 537         struct fib6_nh *nh;
 538
 539         if (!oif && ipv6_addr_any(saddr)) {
 540                 if (unlikely(f6i->nh)) {
 541                         nh = nexthop_fib6_nh(f6i->nh);
 542                         if (nexthop_is_blackhole(f6i->nh))
 543                                 goto out_blackhole;
 544                 } else {
 545                         nh = f6i->fib6_nh;
 546                 }
 547                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 548                         goto out;
 549         }
 550
 551         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 552                 bool matched = false;
 553
 554                 if (unlikely(spf6i->nh)) {
 555                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 556                                               oif, flags);
 557                         if (nh)
 558                                 matched = true;
 559                 } else {
 560                         nh = spf6i->fib6_nh;
 561                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 562                                 matched = true;
 563                 }
 564                 if (matched) {
 565                         res->f6i = spf6i;
 566                         goto out;
 567                 }
 568         }
 569
 570         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 571                 res->f6i = net->ipv6.fib6_null_entry;
 572                 nh = res->f6i->fib6_nh;
 573                 goto out;
 574         }
 575
 576         if (unlikely(f6i->nh)) {
 577                 nh = nexthop_fib6_nh(f6i->nh);
 578                 if (nexthop_is_blackhole(f6i->nh))
 579                         goto out_blackhole;
 580         } else {
 581                 nh = f6i->fib6_nh;
 582         }
 583
 584         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 585                 res->f6i = net->ipv6.fib6_null_entry;
 586                 nh = res->f6i->fib6_nh;
 587         }
 588 out:
 589         res->nh = nh;
 590         res->fib6_type = res->f6i->fib6_type;
 591         res->fib6_flags = res->f6i->fib6_flags;
 592         return;
 593
 594 out_blackhole:
 595         res->fib6_flags |= RTF_REJECT;
 596         res->fib6_type = RTN_BLACKHOLE;
 597         res->nh = nh;
 598 }
 599
 600 #ifdef CONFIG_IPV6_ROUTER_PREF
 601 struct __rt6_probe_work {
 602         struct work_struct work;
 603         struct in6_addr target;
 604         struct net_device *dev;
 605 };
 606
 607 static void rt6_probe_deferred(struct work_struct *w)
 608 {
 609         struct in6_addr mcaddr;
 610         struct __rt6_probe_work *work =
 611                 container_of(w, struct __rt6_probe_work, work);
 612
 613         addrconf_addr_solict_mult(&work->target, &mcaddr);
 614         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 615         dev_put(work->dev);
 616         kfree(work);
 617 }
 618
 619 static void rt6_probe(struct fib6_nh *fib6_nh)
 620 {
 621         struct __rt6_probe_work *work = NULL;
 622         const struct in6_addr *nh_gw;
 623         struct neighbour *neigh;
 624         struct net_device *dev;
 625         struct inet6_dev *idev;
 626
 627         /*
 628          * Okay, this does not seem to be appropriate
 629          * for now, however, we need to check if it
 630          * is really so; aka Router Reachability Probing.
 631          *
 632          * Router Reachability Probe MUST be rate-limited
 633          * to no more than one per minute.
 634          */
 635         if (fib6_nh->fib_nh_gw_family)
 636                 return;
 637
 638         nh_gw = &fib6_nh->fib_nh_gw6;
 639         dev = fib6_nh->fib_nh_dev;
 640         rcu_read_lock_bh();
 641         idev = __in6_dev_get(dev);
 642         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 643         if (neigh) {
 644                 if (neigh->nud_state & NUD_VALID)
 645                         goto out;
 646
 647                 write_lock(&neigh->lock);
 648                 if (!(neigh->nud_state & NUD_VALID) &&
 649                     time_after(jiffies,
 650                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 651                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 652                         if (work)
 653                                 __neigh_set_probe_once(neigh);
 654                 }
 655                 write_unlock(&neigh->lock);
 656         } else if (time_after(jiffies, fib6_nh->last_probe +
 657                                        idev->cnf.rtr_probe_interval)) {
 658                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 659         }
 660
 661         if (work) {
 662                 fib6_nh->last_probe = jiffies;
 663                 INIT_WORK(&work->work, rt6_probe_deferred);
 664                 work->target = *nh_gw;
 665                 dev_hold(dev);
 666                 work->dev = dev;
 667                 schedule_work(&work->work);
 668         }
 669
 670 out:
 671         rcu_read_unlock_bh();
 672 }
 673 #else
 674 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 675 {
 676 }
 677 #endif
 678
 679 /*
 680  * Default Router Selection (RFC 2461 6.3.6)
 681  */
 682 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 683 {
 684         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 685         struct neighbour *neigh;
 686
 687         rcu_read_lock_bh();
 688         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 689                                           &fib6_nh->fib_nh_gw6);
 690         if (neigh) {
 691                 read_lock(&neigh->lock);
 692                 if (neigh->nud_state & NUD_VALID)
 693                         ret = RT6_NUD_SUCCEED;
 694 #ifdef CONFIG_IPV6_ROUTER_PREF
 695                 else if (!(neigh->nud_state & NUD_FAILED))
 696                         ret = RT6_NUD_SUCCEED;
 697                 else
 698                         ret = RT6_NUD_FAIL_PROBE;
 699 #endif
 700                 read_unlock(&neigh->lock);
 701         } else {
 702                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 703                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 704         }
 705         rcu_read_unlock_bh();
 706
 707         return ret;
 708 }
 709
 710 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 711                            int strict)
 712 {
 713         int m = 0;
 714
 715         if (!oif || nh->fib_nh_dev->ifindex == oif)
 716                 m = 2;
 717
 718         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 719                 return RT6_NUD_FAIL_HARD;
 720 #ifdef CONFIG_IPV6_ROUTER_PREF
 721         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 722 #endif
 723         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 724             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 725                 int n = rt6_check_neigh(nh);
 726                 if (n < 0)
 727                         return n;
 728         }
 729         return m;
 730 }
 731
 732 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 733                        int oif, int strict, int *mpri, bool *do_rr)
 734 {
 735         bool match_do_rr = false;
 736         bool rc = false;
 737         int m;
 738
 739         if (nh->fib_nh_flags & RTNH_F_DEAD)
 740                 goto out;
 741
 742         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 743             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 744             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 745                 goto out;
 746
 747         m = rt6_score_route(nh, fib6_flags, oif, strict);
 748         if (m == RT6_NUD_FAIL_DO_RR) {
 749                 match_do_rr = true;
 750                 m = 0; /* lowest valid score */
 751         } else if (m == RT6_NUD_FAIL_HARD) {
 752                 goto out;
 753         }
 754
 755         if (strict & RT6_LOOKUP_F_REACHABLE)
 756                 rt6_probe(nh);
 757
 758         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 759         if (m > *mpri) {
 760                 *do_rr = match_do_rr;
 761                 *mpri = m;
 762                 rc = true;
 763         }
 764 out:
 765         return rc;
 766 }
 767
 768 struct fib6_nh_frl_arg {
 769         u32             flags;
 770         int             oif;
 771         int             strict;
 772         int             *mpri;
 773         bool            *do_rr;
 774         struct fib6_nh  *nh;
 775 };
 776
 777 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 778 {
 779         struct fib6_nh_frl_arg *arg = _arg;
 780
 781         arg->nh = nh;
 782         return find_match(nh, arg->flags, arg->oif, arg->strict,
 783                           arg->mpri, arg->do_rr);
 784 }
 785
 786 static void __find_rr_leaf(struct fib6_info *f6i_start,
 787                            struct fib6_info *nomatch, u32 metric,
 788                            struct fib6_result *res, struct fib6_info **cont,
 789                            int oif, int strict, bool *do_rr, int *mpri)
 790 {
 791         struct fib6_info *f6i;
 792
 793         for (f6i = f6i_start;
 794              f6i && f6i != nomatch;
 795              f6i = rcu_dereference(f6i->fib6_next)) {
 796                 bool matched = false;
 797                 struct fib6_nh *nh;
 798
 799                 if (cont && f6i->fib6_metric != metric) {
 800                         *cont = f6i;
 801                         return;
 802                 }
 803
 804                 if (fib6_check_expired(f6i))
 805                         continue;
 806
 807                 if (unlikely(f6i->nh)) {
 808                         struct fib6_nh_frl_arg arg = {
 809                                 .flags  = f6i->fib6_flags,
 810                                 .oif    = oif,
 811                                 .strict = strict,
 812                                 .mpri   = mpri,
 813                                 .do_rr  = do_rr
 814                         };
 815
 816                         if (nexthop_is_blackhole(f6i->nh)) {
 817                                 res->fib6_flags = RTF_REJECT;
 818                                 res->fib6_type = RTN_BLACKHOLE;
 819                                 res->f6i = f6i;
 820                                 res->nh = nexthop_fib6_nh(f6i->nh);
 821                                 return;
 822                         }
 823                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 824                                                      &arg)) {
 825                                 matched = true;
 826                                 nh = arg.nh;
 827                         }
 828                 } else {
 829                         nh = f6i->fib6_nh;
 830                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 831                                        mpri, do_rr))
 832                                 matched = true;
 833                 }
 834                 if (matched) {
 835                         res->f6i = f6i;
 836                         res->nh = nh;
 837                         res->fib6_flags = f6i->fib6_flags;
 838                         res->fib6_type = f6i->fib6_type;
 839                 }
 840         }
 841 }
 842
 843 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 844                          struct fib6_info *rr_head, int oif, int strict,
 845                          bool *do_rr, struct fib6_result *res)
 846 {
 847         u32 metric = rr_head->fib6_metric;
 848         struct fib6_info *cont = NULL;
 849         int mpri = -1;
 850
 851         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 852                        oif, strict, do_rr, &mpri);
 853
 854         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 855                        oif, strict, do_rr, &mpri);
 856
 857         if (res->f6i || !cont)
 858                 return;
 859
 860         __find_rr_leaf(cont, NULL, metric, res, NULL,
 861                        oif, strict, do_rr, &mpri);
 862 }
 863
 864 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 865                        struct fib6_result *res, int strict)
 866 {
 867         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 868         struct fib6_info *rt0;
 869         bool do_rr = false;
 870         int key_plen;
 871
 872         /* make sure this function or its helpers sets f6i */
 873         res->f6i = NULL;
 874
 875         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 876                 goto out;
 877
 878         rt0 = rcu_dereference(fn->rr_ptr);
 879         if (!rt0)
 880                 rt0 = leaf;
 881
 882         /* Double check to make sure fn is not an intermediate node
 883          * and fn->leaf does not points to its child's leaf
 884          * (This might happen if all routes under fn are deleted from
 885          * the tree and fib6_repair_tree() is called on the node.)
 886          */
 887         key_plen = rt0->fib6_dst.plen;
 888 #ifdef CONFIG_IPV6_SUBTREES
 889         if (rt0->fib6_src.plen)
 890                 key_plen = rt0->fib6_src.plen;
 891 #endif
 892         if (fn->fn_bit != key_plen)
 893                 goto out;
 894
 895         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 896         if (do_rr) {
 897                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 898
 899                 /* no entries matched; do round-robin */
 900                 if (!next || next->fib6_metric != rt0->fib6_metric)
 901                         next = leaf;
 902
 903                 if (next != rt0) {
 904                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 905                         /* make sure next is not being deleted from the tree */
 906                         if (next->fib6_node)
 907                                 rcu_assign_pointer(fn->rr_ptr, next);
 908                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 909                 }
 910         }
 911
 912 out:
 913         if (!res->f6i) {
 914                 res->f6i = net->ipv6.fib6_null_entry;
 915                 res->nh = res->f6i->fib6_nh;
 916                 res->fib6_flags = res->f6i->fib6_flags;
 917                 res->fib6_type = res->f6i->fib6_type;
 918         }
 919 }
 920
 921 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 922 {
 923         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 924                res->nh->fib_nh_gw_family;
 925 }
 926
 927 #ifdef CONFIG_IPV6_ROUTE_INFO
 928 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 929                   const struct in6_addr *gwaddr)
 930 {
 931         struct net *net = dev_net(dev);
 932         struct route_info *rinfo = (struct route_info *) opt;
 933         struct in6_addr prefix_buf, *prefix;
 934         unsigned int pref;
 935         unsigned long lifetime;
 936         struct fib6_info *rt;
 937
 938         if (len < sizeof(struct route_info)) {
 939                 return -EINVAL;
 940         }
 941
 942         /* Sanity check for prefix_len and length */
 943         if (rinfo->length > 3) {
 944                 return -EINVAL;
 945         } else if (rinfo->prefix_len > 128) {
 946                 return -EINVAL;
 947         } else if (rinfo->prefix_len > 64) {
 948                 if (rinfo->length < 2) {
 949                         return -EINVAL;
 950                 }
 951         } else if (rinfo->prefix_len > 0) {
 952                 if (rinfo->length < 1) {
 953                         return -EINVAL;
 954                 }
 955         }
 956
 957         pref = rinfo->route_pref;
 958         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 959                 return -EINVAL;
 960
 961         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 962
 963         if (rinfo->length == 3)
 964                 prefix = (struct in6_addr *)rinfo->prefix;
 965         else {
 966                 /* this function is safe */
 967                 ipv6_addr_prefix(&prefix_buf,
 968                                  (struct in6_addr *)rinfo->prefix,
 969                                  rinfo->prefix_len);
 970                 prefix = &prefix_buf;
 971         }
 972
 973         if (rinfo->prefix_len == 0)
 974                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 975         else
 976                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 977                                         gwaddr, dev);
 978
 979         if (rt && !lifetime) {
 980                 ip6_del_rt(net, rt);
 981                 rt = NULL;
 982         }
 983
 984         if (!rt && lifetime)
 985                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 986                                         dev, pref);
 987         else if (rt)
 988                 rt->fib6_flags = RTF_ROUTEINFO |
 989                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 990
 991         if (rt) {
 992                 if (!addrconf_finite_timeout(lifetime))
 993                         fib6_clean_expires(rt);
 994                 else
 995                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 996
 997                 fib6_info_release(rt);
 998         }
 999         return 0;
1000 }
1001 #endif
1002
1003 /*
1004  *      Misc support functions
1005  */
1006
1007 /* called with rcu_lock held */
1008 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1009 {
1010         struct net_device *dev = res->nh->fib_nh_dev;
1011
1012         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1013                 /* for copies of local routes, dst->dev needs to be the
1014                  * device if it is a master device, the master device if
1015                  * device is enslaved, and the loopback as the default
1016                  */
1017                 if (netif_is_l3_slave(dev) &&
1018                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1019                         dev = l3mdev_master_dev_rcu(dev);
1020                 else if (!netif_is_l3_master(dev))
1021                         dev = dev_net(dev)->loopback_dev;
1022                 /* last case is netif_is_l3_master(dev) is true in which
1023                  * case we want dev returned to be dev
1024                  */
1025         }
1026
1027         return dev;
1028 }
1029
1030 static const int fib6_prop[RTN_MAX + 1] = {
1031         [RTN_UNSPEC]    = 0,
1032         [RTN_UNICAST]   = 0,
1033         [RTN_LOCAL]     = 0,
1034         [RTN_BROADCAST] = 0,
1035         [RTN_ANYCAST]   = 0,
1036         [RTN_MULTICAST] = 0,
1037         [RTN_BLACKHOLE] = -EINVAL,
1038         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1039         [RTN_PROHIBIT]  = -EACCES,
1040         [RTN_THROW]     = -EAGAIN,
1041         [RTN_NAT]       = -EINVAL,
1042         [RTN_XRESOLVE]  = -EINVAL,
1043 };
1044
1045 static int ip6_rt_type_to_error(u8 fib6_type)
1046 {
1047         return fib6_prop[fib6_type];
1048 }
1049
1050 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1051 {
1052         unsigned short flags = 0;
1053
1054         if (rt->dst_nocount)
1055                 flags |= DST_NOCOUNT;
1056         if (rt->dst_nopolicy)
1057                 flags |= DST_NOPOLICY;
1058         if (rt->dst_host)
1059                 flags |= DST_HOST;
1060
1061         return flags;
1062 }
1063
1064 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1065 {
1066         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1067
1068         switch (fib6_type) {
1069         case RTN_BLACKHOLE:
1070                 rt->dst.output = dst_discard_out;
1071                 rt->dst.input = dst_discard;
1072                 break;
1073         case RTN_PROHIBIT:
1074                 rt->dst.output = ip6_pkt_prohibit_out;
1075                 rt->dst.input = ip6_pkt_prohibit;
1076                 break;
1077         case RTN_THROW:
1078         case RTN_UNREACHABLE:
1079         default:
1080                 rt->dst.output = ip6_pkt_discard_out;
1081                 rt->dst.input = ip6_pkt_discard;
1082                 break;
1083         }
1084 }
1085
1086 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1087 {
1088         struct fib6_info *f6i = res->f6i;
1089
1090         if (res->fib6_flags & RTF_REJECT) {
1091                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1092                 return;
1093         }
1094
1095         rt->dst.error = 0;
1096         rt->dst.output = ip6_output;
1097
1098         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1099                 rt->dst.input = ip6_input;
1100         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1101                 rt->dst.input = ip6_mc_input;
1102         } else {
1103                 rt->dst.input = ip6_forward;
1104         }
1105
1106         if (res->nh->fib_nh_lws) {
1107                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1108                 lwtunnel_set_redirect(&rt->dst);
1109         }
1110
1111         rt->dst.lastuse = jiffies;
1112 }
1113
1114 /* Caller must already hold reference to @from */
1115 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1116 {
1117         rt->rt6i_flags &= ~RTF_EXPIRES;
1118         rcu_assign_pointer(rt->from, from);
1119         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1120 }
1121
1122 /* Caller must already hold reference to f6i in result */
1123 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1124 {
1125         const struct fib6_nh *nh = res->nh;
1126         const struct net_device *dev = nh->fib_nh_dev;
1127         struct fib6_info *f6i = res->f6i;
1128
1129         ip6_rt_init_dst(rt, res);
1130
1131         rt->rt6i_dst = f6i->fib6_dst;
1132         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1133         rt->rt6i_flags = res->fib6_flags;
1134         if (nh->fib_nh_gw_family) {
1135                 rt->rt6i_gateway = nh->fib_nh_gw6;
1136                 rt->rt6i_flags |= RTF_GATEWAY;
1137         }
1138         rt6_set_from(rt, f6i);
1139 #ifdef CONFIG_IPV6_SUBTREES
1140         rt->rt6i_src = f6i->fib6_src;
1141 #endif
1142 }
1143
1144 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1145                                         struct in6_addr *saddr)
1146 {
1147         struct fib6_node *pn, *sn;
1148         while (1) {
1149                 if (fn->fn_flags & RTN_TL_ROOT)
1150                         return NULL;
1151                 pn = rcu_dereference(fn->parent);
1152                 sn = FIB6_SUBTREE(pn);
1153                 if (sn && sn != fn)
1154                         fn = fib6_node_lookup(sn, NULL, saddr);
1155                 else
1156                         fn = pn;
1157                 if (fn->fn_flags & RTN_RTINFO)
1158                         return fn;
1159         }
1160 }
1161
1162 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1163 {
1164         struct rt6_info *rt = *prt;
1165
1166         if (dst_hold_safe(&rt->dst))
1167                 return true;
1168         if (net) {
1169                 rt = net->ipv6.ip6_null_entry;
1170                 dst_hold(&rt->dst);
1171         } else {
1172                 rt = NULL;
1173         }
1174         *prt = rt;
1175         return false;
1176 }
1177
1178 /* called with rcu_lock held */
1179 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1180 {
1181         struct net_device *dev = res->nh->fib_nh_dev;
1182         struct fib6_info *f6i = res->f6i;
1183         unsigned short flags;
1184         struct rt6_info *nrt;
1185
1186         if (!fib6_info_hold_safe(f6i))
1187                 goto fallback;
1188
1189         flags = fib6_info_dst_flags(f6i);
1190         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1191         if (!nrt) {
1192                 fib6_info_release(f6i);
1193                 goto fallback;
1194         }
1195
1196         ip6_rt_copy_init(nrt, res);
1197         return nrt;
1198
1199 fallback:
1200         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1201         dst_hold(&nrt->dst);
1202         return nrt;
1203 }
1204
1205 static struct rt6_info *ip6_pol_route_lookup(struct net *net,
1206                                              struct fib6_table *table,
1207                                              struct flowi6 *fl6,
1208                                              const struct sk_buff *skb,
1209                                              int flags)
1210 {
1211         struct fib6_result res = {};
1212         struct fib6_node *fn;
1213         struct rt6_info *rt;
1214
1215         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1216                 flags &= ~RT6_LOOKUP_F_IFACE;
1217
1218         rcu_read_lock();
1219         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1220 restart:
1221         res.f6i = rcu_dereference(fn->leaf);
1222         if (!res.f6i)
1223                 res.f6i = net->ipv6.fib6_null_entry;
1224         else
1225                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1226                                  flags);
1227
1228         if (res.f6i == net->ipv6.fib6_null_entry) {
1229                 fn = fib6_backtrack(fn, &fl6->saddr);
1230                 if (fn)
1231                         goto restart;
1232
1233                 rt = net->ipv6.ip6_null_entry;
1234                 dst_hold(&rt->dst);
1235                 goto out;
1236         } else if (res.fib6_flags & RTF_REJECT) {
1237                 goto do_create;
1238         }
1239
1240         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1241                          fl6->flowi6_oif != 0, skb, flags);
1242
1243         /* Search through exception table */
1244         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1245         if (rt) {
1246                 if (ip6_hold_safe(net, &rt))
1247                         dst_use_noref(&rt->dst, jiffies);
1248         } else {
1249 do_create:
1250                 rt = ip6_create_rt_rcu(&res);
1251         }
1252
1253 out:
1254         trace_fib6_table_lookup(net, &res, table, fl6);
1255
1256         rcu_read_unlock();
1257
1258         return rt;
1259 }
1260
1261 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1262                                    const struct sk_buff *skb, int flags)
1263 {
1264         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1265 }
1266 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1267
1268 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1269                             const struct in6_addr *saddr, int oif,
1270                             const struct sk_buff *skb, int strict)
1271 {
1272         struct flowi6 fl6 = {
1273                 .flowi6_oif = oif,
1274                 .daddr = *daddr,
1275         };
1276         struct dst_entry *dst;
1277         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1278
1279         if (saddr) {
1280                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1281                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1282         }
1283
1284         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1285         if (dst->error == 0)
1286                 return (struct rt6_info *) dst;
1287
1288         dst_release(dst);
1289
1290         return NULL;
1291 }
1292 EXPORT_SYMBOL(rt6_lookup);
1293
1294 /* ip6_ins_rt is called with FREE table->tb6_lock.
1295  * It takes new route entry, the addition fails by any reason the
1296  * route is released.
1297  * Caller must hold dst before calling it.
1298  */
1299
1300 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1301                         struct netlink_ext_ack *extack)
1302 {
1303         int err;
1304         struct fib6_table *table;
1305
1306         table = rt->fib6_table;
1307         spin_lock_bh(&table->tb6_lock);
1308         err = fib6_add(&table->tb6_root, rt, info, extack);
1309         spin_unlock_bh(&table->tb6_lock);
1310
1311         return err;
1312 }
1313
1314 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1315 {
1316         struct nl_info info = { .nl_net = net, };
1317
1318         return __ip6_ins_rt(rt, &info, NULL);
1319 }
1320
1321 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1322                                            const struct in6_addr *daddr,
1323                                            const struct in6_addr *saddr)
1324 {
1325         struct fib6_info *f6i = res->f6i;
1326         struct net_device *dev;
1327         struct rt6_info *rt;
1328
1329         /*
1330          *      Clone the route.
1331          */
1332
1333         if (!fib6_info_hold_safe(f6i))
1334                 return NULL;
1335
1336         dev = ip6_rt_get_dev_rcu(res);
1337         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1338         if (!rt) {
1339                 fib6_info_release(f6i);
1340                 return NULL;
1341         }
1342
1343         ip6_rt_copy_init(rt, res);
1344         rt->rt6i_flags |= RTF_CACHE;
1345         rt->dst.flags |= DST_HOST;
1346         rt->rt6i_dst.addr = *daddr;
1347         rt->rt6i_dst.plen = 128;
1348
1349         if (!rt6_is_gw_or_nonexthop(res)) {
1350                 if (f6i->fib6_dst.plen != 128 &&
1351                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1352                         rt->rt6i_flags |= RTF_ANYCAST;
1353 #ifdef CONFIG_IPV6_SUBTREES
1354                 if (rt->rt6i_src.plen && saddr) {
1355                         rt->rt6i_src.addr = *saddr;
1356                         rt->rt6i_src.plen = 128;
1357                 }
1358 #endif
1359         }
1360
1361         return rt;
1362 }
1363
1364 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1365 {
1366         struct fib6_info *f6i = res->f6i;
1367         unsigned short flags = fib6_info_dst_flags(f6i);
1368         struct net_device *dev;
1369         struct rt6_info *pcpu_rt;
1370
1371         if (!fib6_info_hold_safe(f6i))
1372                 return NULL;
1373
1374         rcu_read_lock();
1375         dev = ip6_rt_get_dev_rcu(res);
1376         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags);
1377         rcu_read_unlock();
1378         if (!pcpu_rt) {
1379                 fib6_info_release(f6i);
1380                 return NULL;
1381         }
1382         ip6_rt_copy_init(pcpu_rt, res);
1383         pcpu_rt->rt6i_flags |= RTF_PCPU;
1384         return pcpu_rt;
1385 }
1386
1387 /* It should be called with rcu_read_lock() acquired */
1388 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1389 {
1390         struct rt6_info *pcpu_rt;
1391
1392         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1393
1394         return pcpu_rt;
1395 }
1396
1397 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1398                                             const struct fib6_result *res)
1399 {
1400         struct rt6_info *pcpu_rt, *prev, **p;
1401
1402         pcpu_rt = ip6_rt_pcpu_alloc(res);
1403         if (!pcpu_rt)
1404                 return NULL;
1405
1406         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1407         prev = cmpxchg(p, NULL, pcpu_rt);
1408         BUG_ON(prev);
1409
1410         if (res->f6i->fib6_destroying) {
1411                 struct fib6_info *from;
1412
1413                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1414                 fib6_info_release(from);
1415         }
1416
1417         return pcpu_rt;
1418 }
1419
1420 /* exception hash table implementation
1421  */
1422 static DEFINE_SPINLOCK(rt6_exception_lock);
1423
1424 /* Remove rt6_ex from hash table and free the memory
1425  * Caller must hold rt6_exception_lock
1426  */
1427 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1428                                  struct rt6_exception *rt6_ex)
1429 {
1430         struct fib6_info *from;
1431         struct net *net;
1432
1433         if (!bucket || !rt6_ex)
1434                 return;
1435
1436         net = dev_net(rt6_ex->rt6i->dst.dev);
1437         net->ipv6.rt6_stats->fib_rt_cache--;
1438
1439         /* purge completely the exception to allow releasing the held resources:
1440          * some [sk] cache may keep the dst around for unlimited time
1441          */
1442         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1443         fib6_info_release(from);
1444         dst_dev_put(&rt6_ex->rt6i->dst);
1445
1446         hlist_del_rcu(&rt6_ex->hlist);
1447         dst_release(&rt6_ex->rt6i->dst);
1448         kfree_rcu(rt6_ex, rcu);
1449         WARN_ON_ONCE(!bucket->depth);
1450         bucket->depth--;
1451 }
1452
1453 /* Remove oldest rt6_ex in bucket and free the memory
1454  * Caller must hold rt6_exception_lock
1455  */
1456 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1457 {
1458         struct rt6_exception *rt6_ex, *oldest = NULL;
1459
1460         if (!bucket)
1461                 return;
1462
1463         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1464                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1465                         oldest = rt6_ex;
1466         }
1467         rt6_remove_exception(bucket, oldest);
1468 }
1469
1470 static u32 rt6_exception_hash(const struct in6_addr *dst,
1471                               const struct in6_addr *src)
1472 {
1473         static u32 seed __read_mostly;
1474         u32 val;
1475
1476         net_get_random_once(&seed, sizeof(seed));
1477         val = jhash(dst, sizeof(*dst), seed);
1478
1479 #ifdef CONFIG_IPV6_SUBTREES
1480         if (src)
1481                 val = jhash(src, sizeof(*src), val);
1482 #endif
1483         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1484 }
1485
1486 /* Helper function to find the cached rt in the hash table
1487  * and update bucket pointer to point to the bucket for this
1488  * (daddr, saddr) pair
1489  * Caller must hold rt6_exception_lock
1490  */
1491 static struct rt6_exception *
1492 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1493                               const struct in6_addr *daddr,
1494                               const struct in6_addr *saddr)
1495 {
1496         struct rt6_exception *rt6_ex;
1497         u32 hval;
1498
1499         if (!(*bucket) || !daddr)
1500                 return NULL;
1501
1502         hval = rt6_exception_hash(daddr, saddr);
1503         *bucket += hval;
1504
1505         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1506                 struct rt6_info *rt6 = rt6_ex->rt6i;
1507                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1508
1509 #ifdef CONFIG_IPV6_SUBTREES
1510                 if (matched && saddr)
1511                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1512 #endif
1513                 if (matched)
1514                         return rt6_ex;
1515         }
1516         return NULL;
1517 }
1518
1519 /* Helper function to find the cached rt in the hash table
1520  * and update bucket pointer to point to the bucket for this
1521  * (daddr, saddr) pair
1522  * Caller must hold rcu_read_lock()
1523  */
1524 static struct rt6_exception *
1525 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1526                          const struct in6_addr *daddr,
1527                          const struct in6_addr *saddr)
1528 {
1529         struct rt6_exception *rt6_ex;
1530         u32 hval;
1531
1532         WARN_ON_ONCE(!rcu_read_lock_held());
1533
1534         if (!(*bucket) || !daddr)
1535                 return NULL;
1536
1537         hval = rt6_exception_hash(daddr, saddr);
1538         *bucket += hval;
1539
1540         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1541                 struct rt6_info *rt6 = rt6_ex->rt6i;
1542                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1543
1544 #ifdef CONFIG_IPV6_SUBTREES
1545                 if (matched && saddr)
1546                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1547 #endif
1548                 if (matched)
1549                         return rt6_ex;
1550         }
1551         return NULL;
1552 }
1553
1554 static unsigned int fib6_mtu(const struct fib6_result *res)
1555 {
1556         const struct fib6_nh *nh = res->nh;
1557         unsigned int mtu;
1558
1559         if (res->f6i->fib6_pmtu) {
1560                 mtu = res->f6i->fib6_pmtu;
1561         } else {
1562                 struct net_device *dev = nh->fib_nh_dev;
1563                 struct inet6_dev *idev;
1564
1565                 rcu_read_lock();
1566                 idev = __in6_dev_get(dev);
1567                 mtu = idev->cnf.mtu6;
1568                 rcu_read_unlock();
1569         }
1570
1571         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1572
1573         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1574 }
1575
1576 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1577
1578 /* used when the flushed bit is not relevant, only access to the bucket
1579  * (ie., all bucket users except rt6_insert_exception);
1580  *
1581  * called under rcu lock; sometimes called with rt6_exception_lock held
1582  */
1583 static
1584 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1585                                                        spinlock_t *lock)
1586 {
1587         struct rt6_exception_bucket *bucket;
1588
1589         if (lock)
1590                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1591                                                    lockdep_is_held(lock));
1592         else
1593                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1594
1595         /* remove bucket flushed bit if set */
1596         if (bucket) {
1597                 unsigned long p = (unsigned long)bucket;
1598
1599                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1600                 bucket = (struct rt6_exception_bucket *)p;
1601         }
1602
1603         return bucket;
1604 }
1605
1606 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1607 {
1608         unsigned long p = (unsigned long)bucket;
1609
1610         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1611 }
1612
1613 /* called with rt6_exception_lock held */
1614 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1615                                               spinlock_t *lock)
1616 {
1617         struct rt6_exception_bucket *bucket;
1618         unsigned long p;
1619
1620         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1621                                            lockdep_is_held(lock));
1622
1623         p = (unsigned long)bucket;
1624         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1625         bucket = (struct rt6_exception_bucket *)p;
1626         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1627 }
1628
1629 static int rt6_insert_exception(struct rt6_info *nrt,
1630                                 const struct fib6_result *res)
1631 {
1632         struct net *net = dev_net(nrt->dst.dev);
1633         struct rt6_exception_bucket *bucket;
1634         struct fib6_info *f6i = res->f6i;
1635         struct in6_addr *src_key = NULL;
1636         struct rt6_exception *rt6_ex;
1637         struct fib6_nh *nh = res->nh;
1638         int err = 0;
1639
1640         spin_lock_bh(&rt6_exception_lock);
1641
1642         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1643                                           lockdep_is_held(&rt6_exception_lock));
1644         if (!bucket) {
1645                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1646                                  GFP_ATOMIC);
1647                 if (!bucket) {
1648                         err = -ENOMEM;
1649                         goto out;
1650                 }
1651                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1652         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1653                 err = -EINVAL;
1654                 goto out;
1655         }
1656
1657 #ifdef CONFIG_IPV6_SUBTREES
1658         /* fib6_src.plen != 0 indicates f6i is in subtree
1659          * and exception table is indexed by a hash of
1660          * both fib6_dst and fib6_src.
1661          * Otherwise, the exception table is indexed by
1662          * a hash of only fib6_dst.
1663          */
1664         if (f6i->fib6_src.plen)
1665                 src_key = &nrt->rt6i_src.addr;
1666 #endif
1667         /* rt6_mtu_change() might lower mtu on f6i.
1668          * Only insert this exception route if its mtu
1669          * is less than f6i's mtu value.
1670          */
1671         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1672                 err = -EINVAL;
1673                 goto out;
1674         }
1675
1676         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1677                                                src_key);
1678         if (rt6_ex)
1679                 rt6_remove_exception(bucket, rt6_ex);
1680
1681         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1682         if (!rt6_ex) {
1683                 err = -ENOMEM;
1684                 goto out;
1685         }
1686         rt6_ex->rt6i = nrt;
1687         rt6_ex->stamp = jiffies;
1688         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1689         bucket->depth++;
1690         net->ipv6.rt6_stats->fib_rt_cache++;
1691
1692         if (bucket->depth > FIB6_MAX_DEPTH)
1693                 rt6_exception_remove_oldest(bucket);
1694
1695 out:
1696         spin_unlock_bh(&rt6_exception_lock);
1697
1698         /* Update fn->fn_sernum to invalidate all cached dst */
1699         if (!err) {
1700                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1701                 fib6_update_sernum(net, f6i);
1702                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1703                 fib6_force_start_gc(net);
1704         }
1705
1706         return err;
1707 }
1708
1709 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1710 {
1711         struct rt6_exception_bucket *bucket;
1712         struct rt6_exception *rt6_ex;
1713         struct hlist_node *tmp;
1714         int i;
1715
1716         spin_lock_bh(&rt6_exception_lock);
1717
1718         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1719         if (!bucket)
1720                 goto out;
1721
1722         /* Prevent rt6_insert_exception() to recreate the bucket list */
1723         if (!from)
1724                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1725
1726         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1727                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1728                         if (!from ||
1729                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1730                                 rt6_remove_exception(bucket, rt6_ex);
1731                 }
1732                 WARN_ON_ONCE(!from && bucket->depth);
1733                 bucket++;
1734         }
1735 out:
1736         spin_unlock_bh(&rt6_exception_lock);
1737 }
1738
1739 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1740 {
1741         struct fib6_info *f6i = arg;
1742
1743         fib6_nh_flush_exceptions(nh, f6i);
1744
1745         return 0;
1746 }
1747
1748 void rt6_flush_exceptions(struct fib6_info *f6i)
1749 {
1750         if (f6i->nh)
1751                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1752                                          f6i);
1753         else
1754                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1755 }
1756
1757 /* Find cached rt in the hash table inside passed in rt
1758  * Caller has to hold rcu_read_lock()
1759  */
1760 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1761                                            const struct in6_addr *daddr,
1762                                            const struct in6_addr *saddr)
1763 {
1764         const struct in6_addr *src_key = NULL;
1765         struct rt6_exception_bucket *bucket;
1766         struct rt6_exception *rt6_ex;
1767         struct rt6_info *ret = NULL;
1768
1769 #ifdef CONFIG_IPV6_SUBTREES
1770         /* fib6i_src.plen != 0 indicates f6i is in subtree
1771          * and exception table is indexed by a hash of
1772          * both fib6_dst and fib6_src.
1773          * However, the src addr used to create the hash
1774          * might not be exactly the passed in saddr which
1775          * is a /128 addr from the flow.
1776          * So we need to use f6i->fib6_src to redo lookup
1777          * if the passed in saddr does not find anything.
1778          * (See the logic in ip6_rt_cache_alloc() on how
1779          * rt->rt6i_src is updated.)
1780          */
1781         if (res->f6i->fib6_src.plen)
1782                 src_key = saddr;
1783 find_ex:
1784 #endif
1785         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1786         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1787
1788         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1789                 ret = rt6_ex->rt6i;
1790
1791 #ifdef CONFIG_IPV6_SUBTREES
1792         /* Use fib6_src as src_key and redo lookup */
1793         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1794                 src_key = &res->f6i->fib6_src.addr;
1795                 goto find_ex;
1796         }
1797 #endif
1798
1799         return ret;
1800 }
1801
1802 /* Remove the passed in cached rt from the hash table that contains it */
1803 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1804                                     const struct rt6_info *rt)
1805 {
1806         const struct in6_addr *src_key = NULL;
1807         struct rt6_exception_bucket *bucket;
1808         struct rt6_exception *rt6_ex;
1809         int err;
1810
1811         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1812                 return -ENOENT;
1813
1814         spin_lock_bh(&rt6_exception_lock);
1815         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1816
1817 #ifdef CONFIG_IPV6_SUBTREES
1818         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1819          * and exception table is indexed by a hash of
1820          * both rt6i_dst and rt6i_src.
1821          * Otherwise, the exception table is indexed by
1822          * a hash of only rt6i_dst.
1823          */
1824         if (plen)
1825                 src_key = &rt->rt6i_src.addr;
1826 #endif
1827         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1828                                                &rt->rt6i_dst.addr,
1829                                                src_key);
1830         if (rt6_ex) {
1831                 rt6_remove_exception(bucket, rt6_ex);
1832                 err = 0;
1833         } else {
1834                 err = -ENOENT;
1835         }
1836
1837         spin_unlock_bh(&rt6_exception_lock);
1838         return err;
1839 }
1840
1841 struct fib6_nh_excptn_arg {
1842         struct rt6_info *rt;
1843         int             plen;
1844 };
1845
1846 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1847 {
1848         struct fib6_nh_excptn_arg *arg = _arg;
1849         int err;
1850
1851         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1852         if (err == 0)
1853                 return 1;
1854
1855         return 0;
1856 }
1857
1858 static int rt6_remove_exception_rt(struct rt6_info *rt)
1859 {
1860         struct fib6_info *from;
1861
1862         from = rcu_dereference(rt->from);
1863         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1864                 return -EINVAL;
1865
1866         if (from->nh) {
1867                 struct fib6_nh_excptn_arg arg = {
1868                         .rt = rt,
1869                         .plen = from->fib6_src.plen
1870                 };
1871                 int rc;
1872
1873                 /* rc = 1 means an entry was found */
1874                 rc = nexthop_for_each_fib6_nh(from->nh,
1875                                               rt6_nh_remove_exception_rt,
1876                                               &arg);
1877                 return rc ? 0 : -ENOENT;
1878         }
1879
1880         return fib6_nh_remove_exception(from->fib6_nh,
1881                                         from->fib6_src.plen, rt);
1882 }
1883
1884 /* Find rt6_ex which contains the passed in rt cache and
1885  * refresh its stamp
1886  */
1887 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1888                                      const struct rt6_info *rt)
1889 {
1890         const struct in6_addr *src_key = NULL;
1891         struct rt6_exception_bucket *bucket;
1892         struct rt6_exception *rt6_ex;
1893
1894         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1895 #ifdef CONFIG_IPV6_SUBTREES
1896         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1897          * and exception table is indexed by a hash of
1898          * both rt6i_dst and rt6i_src.
1899          * Otherwise, the exception table is indexed by
1900          * a hash of only rt6i_dst.
1901          */
1902         if (plen)
1903                 src_key = &rt->rt6i_src.addr;
1904 #endif
1905         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1906         if (rt6_ex)
1907                 rt6_ex->stamp = jiffies;
1908 }
1909
1910 struct fib6_nh_match_arg {
1911         const struct net_device *dev;
1912         const struct in6_addr   *gw;
1913         struct fib6_nh          *match;
1914 };
1915
1916 /* determine if fib6_nh has given device and gateway */
1917 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1918 {
1919         struct fib6_nh_match_arg *arg = _arg;
1920
1921         if (arg->dev != nh->fib_nh_dev ||
1922             (arg->gw && !nh->fib_nh_gw_family) ||
1923             (!arg->gw && nh->fib_nh_gw_family) ||
1924             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1925                 return 0;
1926
1927         arg->match = nh;
1928
1929         /* found a match, break the loop */
1930         return 1;
1931 }
1932
1933 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1934 {
1935         struct fib6_info *from;
1936         struct fib6_nh *fib6_nh;
1937
1938         rcu_read_lock();
1939
1940         from = rcu_dereference(rt->from);
1941         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1942                 goto unlock;
1943
1944         if (from->nh) {
1945                 struct fib6_nh_match_arg arg = {
1946                         .dev = rt->dst.dev,
1947                         .gw = &rt->rt6i_gateway,
1948                 };
1949
1950                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1951
1952                 if (!arg.match)
1953                         return;
1954                 fib6_nh = arg.match;
1955         } else {
1956                 fib6_nh = from->fib6_nh;
1957         }
1958         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1959 unlock:
1960         rcu_read_unlock();
1961 }
1962
1963 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1964                                          struct rt6_info *rt, int mtu)
1965 {
1966         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1967          * lowest MTU in the path: always allow updating the route PMTU to
1968          * reflect PMTU decreases.
1969          *
1970          * If the new MTU is higher, and the route PMTU is equal to the local
1971          * MTU, this means the old MTU is the lowest in the path, so allow
1972          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1973          * handle this.
1974          */
1975
1976         if (dst_mtu(&rt->dst) >= mtu)
1977                 return true;
1978
1979         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1980                 return true;
1981
1982         return false;
1983 }
1984
1985 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
1986                                        const struct fib6_nh *nh, int mtu)
1987 {
1988         struct rt6_exception_bucket *bucket;
1989         struct rt6_exception *rt6_ex;
1990         int i;
1991
1992         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1993         if (!bucket)
1994                 return;
1995
1996         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1997                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1998                         struct rt6_info *entry = rt6_ex->rt6i;
1999
2000                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2001                          * route), the metrics of its rt->from have already
2002                          * been updated.
2003                          */
2004                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2005                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2006                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2007                 }
2008                 bucket++;
2009         }
2010 }
2011
2012 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2013
2014 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2015                                             const struct in6_addr *gateway)
2016 {
2017         struct rt6_exception_bucket *bucket;
2018         struct rt6_exception *rt6_ex;
2019         struct hlist_node *tmp;
2020         int i;
2021
2022         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2023                 return;
2024
2025         spin_lock_bh(&rt6_exception_lock);
2026         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2027         if (bucket) {
2028                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2029                         hlist_for_each_entry_safe(rt6_ex, tmp,
2030                                                   &bucket->chain, hlist) {
2031                                 struct rt6_info *entry = rt6_ex->rt6i;
2032
2033                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2034                                     RTF_CACHE_GATEWAY &&
2035                                     ipv6_addr_equal(gateway,
2036                                                     &entry->rt6i_gateway)) {
2037                                         rt6_remove_exception(bucket, rt6_ex);
2038                                 }
2039                         }
2040                         bucket++;
2041                 }
2042         }
2043
2044         spin_unlock_bh(&rt6_exception_lock);
2045 }
2046
2047 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2048                                       struct rt6_exception *rt6_ex,
2049                                       struct fib6_gc_args *gc_args,
2050                                       unsigned long now)
2051 {
2052         struct rt6_info *rt = rt6_ex->rt6i;
2053
2054         /* we are pruning and obsoleting aged-out and non gateway exceptions
2055          * even if others have still references to them, so that on next
2056          * dst_check() such references can be dropped.
2057          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2058          * expired, independently from their aging, as per RFC 8201 section 4
2059          */
2060         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2061                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2062                         RT6_TRACE("aging clone %p\n", rt);
2063                         rt6_remove_exception(bucket, rt6_ex);
2064                         return;
2065                 }
2066         } else if (time_after(jiffies, rt->dst.expires)) {
2067                 RT6_TRACE("purging expired route %p\n", rt);
2068                 rt6_remove_exception(bucket, rt6_ex);
2069                 return;
2070         }
2071
2072         if (rt->rt6i_flags & RTF_GATEWAY) {
2073                 struct neighbour *neigh;
2074                 __u8 neigh_flags = 0;
2075
2076                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2077                 if (neigh)
2078                         neigh_flags = neigh->flags;
2079
2080                 if (!(neigh_flags & NTF_ROUTER)) {
2081                         RT6_TRACE("purging route %p via non-router but gateway\n",
2082                                   rt);
2083                         rt6_remove_exception(bucket, rt6_ex);
2084                         return;
2085                 }
2086         }
2087
2088         gc_args->more++;
2089 }
2090
2091 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2092                                    struct fib6_gc_args *gc_args,
2093                                    unsigned long now)
2094 {
2095         struct rt6_exception_bucket *bucket;
2096         struct rt6_exception *rt6_ex;
2097         struct hlist_node *tmp;
2098         int i;
2099
2100         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2101                 return;
2102
2103         rcu_read_lock_bh();
2104         spin_lock(&rt6_exception_lock);
2105         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2106         if (bucket) {
2107                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2108                         hlist_for_each_entry_safe(rt6_ex, tmp,
2109                                                   &bucket->chain, hlist) {
2110                                 rt6_age_examine_exception(bucket, rt6_ex,
2111                                                           gc_args, now);
2112                         }
2113                         bucket++;
2114                 }
2115         }
2116         spin_unlock(&rt6_exception_lock);
2117         rcu_read_unlock_bh();
2118 }
2119
2120 struct fib6_nh_age_excptn_arg {
2121         struct fib6_gc_args     *gc_args;
2122         unsigned long           now;
2123 };
2124
2125 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2126 {
2127         struct fib6_nh_age_excptn_arg *arg = _arg;
2128
2129         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2130         return 0;
2131 }
2132
2133 void rt6_age_exceptions(struct fib6_info *f6i,
2134                         struct fib6_gc_args *gc_args,
2135                         unsigned long now)
2136 {
2137         if (f6i->nh) {
2138                 struct fib6_nh_age_excptn_arg arg = {
2139                         .gc_args = gc_args,
2140                         .now = now
2141                 };
2142
2143                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2144                                          &arg);
2145         } else {
2146                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2147         }
2148 }
2149
2150 /* must be called with rcu lock held */
2151 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2152                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2153 {
2154         struct fib6_node *fn, *saved_fn;
2155
2156         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2157         saved_fn = fn;
2158
2159         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2160                 oif = 0;
2161
2162 redo_rt6_select:
2163         rt6_select(net, fn, oif, res, strict);
2164         if (res->f6i == net->ipv6.fib6_null_entry) {
2165                 fn = fib6_backtrack(fn, &fl6->saddr);
2166                 if (fn)
2167                         goto redo_rt6_select;
2168                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2169                         /* also consider unreachable route */
2170                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2171                         fn = saved_fn;
2172                         goto redo_rt6_select;
2173                 }
2174         }
2175
2176         trace_fib6_table_lookup(net, res, table, fl6);
2177
2178         return 0;
2179 }
2180
2181 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2182                                int oif, struct flowi6 *fl6,
2183                                const struct sk_buff *skb, int flags)
2184 {
2185         struct fib6_result res = {};
2186         struct rt6_info *rt = NULL;
2187         int strict = 0;
2188
2189         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2190                      !rcu_read_lock_held());
2191
2192         strict |= flags & RT6_LOOKUP_F_IFACE;
2193         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2194         if (net->ipv6.devconf_all->forwarding == 0)
2195                 strict |= RT6_LOOKUP_F_REACHABLE;
2196
2197         rcu_read_lock();
2198
2199         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2200         if (res.f6i == net->ipv6.fib6_null_entry)
2201                 goto out;
2202
2203         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2204
2205         /*Search through exception table */
2206         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2207         if (rt) {
2208                 goto out;
2209         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2210                             !res.nh->fib_nh_gw_family)) {
2211                 /* Create a RTF_CACHE clone which will not be
2212                  * owned by the fib6 tree.  It is for the special case where
2213                  * the daddr in the skb during the neighbor look-up is different
2214                  * from the fl6->daddr used to look-up route here.
2215                  */
2216                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2217
2218                 if (rt) {
2219                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2220                          * As rt6_uncached_list_add() does not consume refcnt,
2221                          * this refcnt is always returned to the caller even
2222                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2223                          */
2224                         rt6_uncached_list_add(rt);
2225                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2226                         rcu_read_unlock();
2227
2228                         return rt;
2229                 }
2230         } else {
2231                 /* Get a percpu copy */
2232                 local_bh_disable();
2233                 rt = rt6_get_pcpu_route(&res);
2234
2235                 if (!rt)
2236                         rt = rt6_make_pcpu_route(net, &res);
2237
2238                 local_bh_enable();
2239         }
2240 out:
2241         if (!rt)
2242                 rt = net->ipv6.ip6_null_entry;
2243         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2244                 ip6_hold_safe(net, &rt);
2245         rcu_read_unlock();
2246
2247         return rt;
2248 }
2249 EXPORT_SYMBOL_GPL(ip6_pol_route);
2250
2251 static struct rt6_info *ip6_pol_route_input(struct net *net,
2252                                             struct fib6_table *table,
2253                                             struct flowi6 *fl6,
2254                                             const struct sk_buff *skb,
2255                                             int flags)
2256 {
2257         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2258 }
2259
2260 struct dst_entry *ip6_route_input_lookup(struct net *net,
2261                                          struct net_device *dev,
2262                                          struct flowi6 *fl6,
2263                                          const struct sk_buff *skb,
2264                                          int flags)
2265 {
2266         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2267                 flags |= RT6_LOOKUP_F_IFACE;
2268
2269         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2270 }
2271 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2272
2273 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2274                                   struct flow_keys *keys,
2275                                   struct flow_keys *flkeys)
2276 {
2277         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2278         const struct ipv6hdr *key_iph = outer_iph;
2279         struct flow_keys *_flkeys = flkeys;
2280         const struct ipv6hdr *inner_iph;
2281         const struct icmp6hdr *icmph;
2282         struct ipv6hdr _inner_iph;
2283         struct icmp6hdr _icmph;
2284
2285         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2286                 goto out;
2287
2288         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2289                                    sizeof(_icmph), &_icmph);
2290         if (!icmph)
2291                 goto out;
2292
2293         if (icmph->icmp6_type != ICMPV6_DEST_UNREACH &&
2294             icmph->icmp6_type != ICMPV6_PKT_TOOBIG &&
2295             icmph->icmp6_type != ICMPV6_TIME_EXCEED &&
2296             icmph->icmp6_type != ICMPV6_PARAMPROB)
2297                 goto out;
2298
2299         inner_iph = skb_header_pointer(skb,
2300                                        skb_transport_offset(skb) + sizeof(*icmph),
2301                                        sizeof(_inner_iph), &_inner_iph);
2302         if (!inner_iph)
2303                 goto out;
2304
2305         key_iph = inner_iph;
2306         _flkeys = NULL;
2307 out:
2308         if (_flkeys) {
2309                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2310                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2311                 keys->tags.flow_label = _flkeys->tags.flow_label;
2312                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2313         } else {
2314                 keys->addrs.v6addrs.src = key_iph->saddr;
2315                 keys->addrs.v6addrs.dst = key_iph->daddr;
2316                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2317                 keys->basic.ip_proto = key_iph->nexthdr;
2318         }
2319 }
2320
2321 /* if skb is set it will be used and fl6 can be NULL */
2322 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2323                        const struct sk_buff *skb, struct flow_keys *flkeys)
2324 {
2325         struct flow_keys hash_keys;
2326         u32 mhash;
2327
2328         switch (ip6_multipath_hash_policy(net)) {
2329         case 0:
2330                 memset(&hash_keys, 0, sizeof(hash_keys));
2331                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2332                 if (skb) {
2333                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2334                 } else {
2335                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2336                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2337                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2338                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2339                 }
2340                 break;
2341         case 1:
2342                 if (skb) {
2343                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2344                         struct flow_keys keys;
2345
2346                         /* short-circuit if we already have L4 hash present */
2347                         if (skb->l4_hash)
2348                                 return skb_get_hash_raw(skb) >> 1;
2349
2350                         memset(&hash_keys, 0, sizeof(hash_keys));
2351
2352                         if (!flkeys) {
2353                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2354                                 flkeys = &keys;
2355                         }
2356                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2357                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2358                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2359                         hash_keys.ports.src = flkeys->ports.src;
2360                         hash_keys.ports.dst = flkeys->ports.dst;
2361                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2362                 } else {
2363                         memset(&hash_keys, 0, sizeof(hash_keys));
2364                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2365                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2366                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2367                         hash_keys.ports.src = fl6->fl6_sport;
2368                         hash_keys.ports.dst = fl6->fl6_dport;
2369                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2370                 }
2371                 break;
2372         }
2373         mhash = flow_hash_from_keys(&hash_keys);
2374
2375         return mhash >> 1;
2376 }
2377
2378 /* Called with rcu held */
2379 void ip6_route_input(struct sk_buff *skb)
2380 {
2381         const struct ipv6hdr *iph = ipv6_hdr(skb);
2382         struct net *net = dev_net(skb->dev);
2383         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2384         struct ip_tunnel_info *tun_info;
2385         struct flowi6 fl6 = {
2386                 .flowi6_iif = skb->dev->ifindex,
2387                 .daddr = iph->daddr,
2388                 .saddr = iph->saddr,
2389                 .flowlabel = ip6_flowinfo(iph),
2390                 .flowi6_mark = skb->mark,
2391                 .flowi6_proto = iph->nexthdr,
2392         };
2393         struct flow_keys *flkeys = NULL, _flkeys;
2394
2395         tun_info = skb_tunnel_info(skb);
2396         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2397                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2398
2399         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2400                 flkeys = &_flkeys;
2401
2402         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2403                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2404         skb_dst_drop(skb);
2405         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2406                                                       &fl6, skb, flags));
2407 }
2408
2409 static struct rt6_info *ip6_pol_route_output(struct net *net,
2410                                              struct fib6_table *table,
2411                                              struct flowi6 *fl6,
2412                                              const struct sk_buff *skb,
2413                                              int flags)
2414 {
2415         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2416 }
2417
2418 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2419                                                const struct sock *sk,
2420                                                struct flowi6 *fl6, int flags)
2421 {
2422         bool any_src;
2423
2424         if (ipv6_addr_type(&fl6->daddr) &
2425             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2426                 struct dst_entry *dst;
2427
2428                 /* This function does not take refcnt on the dst */
2429                 dst = l3mdev_link_scope_lookup(net, fl6);
2430                 if (dst)
2431                         return dst;
2432         }
2433
2434         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2435
2436         flags |= RT6_LOOKUP_F_DST_NOREF;
2437         any_src = ipv6_addr_any(&fl6->saddr);
2438         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2439             (fl6->flowi6_oif && any_src))
2440                 flags |= RT6_LOOKUP_F_IFACE;
2441
2442         if (!any_src)
2443                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2444         else if (sk)
2445                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2446
2447         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2448 }
2449 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2450
2451 struct dst_entry *ip6_route_output_flags(struct net *net,
2452                                          const struct sock *sk,
2453                                          struct flowi6 *fl6,
2454                                          int flags)
2455 {
2456         struct dst_entry *dst;
2457         struct rt6_info *rt6;
2458
2459         rcu_read_lock();
2460         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2461         rt6 = (struct rt6_info *)dst;
2462         /* For dst cached in uncached_list, refcnt is already taken. */
2463         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2464                 dst = &net->ipv6.ip6_null_entry->dst;
2465                 dst_hold(dst);
2466         }
2467         rcu_read_unlock();
2468
2469         return dst;
2470 }
2471 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2472
2473 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2474 {
2475         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2476         struct net_device *loopback_dev = net->loopback_dev;
2477         struct dst_entry *new = NULL;
2478
2479         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2480                        DST_OBSOLETE_DEAD, 0);
2481         if (rt) {
2482                 rt6_info_init(rt);
2483                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2484
2485                 new = &rt->dst;
2486                 new->__use = 1;
2487                 new->input = dst_discard;
2488                 new->output = dst_discard_out;
2489
2490                 dst_copy_metrics(new, &ort->dst);
2491
2492                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2493                 rt->rt6i_gateway = ort->rt6i_gateway;
2494                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2495
2496                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2497 #ifdef CONFIG_IPV6_SUBTREES
2498                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2499 #endif
2500         }
2501
2502         dst_release(dst_orig);
2503         return new ? new : ERR_PTR(-ENOMEM);
2504 }
2505
2506 /*
2507  *      Destination cache support functions
2508  */
2509
2510 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2511 {
2512         u32 rt_cookie = 0;
2513
2514         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2515                 return false;
2516
2517         if (fib6_check_expired(f6i))
2518                 return false;
2519
2520         return true;
2521 }
2522
2523 static struct dst_entry *rt6_check(struct rt6_info *rt,
2524                                    struct fib6_info *from,
2525                                    u32 cookie)
2526 {
2527         u32 rt_cookie = 0;
2528
2529         if ((from && !fib6_get_cookie_safe(from, &rt_cookie)) ||
2530             rt_cookie != cookie)
2531                 return NULL;
2532
2533         if (rt6_check_expired(rt))
2534                 return NULL;
2535
2536         return &rt->dst;
2537 }
2538
2539 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2540                                             struct fib6_info *from,
2541                                             u32 cookie)
2542 {
2543         if (!__rt6_check_expired(rt) &&
2544             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2545             fib6_check(from, cookie))
2546                 return &rt->dst;
2547         else
2548                 return NULL;
2549 }
2550
2551 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2552 {
2553         struct dst_entry *dst_ret;
2554         struct fib6_info *from;
2555         struct rt6_info *rt;
2556
2557         rt = container_of(dst, struct rt6_info, dst);
2558
2559         rcu_read_lock();
2560
2561         /* All IPV6 dsts are created with ->obsolete set to the value
2562          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2563          * into this function always.
2564          */
2565
2566         from = rcu_dereference(rt->from);
2567
2568         if (from && (rt->rt6i_flags & RTF_PCPU ||
2569             unlikely(!list_empty(&rt->rt6i_uncached))))
2570                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2571         else
2572                 dst_ret = rt6_check(rt, from, cookie);
2573
2574         rcu_read_unlock();
2575
2576         return dst_ret;
2577 }
2578
2579 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2580 {
2581         struct rt6_info *rt = (struct rt6_info *) dst;
2582
2583         if (rt) {
2584                 if (rt->rt6i_flags & RTF_CACHE) {
2585                         rcu_read_lock();
2586                         if (rt6_check_expired(rt)) {
2587                                 rt6_remove_exception_rt(rt);
2588                                 dst = NULL;
2589                         }
2590                         rcu_read_unlock();
2591                 } else {
2592                         dst_release(dst);
2593                         dst = NULL;
2594                 }
2595         }
2596         return dst;
2597 }
2598
2599 static void ip6_link_failure(struct sk_buff *skb)
2600 {
2601         struct rt6_info *rt;
2602
2603         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2604
2605         rt = (struct rt6_info *) skb_dst(skb);
2606         if (rt) {
2607                 rcu_read_lock();
2608                 if (rt->rt6i_flags & RTF_CACHE) {
2609                         rt6_remove_exception_rt(rt);
2610                 } else {
2611                         struct fib6_info *from;
2612                         struct fib6_node *fn;
2613
2614                         from = rcu_dereference(rt->from);
2615                         if (from) {
2616                                 fn = rcu_dereference(from->fib6_node);
2617                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2618                                         fn->fn_sernum = -1;
2619                         }
2620                 }
2621                 rcu_read_unlock();
2622         }
2623 }
2624
2625 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2626 {
2627         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2628                 struct fib6_info *from;
2629
2630                 rcu_read_lock();
2631                 from = rcu_dereference(rt0->from);
2632                 if (from)
2633                         rt0->dst.expires = from->expires;
2634                 rcu_read_unlock();
2635         }
2636
2637         dst_set_expires(&rt0->dst, timeout);
2638         rt0->rt6i_flags |= RTF_EXPIRES;
2639 }
2640
2641 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2642 {
2643         struct net *net = dev_net(rt->dst.dev);
2644
2645         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2646         rt->rt6i_flags |= RTF_MODIFIED;
2647         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2648 }
2649
2650 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2651 {
2652         return !(rt->rt6i_flags & RTF_CACHE) &&
2653                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2654 }
2655
2656 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2657                                  const struct ipv6hdr *iph, u32 mtu)
2658 {
2659         const struct in6_addr *daddr, *saddr;
2660         struct rt6_info *rt6 = (struct rt6_info *)dst;
2661
2662         if (dst_metric_locked(dst, RTAX_MTU))
2663                 return;
2664
2665         if (iph) {
2666                 daddr = &iph->daddr;
2667                 saddr = &iph->saddr;
2668         } else if (sk) {
2669                 daddr = &sk->sk_v6_daddr;
2670                 saddr = &inet6_sk(sk)->saddr;
2671         } else {
2672                 daddr = NULL;
2673                 saddr = NULL;
2674         }
2675         dst_confirm_neigh(dst, daddr);
2676         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2677         if (mtu >= dst_mtu(dst))
2678                 return;
2679
2680         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2681                 rt6_do_update_pmtu(rt6, mtu);
2682                 /* update rt6_ex->stamp for cache */
2683                 if (rt6->rt6i_flags & RTF_CACHE)
2684                         rt6_update_exception_stamp_rt(rt6);
2685         } else if (daddr) {
2686                 struct fib6_result res = {};
2687                 struct rt6_info *nrt6;
2688
2689                 rcu_read_lock();
2690                 res.f6i = rcu_dereference(rt6->from);
2691                 if (!res.f6i) {
2692                         rcu_read_unlock();
2693                         return;
2694                 }
2695                 res.fib6_flags = res.f6i->fib6_flags;
2696                 res.fib6_type = res.f6i->fib6_type;
2697
2698                 if (res.f6i->nh) {
2699                         struct fib6_nh_match_arg arg = {
2700                                 .dev = dst->dev,
2701                                 .gw = &rt6->rt6i_gateway,
2702                         };
2703
2704                         nexthop_for_each_fib6_nh(res.f6i->nh,
2705                                                  fib6_nh_find_match, &arg);
2706
2707                         /* fib6_info uses a nexthop that does not have fib6_nh
2708                          * using the dst->dev + gw. Should be impossible.
2709                          */
2710                         if (!arg.match) {
2711                                 rcu_read_unlock();
2712                                 return;
2713                         }
2714
2715                         res.nh = arg.match;
2716                 } else {
2717                         res.nh = res.f6i->fib6_nh;
2718                 }
2719
2720                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2721                 if (nrt6) {
2722                         rt6_do_update_pmtu(nrt6, mtu);
2723                         if (rt6_insert_exception(nrt6, &res))
2724                                 dst_release_immediate(&nrt6->dst);
2725                 }
2726                 rcu_read_unlock();
2727         }
2728 }
2729
2730 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2731                                struct sk_buff *skb, u32 mtu)
2732 {
2733         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu);
2734 }
2735
2736 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2737                      int oif, u32 mark, kuid_t uid)
2738 {
2739         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2740         struct dst_entry *dst;
2741         struct flowi6 fl6 = {
2742                 .flowi6_oif = oif,
2743                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2744                 .daddr = iph->daddr,
2745                 .saddr = iph->saddr,
2746                 .flowlabel = ip6_flowinfo(iph),
2747                 .flowi6_uid = uid,
2748         };
2749
2750         dst = ip6_route_output(net, NULL, &fl6);
2751         if (!dst->error)
2752                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu));
2753         dst_release(dst);
2754 }
2755 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2756
2757 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2758 {
2759         int oif = sk->sk_bound_dev_if;
2760         struct dst_entry *dst;
2761
2762         if (!oif && skb->dev)
2763                 oif = l3mdev_master_ifindex(skb->dev);
2764
2765         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2766
2767         dst = __sk_dst_get(sk);
2768         if (!dst || !dst->obsolete ||
2769             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2770                 return;
2771
2772         bh_lock_sock(sk);
2773         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2774                 ip6_datagram_dst_update(sk, false);
2775         bh_unlock_sock(sk);
2776 }
2777 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2778
2779 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2780                            const struct flowi6 *fl6)
2781 {
2782 #ifdef CONFIG_IPV6_SUBTREES
2783         struct ipv6_pinfo *np = inet6_sk(sk);
2784 #endif
2785
2786         ip6_dst_store(sk, dst,
2787                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2788                       &sk->sk_v6_daddr : NULL,
2789 #ifdef CONFIG_IPV6_SUBTREES
2790                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2791                       &np->saddr :
2792 #endif
2793                       NULL);
2794 }
2795
2796 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2797                                   struct flowi6 *fl6,
2798                                   const struct in6_addr *gw,
2799                                   struct rt6_info **ret)
2800 {
2801         const struct fib6_nh *nh = res->nh;
2802
2803         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2804             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2805                 return false;
2806
2807         /* rt_cache's gateway might be different from its 'parent'
2808          * in the case of an ip redirect.
2809          * So we keep searching in the exception table if the gateway
2810          * is different.
2811          */
2812         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2813                 struct rt6_info *rt_cache;
2814
2815                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2816                 if (rt_cache &&
2817                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2818                         *ret = rt_cache;
2819                         return true;
2820                 }
2821                 return false;
2822         }
2823         return true;
2824 }
2825
2826 struct fib6_nh_rd_arg {
2827         struct fib6_result      *res;
2828         struct flowi6           *fl6;
2829         const struct in6_addr   *gw;
2830         struct rt6_info         **ret;
2831 };
2832
2833 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2834 {
2835         struct fib6_nh_rd_arg *arg = _arg;
2836
2837         arg->res->nh = nh;
2838         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2839 }
2840
2841 /* Handle redirects */
2842 struct ip6rd_flowi {
2843         struct flowi6 fl6;
2844         struct in6_addr gateway;
2845 };
2846
2847 static struct rt6_info *__ip6_route_redirect(struct net *net,
2848                                              struct fib6_table *table,
2849                                              struct flowi6 *fl6,
2850                                              const struct sk_buff *skb,
2851                                              int flags)
2852 {
2853         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2854         struct rt6_info *ret = NULL;
2855         struct fib6_result res = {};
2856         struct fib6_nh_rd_arg arg = {
2857                 .res = &res,
2858                 .fl6 = fl6,
2859                 .gw  = &rdfl->gateway,
2860                 .ret = &ret
2861         };
2862         struct fib6_info *rt;
2863         struct fib6_node *fn;
2864
2865         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2866          * this case we must match on the real ingress device, so reset it
2867          */
2868         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2869                 fl6->flowi6_oif = skb->dev->ifindex;
2870
2871         /* Get the "current" route for this destination and
2872          * check if the redirect has come from appropriate router.
2873          *
2874          * RFC 4861 specifies that redirects should only be
2875          * accepted if they come from the nexthop to the target.
2876          * Due to the way the routes are chosen, this notion
2877          * is a bit fuzzy and one might need to check all possible
2878          * routes.
2879          */
2880
2881         rcu_read_lock();
2882         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2883 restart:
2884         for_each_fib6_node_rt_rcu(fn) {
2885                 res.f6i = rt;
2886                 if (fib6_check_expired(rt))
2887                         continue;
2888                 if (rt->fib6_flags & RTF_REJECT)
2889                         break;
2890                 if (unlikely(rt->nh)) {
2891                         if (nexthop_is_blackhole(rt->nh))
2892                                 continue;
2893                         /* on match, res->nh is filled in and potentially ret */
2894                         if (nexthop_for_each_fib6_nh(rt->nh,
2895                                                      fib6_nh_redirect_match,
2896                                                      &arg))
2897                                 goto out;
2898                 } else {
2899                         res.nh = rt->fib6_nh;
2900                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2901                                                   &ret))
2902                                 goto out;
2903                 }
2904         }
2905
2906         if (!rt)
2907                 rt = net->ipv6.fib6_null_entry;
2908         else if (rt->fib6_flags & RTF_REJECT) {
2909                 ret = net->ipv6.ip6_null_entry;
2910                 goto out;
2911         }
2912
2913         if (rt == net->ipv6.fib6_null_entry) {
2914                 fn = fib6_backtrack(fn, &fl6->saddr);
2915                 if (fn)
2916                         goto restart;
2917         }
2918
2919         res.f6i = rt;
2920         res.nh = rt->fib6_nh;
2921 out:
2922         if (ret) {
2923                 ip6_hold_safe(net, &ret);
2924         } else {
2925                 res.fib6_flags = res.f6i->fib6_flags;
2926                 res.fib6_type = res.f6i->fib6_type;
2927                 ret = ip6_create_rt_rcu(&res);
2928         }
2929
2930         rcu_read_unlock();
2931
2932         trace_fib6_table_lookup(net, &res, table, fl6);
2933         return ret;
2934 };
2935
2936 static struct dst_entry *ip6_route_redirect(struct net *net,
2937                                             const struct flowi6 *fl6,
2938                                             const struct sk_buff *skb,
2939                                             const struct in6_addr *gateway)
2940 {
2941         int flags = RT6_LOOKUP_F_HAS_SADDR;
2942         struct ip6rd_flowi rdfl;
2943
2944         rdfl.fl6 = *fl6;
2945         rdfl.gateway = *gateway;
2946
2947         return fib6_rule_lookup(net, &rdfl.fl6, skb,
2948                                 flags, __ip6_route_redirect);
2949 }
2950
2951 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
2952                   kuid_t uid)
2953 {
2954         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2955         struct dst_entry *dst;
2956         struct flowi6 fl6 = {
2957                 .flowi6_iif = LOOPBACK_IFINDEX,
2958                 .flowi6_oif = oif,
2959                 .flowi6_mark = mark,
2960                 .daddr = iph->daddr,
2961                 .saddr = iph->saddr,
2962                 .flowlabel = ip6_flowinfo(iph),
2963                 .flowi6_uid = uid,
2964         };
2965
2966         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
2967         rt6_do_redirect(dst, NULL, skb);
2968         dst_release(dst);
2969 }
2970 EXPORT_SYMBOL_GPL(ip6_redirect);
2971
2972 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
2973 {
2974         const struct ipv6hdr *iph = ipv6_hdr(skb);
2975         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
2976         struct dst_entry *dst;
2977         struct flowi6 fl6 = {
2978                 .flowi6_iif = LOOPBACK_IFINDEX,
2979                 .flowi6_oif = oif,
2980                 .daddr = msg->dest,
2981                 .saddr = iph->daddr,
2982                 .flowi6_uid = sock_net_uid(net, NULL),
2983         };
2984
2985         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
2986         rt6_do_redirect(dst, NULL, skb);
2987         dst_release(dst);
2988 }
2989
2990 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
2991 {
2992         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
2993                      sk->sk_uid);
2994 }
2995 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
2996
2997 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
2998 {
2999         struct net_device *dev = dst->dev;
3000         unsigned int mtu = dst_mtu(dst);
3001         struct net *net = dev_net(dev);
3002
3003         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3004
3005         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3006                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3007
3008         /*
3009          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3010          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3011          * IPV6_MAXPLEN is also valid and means: "any MSS,
3012          * rely only on pmtu discovery"
3013          */
3014         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3015                 mtu = IPV6_MAXPLEN;
3016         return mtu;
3017 }
3018
3019 static unsigned int ip6_mtu(const struct dst_entry *dst)
3020 {
3021         struct inet6_dev *idev;
3022         unsigned int mtu;
3023
3024         mtu = dst_metric_raw(dst, RTAX_MTU);
3025         if (mtu)
3026                 goto out;
3027
3028         mtu = IPV6_MIN_MTU;
3029
3030         rcu_read_lock();
3031         idev = __in6_dev_get(dst->dev);
3032         if (idev)
3033                 mtu = idev->cnf.mtu6;
3034         rcu_read_unlock();
3035
3036 out:
3037         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3038
3039         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3040 }
3041
3042 /* MTU selection:
3043  * 1. mtu on route is locked - use it
3044  * 2. mtu from nexthop exception
3045  * 3. mtu from egress device
3046  *
3047  * based on ip6_dst_mtu_forward and exception logic of
3048  * rt6_find_cached_rt; called with rcu_read_lock
3049  */
3050 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3051                       const struct in6_addr *daddr,
3052                       const struct in6_addr *saddr)
3053 {
3054         const struct fib6_nh *nh = res->nh;
3055         struct fib6_info *f6i = res->f6i;
3056         struct inet6_dev *idev;
3057         struct rt6_info *rt;
3058         u32 mtu = 0;
3059
3060         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3061                 mtu = f6i->fib6_pmtu;
3062                 if (mtu)
3063                         goto out;
3064         }
3065
3066         rt = rt6_find_cached_rt(res, daddr, saddr);
3067         if (unlikely(rt)) {
3068                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3069         } else {
3070                 struct net_device *dev = nh->fib_nh_dev;
3071
3072                 mtu = IPV6_MIN_MTU;
3073                 idev = __in6_dev_get(dev);
3074                 if (idev && idev->cnf.mtu6 > mtu)
3075                         mtu = idev->cnf.mtu6;
3076         }
3077
3078         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3079 out:
3080         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3081 }
3082
3083 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3084                                   struct flowi6 *fl6)
3085 {
3086         struct dst_entry *dst;
3087         struct rt6_info *rt;
3088         struct inet6_dev *idev = in6_dev_get(dev);
3089         struct net *net = dev_net(dev);
3090
3091         if (unlikely(!idev))
3092                 return ERR_PTR(-ENODEV);
3093
3094         rt = ip6_dst_alloc(net, dev, 0);
3095         if (unlikely(!rt)) {
3096                 in6_dev_put(idev);
3097                 dst = ERR_PTR(-ENOMEM);
3098                 goto out;
3099         }
3100
3101         rt->dst.flags |= DST_HOST;
3102         rt->dst.input = ip6_input;
3103         rt->dst.output  = ip6_output;
3104         rt->rt6i_gateway  = fl6->daddr;
3105         rt->rt6i_dst.addr = fl6->daddr;
3106         rt->rt6i_dst.plen = 128;
3107         rt->rt6i_idev     = idev;
3108         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3109
3110         /* Add this dst into uncached_list so that rt6_disable_ip() can
3111          * do proper release of the net_device
3112          */
3113         rt6_uncached_list_add(rt);
3114         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3115
3116         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3117
3118 out:
3119         return dst;
3120 }
3121
3122 static int ip6_dst_gc(struct dst_ops *ops)
3123 {
3124         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3125         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3126         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3127         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3128         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3129         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3130         int entries;
3131
3132         entries = dst_entries_get_fast(ops);
3133         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3134             entries <= rt_max_size)
3135                 goto out;
3136
3137         net->ipv6.ip6_rt_gc_expire++;
3138         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3139         entries = dst_entries_get_slow(ops);
3140         if (entries < ops->gc_thresh)
3141                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3142 out:
3143         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3144         return entries > rt_max_size;
3145 }
3146
3147 static struct rt6_info *ip6_nh_lookup_table(struct net *net,
3148                                             struct fib6_config *cfg,
3149                                             const struct in6_addr *gw_addr,
3150                                             u32 tbid, int flags)
3151 {
3152         struct flowi6 fl6 = {
3153                 .flowi6_oif = cfg->fc_ifindex,
3154                 .daddr = *gw_addr,
3155                 .saddr = cfg->fc_prefsrc,
3156         };
3157         struct fib6_table *table;
3158         struct rt6_info *rt;
3159
3160         table = fib6_get_table(net, tbid);
3161         if (!table)
3162                 return NULL;
3163
3164         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3165                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3166
3167         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3168         rt = ip6_pol_route(net, table, cfg->fc_ifindex, &fl6, NULL, flags);
3169
3170         /* if table lookup failed, fall back to full lookup */
3171         if (rt == net->ipv6.ip6_null_entry) {
3172                 ip6_rt_put(rt);
3173                 rt = NULL;
3174         }
3175
3176         return rt;
3177 }
3178
3179 static int ip6_route_check_nh_onlink(struct net *net,
3180                                      struct fib6_config *cfg,
3181                                      const struct net_device *dev,
3182                                      struct netlink_ext_ack *extack)
3183 {
3184         u32 tbid = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
3185         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3186         u32 flags = RTF_LOCAL | RTF_ANYCAST | RTF_REJECT;
3187         struct fib6_info *from;
3188         struct rt6_info *grt;
3189         int err;
3190
3191         err = 0;
3192         grt = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0);
3193         if (grt) {
3194                 rcu_read_lock();
3195                 from = rcu_dereference(grt->from);
3196                 if (!grt->dst.error &&
3197                     /* ignore match if it is the default route */
3198                     from && !ipv6_addr_any(&from->fib6_dst.addr) &&
3199                     (grt->rt6i_flags & flags || dev != grt->dst.dev)) {
3200                         NL_SET_ERR_MSG(extack,
3201                                        "Nexthop has invalid gateway or device mismatch");
3202                         err = -EINVAL;
3203                 }
3204                 rcu_read_unlock();
3205
3206                 ip6_rt_put(grt);
3207         }
3208
3209         return err;
3210 }
3211
3212 static int ip6_route_check_nh(struct net *net,
3213                               struct fib6_config *cfg,
3214                               struct net_device **_dev,
3215                               struct inet6_dev **idev)
3216 {
3217         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3218         struct net_device *dev = _dev ? *_dev : NULL;
3219         struct rt6_info *grt = NULL;
3220         int err = -EHOSTUNREACH;
3221
3222         if (cfg->fc_table) {
3223                 int flags = RT6_LOOKUP_F_IFACE;
3224
3225                 grt = ip6_nh_lookup_table(net, cfg, gw_addr,
3226                                           cfg->fc_table, flags);
3227                 if (grt) {
3228                         if (grt->rt6i_flags & RTF_GATEWAY ||
3229                             (dev && dev != grt->dst.dev)) {
3230                                 ip6_rt_put(grt);
3231                                 grt = NULL;
3232                         }
3233                 }
3234         }
3235
3236         if (!grt)
3237                 grt = rt6_lookup(net, gw_addr, NULL, cfg->fc_ifindex, NULL, 1);
3238
3239         if (!grt)
3240                 goto out;
3241
3242         if (dev) {
3243                 if (dev != grt->dst.dev) {
3244                         ip6_rt_put(grt);
3245                         goto out;
3246                 }
3247         } else {
3248                 *_dev = dev = grt->dst.dev;
3249                 *idev = grt->rt6i_idev;
3250                 dev_hold(dev);
3251                 in6_dev_hold(grt->rt6i_idev);
3252         }
3253
3254         if (!(grt->rt6i_flags & RTF_GATEWAY))
3255                 err = 0;
3256
3257         ip6_rt_put(grt);
3258
3259 out:
3260         return err;
3261 }
3262
3263 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3264                            struct net_device **_dev, struct inet6_dev **idev,
3265                            struct netlink_ext_ack *extack)
3266 {
3267         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3268         int gwa_type = ipv6_addr_type(gw_addr);
3269         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3270         const struct net_device *dev = *_dev;
3271         bool need_addr_check = !dev;
3272         int err = -EINVAL;
3273
3274         /* if gw_addr is local we will fail to detect this in case
3275          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3276          * will return already-added prefix route via interface that
3277          * prefix route was assigned to, which might be non-loopback.
3278          */
3279         if (dev &&
3280             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3281                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3282                 goto out;
3283         }
3284
3285         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3286                 /* IPv6 strictly inhibits using not link-local
3287                  * addresses as nexthop address.
3288                  * Otherwise, router will not able to send redirects.
3289                  * It is very good, but in some (rare!) circumstances
3290                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3291                  * some exceptions. --ANK
3292                  * We allow IPv4-mapped nexthops to support RFC4798-type
3293                  * addressing
3294                  */
3295                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3296                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3297                         goto out;
3298                 }
3299
3300                 if (cfg->fc_flags & RTNH_F_ONLINK)
3301                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3302                 else
3303                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3304
3305                 if (err)
3306                         goto out;
3307         }
3308
3309         /* reload in case device was changed */
3310         dev = *_dev;
3311
3312         err = -EINVAL;
3313         if (!dev) {
3314                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3315                 goto out;
3316         } else if (dev->flags & IFF_LOOPBACK) {
3317                 NL_SET_ERR_MSG(extack,
3318                                "Egress device can not be loopback device for this route");
3319                 goto out;
3320         }
3321
3322         /* if we did not check gw_addr above, do so now that the
3323          * egress device has been resolved.
3324          */
3325         if (need_addr_check &&
3326             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3327                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3328                 goto out;
3329         }
3330
3331         err = 0;
3332 out:
3333         return err;
3334 }
3335
3336 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3337 {
3338         if ((flags & RTF_REJECT) ||
3339             (dev && (dev->flags & IFF_LOOPBACK) &&
3340              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3341              !(flags & RTF_LOCAL)))
3342                 return true;
3343
3344         return false;
3345 }
3346
3347 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3348                  struct fib6_config *cfg, gfp_t gfp_flags,
3349                  struct netlink_ext_ack *extack)
3350 {
3351         struct net_device *dev = NULL;
3352         struct inet6_dev *idev = NULL;
3353         int addr_type;
3354         int err;
3355
3356         fib6_nh->fib_nh_family = AF_INET6;
3357
3358         err = -ENODEV;
3359         if (cfg->fc_ifindex) {
3360                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3361                 if (!dev)
3362                         goto out;
3363                 idev = in6_dev_get(dev);
3364                 if (!idev)
3365                         goto out;
3366         }
3367
3368         if (cfg->fc_flags & RTNH_F_ONLINK) {
3369                 if (!dev) {
3370                         NL_SET_ERR_MSG(extack,
3371                                        "Nexthop device required for onlink");
3372                         goto out;
3373                 }
3374
3375                 if (!(dev->flags & IFF_UP)) {
3376                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3377                         err = -ENETDOWN;
3378                         goto out;
3379                 }
3380
3381                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3382         }
3383
3384         fib6_nh->fib_nh_weight = 1;
3385
3386         /* We cannot add true routes via loopback here,
3387          * they would result in kernel looping; promote them to reject routes
3388          */
3389         addr_type = ipv6_addr_type(&cfg->fc_dst);
3390         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3391                 /* hold loopback dev/idev if we haven't done so. */
3392                 if (dev != net->loopback_dev) {
3393                         if (dev) {
3394                                 dev_put(dev);
3395                                 in6_dev_put(idev);
3396                         }
3397                         dev = net->loopback_dev;
3398                         dev_hold(dev);
3399                         idev = in6_dev_get(dev);
3400                         if (!idev) {
3401                                 err = -ENODEV;
3402                                 goto out;
3403                         }
3404                 }
3405                 goto pcpu_alloc;
3406         }
3407
3408         if (cfg->fc_flags & RTF_GATEWAY) {
3409                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3410                 if (err)
3411                         goto out;
3412
3413                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3414                 fib6_nh->fib_nh_gw_family = AF_INET6;
3415         }
3416
3417         err = -ENODEV;
3418         if (!dev)
3419                 goto out;
3420
3421         if (idev->cnf.disable_ipv6) {
3422                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3423                 err = -EACCES;
3424                 goto out;
3425         }
3426
3427         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3428                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3429                 err = -ENETDOWN;
3430                 goto out;
3431         }
3432
3433         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3434             !netif_carrier_ok(dev))
3435                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3436
3437         err = fib_nh_common_init(&fib6_nh->nh_common, cfg->fc_encap,
3438                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3439         if (err)
3440                 goto out;
3441
3442 pcpu_alloc:
3443         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3444         if (!fib6_nh->rt6i_pcpu) {
3445                 err = -ENOMEM;
3446                 goto out;
3447         }
3448
3449         fib6_nh->fib_nh_dev = dev;
3450         fib6_nh->fib_nh_oif = dev->ifindex;
3451         err = 0;
3452 out:
3453         if (idev)
3454                 in6_dev_put(idev);
3455
3456         if (err) {
3457                 lwtstate_put(fib6_nh->fib_nh_lws);
3458                 fib6_nh->fib_nh_lws = NULL;
3459                 if (dev)
3460                         dev_put(dev);
3461         }
3462
3463         return err;
3464 }
3465
3466 void fib6_nh_release(struct fib6_nh *fib6_nh)
3467 {
3468         struct rt6_exception_bucket *bucket;
3469
3470         rcu_read_lock();
3471
3472         fib6_nh_flush_exceptions(fib6_nh, NULL);
3473         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3474         if (bucket) {
3475                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3476                 kfree(bucket);
3477         }
3478
3479         rcu_read_unlock();
3480
3481         if (fib6_nh->rt6i_pcpu) {
3482                 int cpu;
3483
3484                 for_each_possible_cpu(cpu) {
3485                         struct rt6_info **ppcpu_rt;
3486                         struct rt6_info *pcpu_rt;
3487
3488                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3489                         pcpu_rt = *ppcpu_rt;
3490                         if (pcpu_rt) {
3491                                 dst_dev_put(&pcpu_rt->dst);
3492                                 dst_release(&pcpu_rt->dst);
3493                                 *ppcpu_rt = NULL;
3494                         }
3495                 }
3496
3497                 free_percpu(fib6_nh->rt6i_pcpu);
3498         }
3499
3500         fib_nh_common_release(&fib6_nh->nh_common);
3501 }
3502
3503 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3504                                               gfp_t gfp_flags,
3505                                               struct netlink_ext_ack *extack)
3506 {
3507         struct net *net = cfg->fc_nlinfo.nl_net;
3508         struct fib6_info *rt = NULL;
3509         struct nexthop *nh = NULL;
3510         struct fib6_table *table;
3511         struct fib6_nh *fib6_nh;
3512         int err = -EINVAL;
3513         int addr_type;
3514
3515         /* RTF_PCPU is an internal flag; can not be set by userspace */
3516         if (cfg->fc_flags & RTF_PCPU) {
3517                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3518                 goto out;
3519         }
3520
3521         /* RTF_CACHE is an internal flag; can not be set by userspace */
3522         if (cfg->fc_flags & RTF_CACHE) {
3523                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3524                 goto out;
3525         }
3526
3527         if (cfg->fc_type > RTN_MAX) {
3528                 NL_SET_ERR_MSG(extack, "Invalid route type");
3529                 goto out;
3530         }
3531
3532         if (cfg->fc_dst_len > 128) {
3533                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3534                 goto out;
3535         }
3536         if (cfg->fc_src_len > 128) {
3537                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3538                 goto out;
3539         }
3540 #ifndef CONFIG_IPV6_SUBTREES
3541         if (cfg->fc_src_len) {
3542                 NL_SET_ERR_MSG(extack,
3543                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3544                 goto out;
3545         }
3546 #endif
3547         if (cfg->fc_nh_id) {
3548                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3549                 if (!nh) {
3550                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3551                         goto out;
3552                 }
3553                 err = fib6_check_nexthop(nh, cfg, extack);
3554                 if (err)
3555                         goto out;
3556         }
3557
3558         err = -ENOBUFS;
3559         if (cfg->fc_nlinfo.nlh &&
3560             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3561                 table = fib6_get_table(net, cfg->fc_table);
3562                 if (!table) {
3563                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3564                         table = fib6_new_table(net, cfg->fc_table);
3565                 }
3566         } else {
3567                 table = fib6_new_table(net, cfg->fc_table);
3568         }
3569
3570         if (!table)
3571                 goto out;
3572
3573         err = -ENOMEM;
3574         rt = fib6_info_alloc(gfp_flags, !nh);
3575         if (!rt)
3576                 goto out;
3577
3578         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3579                                                extack);
3580         if (IS_ERR(rt->fib6_metrics)) {
3581                 err = PTR_ERR(rt->fib6_metrics);
3582                 /* Do not leave garbage there. */
3583                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3584                 goto out;
3585         }
3586
3587         if (cfg->fc_flags & RTF_ADDRCONF)
3588                 rt->dst_nocount = true;
3589
3590         if (cfg->fc_flags & RTF_EXPIRES)
3591                 fib6_set_expires(rt, jiffies +
3592                                 clock_t_to_jiffies(cfg->fc_expires));
3593         else
3594                 fib6_clean_expires(rt);
3595
3596         if (cfg->fc_protocol == RTPROT_UNSPEC)
3597                 cfg->fc_protocol = RTPROT_BOOT;
3598         rt->fib6_protocol = cfg->fc_protocol;
3599
3600         rt->fib6_table = table;
3601         rt->fib6_metric = cfg->fc_metric;
3602         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3603         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3604
3605         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3606         rt->fib6_dst.plen = cfg->fc_dst_len;
3607         if (rt->fib6_dst.plen == 128)
3608                 rt->dst_host = true;
3609
3610 #ifdef CONFIG_IPV6_SUBTREES
3611         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3612         rt->fib6_src.plen = cfg->fc_src_len;
3613 #endif
3614         if (nh) {
3615                 if (!nexthop_get(nh)) {
3616                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3617                         goto out;
3618                 }
3619                 if (rt->fib6_src.plen) {
3620                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3621                         goto out;
3622                 }
3623                 rt->nh = nh;
3624                 fib6_nh = nexthop_fib6_nh(rt->nh);
3625         } else {
3626                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3627                 if (err)
3628                         goto out;
3629
3630                 fib6_nh = rt->fib6_nh;
3631
3632                 /* We cannot add true routes via loopback here, they would
3633                  * result in kernel looping; promote them to reject routes
3634                  */
3635                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3636                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3637                                    addr_type))
3638                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3639         }
3640
3641         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3642                 struct net_device *dev = fib6_nh->fib_nh_dev;
3643
3644                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3645                         NL_SET_ERR_MSG(extack, "Invalid source address");
3646                         err = -EINVAL;
3647                         goto out;
3648                 }
3649                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3650                 rt->fib6_prefsrc.plen = 128;
3651         } else
3652                 rt->fib6_prefsrc.plen = 0;
3653
3654         return rt;
3655 out:
3656         fib6_info_release(rt);
3657         return ERR_PTR(err);
3658 }
3659
3660 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3661                   struct netlink_ext_ack *extack)
3662 {
3663         struct fib6_info *rt;
3664         int err;
3665
3666         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3667         if (IS_ERR(rt))
3668                 return PTR_ERR(rt);
3669
3670         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3671         fib6_info_release(rt);
3672
3673         return err;
3674 }
3675
3676 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3677 {
3678         struct net *net = info->nl_net;
3679         struct fib6_table *table;
3680         int err;
3681
3682         if (rt == net->ipv6.fib6_null_entry) {
3683                 err = -ENOENT;
3684                 goto out;
3685         }
3686
3687         table = rt->fib6_table;
3688         spin_lock_bh(&table->tb6_lock);
3689         err = fib6_del(rt, info);
3690         spin_unlock_bh(&table->tb6_lock);
3691
3692 out:
3693         fib6_info_release(rt);
3694         return err;
3695 }
3696
3697 int ip6_del_rt(struct net *net, struct fib6_info *rt)
3698 {
3699         struct nl_info info = { .nl_net = net };
3700
3701         return __ip6_del_rt(rt, &info);
3702 }
3703
3704 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3705 {
3706         struct nl_info *info = &cfg->fc_nlinfo;
3707         struct net *net = info->nl_net;
3708         struct sk_buff *skb = NULL;
3709         struct fib6_table *table;
3710         int err = -ENOENT;
3711
3712         if (rt == net->ipv6.fib6_null_entry)
3713                 goto out_put;
3714         table = rt->fib6_table;
3715         spin_lock_bh(&table->tb6_lock);
3716
3717         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3718                 struct fib6_info *sibling, *next_sibling;
3719
3720                 /* prefer to send a single notification with all hops */
3721                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3722                 if (skb) {
3723                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3724
3725                         if (rt6_fill_node(net, skb, rt, NULL,
3726                                           NULL, NULL, 0, RTM_DELROUTE,
3727                                           info->portid, seq, 0) < 0) {
3728                                 kfree_skb(skb);
3729                                 skb = NULL;
3730                         } else
3731                                 info->skip_notify = 1;
3732                 }
3733
3734                 info->skip_notify_kernel = 1;
3735                 call_fib6_multipath_entry_notifiers(net,
3736                                                     FIB_EVENT_ENTRY_DEL,
3737                                                     rt,
3738                                                     rt->fib6_nsiblings,
3739                                                     NULL);
3740                 list_for_each_entry_safe(sibling, next_sibling,
3741                                          &rt->fib6_siblings,
3742                                          fib6_siblings) {
3743                         err = fib6_del(sibling, info);
3744                         if (err)
3745                                 goto out_unlock;
3746                 }
3747         }
3748
3749         err = fib6_del(rt, info);
3750 out_unlock:
3751         spin_unlock_bh(&table->tb6_lock);
3752 out_put:
3753         fib6_info_release(rt);
3754
3755         if (skb) {
3756                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3757                             info->nlh, gfp_any());
3758         }
3759         return err;
3760 }
3761
3762 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3763 {
3764         int rc = -ESRCH;
3765
3766         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3767                 goto out;
3768
3769         if (cfg->fc_flags & RTF_GATEWAY &&
3770             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3771                 goto out;
3772
3773         rc = rt6_remove_exception_rt(rt);
3774 out:
3775         return rc;
3776 }
3777
3778 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3779                              struct fib6_nh *nh)
3780 {
3781         struct fib6_result res = {
3782                 .f6i = rt,
3783                 .nh = nh,
3784         };
3785         struct rt6_info *rt_cache;
3786
3787         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3788         if (rt_cache)
3789                 return __ip6_del_cached_rt(rt_cache, cfg);
3790
3791         return 0;
3792 }
3793
3794 struct fib6_nh_del_cached_rt_arg {
3795         struct fib6_config *cfg;
3796         struct fib6_info *f6i;
3797 };
3798
3799 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3800 {
3801         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3802         int rc;
3803
3804         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3805         return rc != -ESRCH ? rc : 0;
3806 }
3807
3808 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3809 {
3810         struct fib6_nh_del_cached_rt_arg arg = {
3811                 .cfg = cfg,
3812                 .f6i = f6i
3813         };
3814
3815         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3816 }
3817
3818 static int ip6_route_del(struct fib6_config *cfg,
3819                          struct netlink_ext_ack *extack)
3820 {
3821         struct fib6_table *table;
3822         struct fib6_info *rt;
3823         struct fib6_node *fn;
3824         int err = -ESRCH;
3825
3826         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3827         if (!table) {
3828                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3829                 return err;
3830         }
3831
3832         rcu_read_lock();
3833
3834         fn = fib6_locate(&table->tb6_root,
3835                          &cfg->fc_dst, cfg->fc_dst_len,
3836                          &cfg->fc_src, cfg->fc_src_len,
3837                          !(cfg->fc_flags & RTF_CACHE));
3838
3839         if (fn) {
3840                 for_each_fib6_node_rt_rcu(fn) {
3841                         struct fib6_nh *nh;
3842
3843                         if (rt->nh && cfg->fc_nh_id &&
3844                             rt->nh->id != cfg->fc_nh_id)
3845                                 continue;
3846
3847                         if (cfg->fc_flags & RTF_CACHE) {
3848                                 int rc = 0;
3849
3850                                 if (rt->nh) {
3851                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3852                                 } else if (cfg->fc_nh_id) {
3853                                         continue;
3854                                 } else {
3855                                         nh = rt->fib6_nh;
3856                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3857                                 }
3858                                 if (rc != -ESRCH) {
3859                                         rcu_read_unlock();
3860                                         return rc;
3861                                 }
3862                                 continue;
3863                         }
3864
3865                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3866                                 continue;
3867                         if (cfg->fc_protocol &&
3868                             cfg->fc_protocol != rt->fib6_protocol)
3869                                 continue;
3870
3871                         if (rt->nh) {
3872                                 if (!fib6_info_hold_safe(rt))
3873                                         continue;
3874                                 rcu_read_unlock();
3875
3876                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3877                         }
3878                         if (cfg->fc_nh_id)
3879                                 continue;
3880
3881                         nh = rt->fib6_nh;
3882                         if (cfg->fc_ifindex &&
3883                             (!nh->fib_nh_dev ||
3884                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3885                                 continue;
3886                         if (cfg->fc_flags & RTF_GATEWAY &&
3887                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3888                                 continue;
3889                         if (!fib6_info_hold_safe(rt))
3890                                 continue;
3891                         rcu_read_unlock();
3892
3893                         /* if gateway was specified only delete the one hop */
3894                         if (cfg->fc_flags & RTF_GATEWAY)
3895                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3896
3897                         return __ip6_del_rt_siblings(rt, cfg);
3898                 }
3899         }
3900         rcu_read_unlock();
3901
3902         return err;
3903 }
3904
3905 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3906 {
3907         struct netevent_redirect netevent;
3908         struct rt6_info *rt, *nrt = NULL;
3909         struct fib6_result res = {};
3910         struct ndisc_options ndopts;
3911         struct inet6_dev *in6_dev;
3912         struct neighbour *neigh;
3913         struct rd_msg *msg;
3914         int optlen, on_link;
3915         u8 *lladdr;
3916
3917         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
3918         optlen -= sizeof(*msg);
3919
3920         if (optlen < 0) {
3921                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
3922                 return;
3923         }
3924
3925         msg = (struct rd_msg *)icmp6_hdr(skb);
3926
3927         if (ipv6_addr_is_multicast(&msg->dest)) {
3928                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
3929                 return;
3930         }
3931
3932         on_link = 0;
3933         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
3934                 on_link = 1;
3935         } else if (ipv6_addr_type(&msg->target) !=
3936                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
3937                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
3938                 return;
3939         }
3940
3941         in6_dev = __in6_dev_get(skb->dev);
3942         if (!in6_dev)
3943                 return;
3944         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
3945                 return;
3946
3947         /* RFC2461 8.1:
3948          *      The IP source address of the Redirect MUST be the same as the current
3949          *      first-hop router for the specified ICMP Destination Address.
3950          */
3951
3952         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
3953                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
3954                 return;
3955         }
3956
3957         lladdr = NULL;
3958         if (ndopts.nd_opts_tgt_lladdr) {
3959                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
3960                                              skb->dev);
3961                 if (!lladdr) {
3962                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
3963                         return;
3964                 }
3965         }
3966
3967         rt = (struct rt6_info *) dst;
3968         if (rt->rt6i_flags & RTF_REJECT) {
3969                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
3970                 return;
3971         }
3972
3973         /* Redirect received -> path was valid.
3974          * Look, redirects are sent only in response to data packets,
3975          * so that this nexthop apparently is reachable. --ANK
3976          */
3977         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
3978
3979         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
3980         if (!neigh)
3981                 return;
3982
3983         /*
3984          *      We have finally decided to accept it.
3985          */
3986
3987         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
3988                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
3989                      NEIGH_UPDATE_F_OVERRIDE|
3990                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
3991                                      NEIGH_UPDATE_F_ISROUTER)),
3992                      NDISC_REDIRECT, &ndopts);
3993
3994         rcu_read_lock();
3995         res.f6i = rcu_dereference(rt->from);
3996         if (!res.f6i)
3997                 goto out;
3998
3999         if (res.f6i->nh) {
4000                 struct fib6_nh_match_arg arg = {
4001                         .dev = dst->dev,
4002                         .gw = &rt->rt6i_gateway,
4003                 };
4004
4005                 nexthop_for_each_fib6_nh(res.f6i->nh,
4006                                          fib6_nh_find_match, &arg);
4007
4008                 /* fib6_info uses a nexthop that does not have fib6_nh
4009                  * using the dst->dev. Should be impossible
4010                  */
4011                 if (!arg.match)
4012                         goto out;
4013                 res.nh = arg.match;
4014         } else {
4015                 res.nh = res.f6i->fib6_nh;
4016         }
4017
4018         res.fib6_flags = res.f6i->fib6_flags;
4019         res.fib6_type = res.f6i->fib6_type;
4020         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4021         if (!nrt)
4022                 goto out;
4023
4024         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4025         if (on_link)
4026                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4027
4028         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4029
4030         /* rt6_insert_exception() will take care of duplicated exceptions */
4031         if (rt6_insert_exception(nrt, &res)) {
4032                 dst_release_immediate(&nrt->dst);
4033                 goto out;
4034         }
4035
4036         netevent.old = &rt->dst;
4037         netevent.new = &nrt->dst;
4038         netevent.daddr = &msg->dest;
4039         netevent.neigh = neigh;
4040         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4041
4042 out:
4043         rcu_read_unlock();
4044         neigh_release(neigh);
4045 }
4046
4047 #ifdef CONFIG_IPV6_ROUTE_INFO
4048 static struct fib6_info *rt6_get_route_info(struct net *net,
4049                                            const struct in6_addr *prefix, int prefixlen,
4050                                            const struct in6_addr *gwaddr,
4051                                            struct net_device *dev)
4052 {
4053         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4054         int ifindex = dev->ifindex;
4055         struct fib6_node *fn;
4056         struct fib6_info *rt = NULL;
4057         struct fib6_table *table;
4058
4059         table = fib6_get_table(net, tb_id);
4060         if (!table)
4061                 return NULL;
4062
4063         rcu_read_lock();
4064         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4065         if (!fn)
4066                 goto out;
4067
4068         for_each_fib6_node_rt_rcu(fn) {
4069                 /* these routes do not use nexthops */
4070                 if (rt->nh)
4071                         continue;
4072                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4073                         continue;
4074                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4075                     !rt->fib6_nh->fib_nh_gw_family)
4076                         continue;
4077                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4078                         continue;
4079                 if (!fib6_info_hold_safe(rt))
4080                         continue;
4081                 break;
4082         }
4083 out:
4084         rcu_read_unlock();
4085         return rt;
4086 }
4087
4088 static struct fib6_info *rt6_add_route_info(struct net *net,
4089                                            const struct in6_addr *prefix, int prefixlen,
4090                                            const struct in6_addr *gwaddr,
4091                                            struct net_device *dev,
4092                                            unsigned int pref)
4093 {
4094         struct fib6_config cfg = {
4095                 .fc_metric      = IP6_RT_PRIO_USER,
4096                 .fc_ifindex     = dev->ifindex,
4097                 .fc_dst_len     = prefixlen,
4098                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4099                                   RTF_UP | RTF_PREF(pref),
4100                 .fc_protocol = RTPROT_RA,
4101                 .fc_type = RTN_UNICAST,
4102                 .fc_nlinfo.portid = 0,
4103                 .fc_nlinfo.nlh = NULL,
4104                 .fc_nlinfo.nl_net = net,
4105         };
4106
4107         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
4108         cfg.fc_dst = *prefix;
4109         cfg.fc_gateway = *gwaddr;
4110
4111         /* We should treat it as a default route if prefix length is 0. */
4112         if (!prefixlen)
4113                 cfg.fc_flags |= RTF_DEFAULT;
4114
4115         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4116
4117         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4118 }
4119 #endif
4120
4121 struct fib6_info *rt6_get_dflt_router(struct net *net,
4122                                      const struct in6_addr *addr,
4123                                      struct net_device *dev)
4124 {
4125         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4126         struct fib6_info *rt;
4127         struct fib6_table *table;
4128
4129         table = fib6_get_table(net, tb_id);
4130         if (!table)
4131                 return NULL;
4132
4133         rcu_read_lock();
4134         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4135                 struct fib6_nh *nh;
4136
4137                 /* RA routes do not use nexthops */
4138                 if (rt->nh)
4139                         continue;
4140
4141                 nh = rt->fib6_nh;
4142                 if (dev == nh->fib_nh_dev &&
4143                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4144                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4145                         break;
4146         }
4147         if (rt && !fib6_info_hold_safe(rt))
4148                 rt = NULL;
4149         rcu_read_unlock();
4150         return rt;
4151 }
4152
4153 struct fib6_info *rt6_add_dflt_router(struct net *net,
4154                                      const struct in6_addr *gwaddr,
4155                                      struct net_device *dev,
4156                                      unsigned int pref)
4157 {
4158         struct fib6_config cfg = {
4159                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4160                 .fc_metric      = IP6_RT_PRIO_USER,
4161                 .fc_ifindex     = dev->ifindex,
4162                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4163                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4164                 .fc_protocol = RTPROT_RA,
4165                 .fc_type = RTN_UNICAST,
4166                 .fc_nlinfo.portid = 0,
4167                 .fc_nlinfo.nlh = NULL,
4168                 .fc_nlinfo.nl_net = net,
4169         };
4170
4171         cfg.fc_gateway = *gwaddr;
4172
4173         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4174                 struct fib6_table *table;
4175
4176                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4177                 if (table)
4178                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4179         }
4180
4181         return rt6_get_dflt_router(net, gwaddr, dev);
4182 }
4183
4184 static void __rt6_purge_dflt_routers(struct net *net,
4185                                      struct fib6_table *table)
4186 {
4187         struct fib6_info *rt;
4188
4189 restart:
4190         rcu_read_lock();
4191         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4192                 struct net_device *dev = fib6_info_nh_dev(rt);
4193                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4194
4195                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4196                     (!idev || idev->cnf.accept_ra != 2) &&
4197                     fib6_info_hold_safe(rt)) {
4198                         rcu_read_unlock();
4199                         ip6_del_rt(net, rt);
4200                         goto restart;
4201                 }
4202         }
4203         rcu_read_unlock();
4204
4205         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4206 }
4207
4208 void rt6_purge_dflt_routers(struct net *net)
4209 {
4210         struct fib6_table *table;
4211         struct hlist_head *head;
4212         unsigned int h;
4213
4214         rcu_read_lock();
4215
4216         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4217                 head = &net->ipv6.fib_table_hash[h];
4218                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4219                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4220                                 __rt6_purge_dflt_routers(net, table);
4221                 }
4222         }
4223
4224         rcu_read_unlock();
4225 }
4226
4227 static void rtmsg_to_fib6_config(struct net *net,
4228                                  struct in6_rtmsg *rtmsg,
4229                                  struct fib6_config *cfg)
4230 {
4231         *cfg = (struct fib6_config){
4232                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4233                          : RT6_TABLE_MAIN,
4234                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4235                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4236                 .fc_expires = rtmsg->rtmsg_info,
4237                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4238                 .fc_src_len = rtmsg->rtmsg_src_len,
4239                 .fc_flags = rtmsg->rtmsg_flags,
4240                 .fc_type = rtmsg->rtmsg_type,
4241
4242                 .fc_nlinfo.nl_net = net,
4243
4244                 .fc_dst = rtmsg->rtmsg_dst,
4245                 .fc_src = rtmsg->rtmsg_src,
4246                 .fc_gateway = rtmsg->rtmsg_gateway,
4247         };
4248 }
4249
4250 int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4251 {
4252         struct fib6_config cfg;
4253         struct in6_rtmsg rtmsg;
4254         int err;
4255
4256         switch (cmd) {
4257         case SIOCADDRT:         /* Add a route */
4258         case SIOCDELRT:         /* Delete a route */
4259                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4260                         return -EPERM;
4261                 err = copy_from_user(&rtmsg, arg,
4262                                      sizeof(struct in6_rtmsg));
4263                 if (err)
4264                         return -EFAULT;
4265
4266                 rtmsg_to_fib6_config(net, &rtmsg, &cfg);
4267
4268                 rtnl_lock();
4269                 switch (cmd) {
4270                 case SIOCADDRT:
4271                         err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4272                         break;
4273                 case SIOCDELRT:
4274                         err = ip6_route_del(&cfg, NULL);
4275                         break;
4276                 default:
4277                         err = -EINVAL;
4278                 }
4279                 rtnl_unlock();
4280
4281                 return err;
4282         }
4283
4284         return -EINVAL;
4285 }
4286
4287 /*
4288  *      Drop the packet on the floor
4289  */
4290
4291 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4292 {
4293         struct dst_entry *dst = skb_dst(skb);
4294         struct net *net = dev_net(dst->dev);
4295         struct inet6_dev *idev;
4296         int type;
4297
4298         if (netif_is_l3_master(skb->dev) &&
4299             dst->dev == net->loopback_dev)
4300                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4301         else
4302                 idev = ip6_dst_idev(dst);
4303
4304         switch (ipstats_mib_noroutes) {
4305         case IPSTATS_MIB_INNOROUTES:
4306                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4307                 if (type == IPV6_ADDR_ANY) {
4308                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4309                         break;
4310                 }
4311                 /* FALLTHROUGH */
4312         case IPSTATS_MIB_OUTNOROUTES:
4313                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4314                 break;
4315         }
4316
4317         /* Start over by dropping the dst for l3mdev case */
4318         if (netif_is_l3_master(skb->dev))
4319                 skb_dst_drop(skb);
4320
4321         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4322         kfree_skb(skb);
4323         return 0;
4324 }
4325
4326 static int ip6_pkt_discard(struct sk_buff *skb)
4327 {
4328         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4329 }
4330
4331 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4332 {
4333         skb->dev = skb_dst(skb)->dev;
4334         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4335 }
4336
4337 static int ip6_pkt_prohibit(struct sk_buff *skb)
4338 {
4339         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4340 }
4341
4342 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4343 {
4344         skb->dev = skb_dst(skb)->dev;
4345         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4346 }
4347
4348 /*
4349  *      Allocate a dst for local (unicast / anycast) address.
4350  */
4351
4352 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4353                                      struct inet6_dev *idev,
4354                                      const struct in6_addr *addr,
4355                                      bool anycast, gfp_t gfp_flags)
4356 {
4357         struct fib6_config cfg = {
4358                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4359                 .fc_ifindex = idev->dev->ifindex,
4360                 .fc_flags = RTF_UP | RTF_ADDRCONF | RTF_NONEXTHOP,
4361                 .fc_dst = *addr,
4362                 .fc_dst_len = 128,
4363                 .fc_protocol = RTPROT_KERNEL,
4364                 .fc_nlinfo.nl_net = net,
4365                 .fc_ignore_dev_down = true,
4366         };
4367
4368         if (anycast) {
4369                 cfg.fc_type = RTN_ANYCAST;
4370                 cfg.fc_flags |= RTF_ANYCAST;
4371         } else {
4372                 cfg.fc_type = RTN_LOCAL;
4373                 cfg.fc_flags |= RTF_LOCAL;
4374         }
4375
4376         return ip6_route_info_create(&cfg, gfp_flags, NULL);
4377 }
4378
4379 /* remove deleted ip from prefsrc entries */
4380 struct arg_dev_net_ip {
4381         struct net_device *dev;
4382         struct net *net;
4383         struct in6_addr *addr;
4384 };
4385
4386 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4387 {
4388         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4389         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4390         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4391
4392         if (!rt->nh &&
4393             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4394             rt != net->ipv6.fib6_null_entry &&
4395             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4396                 spin_lock_bh(&rt6_exception_lock);
4397                 /* remove prefsrc entry */
4398                 rt->fib6_prefsrc.plen = 0;
4399                 spin_unlock_bh(&rt6_exception_lock);
4400         }
4401         return 0;
4402 }
4403
4404 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4405 {
4406         struct net *net = dev_net(ifp->idev->dev);
4407         struct arg_dev_net_ip adni = {
4408                 .dev = ifp->idev->dev,
4409                 .net = net,
4410                 .addr = &ifp->addr,
4411         };
4412         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4413 }
4414
4415 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4416
4417 /* Remove routers and update dst entries when gateway turn into host. */
4418 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4419 {
4420         struct in6_addr *gateway = (struct in6_addr *)arg;
4421         struct fib6_nh *nh;
4422
4423         /* RA routes do not use nexthops */
4424         if (rt->nh)
4425                 return 0;
4426
4427         nh = rt->fib6_nh;
4428         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4429             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4430                 return -1;
4431
4432         /* Further clean up cached routes in exception table.
4433          * This is needed because cached route may have a different
4434          * gateway than its 'parent' in the case of an ip redirect.
4435          */
4436         fib6_nh_exceptions_clean_tohost(nh, gateway);
4437
4438         return 0;
4439 }
4440
4441 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4442 {
4443         fib6_clean_all(net, fib6_clean_tohost, gateway);
4444 }
4445
4446 struct arg_netdev_event {
4447         const struct net_device *dev;
4448         union {
4449                 unsigned char nh_flags;
4450                 unsigned long event;
4451         };
4452 };
4453
4454 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4455 {
4456         struct fib6_info *iter;
4457         struct fib6_node *fn;
4458
4459         fn = rcu_dereference_protected(rt->fib6_node,
4460                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4461         iter = rcu_dereference_protected(fn->leaf,
4462                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4463         while (iter) {
4464                 if (iter->fib6_metric == rt->fib6_metric &&
4465                     rt6_qualify_for_ecmp(iter))
4466                         return iter;
4467                 iter = rcu_dereference_protected(iter->fib6_next,
4468                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4469         }
4470
4471         return NULL;
4472 }
4473
4474 /* only called for fib entries with builtin fib6_nh */
4475 static bool rt6_is_dead(const struct fib6_info *rt)
4476 {
4477         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4478             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4479              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4480                 return true;
4481
4482         return false;
4483 }
4484
4485 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4486 {
4487         struct fib6_info *iter;
4488         int total = 0;
4489
4490         if (!rt6_is_dead(rt))
4491                 total += rt->fib6_nh->fib_nh_weight;
4492
4493         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4494                 if (!rt6_is_dead(iter))
4495                         total += iter->fib6_nh->fib_nh_weight;
4496         }
4497
4498         return total;
4499 }
4500
4501 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4502 {
4503         int upper_bound = -1;
4504
4505         if (!rt6_is_dead(rt)) {
4506                 *weight += rt->fib6_nh->fib_nh_weight;
4507                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4508                                                     total) - 1;
4509         }
4510         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4511 }
4512
4513 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4514 {
4515         struct fib6_info *iter;
4516         int weight = 0;
4517
4518         rt6_upper_bound_set(rt, &weight, total);
4519
4520         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4521                 rt6_upper_bound_set(iter, &weight, total);
4522 }
4523
4524 void rt6_multipath_rebalance(struct fib6_info *rt)
4525 {
4526         struct fib6_info *first;
4527         int total;
4528
4529         /* In case the entire multipath route was marked for flushing,
4530          * then there is no need to rebalance upon the removal of every
4531          * sibling route.
4532          */
4533         if (!rt->fib6_nsiblings || rt->should_flush)
4534                 return;
4535
4536         /* During lookup routes are evaluated in order, so we need to
4537          * make sure upper bounds are assigned from the first sibling
4538          * onwards.
4539          */
4540         first = rt6_multipath_first_sibling(rt);
4541         if (WARN_ON_ONCE(!first))
4542                 return;
4543
4544         total = rt6_multipath_total_weight(first);
4545         rt6_multipath_upper_bound_set(first, total);
4546 }
4547
4548 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4549 {
4550         const struct arg_netdev_event *arg = p_arg;
4551         struct net *net = dev_net(arg->dev);
4552
4553         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4554             rt->fib6_nh->fib_nh_dev == arg->dev) {
4555                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4556                 fib6_update_sernum_upto_root(net, rt);
4557                 rt6_multipath_rebalance(rt);
4558         }
4559
4560         return 0;
4561 }
4562
4563 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4564 {
4565         struct arg_netdev_event arg = {
4566                 .dev = dev,
4567                 {
4568                         .nh_flags = nh_flags,
4569                 },
4570         };
4571
4572         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4573                 arg.nh_flags |= RTNH_F_LINKDOWN;
4574
4575         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4576 }
4577
4578 /* only called for fib entries with inline fib6_nh */
4579 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4580                                    const struct net_device *dev)
4581 {
4582         struct fib6_info *iter;
4583
4584         if (rt->fib6_nh->fib_nh_dev == dev)
4585                 return true;
4586         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4587                 if (iter->fib6_nh->fib_nh_dev == dev)
4588                         return true;
4589
4590         return false;
4591 }
4592
4593 static void rt6_multipath_flush(struct fib6_info *rt)
4594 {
4595         struct fib6_info *iter;
4596
4597         rt->should_flush = 1;
4598         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4599                 iter->should_flush = 1;
4600 }
4601
4602 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4603                                              const struct net_device *down_dev)
4604 {
4605         struct fib6_info *iter;
4606         unsigned int dead = 0;
4607
4608         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4609             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4610                 dead++;
4611         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4612                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4613                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4614                         dead++;
4615
4616         return dead;
4617 }
4618
4619 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4620                                        const struct net_device *dev,
4621                                        unsigned char nh_flags)
4622 {
4623         struct fib6_info *iter;
4624
4625         if (rt->fib6_nh->fib_nh_dev == dev)
4626                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4627         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4628                 if (iter->fib6_nh->fib_nh_dev == dev)
4629                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4630 }
4631
4632 /* called with write lock held for table with rt */
4633 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4634 {
4635         const struct arg_netdev_event *arg = p_arg;
4636         const struct net_device *dev = arg->dev;
4637         struct net *net = dev_net(dev);
4638
4639         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4640                 return 0;
4641
4642         switch (arg->event) {
4643         case NETDEV_UNREGISTER:
4644                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4645         case NETDEV_DOWN:
4646                 if (rt->should_flush)
4647                         return -1;
4648                 if (!rt->fib6_nsiblings)
4649                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4650                 if (rt6_multipath_uses_dev(rt, dev)) {
4651                         unsigned int count;
4652
4653                         count = rt6_multipath_dead_count(rt, dev);
4654                         if (rt->fib6_nsiblings + 1 == count) {
4655                                 rt6_multipath_flush(rt);
4656                                 return -1;
4657                         }
4658                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4659                                                    RTNH_F_LINKDOWN);
4660                         fib6_update_sernum(net, rt);
4661                         rt6_multipath_rebalance(rt);
4662                 }
4663                 return -2;
4664         case NETDEV_CHANGE:
4665                 if (rt->fib6_nh->fib_nh_dev != dev ||
4666                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4667                         break;
4668                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4669                 rt6_multipath_rebalance(rt);
4670                 break;
4671         }
4672
4673         return 0;
4674 }
4675
4676 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4677 {
4678         struct arg_netdev_event arg = {
4679                 .dev = dev,
4680                 {
4681                         .event = event,
4682                 },
4683         };
4684         struct net *net = dev_net(dev);
4685
4686         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4687                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4688         else
4689                 fib6_clean_all(net, fib6_ifdown, &arg);
4690 }
4691
4692 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4693 {
4694         rt6_sync_down_dev(dev, event);
4695         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4696         neigh_ifdown(&nd_tbl, dev);
4697 }
4698
4699 struct rt6_mtu_change_arg {
4700         struct net_device *dev;
4701         unsigned int mtu;
4702         struct fib6_info *f6i;
4703 };
4704
4705 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4706 {
4707         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4708         struct fib6_info *f6i = arg->f6i;
4709
4710         /* For administrative MTU increase, there is no way to discover
4711          * IPv6 PMTU increase, so PMTU increase should be updated here.
4712          * Since RFC 1981 doesn't include administrative MTU increase
4713          * update PMTU increase is a MUST. (i.e. jumbo frame)
4714          */
4715         if (nh->fib_nh_dev == arg->dev) {
4716                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4717                 u32 mtu = f6i->fib6_pmtu;
4718
4719                 if (mtu >= arg->mtu ||
4720                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4721                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4722
4723                 spin_lock_bh(&rt6_exception_lock);
4724                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4725                 spin_unlock_bh(&rt6_exception_lock);
4726         }
4727
4728         return 0;
4729 }
4730
4731 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4732 {
4733         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4734         struct inet6_dev *idev;
4735
4736         /* In IPv6 pmtu discovery is not optional,
4737            so that RTAX_MTU lock cannot disable it.
4738            We still use this lock to block changes
4739            caused by addrconf/ndisc.
4740         */
4741
4742         idev = __in6_dev_get(arg->dev);
4743         if (!idev)
4744                 return 0;
4745
4746         if (fib6_metric_locked(f6i, RTAX_MTU))
4747                 return 0;
4748
4749         arg->f6i = f6i;
4750         if (f6i->nh) {
4751                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4752                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4753                                                 arg);
4754         }
4755
4756         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4757 }
4758
4759 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4760 {
4761         struct rt6_mtu_change_arg arg = {
4762                 .dev = dev,
4763                 .mtu = mtu,
4764         };
4765
4766         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4767 }
4768
4769 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4770         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4771         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4772         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4773         [RTA_OIF]               = { .type = NLA_U32 },
4774         [RTA_IIF]               = { .type = NLA_U32 },
4775         [RTA_PRIORITY]          = { .type = NLA_U32 },
4776         [RTA_METRICS]           = { .type = NLA_NESTED },
4777         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4778         [RTA_PREF]              = { .type = NLA_U8 },
4779         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4780         [RTA_ENCAP]             = { .type = NLA_NESTED },
4781         [RTA_EXPIRES]           = { .type = NLA_U32 },
4782         [RTA_UID]               = { .type = NLA_U32 },
4783         [RTA_MARK]              = { .type = NLA_U32 },
4784         [RTA_TABLE]             = { .type = NLA_U32 },
4785         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4786         [RTA_SPORT]             = { .type = NLA_U16 },
4787         [RTA_DPORT]             = { .type = NLA_U16 },
4788         [RTA_NH_ID]             = { .type = NLA_U32 },
4789 };
4790
4791 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4792                               struct fib6_config *cfg,
4793                               struct netlink_ext_ack *extack)
4794 {
4795         struct rtmsg *rtm;
4796         struct nlattr *tb[RTA_MAX+1];
4797         unsigned int pref;
4798         int err;
4799
4800         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4801                                      rtm_ipv6_policy, extack);
4802         if (err < 0)
4803                 goto errout;
4804
4805         err = -EINVAL;
4806         rtm = nlmsg_data(nlh);
4807
4808         *cfg = (struct fib6_config){
4809                 .fc_table = rtm->rtm_table,
4810                 .fc_dst_len = rtm->rtm_dst_len,
4811                 .fc_src_len = rtm->rtm_src_len,
4812                 .fc_flags = RTF_UP,
4813                 .fc_protocol = rtm->rtm_protocol,
4814                 .fc_type = rtm->rtm_type,
4815
4816                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4817                 .fc_nlinfo.nlh = nlh,
4818                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4819         };
4820
4821         if (rtm->rtm_type == RTN_UNREACHABLE ||
4822             rtm->rtm_type == RTN_BLACKHOLE ||
4823             rtm->rtm_type == RTN_PROHIBIT ||
4824             rtm->rtm_type == RTN_THROW)
4825                 cfg->fc_flags |= RTF_REJECT;
4826
4827         if (rtm->rtm_type == RTN_LOCAL)
4828                 cfg->fc_flags |= RTF_LOCAL;
4829
4830         if (rtm->rtm_flags & RTM_F_CLONED)
4831                 cfg->fc_flags |= RTF_CACHE;
4832
4833         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4834
4835         if (tb[RTA_NH_ID]) {
4836                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4837                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4838                         NL_SET_ERR_MSG(extack,
4839                                        "Nexthop specification and nexthop id are mutually exclusive");
4840                         goto errout;
4841                 }
4842                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4843         }
4844
4845         if (tb[RTA_GATEWAY]) {
4846                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4847                 cfg->fc_flags |= RTF_GATEWAY;
4848         }
4849         if (tb[RTA_VIA]) {
4850                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4851                 goto errout;
4852         }
4853
4854         if (tb[RTA_DST]) {
4855                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4856
4857                 if (nla_len(tb[RTA_DST]) < plen)
4858                         goto errout;
4859
4860                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4861         }
4862
4863         if (tb[RTA_SRC]) {
4864                 int plen = (rtm->rtm_src_len + 7) >> 3;
4865
4866                 if (nla_len(tb[RTA_SRC]) < plen)
4867                         goto errout;
4868
4869                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4870         }
4871
4872         if (tb[RTA_PREFSRC])
4873                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4874
4875         if (tb[RTA_OIF])
4876                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4877
4878         if (tb[RTA_PRIORITY])
4879                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4880
4881         if (tb[RTA_METRICS]) {
4882                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4883                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4884         }
4885
4886         if (tb[RTA_TABLE])
4887                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4888
4889         if (tb[RTA_MULTIPATH]) {
4890                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4891                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4892
4893                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4894                                                      cfg->fc_mp_len, extack);
4895                 if (err < 0)
4896                         goto errout;
4897         }
4898
4899         if (tb[RTA_PREF]) {
4900                 pref = nla_get_u8(tb[RTA_PREF]);
4901                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4902                     pref != ICMPV6_ROUTER_PREF_HIGH)
4903                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4904                 cfg->fc_flags |= RTF_PREF(pref);
4905         }
4906
4907         if (tb[RTA_ENCAP])
4908                 cfg->fc_encap = tb[RTA_ENCAP];
4909
4910         if (tb[RTA_ENCAP_TYPE]) {
4911                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4912
4913                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4914                 if (err < 0)
4915                         goto errout;
4916         }
4917
4918         if (tb[RTA_EXPIRES]) {
4919                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4920
4921                 if (addrconf_finite_timeout(timeout)) {
4922                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
4923                         cfg->fc_flags |= RTF_EXPIRES;
4924                 }
4925         }
4926
4927         err = 0;
4928 errout:
4929         return err;
4930 }
4931
4932 struct rt6_nh {
4933         struct fib6_info *fib6_info;
4934         struct fib6_config r_cfg;
4935         struct list_head next;
4936 };
4937
4938 static int ip6_route_info_append(struct net *net,
4939                                  struct list_head *rt6_nh_list,
4940                                  struct fib6_info *rt,
4941                                  struct fib6_config *r_cfg)
4942 {
4943         struct rt6_nh *nh;
4944         int err = -EEXIST;
4945
4946         list_for_each_entry(nh, rt6_nh_list, next) {
4947                 /* check if fib6_info already exists */
4948                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
4949                         return err;
4950         }
4951
4952         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
4953         if (!nh)
4954                 return -ENOMEM;
4955         nh->fib6_info = rt;
4956         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
4957         list_add_tail(&nh->next, rt6_nh_list);
4958
4959         return 0;
4960 }
4961
4962 static void ip6_route_mpath_notify(struct fib6_info *rt,
4963                                    struct fib6_info *rt_last,
4964                                    struct nl_info *info,
4965                                    __u16 nlflags)
4966 {
4967         /* if this is an APPEND route, then rt points to the first route
4968          * inserted and rt_last points to last route inserted. Userspace
4969          * wants a consistent dump of the route which starts at the first
4970          * nexthop. Since sibling routes are always added at the end of
4971          * the list, find the first sibling of the last route appended
4972          */
4973         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
4974                 rt = list_first_entry(&rt_last->fib6_siblings,
4975                                       struct fib6_info,
4976                                       fib6_siblings);
4977         }
4978
4979         if (rt)
4980                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
4981 }
4982
4983 static int ip6_route_multipath_add(struct fib6_config *cfg,
4984                                    struct netlink_ext_ack *extack)
4985 {
4986         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
4987         struct nl_info *info = &cfg->fc_nlinfo;
4988         enum fib_event_type event_type;
4989         struct fib6_config r_cfg;
4990         struct rtnexthop *rtnh;
4991         struct fib6_info *rt;
4992         struct rt6_nh *err_nh;
4993         struct rt6_nh *nh, *nh_safe;
4994         __u16 nlflags;
4995         int remaining;
4996         int attrlen;
4997         int err = 1;
4998         int nhn = 0;
4999         int replace = (cfg->fc_nlinfo.nlh &&
5000                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5001         LIST_HEAD(rt6_nh_list);
5002
5003         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5004         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5005                 nlflags |= NLM_F_APPEND;
5006
5007         remaining = cfg->fc_mp_len;
5008         rtnh = (struct rtnexthop *)cfg->fc_mp;
5009
5010         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5011          * fib6_info structs per nexthop
5012          */
5013         while (rtnh_ok(rtnh, remaining)) {
5014                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5015                 if (rtnh->rtnh_ifindex)
5016                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5017
5018                 attrlen = rtnh_attrlen(rtnh);
5019                 if (attrlen > 0) {
5020                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5021
5022                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5023                         if (nla) {
5024                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5025                                 r_cfg.fc_flags |= RTF_GATEWAY;
5026                         }
5027                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5028                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5029                         if (nla)
5030                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5031                 }
5032
5033                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5034                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5035                 if (IS_ERR(rt)) {
5036                         err = PTR_ERR(rt);
5037                         rt = NULL;
5038                         goto cleanup;
5039                 }
5040                 if (!rt6_qualify_for_ecmp(rt)) {
5041                         err = -EINVAL;
5042                         NL_SET_ERR_MSG(extack,
5043                                        "Device only routes can not be added for IPv6 using the multipath API.");
5044                         fib6_info_release(rt);
5045                         goto cleanup;
5046                 }
5047
5048                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5049
5050                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5051                                             rt, &r_cfg);
5052                 if (err) {
5053                         fib6_info_release(rt);
5054                         goto cleanup;
5055                 }
5056
5057                 rtnh = rtnh_next(rtnh, &remaining);
5058         }
5059
5060         if (list_empty(&rt6_nh_list)) {
5061                 NL_SET_ERR_MSG(extack,
5062                                "Invalid nexthop configuration - no valid nexthops");
5063                 return -EINVAL;
5064         }
5065
5066         /* for add and replace send one notification with all nexthops.
5067          * Skip the notification in fib6_add_rt2node and send one with
5068          * the full route when done
5069          */
5070         info->skip_notify = 1;
5071
5072         /* For add and replace, send one notification with all nexthops. For
5073          * append, send one notification with all appended nexthops.
5074          */
5075         info->skip_notify_kernel = 1;
5076
5077         err_nh = NULL;
5078         list_for_each_entry(nh, &rt6_nh_list, next) {
5079                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5080                 fib6_info_release(nh->fib6_info);
5081
5082                 if (!err) {
5083                         /* save reference to last route successfully inserted */
5084                         rt_last = nh->fib6_info;
5085
5086                         /* save reference to first route for notification */
5087                         if (!rt_notif)
5088                                 rt_notif = nh->fib6_info;
5089                 }
5090
5091                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5092                 nh->fib6_info = NULL;
5093                 if (err) {
5094                         if (replace && nhn)
5095                                 NL_SET_ERR_MSG_MOD(extack,
5096                                                    "multipath route replace failed (check consistency of installed routes)");
5097                         err_nh = nh;
5098                         goto add_errout;
5099                 }
5100
5101                 /* Because each route is added like a single route we remove
5102                  * these flags after the first nexthop: if there is a collision,
5103                  * we have already failed to add the first nexthop:
5104                  * fib6_add_rt2node() has rejected it; when replacing, old
5105                  * nexthops have been replaced by first new, the rest should
5106                  * be added to it.
5107                  */
5108                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5109                                                      NLM_F_REPLACE);
5110                 nhn++;
5111         }
5112
5113         event_type = replace ? FIB_EVENT_ENTRY_REPLACE : FIB_EVENT_ENTRY_ADD;
5114         err = call_fib6_multipath_entry_notifiers(info->nl_net, event_type,
5115                                                   rt_notif, nhn - 1, extack);
5116         if (err) {
5117                 /* Delete all the siblings that were just added */
5118                 err_nh = NULL;
5119                 goto add_errout;
5120         }
5121
5122         /* success ... tell user about new route */
5123         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5124         goto cleanup;
5125
5126 add_errout:
5127         /* send notification for routes that were added so that
5128          * the delete notifications sent by ip6_route_del are
5129          * coherent
5130          */
5131         if (rt_notif)
5132                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5133
5134         /* Delete routes that were already added */
5135         list_for_each_entry(nh, &rt6_nh_list, next) {
5136                 if (err_nh == nh)
5137                         break;
5138                 ip6_route_del(&nh->r_cfg, extack);
5139         }
5140
5141 cleanup:
5142         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5143                 if (nh->fib6_info)
5144                         fib6_info_release(nh->fib6_info);
5145                 list_del(&nh->next);
5146                 kfree(nh);
5147         }
5148
5149         return err;
5150 }
5151
5152 static int ip6_route_multipath_del(struct fib6_config *cfg,
5153                                    struct netlink_ext_ack *extack)
5154 {
5155         struct fib6_config r_cfg;
5156         struct rtnexthop *rtnh;
5157         int remaining;
5158         int attrlen;
5159         int err = 1, last_err = 0;
5160
5161         remaining = cfg->fc_mp_len;
5162         rtnh = (struct rtnexthop *)cfg->fc_mp;
5163
5164         /* Parse a Multipath Entry */
5165         while (rtnh_ok(rtnh, remaining)) {
5166                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5167                 if (rtnh->rtnh_ifindex)
5168                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5169
5170                 attrlen = rtnh_attrlen(rtnh);
5171                 if (attrlen > 0) {
5172                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5173
5174                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5175                         if (nla) {
5176                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5177                                 r_cfg.fc_flags |= RTF_GATEWAY;
5178                         }
5179                 }
5180                 err = ip6_route_del(&r_cfg, extack);
5181                 if (err)
5182                         last_err = err;
5183
5184                 rtnh = rtnh_next(rtnh, &remaining);
5185         }
5186
5187         return last_err;
5188 }
5189
5190 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5191                               struct netlink_ext_ack *extack)
5192 {
5193         struct fib6_config cfg;
5194         int err;
5195
5196         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5197         if (err < 0)
5198                 return err;
5199
5200         if (cfg.fc_nh_id &&
5201             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5202                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5203                 return -EINVAL;
5204         }
5205
5206         if (cfg.fc_mp)
5207                 return ip6_route_multipath_del(&cfg, extack);
5208         else {
5209                 cfg.fc_delete_all_nh = 1;
5210                 return ip6_route_del(&cfg, extack);
5211         }
5212 }
5213
5214 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5215                               struct netlink_ext_ack *extack)
5216 {
5217         struct fib6_config cfg;
5218         int err;
5219
5220         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5221         if (err < 0)
5222                 return err;
5223
5224         if (cfg.fc_metric == 0)
5225                 cfg.fc_metric = IP6_RT_PRIO_USER;
5226
5227         if (cfg.fc_mp)
5228                 return ip6_route_multipath_add(&cfg, extack);
5229         else
5230                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5231 }
5232
5233 /* add the overhead of this fib6_nh to nexthop_len */
5234 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5235 {
5236         int *nexthop_len = arg;
5237
5238         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5239                      + NLA_ALIGN(sizeof(struct rtnexthop))
5240                      + nla_total_size(16); /* RTA_GATEWAY */
5241
5242         if (nh->fib_nh_lws) {
5243                 /* RTA_ENCAP_TYPE */
5244                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5245                 /* RTA_ENCAP */
5246                 *nexthop_len += nla_total_size(2);
5247         }
5248
5249         return 0;
5250 }
5251
5252 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5253 {
5254         int nexthop_len;
5255
5256         if (f6i->nh) {
5257                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5258                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5259                                          &nexthop_len);
5260         } else {
5261                 struct fib6_nh *nh = f6i->fib6_nh;
5262
5263                 nexthop_len = 0;
5264                 if (f6i->fib6_nsiblings) {
5265                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5266                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5267                                     + nla_total_size(16) /* RTA_GATEWAY */
5268                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5269
5270                         nexthop_len *= f6i->fib6_nsiblings;
5271                 }
5272                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5273         }
5274
5275         return NLMSG_ALIGN(sizeof(struct rtmsg))
5276                + nla_total_size(16) /* RTA_SRC */
5277                + nla_total_size(16) /* RTA_DST */
5278                + nla_total_size(16) /* RTA_GATEWAY */
5279                + nla_total_size(16) /* RTA_PREFSRC */
5280                + nla_total_size(4) /* RTA_TABLE */
5281                + nla_total_size(4) /* RTA_IIF */
5282                + nla_total_size(4) /* RTA_OIF */
5283                + nla_total_size(4) /* RTA_PRIORITY */
5284                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5285                + nla_total_size(sizeof(struct rta_cacheinfo))
5286                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5287                + nla_total_size(1) /* RTA_PREF */
5288                + nexthop_len;
5289 }
5290
5291 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5292                                  unsigned char *flags)
5293 {
5294         if (nexthop_is_multipath(nh)) {
5295                 struct nlattr *mp;
5296
5297                 mp = nla_nest_start(skb, RTA_MULTIPATH);
5298                 if (!mp)
5299                         goto nla_put_failure;
5300
5301                 if (nexthop_mpath_fill_node(skb, nh))
5302                         goto nla_put_failure;
5303
5304                 nla_nest_end(skb, mp);
5305         } else {
5306                 struct fib6_nh *fib6_nh;
5307
5308                 fib6_nh = nexthop_fib6_nh(nh);
5309                 if (fib_nexthop_info(skb, &fib6_nh->nh_common,
5310                                      flags, false) < 0)
5311                         goto nla_put_failure;
5312         }
5313
5314         return 0;
5315
5316 nla_put_failure:
5317         return -EMSGSIZE;
5318 }
5319
5320 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5321                          struct fib6_info *rt, struct dst_entry *dst,
5322                          struct in6_addr *dest, struct in6_addr *src,
5323                          int iif, int type, u32 portid, u32 seq,
5324                          unsigned int flags)
5325 {
5326         struct rt6_info *rt6 = (struct rt6_info *)dst;
5327         struct rt6key *rt6_dst, *rt6_src;
5328         u32 *pmetrics, table, rt6_flags;
5329         unsigned char nh_flags = 0;
5330         struct nlmsghdr *nlh;
5331         struct rtmsg *rtm;
5332         long expires = 0;
5333
5334         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5335         if (!nlh)
5336                 return -EMSGSIZE;
5337
5338         if (rt6) {
5339                 rt6_dst = &rt6->rt6i_dst;
5340                 rt6_src = &rt6->rt6i_src;
5341                 rt6_flags = rt6->rt6i_flags;
5342         } else {
5343                 rt6_dst = &rt->fib6_dst;
5344                 rt6_src = &rt->fib6_src;
5345                 rt6_flags = rt->fib6_flags;
5346         }
5347
5348         rtm = nlmsg_data(nlh);
5349         rtm->rtm_family = AF_INET6;
5350         rtm->rtm_dst_len = rt6_dst->plen;
5351         rtm->rtm_src_len = rt6_src->plen;
5352         rtm->rtm_tos = 0;
5353         if (rt->fib6_table)
5354                 table = rt->fib6_table->tb6_id;
5355         else
5356                 table = RT6_TABLE_UNSPEC;
5357         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5358         if (nla_put_u32(skb, RTA_TABLE, table))
5359                 goto nla_put_failure;
5360
5361         rtm->rtm_type = rt->fib6_type;
5362         rtm->rtm_flags = 0;
5363         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5364         rtm->rtm_protocol = rt->fib6_protocol;
5365
5366         if (rt6_flags & RTF_CACHE)
5367                 rtm->rtm_flags |= RTM_F_CLONED;
5368
5369         if (dest) {
5370                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5371                         goto nla_put_failure;
5372                 rtm->rtm_dst_len = 128;
5373         } else if (rtm->rtm_dst_len)
5374                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5375                         goto nla_put_failure;
5376 #ifdef CONFIG_IPV6_SUBTREES
5377         if (src) {
5378                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5379                         goto nla_put_failure;
5380                 rtm->rtm_src_len = 128;
5381         } else if (rtm->rtm_src_len &&
5382                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5383                 goto nla_put_failure;
5384 #endif
5385         if (iif) {
5386 #ifdef CONFIG_IPV6_MROUTE
5387                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5388                         int err = ip6mr_get_route(net, skb, rtm, portid);
5389
5390                         if (err == 0)
5391                                 return 0;
5392                         if (err < 0)
5393                                 goto nla_put_failure;
5394                 } else
5395 #endif
5396                         if (nla_put_u32(skb, RTA_IIF, iif))
5397                                 goto nla_put_failure;
5398         } else if (dest) {
5399                 struct in6_addr saddr_buf;
5400                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5401                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5402                         goto nla_put_failure;
5403         }
5404
5405         if (rt->fib6_prefsrc.plen) {
5406                 struct in6_addr saddr_buf;
5407                 saddr_buf = rt->fib6_prefsrc.addr;
5408                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5409                         goto nla_put_failure;
5410         }
5411
5412         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5413         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5414                 goto nla_put_failure;
5415
5416         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5417                 goto nla_put_failure;
5418
5419         /* For multipath routes, walk the siblings list and add
5420          * each as a nexthop within RTA_MULTIPATH.
5421          */
5422         if (rt6) {
5423                 if (rt6_flags & RTF_GATEWAY &&
5424                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5425                         goto nla_put_failure;
5426
5427                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5428                         goto nla_put_failure;
5429         } else if (rt->fib6_nsiblings) {
5430                 struct fib6_info *sibling, *next_sibling;
5431                 struct nlattr *mp;
5432
5433                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5434                 if (!mp)
5435                         goto nla_put_failure;
5436
5437                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5438                                     rt->fib6_nh->fib_nh_weight) < 0)
5439                         goto nla_put_failure;
5440
5441                 list_for_each_entry_safe(sibling, next_sibling,
5442                                          &rt->fib6_siblings, fib6_siblings) {
5443                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5444                                             sibling->fib6_nh->fib_nh_weight) < 0)
5445                                 goto nla_put_failure;
5446                 }
5447
5448                 nla_nest_end(skb, mp);
5449         } else if (rt->nh) {
5450                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5451                         goto nla_put_failure;
5452
5453                 if (nexthop_is_blackhole(rt->nh))
5454                         rtm->rtm_type = RTN_BLACKHOLE;
5455
5456                 if (rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5457                         goto nla_put_failure;
5458
5459                 rtm->rtm_flags |= nh_flags;
5460         } else {
5461                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common,
5462                                      &nh_flags, false) < 0)
5463                         goto nla_put_failure;
5464
5465                 rtm->rtm_flags |= nh_flags;
5466         }
5467
5468         if (rt6_flags & RTF_EXPIRES) {
5469                 expires = dst ? dst->expires : rt->expires;
5470                 expires -= jiffies;
5471         }
5472
5473         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5474                 goto nla_put_failure;
5475
5476         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5477                 goto nla_put_failure;
5478
5479
5480         nlmsg_end(skb, nlh);
5481         return 0;
5482
5483 nla_put_failure:
5484         nlmsg_cancel(skb, nlh);
5485         return -EMSGSIZE;
5486 }
5487
5488 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5489 {
5490         const struct net_device *dev = arg;
5491
5492         if (nh->fib_nh_dev == dev)
5493                 return 1;
5494
5495         return 0;
5496 }
5497
5498 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5499                                const struct net_device *dev)
5500 {
5501         if (f6i->nh) {
5502                 struct net_device *_dev = (struct net_device *)dev;
5503
5504                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5505                                                   fib6_info_nh_uses_dev,
5506                                                   _dev);
5507         }
5508
5509         if (f6i->fib6_nh->fib_nh_dev == dev)
5510                 return true;
5511
5512         if (f6i->fib6_nsiblings) {
5513                 struct fib6_info *sibling, *next_sibling;
5514
5515                 list_for_each_entry_safe(sibling, next_sibling,
5516                                          &f6i->fib6_siblings, fib6_siblings) {
5517                         if (sibling->fib6_nh->fib_nh_dev == dev)
5518                                 return true;
5519                 }
5520         }
5521
5522         return false;
5523 }
5524
5525 struct fib6_nh_exception_dump_walker {
5526         struct rt6_rtnl_dump_arg *dump;
5527         struct fib6_info *rt;
5528         unsigned int flags;
5529         unsigned int skip;
5530         unsigned int count;
5531 };
5532
5533 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5534 {
5535         struct fib6_nh_exception_dump_walker *w = arg;
5536         struct rt6_rtnl_dump_arg *dump = w->dump;
5537         struct rt6_exception_bucket *bucket;
5538         struct rt6_exception *rt6_ex;
5539         int i, err;
5540
5541         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5542         if (!bucket)
5543                 return 0;
5544
5545         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5546                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5547                         if (w->skip) {
5548                                 w->skip--;
5549                                 continue;
5550                         }
5551
5552                         /* Expiration of entries doesn't bump sernum, insertion
5553                          * does. Removal is triggered by insertion, so we can
5554                          * rely on the fact that if entries change between two
5555                          * partial dumps, this node is scanned again completely,
5556                          * see rt6_insert_exception() and fib6_dump_table().
5557                          *
5558                          * Count expired entries we go through as handled
5559                          * entries that we'll skip next time, in case of partial
5560                          * node dump. Otherwise, if entries expire meanwhile,
5561                          * we'll skip the wrong amount.
5562                          */
5563                         if (rt6_check_expired(rt6_ex->rt6i)) {
5564                                 w->count++;
5565                                 continue;
5566                         }
5567
5568                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5569                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5570                                             RTM_NEWROUTE,
5571                                             NETLINK_CB(dump->cb->skb).portid,
5572                                             dump->cb->nlh->nlmsg_seq, w->flags);
5573                         if (err)
5574                                 return err;
5575
5576                         w->count++;
5577                 }
5578                 bucket++;
5579         }
5580
5581         return 0;
5582 }
5583
5584 /* Return -1 if done with node, number of handled routes on partial dump */
5585 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5586 {
5587         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5588         struct fib_dump_filter *filter = &arg->filter;
5589         unsigned int flags = NLM_F_MULTI;
5590         struct net *net = arg->net;
5591         int count = 0;
5592
5593         if (rt == net->ipv6.fib6_null_entry)
5594                 return -1;
5595
5596         if ((filter->flags & RTM_F_PREFIX) &&
5597             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5598                 /* success since this is not a prefix route */
5599                 return -1;
5600         }
5601         if (filter->filter_set &&
5602             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5603              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5604              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5605                 return -1;
5606         }
5607
5608         if (filter->filter_set ||
5609             !filter->dump_routes || !filter->dump_exceptions) {
5610                 flags |= NLM_F_DUMP_FILTERED;
5611         }
5612
5613         if (filter->dump_routes) {
5614                 if (skip) {
5615                         skip--;
5616                 } else {
5617                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5618                                           0, RTM_NEWROUTE,
5619                                           NETLINK_CB(arg->cb->skb).portid,
5620                                           arg->cb->nlh->nlmsg_seq, flags)) {
5621                                 return 0;
5622                         }
5623                         count++;
5624                 }
5625         }
5626
5627         if (filter->dump_exceptions) {
5628                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5629                                                            .rt = rt,
5630                                                            .flags = flags,
5631                                                            .skip = skip,
5632                                                            .count = 0 };
5633                 int err;
5634
5635                 if (rt->nh) {
5636                         err = nexthop_for_each_fib6_nh(rt->nh,
5637                                                        rt6_nh_dump_exceptions,
5638                                                        &w);
5639                 } else {
5640                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5641                 }
5642
5643                 if (err)
5644                         return count += w.count;
5645         }
5646
5647         return -1;
5648 }
5649
5650 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5651                                         const struct nlmsghdr *nlh,
5652                                         struct nlattr **tb,
5653                                         struct netlink_ext_ack *extack)
5654 {
5655         struct rtmsg *rtm;
5656         int i, err;
5657
5658         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5659                 NL_SET_ERR_MSG_MOD(extack,
5660                                    "Invalid header for get route request");
5661                 return -EINVAL;
5662         }
5663
5664         if (!netlink_strict_get_check(skb))
5665                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5666                                               rtm_ipv6_policy, extack);
5667
5668         rtm = nlmsg_data(nlh);
5669         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5670             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5671             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5672             rtm->rtm_type) {
5673                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5674                 return -EINVAL;
5675         }
5676         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5677                 NL_SET_ERR_MSG_MOD(extack,
5678                                    "Invalid flags for get route request");
5679                 return -EINVAL;
5680         }
5681
5682         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5683                                             rtm_ipv6_policy, extack);
5684         if (err)
5685                 return err;
5686
5687         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5688             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5689                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5690                 return -EINVAL;
5691         }
5692
5693         for (i = 0; i <= RTA_MAX; i++) {
5694                 if (!tb[i])
5695                         continue;
5696
5697                 switch (i) {
5698                 case RTA_SRC:
5699                 case RTA_DST:
5700                 case RTA_IIF:
5701                 case RTA_OIF:
5702                 case RTA_MARK:
5703                 case RTA_UID:
5704                 case RTA_SPORT:
5705                 case RTA_DPORT:
5706                 case RTA_IP_PROTO:
5707                         break;
5708                 default:
5709                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5710                         return -EINVAL;
5711                 }
5712         }
5713
5714         return 0;
5715 }
5716
5717 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5718                               struct netlink_ext_ack *extack)
5719 {
5720         struct net *net = sock_net(in_skb->sk);
5721         struct nlattr *tb[RTA_MAX+1];
5722         int err, iif = 0, oif = 0;
5723         struct fib6_info *from;
5724         struct dst_entry *dst;
5725         struct rt6_info *rt;
5726         struct sk_buff *skb;
5727         struct rtmsg *rtm;
5728         struct flowi6 fl6 = {};
5729         bool fibmatch;
5730
5731         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5732         if (err < 0)
5733                 goto errout;
5734
5735         err = -EINVAL;
5736         rtm = nlmsg_data(nlh);
5737         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5738         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5739
5740         if (tb[RTA_SRC]) {
5741                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5742                         goto errout;
5743
5744                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5745         }
5746
5747         if (tb[RTA_DST]) {
5748                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5749                         goto errout;
5750
5751                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5752         }
5753
5754         if (tb[RTA_IIF])
5755                 iif = nla_get_u32(tb[RTA_IIF]);
5756
5757         if (tb[RTA_OIF])
5758                 oif = nla_get_u32(tb[RTA_OIF]);
5759
5760         if (tb[RTA_MARK])
5761                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5762
5763         if (tb[RTA_UID])
5764                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5765                                            nla_get_u32(tb[RTA_UID]));
5766         else
5767                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5768
5769         if (tb[RTA_SPORT])
5770                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5771
5772         if (tb[RTA_DPORT])
5773                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5774
5775         if (tb[RTA_IP_PROTO]) {
5776                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5777                                                   &fl6.flowi6_proto, AF_INET6,
5778                                                   extack);
5779                 if (err)
5780                         goto errout;
5781         }
5782
5783         if (iif) {
5784                 struct net_device *dev;
5785                 int flags = 0;
5786
5787                 rcu_read_lock();
5788
5789                 dev = dev_get_by_index_rcu(net, iif);
5790                 if (!dev) {
5791                         rcu_read_unlock();
5792                         err = -ENODEV;
5793                         goto errout;
5794                 }
5795
5796                 fl6.flowi6_iif = iif;
5797
5798                 if (!ipv6_addr_any(&fl6.saddr))
5799                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5800
5801                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5802
5803                 rcu_read_unlock();
5804         } else {
5805                 fl6.flowi6_oif = oif;
5806
5807                 dst = ip6_route_output(net, NULL, &fl6);
5808         }
5809
5810
5811         rt = container_of(dst, struct rt6_info, dst);
5812         if (rt->dst.error) {
5813                 err = rt->dst.error;
5814                 ip6_rt_put(rt);
5815                 goto errout;
5816         }
5817
5818         if (rt == net->ipv6.ip6_null_entry) {
5819                 err = rt->dst.error;
5820                 ip6_rt_put(rt);
5821                 goto errout;
5822         }
5823
5824         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5825         if (!skb) {
5826                 ip6_rt_put(rt);
5827                 err = -ENOBUFS;
5828                 goto errout;
5829         }
5830
5831         skb_dst_set(skb, &rt->dst);
5832
5833         rcu_read_lock();
5834         from = rcu_dereference(rt->from);
5835         if (from) {
5836                 if (fibmatch)
5837                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5838                                             iif, RTM_NEWROUTE,
5839                                             NETLINK_CB(in_skb).portid,
5840                                             nlh->nlmsg_seq, 0);
5841                 else
5842                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5843                                             &fl6.saddr, iif, RTM_NEWROUTE,
5844                                             NETLINK_CB(in_skb).portid,
5845                                             nlh->nlmsg_seq, 0);
5846         } else {
5847                 err = -ENETUNREACH;
5848         }
5849         rcu_read_unlock();
5850
5851         if (err < 0) {
5852                 kfree_skb(skb);
5853                 goto errout;
5854         }
5855
5856         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5857 errout:
5858         return err;
5859 }
5860
5861 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
5862                      unsigned int nlm_flags)
5863 {
5864         struct sk_buff *skb;
5865         struct net *net = info->nl_net;
5866         u32 seq;
5867         int err;
5868
5869         err = -ENOBUFS;
5870         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5871
5872         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5873         if (!skb)
5874                 goto errout;
5875
5876         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5877                             event, info->portid, seq, nlm_flags);
5878         if (err < 0) {
5879                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5880                 WARN_ON(err == -EMSGSIZE);
5881                 kfree_skb(skb);
5882                 goto errout;
5883         }
5884         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5885                     info->nlh, gfp_any());
5886         return;
5887 errout:
5888         if (err < 0)
5889                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5890 }
5891
5892 void fib6_rt_update(struct net *net, struct fib6_info *rt,
5893                     struct nl_info *info)
5894 {
5895         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5896         struct sk_buff *skb;
5897         int err = -ENOBUFS;
5898
5899         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
5900          * is implemented and supported for nexthop objects
5901          */
5902         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
5903
5904         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5905         if (!skb)
5906                 goto errout;
5907
5908         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5909                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
5910         if (err < 0) {
5911                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5912                 WARN_ON(err == -EMSGSIZE);
5913                 kfree_skb(skb);
5914                 goto errout;
5915         }
5916         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5917                     info->nlh, gfp_any());
5918         return;
5919 errout:
5920         if (err < 0)
5921                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5922 }
5923
5924 static int ip6_route_dev_notify(struct notifier_block *this,
5925                                 unsigned long event, void *ptr)
5926 {
5927         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
5928         struct net *net = dev_net(dev);
5929
5930         if (!(dev->flags & IFF_LOOPBACK))
5931                 return NOTIFY_OK;
5932
5933         if (event == NETDEV_REGISTER) {
5934                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
5935                 net->ipv6.ip6_null_entry->dst.dev = dev;
5936                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
5937 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5938                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
5939                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
5940                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
5941                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
5942 #endif
5943          } else if (event == NETDEV_UNREGISTER &&
5944                     dev->reg_state != NETREG_UNREGISTERED) {
5945                 /* NETDEV_UNREGISTER could be fired for multiple times by
5946                  * netdev_wait_allrefs(). Make sure we only call this once.
5947                  */
5948                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
5949 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5950                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
5951                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
5952 #endif
5953         }
5954
5955         return NOTIFY_OK;
5956 }
5957
5958 /*
5959  *      /proc
5960  */
5961
5962 #ifdef CONFIG_PROC_FS
5963 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
5964 {
5965         struct net *net = (struct net *)seq->private;
5966         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
5967                    net->ipv6.rt6_stats->fib_nodes,
5968                    net->ipv6.rt6_stats->fib_route_nodes,
5969                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
5970                    net->ipv6.rt6_stats->fib_rt_entries,
5971                    net->ipv6.rt6_stats->fib_rt_cache,
5972                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
5973                    net->ipv6.rt6_stats->fib_discarded_routes);
5974
5975         return 0;
5976 }
5977 #endif  /* CONFIG_PROC_FS */
5978
5979 #ifdef CONFIG_SYSCTL
5980
5981 static
5982 int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
5983                               void __user *buffer, size_t *lenp, loff_t *ppos)
5984 {
5985         struct net *net;
5986         int delay;
5987         int ret;
5988         if (!write)
5989                 return -EINVAL;
5990
5991         net = (struct net *)ctl->extra1;
5992         delay = net->ipv6.sysctl.flush_delay;
5993         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
5994         if (ret)
5995                 return ret;
5996
5997         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
5998         return 0;
5999 }
6000
6001 static int zero;
6002 static int one = 1;
6003
6004 static struct ctl_table ipv6_route_table_template[] = {
6005         {
6006                 .procname       =       "flush",
6007                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6008                 .maxlen         =       sizeof(int),
6009                 .mode           =       0200,
6010                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6011         },
6012         {
6013                 .procname       =       "gc_thresh",
6014                 .data           =       &ip6_dst_ops_template.gc_thresh,
6015                 .maxlen         =       sizeof(int),
6016                 .mode           =       0644,
6017                 .proc_handler   =       proc_dointvec,
6018         },
6019         {
6020                 .procname       =       "max_size",
6021                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6022                 .maxlen         =       sizeof(int),
6023                 .mode           =       0644,
6024                 .proc_handler   =       proc_dointvec,
6025         },
6026         {
6027                 .procname       =       "gc_min_interval",
6028                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6029                 .maxlen         =       sizeof(int),
6030                 .mode           =       0644,
6031                 .proc_handler   =       proc_dointvec_jiffies,
6032         },
6033         {
6034                 .procname       =       "gc_timeout",
6035                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6036                 .maxlen         =       sizeof(int),
6037                 .mode           =       0644,
6038                 .proc_handler   =       proc_dointvec_jiffies,
6039         },
6040         {
6041                 .procname       =       "gc_interval",
6042                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6043                 .maxlen         =       sizeof(int),
6044                 .mode           =       0644,
6045                 .proc_handler   =       proc_dointvec_jiffies,
6046         },
6047         {
6048                 .procname       =       "gc_elasticity",
6049                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6050                 .maxlen         =       sizeof(int),
6051                 .mode           =       0644,
6052                 .proc_handler   =       proc_dointvec,
6053         },
6054         {
6055                 .procname       =       "mtu_expires",
6056                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6057                 .maxlen         =       sizeof(int),
6058                 .mode           =       0644,
6059                 .proc_handler   =       proc_dointvec_jiffies,
6060         },
6061         {
6062                 .procname       =       "min_adv_mss",
6063                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6064                 .maxlen         =       sizeof(int),
6065                 .mode           =       0644,
6066                 .proc_handler   =       proc_dointvec,
6067         },
6068         {
6069                 .procname       =       "gc_min_interval_ms",
6070                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6071                 .maxlen         =       sizeof(int),
6072                 .mode           =       0644,
6073                 .proc_handler   =       proc_dointvec_ms_jiffies,
6074         },
6075         {
6076                 .procname       =       "skip_notify_on_dev_down",
6077                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6078                 .maxlen         =       sizeof(int),
6079                 .mode           =       0644,
6080                 .proc_handler   =       proc_dointvec,
6081                 .extra1         =       &zero,
6082                 .extra2         =       &one,
6083         },
6084         { }
6085 };
6086
6087 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6088 {
6089         struct ctl_table *table;
6090
6091         table = kmemdup(ipv6_route_table_template,
6092                         sizeof(ipv6_route_table_template),
6093                         GFP_KERNEL);
6094
6095         if (table) {
6096                 table[0].data = &net->ipv6.sysctl.flush_delay;
6097                 table[0].extra1 = net;
6098                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6099                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6100                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6101                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6102                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6103                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6104                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6105                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6106                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6107                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6108
6109                 /* Don't export sysctls to unprivileged users */
6110                 if (net->user_ns != &init_user_ns)
6111                         table[0].procname = NULL;
6112         }
6113
6114         return table;
6115 }
6116 #endif
6117
6118 static int __net_init ip6_route_net_init(struct net *net)
6119 {
6120         int ret = -ENOMEM;
6121
6122         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6123                sizeof(net->ipv6.ip6_dst_ops));
6124
6125         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6126                 goto out_ip6_dst_ops;
6127
6128         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6129         if (!net->ipv6.fib6_null_entry)
6130                 goto out_ip6_dst_entries;
6131         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6132                sizeof(*net->ipv6.fib6_null_entry));
6133
6134         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6135                                            sizeof(*net->ipv6.ip6_null_entry),
6136                                            GFP_KERNEL);
6137         if (!net->ipv6.ip6_null_entry)
6138                 goto out_fib6_null_entry;
6139         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6140         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6141                          ip6_template_metrics, true);
6142         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6143
6144 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6145         net->ipv6.fib6_has_custom_rules = false;
6146         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6147                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6148                                                GFP_KERNEL);
6149         if (!net->ipv6.ip6_prohibit_entry)
6150                 goto out_ip6_null_entry;
6151         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6152         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6153                          ip6_template_metrics, true);
6154         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6155
6156         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6157                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6158                                                GFP_KERNEL);
6159         if (!net->ipv6.ip6_blk_hole_entry)
6160                 goto out_ip6_prohibit_entry;
6161         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6162         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6163                          ip6_template_metrics, true);
6164         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6165 #endif
6166
6167         net->ipv6.sysctl.flush_delay = 0;
6168         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6169         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6170         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6171         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6172         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6173         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6174         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6175         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6176
6177         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6178
6179         ret = 0;
6180 out:
6181         return ret;
6182
6183 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6184 out_ip6_prohibit_entry:
6185         kfree(net->ipv6.ip6_prohibit_entry);
6186 out_ip6_null_entry:
6187         kfree(net->ipv6.ip6_null_entry);
6188 #endif
6189 out_fib6_null_entry:
6190         kfree(net->ipv6.fib6_null_entry);
6191 out_ip6_dst_entries:
6192         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6193 out_ip6_dst_ops:
6194         goto out;
6195 }
6196
6197 static void __net_exit ip6_route_net_exit(struct net *net)
6198 {
6199         kfree(net->ipv6.fib6_null_entry);
6200         kfree(net->ipv6.ip6_null_entry);
6201 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6202         kfree(net->ipv6.ip6_prohibit_entry);
6203         kfree(net->ipv6.ip6_blk_hole_entry);
6204 #endif
6205         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6206 }
6207
6208 static int __net_init ip6_route_net_init_late(struct net *net)
6209 {
6210 #ifdef CONFIG_PROC_FS
6211         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6212                         sizeof(struct ipv6_route_iter));
6213         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6214                         rt6_stats_seq_show, NULL);
6215 #endif
6216         return 0;
6217 }
6218
6219 static void __net_exit ip6_route_net_exit_late(struct net *net)
6220 {
6221 #ifdef CONFIG_PROC_FS
6222         remove_proc_entry("ipv6_route", net->proc_net);
6223         remove_proc_entry("rt6_stats", net->proc_net);
6224 #endif
6225 }
6226
6227 static struct pernet_operations ip6_route_net_ops = {
6228         .init = ip6_route_net_init,
6229         .exit = ip6_route_net_exit,
6230 };
6231
6232 static int __net_init ipv6_inetpeer_init(struct net *net)
6233 {
6234         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6235
6236         if (!bp)
6237                 return -ENOMEM;
6238         inet_peer_base_init(bp);
6239         net->ipv6.peers = bp;
6240         return 0;
6241 }
6242
6243 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6244 {
6245         struct inet_peer_base *bp = net->ipv6.peers;
6246
6247         net->ipv6.peers = NULL;
6248         inetpeer_invalidate_tree(bp);
6249         kfree(bp);
6250 }
6251
6252 static struct pernet_operations ipv6_inetpeer_ops = {
6253         .init   =       ipv6_inetpeer_init,
6254         .exit   =       ipv6_inetpeer_exit,
6255 };
6256
6257 static struct pernet_operations ip6_route_net_late_ops = {
6258         .init = ip6_route_net_init_late,
6259         .exit = ip6_route_net_exit_late,
6260 };
6261
6262 static struct notifier_block ip6_route_dev_notifier = {
6263         .notifier_call = ip6_route_dev_notify,
6264         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6265 };
6266
6267 void __init ip6_route_init_special_entries(void)
6268 {
6269         /* Registering of the loopback is done before this portion of code,
6270          * the loopback reference in rt6_info will not be taken, do it
6271          * manually for init_net */
6272         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6273         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6274         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6275   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6276         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6277         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6278         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6279         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6280   #endif
6281 }
6282
6283 int __init ip6_route_init(void)
6284 {
6285         int ret;
6286         int cpu;
6287
6288         ret = -ENOMEM;
6289         ip6_dst_ops_template.kmem_cachep =
6290                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6291                                   SLAB_HWCACHE_ALIGN, NULL);
6292         if (!ip6_dst_ops_template.kmem_cachep)
6293                 goto out;
6294
6295         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6296         if (ret)
6297                 goto out_kmem_cache;
6298
6299         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6300         if (ret)
6301                 goto out_dst_entries;
6302
6303         ret = register_pernet_subsys(&ip6_route_net_ops);
6304         if (ret)
6305                 goto out_register_inetpeer;
6306
6307         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6308
6309         ret = fib6_init();
6310         if (ret)
6311                 goto out_register_subsys;
6312
6313         ret = xfrm6_init();
6314         if (ret)
6315                 goto out_fib6_init;
6316
6317         ret = fib6_rules_init();
6318         if (ret)
6319                 goto xfrm6_init;
6320
6321         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6322         if (ret)
6323                 goto fib6_rules_init;
6324
6325         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6326                                    inet6_rtm_newroute, NULL, 0);
6327         if (ret < 0)
6328                 goto out_register_late_subsys;
6329
6330         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6331                                    inet6_rtm_delroute, NULL, 0);
6332         if (ret < 0)
6333                 goto out_register_late_subsys;
6334
6335         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6336                                    inet6_rtm_getroute, NULL,
6337                                    RTNL_FLAG_DOIT_UNLOCKED);
6338         if (ret < 0)
6339                 goto out_register_late_subsys;
6340
6341         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6342         if (ret)
6343                 goto out_register_late_subsys;
6344
6345         for_each_possible_cpu(cpu) {
6346                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6347
6348                 INIT_LIST_HEAD(&ul->head);
6349                 spin_lock_init(&ul->lock);
6350         }
6351
6352 out:
6353         return ret;
6354
6355 out_register_late_subsys:
6356         rtnl_unregister_all(PF_INET6);
6357         unregister_pernet_subsys(&ip6_route_net_late_ops);
6358 fib6_rules_init:
6359         fib6_rules_cleanup();
6360 xfrm6_init:
6361         xfrm6_fini();
6362 out_fib6_init:
6363         fib6_gc_cleanup();
6364 out_register_subsys:
6365         unregister_pernet_subsys(&ip6_route_net_ops);
6366 out_register_inetpeer:
6367         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6368 out_dst_entries:
6369         dst_entries_destroy(&ip6_dst_blackhole_ops);
6370 out_kmem_cache:
6371         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6372         goto out;
6373 }
6374
6375 void ip6_route_cleanup(void)
6376 {
6377         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6378         unregister_pernet_subsys(&ip6_route_net_late_ops);
6379         fib6_rules_cleanup();
6380         xfrm6_fini();
6381         fib6_gc_cleanup();
6382         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6383         unregister_pernet_subsys(&ip6_route_net_ops);
6384         dst_entries_destroy(&ip6_dst_blackhole_ops);
6385         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6386 }