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 *rt);
 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         if (pcpu_rt)
1395                 ip6_hold_safe(NULL, &pcpu_rt);
1396
1397         return pcpu_rt;
1398 }
1399
1400 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1401                                             const struct fib6_result *res)
1402 {
1403         struct rt6_info *pcpu_rt, *prev, **p;
1404
1405         pcpu_rt = ip6_rt_pcpu_alloc(res);
1406         if (!pcpu_rt) {
1407                 dst_hold(&net->ipv6.ip6_null_entry->dst);
1408                 return net->ipv6.ip6_null_entry;
1409         }
1410
1411         dst_hold(&pcpu_rt->dst);
1412         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1413         prev = cmpxchg(p, NULL, pcpu_rt);
1414         BUG_ON(prev);
1415
1416         if (res->f6i->fib6_destroying) {
1417                 struct fib6_info *from;
1418
1419                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1420                 fib6_info_release(from);
1421         }
1422
1423         return pcpu_rt;
1424 }
1425
1426 /* exception hash table implementation
1427  */
1428 static DEFINE_SPINLOCK(rt6_exception_lock);
1429
1430 /* Remove rt6_ex from hash table and free the memory
1431  * Caller must hold rt6_exception_lock
1432  */
1433 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1434                                  struct rt6_exception *rt6_ex)
1435 {
1436         struct fib6_info *from;
1437         struct net *net;
1438
1439         if (!bucket || !rt6_ex)
1440                 return;
1441
1442         net = dev_net(rt6_ex->rt6i->dst.dev);
1443         net->ipv6.rt6_stats->fib_rt_cache--;
1444
1445         /* purge completely the exception to allow releasing the held resources:
1446          * some [sk] cache may keep the dst around for unlimited time
1447          */
1448         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1449         fib6_info_release(from);
1450         dst_dev_put(&rt6_ex->rt6i->dst);
1451
1452         hlist_del_rcu(&rt6_ex->hlist);
1453         dst_release(&rt6_ex->rt6i->dst);
1454         kfree_rcu(rt6_ex, rcu);
1455         WARN_ON_ONCE(!bucket->depth);
1456         bucket->depth--;
1457 }
1458
1459 /* Remove oldest rt6_ex in bucket and free the memory
1460  * Caller must hold rt6_exception_lock
1461  */
1462 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1463 {
1464         struct rt6_exception *rt6_ex, *oldest = NULL;
1465
1466         if (!bucket)
1467                 return;
1468
1469         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1470                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1471                         oldest = rt6_ex;
1472         }
1473         rt6_remove_exception(bucket, oldest);
1474 }
1475
1476 static u32 rt6_exception_hash(const struct in6_addr *dst,
1477                               const struct in6_addr *src)
1478 {
1479         static u32 seed __read_mostly;
1480         u32 val;
1481
1482         net_get_random_once(&seed, sizeof(seed));
1483         val = jhash(dst, sizeof(*dst), seed);
1484
1485 #ifdef CONFIG_IPV6_SUBTREES
1486         if (src)
1487                 val = jhash(src, sizeof(*src), val);
1488 #endif
1489         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1490 }
1491
1492 /* Helper function to find the cached rt in the hash table
1493  * and update bucket pointer to point to the bucket for this
1494  * (daddr, saddr) pair
1495  * Caller must hold rt6_exception_lock
1496  */
1497 static struct rt6_exception *
1498 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1499                               const struct in6_addr *daddr,
1500                               const struct in6_addr *saddr)
1501 {
1502         struct rt6_exception *rt6_ex;
1503         u32 hval;
1504
1505         if (!(*bucket) || !daddr)
1506                 return NULL;
1507
1508         hval = rt6_exception_hash(daddr, saddr);
1509         *bucket += hval;
1510
1511         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1512                 struct rt6_info *rt6 = rt6_ex->rt6i;
1513                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1514
1515 #ifdef CONFIG_IPV6_SUBTREES
1516                 if (matched && saddr)
1517                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1518 #endif
1519                 if (matched)
1520                         return rt6_ex;
1521         }
1522         return NULL;
1523 }
1524
1525 /* Helper function to find the cached rt in the hash table
1526  * and update bucket pointer to point to the bucket for this
1527  * (daddr, saddr) pair
1528  * Caller must hold rcu_read_lock()
1529  */
1530 static struct rt6_exception *
1531 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1532                          const struct in6_addr *daddr,
1533                          const struct in6_addr *saddr)
1534 {
1535         struct rt6_exception *rt6_ex;
1536         u32 hval;
1537
1538         WARN_ON_ONCE(!rcu_read_lock_held());
1539
1540         if (!(*bucket) || !daddr)
1541                 return NULL;
1542
1543         hval = rt6_exception_hash(daddr, saddr);
1544         *bucket += hval;
1545
1546         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1547                 struct rt6_info *rt6 = rt6_ex->rt6i;
1548                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1549
1550 #ifdef CONFIG_IPV6_SUBTREES
1551                 if (matched && saddr)
1552                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1553 #endif
1554                 if (matched)
1555                         return rt6_ex;
1556         }
1557         return NULL;
1558 }
1559
1560 static unsigned int fib6_mtu(const struct fib6_result *res)
1561 {
1562         const struct fib6_nh *nh = res->nh;
1563         unsigned int mtu;
1564
1565         if (res->f6i->fib6_pmtu) {
1566                 mtu = res->f6i->fib6_pmtu;
1567         } else {
1568                 struct net_device *dev = nh->fib_nh_dev;
1569                 struct inet6_dev *idev;
1570
1571                 rcu_read_lock();
1572                 idev = __in6_dev_get(dev);
1573                 mtu = idev->cnf.mtu6;
1574                 rcu_read_unlock();
1575         }
1576
1577         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1578
1579         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1580 }
1581
1582 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1583
1584 /* used when the flushed bit is not relevant, only access to the bucket
1585  * (ie., all bucket users except rt6_insert_exception);
1586  *
1587  * called under rcu lock; sometimes called with rt6_exception_lock held
1588  */
1589 static
1590 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1591                                                        spinlock_t *lock)
1592 {
1593         struct rt6_exception_bucket *bucket;
1594
1595         if (lock)
1596                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1597                                                    lockdep_is_held(lock));
1598         else
1599                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1600
1601         /* remove bucket flushed bit if set */
1602         if (bucket) {
1603                 unsigned long p = (unsigned long)bucket;
1604
1605                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1606                 bucket = (struct rt6_exception_bucket *)p;
1607         }
1608
1609         return bucket;
1610 }
1611
1612 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1613 {
1614         unsigned long p = (unsigned long)bucket;
1615
1616         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1617 }
1618
1619 /* called with rt6_exception_lock held */
1620 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1621                                               spinlock_t *lock)
1622 {
1623         struct rt6_exception_bucket *bucket;
1624         unsigned long p;
1625
1626         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1627                                            lockdep_is_held(lock));
1628
1629         p = (unsigned long)bucket;
1630         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1631         bucket = (struct rt6_exception_bucket *)p;
1632         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1633 }
1634
1635 static int rt6_insert_exception(struct rt6_info *nrt,
1636                                 const struct fib6_result *res)
1637 {
1638         struct net *net = dev_net(nrt->dst.dev);
1639         struct rt6_exception_bucket *bucket;
1640         struct fib6_info *f6i = res->f6i;
1641         struct in6_addr *src_key = NULL;
1642         struct rt6_exception *rt6_ex;
1643         struct fib6_nh *nh = res->nh;
1644         int err = 0;
1645
1646         spin_lock_bh(&rt6_exception_lock);
1647
1648         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1649                                           lockdep_is_held(&rt6_exception_lock));
1650         if (!bucket) {
1651                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1652                                  GFP_ATOMIC);
1653                 if (!bucket) {
1654                         err = -ENOMEM;
1655                         goto out;
1656                 }
1657                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1658         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1659                 err = -EINVAL;
1660                 goto out;
1661         }
1662
1663 #ifdef CONFIG_IPV6_SUBTREES
1664         /* fib6_src.plen != 0 indicates f6i is in subtree
1665          * and exception table is indexed by a hash of
1666          * both fib6_dst and fib6_src.
1667          * Otherwise, the exception table is indexed by
1668          * a hash of only fib6_dst.
1669          */
1670         if (f6i->fib6_src.plen)
1671                 src_key = &nrt->rt6i_src.addr;
1672 #endif
1673         /* rt6_mtu_change() might lower mtu on f6i.
1674          * Only insert this exception route if its mtu
1675          * is less than f6i's mtu value.
1676          */
1677         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1678                 err = -EINVAL;
1679                 goto out;
1680         }
1681
1682         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1683                                                src_key);
1684         if (rt6_ex)
1685                 rt6_remove_exception(bucket, rt6_ex);
1686
1687         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1688         if (!rt6_ex) {
1689                 err = -ENOMEM;
1690                 goto out;
1691         }
1692         rt6_ex->rt6i = nrt;
1693         rt6_ex->stamp = jiffies;
1694         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1695         bucket->depth++;
1696         net->ipv6.rt6_stats->fib_rt_cache++;
1697
1698         if (bucket->depth > FIB6_MAX_DEPTH)
1699                 rt6_exception_remove_oldest(bucket);
1700
1701 out:
1702         spin_unlock_bh(&rt6_exception_lock);
1703
1704         /* Update fn->fn_sernum to invalidate all cached dst */
1705         if (!err) {
1706                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1707                 fib6_update_sernum(net, f6i);
1708                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1709                 fib6_force_start_gc(net);
1710         }
1711
1712         return err;
1713 }
1714
1715 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1716 {
1717         struct rt6_exception_bucket *bucket;
1718         struct rt6_exception *rt6_ex;
1719         struct hlist_node *tmp;
1720         int i;
1721
1722         spin_lock_bh(&rt6_exception_lock);
1723
1724         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1725         if (!bucket)
1726                 goto out;
1727
1728         /* Prevent rt6_insert_exception() to recreate the bucket list */
1729         if (!from)
1730                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1731
1732         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1733                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1734                         if (!from ||
1735                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1736                                 rt6_remove_exception(bucket, rt6_ex);
1737                 }
1738                 WARN_ON_ONCE(!from && bucket->depth);
1739                 bucket++;
1740         }
1741 out:
1742         spin_unlock_bh(&rt6_exception_lock);
1743 }
1744
1745 void rt6_flush_exceptions(struct fib6_info *f6i)
1746 {
1747         fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1748 }
1749
1750 /* Find cached rt in the hash table inside passed in rt
1751  * Caller has to hold rcu_read_lock()
1752  */
1753 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1754                                            const struct in6_addr *daddr,
1755                                            const struct in6_addr *saddr)
1756 {
1757         const struct in6_addr *src_key = NULL;
1758         struct rt6_exception_bucket *bucket;
1759         struct rt6_exception *rt6_ex;
1760         struct rt6_info *ret = NULL;
1761
1762 #ifdef CONFIG_IPV6_SUBTREES
1763         /* fib6i_src.plen != 0 indicates f6i is in subtree
1764          * and exception table is indexed by a hash of
1765          * both fib6_dst and fib6_src.
1766          * However, the src addr used to create the hash
1767          * might not be exactly the passed in saddr which
1768          * is a /128 addr from the flow.
1769          * So we need to use f6i->fib6_src to redo lookup
1770          * if the passed in saddr does not find anything.
1771          * (See the logic in ip6_rt_cache_alloc() on how
1772          * rt->rt6i_src is updated.)
1773          */
1774         if (res->f6i->fib6_src.plen)
1775                 src_key = saddr;
1776 find_ex:
1777 #endif
1778         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1779         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1780
1781         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1782                 ret = rt6_ex->rt6i;
1783
1784 #ifdef CONFIG_IPV6_SUBTREES
1785         /* Use fib6_src as src_key and redo lookup */
1786         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1787                 src_key = &res->f6i->fib6_src.addr;
1788                 goto find_ex;
1789         }
1790 #endif
1791
1792         return ret;
1793 }
1794
1795 /* Remove the passed in cached rt from the hash table that contains it */
1796 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1797                                     const struct rt6_info *rt)
1798 {
1799         const struct in6_addr *src_key = NULL;
1800         struct rt6_exception_bucket *bucket;
1801         struct rt6_exception *rt6_ex;
1802         int err;
1803
1804         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1805                 return -ENOENT;
1806
1807         spin_lock_bh(&rt6_exception_lock);
1808         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1809
1810 #ifdef CONFIG_IPV6_SUBTREES
1811         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1812          * and exception table is indexed by a hash of
1813          * both rt6i_dst and rt6i_src.
1814          * Otherwise, the exception table is indexed by
1815          * a hash of only rt6i_dst.
1816          */
1817         if (plen)
1818                 src_key = &rt->rt6i_src.addr;
1819 #endif
1820         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1821                                                &rt->rt6i_dst.addr,
1822                                                src_key);
1823         if (rt6_ex) {
1824                 rt6_remove_exception(bucket, rt6_ex);
1825                 err = 0;
1826         } else {
1827                 err = -ENOENT;
1828         }
1829
1830         spin_unlock_bh(&rt6_exception_lock);
1831         return err;
1832 }
1833
1834 static int rt6_remove_exception_rt(struct rt6_info *rt)
1835 {
1836         struct fib6_info *from;
1837
1838         from = rcu_dereference(rt->from);
1839         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1840                 return -EINVAL;
1841
1842         return fib6_nh_remove_exception(from->fib6_nh,
1843                                         from->fib6_src.plen, rt);
1844 }
1845
1846 /* Find rt6_ex which contains the passed in rt cache and
1847  * refresh its stamp
1848  */
1849 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1850                                      const struct rt6_info *rt)
1851 {
1852         const struct in6_addr *src_key = NULL;
1853         struct rt6_exception_bucket *bucket;
1854         struct rt6_exception *rt6_ex;
1855
1856         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1857 #ifdef CONFIG_IPV6_SUBTREES
1858         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1859          * and exception table is indexed by a hash of
1860          * both rt6i_dst and rt6i_src.
1861          * Otherwise, the exception table is indexed by
1862          * a hash of only rt6i_dst.
1863          */
1864         if (plen)
1865                 src_key = &rt->rt6i_src.addr;
1866 #endif
1867         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1868         if (rt6_ex)
1869                 rt6_ex->stamp = jiffies;
1870 }
1871
1872 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1873 {
1874         struct fib6_info *from;
1875
1876         rcu_read_lock();
1877
1878         from = rcu_dereference(rt->from);
1879         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1880                 goto unlock;
1881
1882         fib6_nh_update_exception(from->fib6_nh, from->fib6_src.plen, rt);
1883 unlock:
1884         rcu_read_unlock();
1885 }
1886
1887 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1888                                          struct rt6_info *rt, int mtu)
1889 {
1890         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1891          * lowest MTU in the path: always allow updating the route PMTU to
1892          * reflect PMTU decreases.
1893          *
1894          * If the new MTU is higher, and the route PMTU is equal to the local
1895          * MTU, this means the old MTU is the lowest in the path, so allow
1896          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1897          * handle this.
1898          */
1899
1900         if (dst_mtu(&rt->dst) >= mtu)
1901                 return true;
1902
1903         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1904                 return true;
1905
1906         return false;
1907 }
1908
1909 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
1910                                        const struct fib6_nh *nh, int mtu)
1911 {
1912         struct rt6_exception_bucket *bucket;
1913         struct rt6_exception *rt6_ex;
1914         int i;
1915
1916         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1917         if (!bucket)
1918                 return;
1919
1920         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1921                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1922                         struct rt6_info *entry = rt6_ex->rt6i;
1923
1924                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
1925                          * route), the metrics of its rt->from have already
1926                          * been updated.
1927                          */
1928                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
1929                             rt6_mtu_change_route_allowed(idev, entry, mtu))
1930                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
1931                 }
1932                 bucket++;
1933         }
1934 }
1935
1936 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
1937
1938 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
1939                                             const struct in6_addr *gateway)
1940 {
1941         struct rt6_exception_bucket *bucket;
1942         struct rt6_exception *rt6_ex;
1943         struct hlist_node *tmp;
1944         int i;
1945
1946         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1947                 return;
1948
1949         spin_lock_bh(&rt6_exception_lock);
1950         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1951         if (bucket) {
1952                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1953                         hlist_for_each_entry_safe(rt6_ex, tmp,
1954                                                   &bucket->chain, hlist) {
1955                                 struct rt6_info *entry = rt6_ex->rt6i;
1956
1957                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
1958                                     RTF_CACHE_GATEWAY &&
1959                                     ipv6_addr_equal(gateway,
1960                                                     &entry->rt6i_gateway)) {
1961                                         rt6_remove_exception(bucket, rt6_ex);
1962                                 }
1963                         }
1964                         bucket++;
1965                 }
1966         }
1967
1968         spin_unlock_bh(&rt6_exception_lock);
1969 }
1970
1971 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
1972                                       struct rt6_exception *rt6_ex,
1973                                       struct fib6_gc_args *gc_args,
1974                                       unsigned long now)
1975 {
1976         struct rt6_info *rt = rt6_ex->rt6i;
1977
1978         /* we are pruning and obsoleting aged-out and non gateway exceptions
1979          * even if others have still references to them, so that on next
1980          * dst_check() such references can be dropped.
1981          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
1982          * expired, independently from their aging, as per RFC 8201 section 4
1983          */
1984         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
1985                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
1986                         RT6_TRACE("aging clone %p\n", rt);
1987                         rt6_remove_exception(bucket, rt6_ex);
1988                         return;
1989                 }
1990         } else if (time_after(jiffies, rt->dst.expires)) {
1991                 RT6_TRACE("purging expired route %p\n", rt);
1992                 rt6_remove_exception(bucket, rt6_ex);
1993                 return;
1994         }
1995
1996         if (rt->rt6i_flags & RTF_GATEWAY) {
1997                 struct neighbour *neigh;
1998                 __u8 neigh_flags = 0;
1999
2000                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2001                 if (neigh)
2002                         neigh_flags = neigh->flags;
2003
2004                 if (!(neigh_flags & NTF_ROUTER)) {
2005                         RT6_TRACE("purging route %p via non-router but gateway\n",
2006                                   rt);
2007                         rt6_remove_exception(bucket, rt6_ex);
2008                         return;
2009                 }
2010         }
2011
2012         gc_args->more++;
2013 }
2014
2015 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2016                                    struct fib6_gc_args *gc_args,
2017                                    unsigned long now)
2018 {
2019         struct rt6_exception_bucket *bucket;
2020         struct rt6_exception *rt6_ex;
2021         struct hlist_node *tmp;
2022         int i;
2023
2024         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2025                 return;
2026
2027         rcu_read_lock_bh();
2028         spin_lock(&rt6_exception_lock);
2029         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2030         if (bucket) {
2031                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2032                         hlist_for_each_entry_safe(rt6_ex, tmp,
2033                                                   &bucket->chain, hlist) {
2034                                 rt6_age_examine_exception(bucket, rt6_ex,
2035                                                           gc_args, now);
2036                         }
2037                         bucket++;
2038                 }
2039         }
2040         spin_unlock(&rt6_exception_lock);
2041         rcu_read_unlock_bh();
2042 }
2043
2044 void rt6_age_exceptions(struct fib6_info *f6i,
2045                         struct fib6_gc_args *gc_args,
2046                         unsigned long now)
2047 {
2048         fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2049 }
2050
2051 /* must be called with rcu lock held */
2052 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2053                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2054 {
2055         struct fib6_node *fn, *saved_fn;
2056
2057         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2058         saved_fn = fn;
2059
2060         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2061                 oif = 0;
2062
2063 redo_rt6_select:
2064         rt6_select(net, fn, oif, res, strict);
2065         if (res->f6i == net->ipv6.fib6_null_entry) {
2066                 fn = fib6_backtrack(fn, &fl6->saddr);
2067                 if (fn)
2068                         goto redo_rt6_select;
2069                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2070                         /* also consider unreachable route */
2071                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2072                         fn = saved_fn;
2073                         goto redo_rt6_select;
2074                 }
2075         }
2076
2077         trace_fib6_table_lookup(net, res, table, fl6);
2078
2079         return 0;
2080 }
2081
2082 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2083                                int oif, struct flowi6 *fl6,
2084                                const struct sk_buff *skb, int flags)
2085 {
2086         struct fib6_result res = {};
2087         struct rt6_info *rt;
2088         int strict = 0;
2089
2090         strict |= flags & RT6_LOOKUP_F_IFACE;
2091         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2092         if (net->ipv6.devconf_all->forwarding == 0)
2093                 strict |= RT6_LOOKUP_F_REACHABLE;
2094
2095         rcu_read_lock();
2096
2097         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2098         if (res.f6i == net->ipv6.fib6_null_entry) {
2099                 rt = net->ipv6.ip6_null_entry;
2100                 rcu_read_unlock();
2101                 dst_hold(&rt->dst);
2102                 return rt;
2103         }
2104
2105         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2106
2107         /*Search through exception table */
2108         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2109         if (rt) {
2110                 if (ip6_hold_safe(net, &rt))
2111                         dst_use_noref(&rt->dst, jiffies);
2112
2113                 rcu_read_unlock();
2114                 return rt;
2115         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2116                             !res.nh->fib_nh_gw_family)) {
2117                 /* Create a RTF_CACHE clone which will not be
2118                  * owned by the fib6 tree.  It is for the special case where
2119                  * the daddr in the skb during the neighbor look-up is different
2120                  * from the fl6->daddr used to look-up route here.
2121                  */
2122                 struct rt6_info *uncached_rt;
2123
2124                 uncached_rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2125
2126                 rcu_read_unlock();
2127
2128                 if (uncached_rt) {
2129                         /* Uncached_rt's refcnt is taken during ip6_rt_cache_alloc()
2130                          * No need for another dst_hold()
2131                          */
2132                         rt6_uncached_list_add(uncached_rt);
2133                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2134                 } else {
2135                         uncached_rt = net->ipv6.ip6_null_entry;
2136                         dst_hold(&uncached_rt->dst);
2137                 }
2138
2139                 return uncached_rt;
2140         } else {
2141                 /* Get a percpu copy */
2142
2143                 struct rt6_info *pcpu_rt;
2144
2145                 local_bh_disable();
2146                 pcpu_rt = rt6_get_pcpu_route(&res);
2147
2148                 if (!pcpu_rt)
2149                         pcpu_rt = rt6_make_pcpu_route(net, &res);
2150
2151                 local_bh_enable();
2152                 rcu_read_unlock();
2153
2154                 return pcpu_rt;
2155         }
2156 }
2157 EXPORT_SYMBOL_GPL(ip6_pol_route);
2158
2159 static struct rt6_info *ip6_pol_route_input(struct net *net,
2160                                             struct fib6_table *table,
2161                                             struct flowi6 *fl6,
2162                                             const struct sk_buff *skb,
2163                                             int flags)
2164 {
2165         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2166 }
2167
2168 struct dst_entry *ip6_route_input_lookup(struct net *net,
2169                                          struct net_device *dev,
2170                                          struct flowi6 *fl6,
2171                                          const struct sk_buff *skb,
2172                                          int flags)
2173 {
2174         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2175                 flags |= RT6_LOOKUP_F_IFACE;
2176
2177         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2178 }
2179 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2180
2181 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2182                                   struct flow_keys *keys,
2183                                   struct flow_keys *flkeys)
2184 {
2185         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2186         const struct ipv6hdr *key_iph = outer_iph;
2187         struct flow_keys *_flkeys = flkeys;
2188         const struct ipv6hdr *inner_iph;
2189         const struct icmp6hdr *icmph;
2190         struct ipv6hdr _inner_iph;
2191         struct icmp6hdr _icmph;
2192
2193         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2194                 goto out;
2195
2196         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2197                                    sizeof(_icmph), &_icmph);
2198         if (!icmph)
2199                 goto out;
2200
2201         if (icmph->icmp6_type != ICMPV6_DEST_UNREACH &&
2202             icmph->icmp6_type != ICMPV6_PKT_TOOBIG &&
2203             icmph->icmp6_type != ICMPV6_TIME_EXCEED &&
2204             icmph->icmp6_type != ICMPV6_PARAMPROB)
2205                 goto out;
2206
2207         inner_iph = skb_header_pointer(skb,
2208                                        skb_transport_offset(skb) + sizeof(*icmph),
2209                                        sizeof(_inner_iph), &_inner_iph);
2210         if (!inner_iph)
2211                 goto out;
2212
2213         key_iph = inner_iph;
2214         _flkeys = NULL;
2215 out:
2216         if (_flkeys) {
2217                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2218                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2219                 keys->tags.flow_label = _flkeys->tags.flow_label;
2220                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2221         } else {
2222                 keys->addrs.v6addrs.src = key_iph->saddr;
2223                 keys->addrs.v6addrs.dst = key_iph->daddr;
2224                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2225                 keys->basic.ip_proto = key_iph->nexthdr;
2226         }
2227 }
2228
2229 /* if skb is set it will be used and fl6 can be NULL */
2230 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2231                        const struct sk_buff *skb, struct flow_keys *flkeys)
2232 {
2233         struct flow_keys hash_keys;
2234         u32 mhash;
2235
2236         switch (ip6_multipath_hash_policy(net)) {
2237         case 0:
2238                 memset(&hash_keys, 0, sizeof(hash_keys));
2239                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2240                 if (skb) {
2241                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2242                 } else {
2243                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2244                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2245                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2246                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2247                 }
2248                 break;
2249         case 1:
2250                 if (skb) {
2251                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2252                         struct flow_keys keys;
2253
2254                         /* short-circuit if we already have L4 hash present */
2255                         if (skb->l4_hash)
2256                                 return skb_get_hash_raw(skb) >> 1;
2257
2258                         memset(&hash_keys, 0, sizeof(hash_keys));
2259
2260                         if (!flkeys) {
2261                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2262                                 flkeys = &keys;
2263                         }
2264                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2265                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2266                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2267                         hash_keys.ports.src = flkeys->ports.src;
2268                         hash_keys.ports.dst = flkeys->ports.dst;
2269                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2270                 } else {
2271                         memset(&hash_keys, 0, sizeof(hash_keys));
2272                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2273                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2274                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2275                         hash_keys.ports.src = fl6->fl6_sport;
2276                         hash_keys.ports.dst = fl6->fl6_dport;
2277                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2278                 }
2279                 break;
2280         }
2281         mhash = flow_hash_from_keys(&hash_keys);
2282
2283         return mhash >> 1;
2284 }
2285
2286 void ip6_route_input(struct sk_buff *skb)
2287 {
2288         const struct ipv6hdr *iph = ipv6_hdr(skb);
2289         struct net *net = dev_net(skb->dev);
2290         int flags = RT6_LOOKUP_F_HAS_SADDR;
2291         struct ip_tunnel_info *tun_info;
2292         struct flowi6 fl6 = {
2293                 .flowi6_iif = skb->dev->ifindex,
2294                 .daddr = iph->daddr,
2295                 .saddr = iph->saddr,
2296                 .flowlabel = ip6_flowinfo(iph),
2297                 .flowi6_mark = skb->mark,
2298                 .flowi6_proto = iph->nexthdr,
2299         };
2300         struct flow_keys *flkeys = NULL, _flkeys;
2301
2302         tun_info = skb_tunnel_info(skb);
2303         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2304                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2305
2306         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2307                 flkeys = &_flkeys;
2308
2309         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2310                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2311         skb_dst_drop(skb);
2312         skb_dst_set(skb,
2313                     ip6_route_input_lookup(net, skb->dev, &fl6, skb, flags));
2314 }
2315
2316 static struct rt6_info *ip6_pol_route_output(struct net *net,
2317                                              struct fib6_table *table,
2318                                              struct flowi6 *fl6,
2319                                              const struct sk_buff *skb,
2320                                              int flags)
2321 {
2322         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2323 }
2324
2325 struct dst_entry *ip6_route_output_flags(struct net *net, const struct sock *sk,
2326                                          struct flowi6 *fl6, int flags)
2327 {
2328         bool any_src;
2329
2330         if (ipv6_addr_type(&fl6->daddr) &
2331             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2332                 struct dst_entry *dst;
2333
2334                 dst = l3mdev_link_scope_lookup(net, fl6);
2335                 if (dst)
2336                         return dst;
2337         }
2338
2339         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2340
2341         any_src = ipv6_addr_any(&fl6->saddr);
2342         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2343             (fl6->flowi6_oif && any_src))
2344                 flags |= RT6_LOOKUP_F_IFACE;
2345
2346         if (!any_src)
2347                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2348         else if (sk)
2349                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2350
2351         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2352 }
2353 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2354
2355 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2356 {
2357         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2358         struct net_device *loopback_dev = net->loopback_dev;
2359         struct dst_entry *new = NULL;
2360
2361         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2362                        DST_OBSOLETE_DEAD, 0);
2363         if (rt) {
2364                 rt6_info_init(rt);
2365                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2366
2367                 new = &rt->dst;
2368                 new->__use = 1;
2369                 new->input = dst_discard;
2370                 new->output = dst_discard_out;
2371
2372                 dst_copy_metrics(new, &ort->dst);
2373
2374                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2375                 rt->rt6i_gateway = ort->rt6i_gateway;
2376                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2377
2378                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2379 #ifdef CONFIG_IPV6_SUBTREES
2380                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2381 #endif
2382         }
2383
2384         dst_release(dst_orig);
2385         return new ? new : ERR_PTR(-ENOMEM);
2386 }
2387
2388 /*
2389  *      Destination cache support functions
2390  */
2391
2392 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2393 {
2394         u32 rt_cookie = 0;
2395
2396         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2397                 return false;
2398
2399         if (fib6_check_expired(f6i))
2400                 return false;
2401
2402         return true;
2403 }
2404
2405 static struct dst_entry *rt6_check(struct rt6_info *rt,
2406                                    struct fib6_info *from,
2407                                    u32 cookie)
2408 {
2409         u32 rt_cookie = 0;
2410
2411         if ((from && !fib6_get_cookie_safe(from, &rt_cookie)) ||
2412             rt_cookie != cookie)
2413                 return NULL;
2414
2415         if (rt6_check_expired(rt))
2416                 return NULL;
2417
2418         return &rt->dst;
2419 }
2420
2421 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2422                                             struct fib6_info *from,
2423                                             u32 cookie)
2424 {
2425         if (!__rt6_check_expired(rt) &&
2426             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2427             fib6_check(from, cookie))
2428                 return &rt->dst;
2429         else
2430                 return NULL;
2431 }
2432
2433 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2434 {
2435         struct dst_entry *dst_ret;
2436         struct fib6_info *from;
2437         struct rt6_info *rt;
2438
2439         rt = container_of(dst, struct rt6_info, dst);
2440
2441         rcu_read_lock();
2442
2443         /* All IPV6 dsts are created with ->obsolete set to the value
2444          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2445          * into this function always.
2446          */
2447
2448         from = rcu_dereference(rt->from);
2449
2450         if (from && (rt->rt6i_flags & RTF_PCPU ||
2451             unlikely(!list_empty(&rt->rt6i_uncached))))
2452                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2453         else
2454                 dst_ret = rt6_check(rt, from, cookie);
2455
2456         rcu_read_unlock();
2457
2458         return dst_ret;
2459 }
2460
2461 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2462 {
2463         struct rt6_info *rt = (struct rt6_info *) dst;
2464
2465         if (rt) {
2466                 if (rt->rt6i_flags & RTF_CACHE) {
2467                         rcu_read_lock();
2468                         if (rt6_check_expired(rt)) {
2469                                 rt6_remove_exception_rt(rt);
2470                                 dst = NULL;
2471                         }
2472                         rcu_read_unlock();
2473                 } else {
2474                         dst_release(dst);
2475                         dst = NULL;
2476                 }
2477         }
2478         return dst;
2479 }
2480
2481 static void ip6_link_failure(struct sk_buff *skb)
2482 {
2483         struct rt6_info *rt;
2484
2485         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2486
2487         rt = (struct rt6_info *) skb_dst(skb);
2488         if (rt) {
2489                 rcu_read_lock();
2490                 if (rt->rt6i_flags & RTF_CACHE) {
2491                         rt6_remove_exception_rt(rt);
2492                 } else {
2493                         struct fib6_info *from;
2494                         struct fib6_node *fn;
2495
2496                         from = rcu_dereference(rt->from);
2497                         if (from) {
2498                                 fn = rcu_dereference(from->fib6_node);
2499                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2500                                         fn->fn_sernum = -1;
2501                         }
2502                 }
2503                 rcu_read_unlock();
2504         }
2505 }
2506
2507 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2508 {
2509         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2510                 struct fib6_info *from;
2511
2512                 rcu_read_lock();
2513                 from = rcu_dereference(rt0->from);
2514                 if (from)
2515                         rt0->dst.expires = from->expires;
2516                 rcu_read_unlock();
2517         }
2518
2519         dst_set_expires(&rt0->dst, timeout);
2520         rt0->rt6i_flags |= RTF_EXPIRES;
2521 }
2522
2523 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2524 {
2525         struct net *net = dev_net(rt->dst.dev);
2526
2527         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2528         rt->rt6i_flags |= RTF_MODIFIED;
2529         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2530 }
2531
2532 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2533 {
2534         return !(rt->rt6i_flags & RTF_CACHE) &&
2535                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2536 }
2537
2538 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2539                                  const struct ipv6hdr *iph, u32 mtu)
2540 {
2541         const struct in6_addr *daddr, *saddr;
2542         struct rt6_info *rt6 = (struct rt6_info *)dst;
2543
2544         if (dst_metric_locked(dst, RTAX_MTU))
2545                 return;
2546
2547         if (iph) {
2548                 daddr = &iph->daddr;
2549                 saddr = &iph->saddr;
2550         } else if (sk) {
2551                 daddr = &sk->sk_v6_daddr;
2552                 saddr = &inet6_sk(sk)->saddr;
2553         } else {
2554                 daddr = NULL;
2555                 saddr = NULL;
2556         }
2557         dst_confirm_neigh(dst, daddr);
2558         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2559         if (mtu >= dst_mtu(dst))
2560                 return;
2561
2562         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2563                 rt6_do_update_pmtu(rt6, mtu);
2564                 /* update rt6_ex->stamp for cache */
2565                 if (rt6->rt6i_flags & RTF_CACHE)
2566                         rt6_update_exception_stamp_rt(rt6);
2567         } else if (daddr) {
2568                 struct fib6_result res = {};
2569                 struct rt6_info *nrt6;
2570
2571                 rcu_read_lock();
2572                 res.f6i = rcu_dereference(rt6->from);
2573                 if (!res.f6i) {
2574                         rcu_read_unlock();
2575                         return;
2576                 }
2577                 res.nh = res.f6i->fib6_nh;
2578                 res.fib6_flags = res.f6i->fib6_flags;
2579                 res.fib6_type = res.f6i->fib6_type;
2580
2581                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2582                 if (nrt6) {
2583                         rt6_do_update_pmtu(nrt6, mtu);
2584                         if (rt6_insert_exception(nrt6, &res))
2585                                 dst_release_immediate(&nrt6->dst);
2586                 }
2587                 rcu_read_unlock();
2588         }
2589 }
2590
2591 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2592                                struct sk_buff *skb, u32 mtu)
2593 {
2594         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu);
2595 }
2596
2597 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2598                      int oif, u32 mark, kuid_t uid)
2599 {
2600         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2601         struct dst_entry *dst;
2602         struct flowi6 fl6 = {
2603                 .flowi6_oif = oif,
2604                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2605                 .daddr = iph->daddr,
2606                 .saddr = iph->saddr,
2607                 .flowlabel = ip6_flowinfo(iph),
2608                 .flowi6_uid = uid,
2609         };
2610
2611         dst = ip6_route_output(net, NULL, &fl6);
2612         if (!dst->error)
2613                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu));
2614         dst_release(dst);
2615 }
2616 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2617
2618 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2619 {
2620         int oif = sk->sk_bound_dev_if;
2621         struct dst_entry *dst;
2622
2623         if (!oif && skb->dev)
2624                 oif = l3mdev_master_ifindex(skb->dev);
2625
2626         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2627
2628         dst = __sk_dst_get(sk);
2629         if (!dst || !dst->obsolete ||
2630             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2631                 return;
2632
2633         bh_lock_sock(sk);
2634         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2635                 ip6_datagram_dst_update(sk, false);
2636         bh_unlock_sock(sk);
2637 }
2638 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2639
2640 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2641                            const struct flowi6 *fl6)
2642 {
2643 #ifdef CONFIG_IPV6_SUBTREES
2644         struct ipv6_pinfo *np = inet6_sk(sk);
2645 #endif
2646
2647         ip6_dst_store(sk, dst,
2648                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2649                       &sk->sk_v6_daddr : NULL,
2650 #ifdef CONFIG_IPV6_SUBTREES
2651                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2652                       &np->saddr :
2653 #endif
2654                       NULL);
2655 }
2656
2657 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2658                                   struct flowi6 *fl6,
2659                                   const struct in6_addr *gw,
2660                                   struct rt6_info **ret)
2661 {
2662         const struct fib6_nh *nh = res->nh;
2663
2664         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2665             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2666                 return false;
2667
2668         /* rt_cache's gateway might be different from its 'parent'
2669          * in the case of an ip redirect.
2670          * So we keep searching in the exception table if the gateway
2671          * is different.
2672          */
2673         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2674                 struct rt6_info *rt_cache;
2675
2676                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2677                 if (rt_cache &&
2678                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2679                         *ret = rt_cache;
2680                         return true;
2681                 }
2682                 return false;
2683         }
2684         return true;
2685 }
2686
2687 /* Handle redirects */
2688 struct ip6rd_flowi {
2689         struct flowi6 fl6;
2690         struct in6_addr gateway;
2691 };
2692
2693 static struct rt6_info *__ip6_route_redirect(struct net *net,
2694                                              struct fib6_table *table,
2695                                              struct flowi6 *fl6,
2696                                              const struct sk_buff *skb,
2697                                              int flags)
2698 {
2699         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2700         struct rt6_info *ret = NULL;
2701         struct fib6_result res = {};
2702         struct fib6_info *rt;
2703         struct fib6_node *fn;
2704
2705         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2706          * this case we must match on the real ingress device, so reset it
2707          */
2708         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2709                 fl6->flowi6_oif = skb->dev->ifindex;
2710
2711         /* Get the "current" route for this destination and
2712          * check if the redirect has come from appropriate router.
2713          *
2714          * RFC 4861 specifies that redirects should only be
2715          * accepted if they come from the nexthop to the target.
2716          * Due to the way the routes are chosen, this notion
2717          * is a bit fuzzy and one might need to check all possible
2718          * routes.
2719          */
2720
2721         rcu_read_lock();
2722         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2723 restart:
2724         for_each_fib6_node_rt_rcu(fn) {
2725                 res.f6i = rt;
2726                 res.nh = rt->fib6_nh;
2727
2728                 if (fib6_check_expired(rt))
2729                         continue;
2730                 if (rt->fib6_flags & RTF_REJECT)
2731                         break;
2732                 if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway, &ret))
2733                         goto out;
2734         }
2735
2736         if (!rt)
2737                 rt = net->ipv6.fib6_null_entry;
2738         else if (rt->fib6_flags & RTF_REJECT) {
2739                 ret = net->ipv6.ip6_null_entry;
2740                 goto out;
2741         }
2742
2743         if (rt == net->ipv6.fib6_null_entry) {
2744                 fn = fib6_backtrack(fn, &fl6->saddr);
2745                 if (fn)
2746                         goto restart;
2747         }
2748
2749         res.f6i = rt;
2750         res.nh = rt->fib6_nh;
2751 out:
2752         if (ret) {
2753                 ip6_hold_safe(net, &ret);
2754         } else {
2755                 res.fib6_flags = res.f6i->fib6_flags;
2756                 res.fib6_type = res.f6i->fib6_type;
2757                 ret = ip6_create_rt_rcu(&res);
2758         }
2759
2760         rcu_read_unlock();
2761
2762         trace_fib6_table_lookup(net, &res, table, fl6);
2763         return ret;
2764 };
2765
2766 static struct dst_entry *ip6_route_redirect(struct net *net,
2767                                             const struct flowi6 *fl6,
2768                                             const struct sk_buff *skb,
2769                                             const struct in6_addr *gateway)
2770 {
2771         int flags = RT6_LOOKUP_F_HAS_SADDR;
2772         struct ip6rd_flowi rdfl;
2773
2774         rdfl.fl6 = *fl6;
2775         rdfl.gateway = *gateway;
2776
2777         return fib6_rule_lookup(net, &rdfl.fl6, skb,
2778                                 flags, __ip6_route_redirect);
2779 }
2780
2781 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
2782                   kuid_t uid)
2783 {
2784         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2785         struct dst_entry *dst;
2786         struct flowi6 fl6 = {
2787                 .flowi6_iif = LOOPBACK_IFINDEX,
2788                 .flowi6_oif = oif,
2789                 .flowi6_mark = mark,
2790                 .daddr = iph->daddr,
2791                 .saddr = iph->saddr,
2792                 .flowlabel = ip6_flowinfo(iph),
2793                 .flowi6_uid = uid,
2794         };
2795
2796         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
2797         rt6_do_redirect(dst, NULL, skb);
2798         dst_release(dst);
2799 }
2800 EXPORT_SYMBOL_GPL(ip6_redirect);
2801
2802 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
2803 {
2804         const struct ipv6hdr *iph = ipv6_hdr(skb);
2805         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
2806         struct dst_entry *dst;
2807         struct flowi6 fl6 = {
2808                 .flowi6_iif = LOOPBACK_IFINDEX,
2809                 .flowi6_oif = oif,
2810                 .daddr = msg->dest,
2811                 .saddr = iph->daddr,
2812                 .flowi6_uid = sock_net_uid(net, NULL),
2813         };
2814
2815         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
2816         rt6_do_redirect(dst, NULL, skb);
2817         dst_release(dst);
2818 }
2819
2820 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
2821 {
2822         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
2823                      sk->sk_uid);
2824 }
2825 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
2826
2827 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
2828 {
2829         struct net_device *dev = dst->dev;
2830         unsigned int mtu = dst_mtu(dst);
2831         struct net *net = dev_net(dev);
2832
2833         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
2834
2835         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
2836                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
2837
2838         /*
2839          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
2840          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
2841          * IPV6_MAXPLEN is also valid and means: "any MSS,
2842          * rely only on pmtu discovery"
2843          */
2844         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
2845                 mtu = IPV6_MAXPLEN;
2846         return mtu;
2847 }
2848
2849 static unsigned int ip6_mtu(const struct dst_entry *dst)
2850 {
2851         struct inet6_dev *idev;
2852         unsigned int mtu;
2853
2854         mtu = dst_metric_raw(dst, RTAX_MTU);
2855         if (mtu)
2856                 goto out;
2857
2858         mtu = IPV6_MIN_MTU;
2859
2860         rcu_read_lock();
2861         idev = __in6_dev_get(dst->dev);
2862         if (idev)
2863                 mtu = idev->cnf.mtu6;
2864         rcu_read_unlock();
2865
2866 out:
2867         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
2868
2869         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
2870 }
2871
2872 /* MTU selection:
2873  * 1. mtu on route is locked - use it
2874  * 2. mtu from nexthop exception
2875  * 3. mtu from egress device
2876  *
2877  * based on ip6_dst_mtu_forward and exception logic of
2878  * rt6_find_cached_rt; called with rcu_read_lock
2879  */
2880 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
2881                       const struct in6_addr *daddr,
2882                       const struct in6_addr *saddr)
2883 {
2884         const struct fib6_nh *nh = res->nh;
2885         struct fib6_info *f6i = res->f6i;
2886         struct inet6_dev *idev;
2887         struct rt6_info *rt;
2888         u32 mtu = 0;
2889
2890         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
2891                 mtu = f6i->fib6_pmtu;
2892                 if (mtu)
2893                         goto out;
2894         }
2895
2896         rt = rt6_find_cached_rt(res, daddr, saddr);
2897         if (unlikely(rt)) {
2898                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
2899         } else {
2900                 struct net_device *dev = nh->fib_nh_dev;
2901
2902                 mtu = IPV6_MIN_MTU;
2903                 idev = __in6_dev_get(dev);
2904                 if (idev && idev->cnf.mtu6 > mtu)
2905                         mtu = idev->cnf.mtu6;
2906         }
2907
2908         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
2909 out:
2910         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
2911 }
2912
2913 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
2914                                   struct flowi6 *fl6)
2915 {
2916         struct dst_entry *dst;
2917         struct rt6_info *rt;
2918         struct inet6_dev *idev = in6_dev_get(dev);
2919         struct net *net = dev_net(dev);
2920
2921         if (unlikely(!idev))
2922                 return ERR_PTR(-ENODEV);
2923
2924         rt = ip6_dst_alloc(net, dev, 0);
2925         if (unlikely(!rt)) {
2926                 in6_dev_put(idev);
2927                 dst = ERR_PTR(-ENOMEM);
2928                 goto out;
2929         }
2930
2931         rt->dst.flags |= DST_HOST;
2932         rt->dst.input = ip6_input;
2933         rt->dst.output  = ip6_output;
2934         rt->rt6i_gateway  = fl6->daddr;
2935         rt->rt6i_dst.addr = fl6->daddr;
2936         rt->rt6i_dst.plen = 128;
2937         rt->rt6i_idev     = idev;
2938         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
2939
2940         /* Add this dst into uncached_list so that rt6_disable_ip() can
2941          * do proper release of the net_device
2942          */
2943         rt6_uncached_list_add(rt);
2944         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2945
2946         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
2947
2948 out:
2949         return dst;
2950 }
2951
2952 static int ip6_dst_gc(struct dst_ops *ops)
2953 {
2954         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
2955         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
2956         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
2957         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
2958         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
2959         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
2960         int entries;
2961
2962         entries = dst_entries_get_fast(ops);
2963         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
2964             entries <= rt_max_size)
2965                 goto out;
2966
2967         net->ipv6.ip6_rt_gc_expire++;
2968         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
2969         entries = dst_entries_get_slow(ops);
2970         if (entries < ops->gc_thresh)
2971                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
2972 out:
2973         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
2974         return entries > rt_max_size;
2975 }
2976
2977 static struct rt6_info *ip6_nh_lookup_table(struct net *net,
2978                                             struct fib6_config *cfg,
2979                                             const struct in6_addr *gw_addr,
2980                                             u32 tbid, int flags)
2981 {
2982         struct flowi6 fl6 = {
2983                 .flowi6_oif = cfg->fc_ifindex,
2984                 .daddr = *gw_addr,
2985                 .saddr = cfg->fc_prefsrc,
2986         };
2987         struct fib6_table *table;
2988         struct rt6_info *rt;
2989
2990         table = fib6_get_table(net, tbid);
2991         if (!table)
2992                 return NULL;
2993
2994         if (!ipv6_addr_any(&cfg->fc_prefsrc))
2995                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2996
2997         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
2998         rt = ip6_pol_route(net, table, cfg->fc_ifindex, &fl6, NULL, flags);
2999
3000         /* if table lookup failed, fall back to full lookup */
3001         if (rt == net->ipv6.ip6_null_entry) {
3002                 ip6_rt_put(rt);
3003                 rt = NULL;
3004         }
3005
3006         return rt;
3007 }
3008
3009 static int ip6_route_check_nh_onlink(struct net *net,
3010                                      struct fib6_config *cfg,
3011                                      const struct net_device *dev,
3012                                      struct netlink_ext_ack *extack)
3013 {
3014         u32 tbid = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
3015         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3016         u32 flags = RTF_LOCAL | RTF_ANYCAST | RTF_REJECT;
3017         struct fib6_info *from;
3018         struct rt6_info *grt;
3019         int err;
3020
3021         err = 0;
3022         grt = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0);
3023         if (grt) {
3024                 rcu_read_lock();
3025                 from = rcu_dereference(grt->from);
3026                 if (!grt->dst.error &&
3027                     /* ignore match if it is the default route */
3028                     from && !ipv6_addr_any(&from->fib6_dst.addr) &&
3029                     (grt->rt6i_flags & flags || dev != grt->dst.dev)) {
3030                         NL_SET_ERR_MSG(extack,
3031                                        "Nexthop has invalid gateway or device mismatch");
3032                         err = -EINVAL;
3033                 }
3034                 rcu_read_unlock();
3035
3036                 ip6_rt_put(grt);
3037         }
3038
3039         return err;
3040 }
3041
3042 static int ip6_route_check_nh(struct net *net,
3043                               struct fib6_config *cfg,
3044                               struct net_device **_dev,
3045                               struct inet6_dev **idev)
3046 {
3047         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3048         struct net_device *dev = _dev ? *_dev : NULL;
3049         struct rt6_info *grt = NULL;
3050         int err = -EHOSTUNREACH;
3051
3052         if (cfg->fc_table) {
3053                 int flags = RT6_LOOKUP_F_IFACE;
3054
3055                 grt = ip6_nh_lookup_table(net, cfg, gw_addr,
3056                                           cfg->fc_table, flags);
3057                 if (grt) {
3058                         if (grt->rt6i_flags & RTF_GATEWAY ||
3059                             (dev && dev != grt->dst.dev)) {
3060                                 ip6_rt_put(grt);
3061                                 grt = NULL;
3062                         }
3063                 }
3064         }
3065
3066         if (!grt)
3067                 grt = rt6_lookup(net, gw_addr, NULL, cfg->fc_ifindex, NULL, 1);
3068
3069         if (!grt)
3070                 goto out;
3071
3072         if (dev) {
3073                 if (dev != grt->dst.dev) {
3074                         ip6_rt_put(grt);
3075                         goto out;
3076                 }
3077         } else {
3078                 *_dev = dev = grt->dst.dev;
3079                 *idev = grt->rt6i_idev;
3080                 dev_hold(dev);
3081                 in6_dev_hold(grt->rt6i_idev);
3082         }
3083
3084         if (!(grt->rt6i_flags & RTF_GATEWAY))
3085                 err = 0;
3086
3087         ip6_rt_put(grt);
3088
3089 out:
3090         return err;
3091 }
3092
3093 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3094                            struct net_device **_dev, struct inet6_dev **idev,
3095                            struct netlink_ext_ack *extack)
3096 {
3097         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3098         int gwa_type = ipv6_addr_type(gw_addr);
3099         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3100         const struct net_device *dev = *_dev;
3101         bool need_addr_check = !dev;
3102         int err = -EINVAL;
3103
3104         /* if gw_addr is local we will fail to detect this in case
3105          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3106          * will return already-added prefix route via interface that
3107          * prefix route was assigned to, which might be non-loopback.
3108          */
3109         if (dev &&
3110             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3111                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3112                 goto out;
3113         }
3114
3115         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3116                 /* IPv6 strictly inhibits using not link-local
3117                  * addresses as nexthop address.
3118                  * Otherwise, router will not able to send redirects.
3119                  * It is very good, but in some (rare!) circumstances
3120                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3121                  * some exceptions. --ANK
3122                  * We allow IPv4-mapped nexthops to support RFC4798-type
3123                  * addressing
3124                  */
3125                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3126                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3127                         goto out;
3128                 }
3129
3130                 if (cfg->fc_flags & RTNH_F_ONLINK)
3131                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3132                 else
3133                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3134
3135                 if (err)
3136                         goto out;
3137         }
3138
3139         /* reload in case device was changed */
3140         dev = *_dev;
3141
3142         err = -EINVAL;
3143         if (!dev) {
3144                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3145                 goto out;
3146         } else if (dev->flags & IFF_LOOPBACK) {
3147                 NL_SET_ERR_MSG(extack,
3148                                "Egress device can not be loopback device for this route");
3149                 goto out;
3150         }
3151
3152         /* if we did not check gw_addr above, do so now that the
3153          * egress device has been resolved.
3154          */
3155         if (need_addr_check &&
3156             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3157                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3158                 goto out;
3159         }
3160
3161         err = 0;
3162 out:
3163         return err;
3164 }
3165
3166 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3167 {
3168         if ((flags & RTF_REJECT) ||
3169             (dev && (dev->flags & IFF_LOOPBACK) &&
3170              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3171              !(flags & RTF_LOCAL)))
3172                 return true;
3173
3174         return false;
3175 }
3176
3177 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3178                  struct fib6_config *cfg, gfp_t gfp_flags,
3179                  struct netlink_ext_ack *extack)
3180 {
3181         struct net_device *dev = NULL;
3182         struct inet6_dev *idev = NULL;
3183         int addr_type;
3184         int err;
3185
3186         fib6_nh->fib_nh_family = AF_INET6;
3187
3188         err = -ENODEV;
3189         if (cfg->fc_ifindex) {
3190                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3191                 if (!dev)
3192                         goto out;
3193                 idev = in6_dev_get(dev);
3194                 if (!idev)
3195                         goto out;
3196         }
3197
3198         if (cfg->fc_flags & RTNH_F_ONLINK) {
3199                 if (!dev) {
3200                         NL_SET_ERR_MSG(extack,
3201                                        "Nexthop device required for onlink");
3202                         goto out;
3203                 }
3204
3205                 if (!(dev->flags & IFF_UP)) {
3206                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3207                         err = -ENETDOWN;
3208                         goto out;
3209                 }
3210
3211                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3212         }
3213
3214         fib6_nh->fib_nh_weight = 1;
3215
3216         /* We cannot add true routes via loopback here,
3217          * they would result in kernel looping; promote them to reject routes
3218          */
3219         addr_type = ipv6_addr_type(&cfg->fc_dst);
3220         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3221                 /* hold loopback dev/idev if we haven't done so. */
3222                 if (dev != net->loopback_dev) {
3223                         if (dev) {
3224                                 dev_put(dev);
3225                                 in6_dev_put(idev);
3226                         }
3227                         dev = net->loopback_dev;
3228                         dev_hold(dev);
3229                         idev = in6_dev_get(dev);
3230                         if (!idev) {
3231                                 err = -ENODEV;
3232                                 goto out;
3233                         }
3234                 }
3235                 goto pcpu_alloc;
3236         }
3237
3238         if (cfg->fc_flags & RTF_GATEWAY) {
3239                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3240                 if (err)
3241                         goto out;
3242
3243                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3244                 fib6_nh->fib_nh_gw_family = AF_INET6;
3245         }
3246
3247         err = -ENODEV;
3248         if (!dev)
3249                 goto out;
3250
3251         if (idev->cnf.disable_ipv6) {
3252                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3253                 err = -EACCES;
3254                 goto out;
3255         }
3256
3257         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3258                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3259                 err = -ENETDOWN;
3260                 goto out;
3261         }
3262
3263         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3264             !netif_carrier_ok(dev))
3265                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3266
3267         err = fib_nh_common_init(&fib6_nh->nh_common, cfg->fc_encap,
3268                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3269         if (err)
3270                 goto out;
3271
3272 pcpu_alloc:
3273         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3274         if (!fib6_nh->rt6i_pcpu) {
3275                 err = -ENOMEM;
3276                 goto out;
3277         }
3278
3279         fib6_nh->fib_nh_dev = dev;
3280         fib6_nh->fib_nh_oif = dev->ifindex;
3281         err = 0;
3282 out:
3283         if (idev)
3284                 in6_dev_put(idev);
3285
3286         if (err) {
3287                 lwtstate_put(fib6_nh->fib_nh_lws);
3288                 fib6_nh->fib_nh_lws = NULL;
3289                 if (dev)
3290                         dev_put(dev);
3291         }
3292
3293         return err;
3294 }
3295
3296 void fib6_nh_release(struct fib6_nh *fib6_nh)
3297 {
3298         struct rt6_exception_bucket *bucket;
3299
3300         rcu_read_lock();
3301
3302         fib6_nh_flush_exceptions(fib6_nh, NULL);
3303         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3304         if (bucket) {
3305                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3306                 kfree(bucket);
3307         }
3308
3309         rcu_read_unlock();
3310
3311         if (fib6_nh->rt6i_pcpu) {
3312                 int cpu;
3313
3314                 for_each_possible_cpu(cpu) {
3315                         struct rt6_info **ppcpu_rt;
3316                         struct rt6_info *pcpu_rt;
3317
3318                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3319                         pcpu_rt = *ppcpu_rt;
3320                         if (pcpu_rt) {
3321                                 dst_dev_put(&pcpu_rt->dst);
3322                                 dst_release(&pcpu_rt->dst);
3323                                 *ppcpu_rt = NULL;
3324                         }
3325                 }
3326
3327                 free_percpu(fib6_nh->rt6i_pcpu);
3328         }
3329
3330         fib_nh_common_release(&fib6_nh->nh_common);
3331 }
3332
3333 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3334                                               gfp_t gfp_flags,
3335                                               struct netlink_ext_ack *extack)
3336 {
3337         struct net *net = cfg->fc_nlinfo.nl_net;
3338         struct fib6_info *rt = NULL;
3339         struct nexthop *nh = NULL;
3340         struct fib6_table *table;
3341         struct fib6_nh *fib6_nh;
3342         int err = -EINVAL;
3343         int addr_type;
3344
3345         /* RTF_PCPU is an internal flag; can not be set by userspace */
3346         if (cfg->fc_flags & RTF_PCPU) {
3347                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3348                 goto out;
3349         }
3350
3351         /* RTF_CACHE is an internal flag; can not be set by userspace */
3352         if (cfg->fc_flags & RTF_CACHE) {
3353                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3354                 goto out;
3355         }
3356
3357         if (cfg->fc_type > RTN_MAX) {
3358                 NL_SET_ERR_MSG(extack, "Invalid route type");
3359                 goto out;
3360         }
3361
3362         if (cfg->fc_dst_len > 128) {
3363                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3364                 goto out;
3365         }
3366         if (cfg->fc_src_len > 128) {
3367                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3368                 goto out;
3369         }
3370 #ifndef CONFIG_IPV6_SUBTREES
3371         if (cfg->fc_src_len) {
3372                 NL_SET_ERR_MSG(extack,
3373                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3374                 goto out;
3375         }
3376 #endif
3377
3378         err = -ENOBUFS;
3379         if (cfg->fc_nlinfo.nlh &&
3380             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3381                 table = fib6_get_table(net, cfg->fc_table);
3382                 if (!table) {
3383                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3384                         table = fib6_new_table(net, cfg->fc_table);
3385                 }
3386         } else {
3387                 table = fib6_new_table(net, cfg->fc_table);
3388         }
3389
3390         if (!table)
3391                 goto out;
3392
3393         err = -ENOMEM;
3394         rt = fib6_info_alloc(gfp_flags, !nh);
3395         if (!rt)
3396                 goto out;
3397
3398         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3399                                                extack);
3400         if (IS_ERR(rt->fib6_metrics)) {
3401                 err = PTR_ERR(rt->fib6_metrics);
3402                 /* Do not leave garbage there. */
3403                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3404                 goto out;
3405         }
3406
3407         if (cfg->fc_flags & RTF_ADDRCONF)
3408                 rt->dst_nocount = true;
3409
3410         if (cfg->fc_flags & RTF_EXPIRES)
3411                 fib6_set_expires(rt, jiffies +
3412                                 clock_t_to_jiffies(cfg->fc_expires));
3413         else
3414                 fib6_clean_expires(rt);
3415
3416         if (cfg->fc_protocol == RTPROT_UNSPEC)
3417                 cfg->fc_protocol = RTPROT_BOOT;
3418         rt->fib6_protocol = cfg->fc_protocol;
3419
3420         rt->fib6_table = table;
3421         rt->fib6_metric = cfg->fc_metric;
3422         rt->fib6_type = cfg->fc_type;
3423         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3424
3425         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3426         rt->fib6_dst.plen = cfg->fc_dst_len;
3427         if (rt->fib6_dst.plen == 128)
3428                 rt->dst_host = true;
3429
3430 #ifdef CONFIG_IPV6_SUBTREES
3431         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3432         rt->fib6_src.plen = cfg->fc_src_len;
3433 #endif
3434         if (nh) {
3435                 if (!nexthop_get(nh)) {
3436                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3437                         goto out;
3438                 }
3439                 if (rt->fib6_src.plen) {
3440                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3441                         goto out;
3442                 }
3443                 rt->nh = nh;
3444                 fib6_nh = nexthop_fib6_nh(rt->nh);
3445         } else {
3446                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3447                 if (err)
3448                         goto out;
3449
3450                 fib6_nh = rt->fib6_nh;
3451
3452                 /* We cannot add true routes via loopback here, they would
3453                  * result in kernel looping; promote them to reject routes
3454                  */
3455                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3456                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3457                                    addr_type))
3458                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3459         }
3460
3461         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3462                 struct net_device *dev = fib6_nh->fib_nh_dev;
3463
3464                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3465                         NL_SET_ERR_MSG(extack, "Invalid source address");
3466                         err = -EINVAL;
3467                         goto out;
3468                 }
3469                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3470                 rt->fib6_prefsrc.plen = 128;
3471         } else
3472                 rt->fib6_prefsrc.plen = 0;
3473
3474         return rt;
3475 out:
3476         fib6_info_release(rt);
3477         return ERR_PTR(err);
3478 }
3479
3480 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3481                   struct netlink_ext_ack *extack)
3482 {
3483         struct fib6_info *rt;
3484         int err;
3485
3486         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3487         if (IS_ERR(rt))
3488                 return PTR_ERR(rt);
3489
3490         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3491         fib6_info_release(rt);
3492
3493         return err;
3494 }
3495
3496 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3497 {
3498         struct net *net = info->nl_net;
3499         struct fib6_table *table;
3500         int err;
3501
3502         if (rt == net->ipv6.fib6_null_entry) {
3503                 err = -ENOENT;
3504                 goto out;
3505         }
3506
3507         table = rt->fib6_table;
3508         spin_lock_bh(&table->tb6_lock);
3509         err = fib6_del(rt, info);
3510         spin_unlock_bh(&table->tb6_lock);
3511
3512 out:
3513         fib6_info_release(rt);
3514         return err;
3515 }
3516
3517 int ip6_del_rt(struct net *net, struct fib6_info *rt)
3518 {
3519         struct nl_info info = { .nl_net = net };
3520
3521         return __ip6_del_rt(rt, &info);
3522 }
3523
3524 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3525 {
3526         struct nl_info *info = &cfg->fc_nlinfo;
3527         struct net *net = info->nl_net;
3528         struct sk_buff *skb = NULL;
3529         struct fib6_table *table;
3530         int err = -ENOENT;
3531
3532         if (rt == net->ipv6.fib6_null_entry)
3533                 goto out_put;
3534         table = rt->fib6_table;
3535         spin_lock_bh(&table->tb6_lock);
3536
3537         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3538                 struct fib6_info *sibling, *next_sibling;
3539
3540                 /* prefer to send a single notification with all hops */
3541                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3542                 if (skb) {
3543                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3544
3545                         if (rt6_fill_node(net, skb, rt, NULL,
3546                                           NULL, NULL, 0, RTM_DELROUTE,
3547                                           info->portid, seq, 0) < 0) {
3548                                 kfree_skb(skb);
3549                                 skb = NULL;
3550                         } else
3551                                 info->skip_notify = 1;
3552                 }
3553
3554                 list_for_each_entry_safe(sibling, next_sibling,
3555                                          &rt->fib6_siblings,
3556                                          fib6_siblings) {
3557                         err = fib6_del(sibling, info);
3558                         if (err)
3559                                 goto out_unlock;
3560                 }
3561         }
3562
3563         err = fib6_del(rt, info);
3564 out_unlock:
3565         spin_unlock_bh(&table->tb6_lock);
3566 out_put:
3567         fib6_info_release(rt);
3568
3569         if (skb) {
3570                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3571                             info->nlh, gfp_any());
3572         }
3573         return err;
3574 }
3575
3576 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3577 {
3578         int rc = -ESRCH;
3579
3580         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3581                 goto out;
3582
3583         if (cfg->fc_flags & RTF_GATEWAY &&
3584             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3585                 goto out;
3586
3587         rc = rt6_remove_exception_rt(rt);
3588 out:
3589         return rc;
3590 }
3591
3592 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3593                              struct fib6_nh *nh)
3594 {
3595         struct fib6_result res = {
3596                 .f6i = rt,
3597                 .nh = nh,
3598         };
3599         struct rt6_info *rt_cache;
3600
3601         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3602         if (rt_cache)
3603                 return __ip6_del_cached_rt(rt_cache, cfg);
3604
3605         return 0;
3606 }
3607
3608 static int ip6_route_del(struct fib6_config *cfg,
3609                          struct netlink_ext_ack *extack)
3610 {
3611         struct fib6_table *table;
3612         struct fib6_info *rt;
3613         struct fib6_node *fn;
3614         int err = -ESRCH;
3615
3616         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3617         if (!table) {
3618                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3619                 return err;
3620         }
3621
3622         rcu_read_lock();
3623
3624         fn = fib6_locate(&table->tb6_root,
3625                          &cfg->fc_dst, cfg->fc_dst_len,
3626                          &cfg->fc_src, cfg->fc_src_len,
3627                          !(cfg->fc_flags & RTF_CACHE));
3628
3629         if (fn) {
3630                 for_each_fib6_node_rt_rcu(fn) {
3631                         struct fib6_nh *nh;
3632
3633                         nh = rt->fib6_nh;
3634                         if (cfg->fc_flags & RTF_CACHE) {
3635                                 int rc;
3636
3637                                 rc = ip6_del_cached_rt(cfg, rt, nh);
3638                                 if (rc != -ESRCH) {
3639                                         rcu_read_unlock();
3640                                         return rc;
3641                                 }
3642                                 continue;
3643                         }
3644
3645                         if (cfg->fc_ifindex &&
3646                             (!nh->fib_nh_dev ||
3647                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3648                                 continue;
3649                         if (cfg->fc_flags & RTF_GATEWAY &&
3650                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3651                                 continue;
3652                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3653                                 continue;
3654                         if (cfg->fc_protocol && cfg->fc_protocol != rt->fib6_protocol)
3655                                 continue;
3656                         if (!fib6_info_hold_safe(rt))
3657                                 continue;
3658                         rcu_read_unlock();
3659
3660                         /* if gateway was specified only delete the one hop */
3661                         if (cfg->fc_flags & RTF_GATEWAY)
3662                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3663
3664                         return __ip6_del_rt_siblings(rt, cfg);
3665                 }
3666         }
3667         rcu_read_unlock();
3668
3669         return err;
3670 }
3671
3672 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3673 {
3674         struct netevent_redirect netevent;
3675         struct rt6_info *rt, *nrt = NULL;
3676         struct fib6_result res = {};
3677         struct ndisc_options ndopts;
3678         struct inet6_dev *in6_dev;
3679         struct neighbour *neigh;
3680         struct rd_msg *msg;
3681         int optlen, on_link;
3682         u8 *lladdr;
3683
3684         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
3685         optlen -= sizeof(*msg);
3686
3687         if (optlen < 0) {
3688                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
3689                 return;
3690         }
3691
3692         msg = (struct rd_msg *)icmp6_hdr(skb);
3693
3694         if (ipv6_addr_is_multicast(&msg->dest)) {
3695                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
3696                 return;
3697         }
3698
3699         on_link = 0;
3700         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
3701                 on_link = 1;
3702         } else if (ipv6_addr_type(&msg->target) !=
3703                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
3704                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
3705                 return;
3706         }
3707
3708         in6_dev = __in6_dev_get(skb->dev);
3709         if (!in6_dev)
3710                 return;
3711         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
3712                 return;
3713
3714         /* RFC2461 8.1:
3715          *      The IP source address of the Redirect MUST be the same as the current
3716          *      first-hop router for the specified ICMP Destination Address.
3717          */
3718
3719         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
3720                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
3721                 return;
3722         }
3723
3724         lladdr = NULL;
3725         if (ndopts.nd_opts_tgt_lladdr) {
3726                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
3727                                              skb->dev);
3728                 if (!lladdr) {
3729                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
3730                         return;
3731                 }
3732         }
3733
3734         rt = (struct rt6_info *) dst;
3735         if (rt->rt6i_flags & RTF_REJECT) {
3736                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
3737                 return;
3738         }
3739
3740         /* Redirect received -> path was valid.
3741          * Look, redirects are sent only in response to data packets,
3742          * so that this nexthop apparently is reachable. --ANK
3743          */
3744         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
3745
3746         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
3747         if (!neigh)
3748                 return;
3749
3750         /*
3751          *      We have finally decided to accept it.
3752          */
3753
3754         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
3755                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
3756                      NEIGH_UPDATE_F_OVERRIDE|
3757                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
3758                                      NEIGH_UPDATE_F_ISROUTER)),
3759                      NDISC_REDIRECT, &ndopts);
3760
3761         rcu_read_lock();
3762         res.f6i = rcu_dereference(rt->from);
3763         if (!res.f6i)
3764                 goto out;
3765
3766         res.nh = res.f6i->fib6_nh;
3767         res.fib6_flags = res.f6i->fib6_flags;
3768         res.fib6_type = res.f6i->fib6_type;
3769         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
3770         if (!nrt)
3771                 goto out;
3772
3773         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
3774         if (on_link)
3775                 nrt->rt6i_flags &= ~RTF_GATEWAY;
3776
3777         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
3778
3779         /* rt6_insert_exception() will take care of duplicated exceptions */
3780         if (rt6_insert_exception(nrt, &res)) {
3781                 dst_release_immediate(&nrt->dst);
3782                 goto out;
3783         }
3784
3785         netevent.old = &rt->dst;
3786         netevent.new = &nrt->dst;
3787         netevent.daddr = &msg->dest;
3788         netevent.neigh = neigh;
3789         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
3790
3791 out:
3792         rcu_read_unlock();
3793         neigh_release(neigh);
3794 }
3795
3796 #ifdef CONFIG_IPV6_ROUTE_INFO
3797 static struct fib6_info *rt6_get_route_info(struct net *net,
3798                                            const struct in6_addr *prefix, int prefixlen,
3799                                            const struct in6_addr *gwaddr,
3800                                            struct net_device *dev)
3801 {
3802         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
3803         int ifindex = dev->ifindex;
3804         struct fib6_node *fn;
3805         struct fib6_info *rt = NULL;
3806         struct fib6_table *table;
3807
3808         table = fib6_get_table(net, tb_id);
3809         if (!table)
3810                 return NULL;
3811
3812         rcu_read_lock();
3813         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
3814         if (!fn)
3815                 goto out;
3816
3817         for_each_fib6_node_rt_rcu(fn) {
3818                 /* these routes do not use nexthops */
3819                 if (rt->nh)
3820                         continue;
3821                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
3822                         continue;
3823                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
3824                     !rt->fib6_nh->fib_nh_gw_family)
3825                         continue;
3826                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
3827                         continue;
3828                 if (!fib6_info_hold_safe(rt))
3829                         continue;
3830                 break;
3831         }
3832 out:
3833         rcu_read_unlock();
3834         return rt;
3835 }
3836
3837 static struct fib6_info *rt6_add_route_info(struct net *net,
3838                                            const struct in6_addr *prefix, int prefixlen,
3839                                            const struct in6_addr *gwaddr,
3840                                            struct net_device *dev,
3841                                            unsigned int pref)
3842 {
3843         struct fib6_config cfg = {
3844                 .fc_metric      = IP6_RT_PRIO_USER,
3845                 .fc_ifindex     = dev->ifindex,
3846                 .fc_dst_len     = prefixlen,
3847                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
3848                                   RTF_UP | RTF_PREF(pref),
3849                 .fc_protocol = RTPROT_RA,
3850                 .fc_type = RTN_UNICAST,
3851                 .fc_nlinfo.portid = 0,
3852                 .fc_nlinfo.nlh = NULL,
3853                 .fc_nlinfo.nl_net = net,
3854         };
3855
3856         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
3857         cfg.fc_dst = *prefix;
3858         cfg.fc_gateway = *gwaddr;
3859
3860         /* We should treat it as a default route if prefix length is 0. */
3861         if (!prefixlen)
3862                 cfg.fc_flags |= RTF_DEFAULT;
3863
3864         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
3865
3866         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
3867 }
3868 #endif
3869
3870 struct fib6_info *rt6_get_dflt_router(struct net *net,
3871                                      const struct in6_addr *addr,
3872                                      struct net_device *dev)
3873 {
3874         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
3875         struct fib6_info *rt;
3876         struct fib6_table *table;
3877
3878         table = fib6_get_table(net, tb_id);
3879         if (!table)
3880                 return NULL;
3881
3882         rcu_read_lock();
3883         for_each_fib6_node_rt_rcu(&table->tb6_root) {
3884                 struct fib6_nh *nh;
3885
3886                 /* RA routes do not use nexthops */
3887                 if (rt->nh)
3888                         continue;
3889
3890                 nh = rt->fib6_nh;
3891                 if (dev == nh->fib_nh_dev &&
3892                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
3893                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
3894                         break;
3895         }
3896         if (rt && !fib6_info_hold_safe(rt))
3897                 rt = NULL;
3898         rcu_read_unlock();
3899         return rt;
3900 }
3901
3902 struct fib6_info *rt6_add_dflt_router(struct net *net,
3903                                      const struct in6_addr *gwaddr,
3904                                      struct net_device *dev,
3905                                      unsigned int pref)
3906 {
3907         struct fib6_config cfg = {
3908                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
3909                 .fc_metric      = IP6_RT_PRIO_USER,
3910                 .fc_ifindex     = dev->ifindex,
3911                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
3912                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
3913                 .fc_protocol = RTPROT_RA,
3914                 .fc_type = RTN_UNICAST,
3915                 .fc_nlinfo.portid = 0,
3916                 .fc_nlinfo.nlh = NULL,
3917                 .fc_nlinfo.nl_net = net,
3918         };
3919
3920         cfg.fc_gateway = *gwaddr;
3921
3922         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
3923                 struct fib6_table *table;
3924
3925                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
3926                 if (table)
3927                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
3928         }
3929
3930         return rt6_get_dflt_router(net, gwaddr, dev);
3931 }
3932
3933 static void __rt6_purge_dflt_routers(struct net *net,
3934                                      struct fib6_table *table)
3935 {
3936         struct fib6_info *rt;
3937
3938 restart:
3939         rcu_read_lock();
3940         for_each_fib6_node_rt_rcu(&table->tb6_root) {
3941                 struct net_device *dev = fib6_info_nh_dev(rt);
3942                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
3943
3944                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
3945                     (!idev || idev->cnf.accept_ra != 2) &&
3946                     fib6_info_hold_safe(rt)) {
3947                         rcu_read_unlock();
3948                         ip6_del_rt(net, rt);
3949                         goto restart;
3950                 }
3951         }
3952         rcu_read_unlock();
3953
3954         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
3955 }
3956
3957 void rt6_purge_dflt_routers(struct net *net)
3958 {
3959         struct fib6_table *table;
3960         struct hlist_head *head;
3961         unsigned int h;
3962
3963         rcu_read_lock();
3964
3965         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
3966                 head = &net->ipv6.fib_table_hash[h];
3967                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
3968                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
3969                                 __rt6_purge_dflt_routers(net, table);
3970                 }
3971         }
3972
3973         rcu_read_unlock();
3974 }
3975
3976 static void rtmsg_to_fib6_config(struct net *net,
3977                                  struct in6_rtmsg *rtmsg,
3978                                  struct fib6_config *cfg)
3979 {
3980         *cfg = (struct fib6_config){
3981                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
3982                          : RT6_TABLE_MAIN,
3983                 .fc_ifindex = rtmsg->rtmsg_ifindex,
3984                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
3985                 .fc_expires = rtmsg->rtmsg_info,
3986                 .fc_dst_len = rtmsg->rtmsg_dst_len,
3987                 .fc_src_len = rtmsg->rtmsg_src_len,
3988                 .fc_flags = rtmsg->rtmsg_flags,
3989                 .fc_type = rtmsg->rtmsg_type,
3990
3991                 .fc_nlinfo.nl_net = net,
3992
3993                 .fc_dst = rtmsg->rtmsg_dst,
3994                 .fc_src = rtmsg->rtmsg_src,
3995                 .fc_gateway = rtmsg->rtmsg_gateway,
3996         };
3997 }
3998
3999 int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4000 {
4001         struct fib6_config cfg;
4002         struct in6_rtmsg rtmsg;
4003         int err;
4004
4005         switch (cmd) {
4006         case SIOCADDRT:         /* Add a route */
4007         case SIOCDELRT:         /* Delete a route */
4008                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4009                         return -EPERM;
4010                 err = copy_from_user(&rtmsg, arg,
4011                                      sizeof(struct in6_rtmsg));
4012                 if (err)
4013                         return -EFAULT;
4014
4015                 rtmsg_to_fib6_config(net, &rtmsg, &cfg);
4016
4017                 rtnl_lock();
4018                 switch (cmd) {
4019                 case SIOCADDRT:
4020                         err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4021                         break;
4022                 case SIOCDELRT:
4023                         err = ip6_route_del(&cfg, NULL);
4024                         break;
4025                 default:
4026                         err = -EINVAL;
4027                 }
4028                 rtnl_unlock();
4029
4030                 return err;
4031         }
4032
4033         return -EINVAL;
4034 }
4035
4036 /*
4037  *      Drop the packet on the floor
4038  */
4039
4040 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4041 {
4042         struct dst_entry *dst = skb_dst(skb);
4043         struct net *net = dev_net(dst->dev);
4044         struct inet6_dev *idev;
4045         int type;
4046
4047         if (netif_is_l3_master(skb->dev) &&
4048             dst->dev == net->loopback_dev)
4049                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4050         else
4051                 idev = ip6_dst_idev(dst);
4052
4053         switch (ipstats_mib_noroutes) {
4054         case IPSTATS_MIB_INNOROUTES:
4055                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4056                 if (type == IPV6_ADDR_ANY) {
4057                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4058                         break;
4059                 }
4060                 /* FALLTHROUGH */
4061         case IPSTATS_MIB_OUTNOROUTES:
4062                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4063                 break;
4064         }
4065
4066         /* Start over by dropping the dst for l3mdev case */
4067         if (netif_is_l3_master(skb->dev))
4068                 skb_dst_drop(skb);
4069
4070         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4071         kfree_skb(skb);
4072         return 0;
4073 }
4074
4075 static int ip6_pkt_discard(struct sk_buff *skb)
4076 {
4077         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4078 }
4079
4080 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4081 {
4082         skb->dev = skb_dst(skb)->dev;
4083         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4084 }
4085
4086 static int ip6_pkt_prohibit(struct sk_buff *skb)
4087 {
4088         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4089 }
4090
4091 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4092 {
4093         skb->dev = skb_dst(skb)->dev;
4094         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4095 }
4096
4097 /*
4098  *      Allocate a dst for local (unicast / anycast) address.
4099  */
4100
4101 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4102                                      struct inet6_dev *idev,
4103                                      const struct in6_addr *addr,
4104                                      bool anycast, gfp_t gfp_flags)
4105 {
4106         struct fib6_config cfg = {
4107                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4108                 .fc_ifindex = idev->dev->ifindex,
4109                 .fc_flags = RTF_UP | RTF_ADDRCONF | RTF_NONEXTHOP,
4110                 .fc_dst = *addr,
4111                 .fc_dst_len = 128,
4112                 .fc_protocol = RTPROT_KERNEL,
4113                 .fc_nlinfo.nl_net = net,
4114                 .fc_ignore_dev_down = true,
4115         };
4116
4117         if (anycast) {
4118                 cfg.fc_type = RTN_ANYCAST;
4119                 cfg.fc_flags |= RTF_ANYCAST;
4120         } else {
4121                 cfg.fc_type = RTN_LOCAL;
4122                 cfg.fc_flags |= RTF_LOCAL;
4123         }
4124
4125         return ip6_route_info_create(&cfg, gfp_flags, NULL);
4126 }
4127
4128 /* remove deleted ip from prefsrc entries */
4129 struct arg_dev_net_ip {
4130         struct net_device *dev;
4131         struct net *net;
4132         struct in6_addr *addr;
4133 };
4134
4135 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4136 {
4137         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4138         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4139         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4140
4141         if (!rt->nh &&
4142             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4143             rt != net->ipv6.fib6_null_entry &&
4144             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4145                 spin_lock_bh(&rt6_exception_lock);
4146                 /* remove prefsrc entry */
4147                 rt->fib6_prefsrc.plen = 0;
4148                 spin_unlock_bh(&rt6_exception_lock);
4149         }
4150         return 0;
4151 }
4152
4153 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4154 {
4155         struct net *net = dev_net(ifp->idev->dev);
4156         struct arg_dev_net_ip adni = {
4157                 .dev = ifp->idev->dev,
4158                 .net = net,
4159                 .addr = &ifp->addr,
4160         };
4161         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4162 }
4163
4164 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4165
4166 /* Remove routers and update dst entries when gateway turn into host. */
4167 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4168 {
4169         struct in6_addr *gateway = (struct in6_addr *)arg;
4170         struct fib6_nh *nh;
4171
4172         /* RA routes do not use nexthops */
4173         if (rt->nh)
4174                 return 0;
4175
4176         nh = rt->fib6_nh;
4177         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4178             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4179                 return -1;
4180
4181         /* Further clean up cached routes in exception table.
4182          * This is needed because cached route may have a different
4183          * gateway than its 'parent' in the case of an ip redirect.
4184          */
4185         fib6_nh_exceptions_clean_tohost(nh, gateway);
4186
4187         return 0;
4188 }
4189
4190 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4191 {
4192         fib6_clean_all(net, fib6_clean_tohost, gateway);
4193 }
4194
4195 struct arg_netdev_event {
4196         const struct net_device *dev;
4197         union {
4198                 unsigned char nh_flags;
4199                 unsigned long event;
4200         };
4201 };
4202
4203 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4204 {
4205         struct fib6_info *iter;
4206         struct fib6_node *fn;
4207
4208         fn = rcu_dereference_protected(rt->fib6_node,
4209                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4210         iter = rcu_dereference_protected(fn->leaf,
4211                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4212         while (iter) {
4213                 if (iter->fib6_metric == rt->fib6_metric &&
4214                     rt6_qualify_for_ecmp(iter))
4215                         return iter;
4216                 iter = rcu_dereference_protected(iter->fib6_next,
4217                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4218         }
4219
4220         return NULL;
4221 }
4222
4223 /* only called for fib entries with builtin fib6_nh */
4224 static bool rt6_is_dead(const struct fib6_info *rt)
4225 {
4226         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4227             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4228              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4229                 return true;
4230
4231         return false;
4232 }
4233
4234 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4235 {
4236         struct fib6_info *iter;
4237         int total = 0;
4238
4239         if (!rt6_is_dead(rt))
4240                 total += rt->fib6_nh->fib_nh_weight;
4241
4242         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4243                 if (!rt6_is_dead(iter))
4244                         total += iter->fib6_nh->fib_nh_weight;
4245         }
4246
4247         return total;
4248 }
4249
4250 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4251 {
4252         int upper_bound = -1;
4253
4254         if (!rt6_is_dead(rt)) {
4255                 *weight += rt->fib6_nh->fib_nh_weight;
4256                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4257                                                     total) - 1;
4258         }
4259         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4260 }
4261
4262 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4263 {
4264         struct fib6_info *iter;
4265         int weight = 0;
4266
4267         rt6_upper_bound_set(rt, &weight, total);
4268
4269         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4270                 rt6_upper_bound_set(iter, &weight, total);
4271 }
4272
4273 void rt6_multipath_rebalance(struct fib6_info *rt)
4274 {
4275         struct fib6_info *first;
4276         int total;
4277
4278         /* In case the entire multipath route was marked for flushing,
4279          * then there is no need to rebalance upon the removal of every
4280          * sibling route.
4281          */
4282         if (!rt->fib6_nsiblings || rt->should_flush)
4283                 return;
4284
4285         /* During lookup routes are evaluated in order, so we need to
4286          * make sure upper bounds are assigned from the first sibling
4287          * onwards.
4288          */
4289         first = rt6_multipath_first_sibling(rt);
4290         if (WARN_ON_ONCE(!first))
4291                 return;
4292
4293         total = rt6_multipath_total_weight(first);
4294         rt6_multipath_upper_bound_set(first, total);
4295 }
4296
4297 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4298 {
4299         const struct arg_netdev_event *arg = p_arg;
4300         struct net *net = dev_net(arg->dev);
4301
4302         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4303             rt->fib6_nh->fib_nh_dev == arg->dev) {
4304                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4305                 fib6_update_sernum_upto_root(net, rt);
4306                 rt6_multipath_rebalance(rt);
4307         }
4308
4309         return 0;
4310 }
4311
4312 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4313 {
4314         struct arg_netdev_event arg = {
4315                 .dev = dev,
4316                 {
4317                         .nh_flags = nh_flags,
4318                 },
4319         };
4320
4321         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4322                 arg.nh_flags |= RTNH_F_LINKDOWN;
4323
4324         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4325 }
4326
4327 /* only called for fib entries with inline fib6_nh */
4328 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4329                                    const struct net_device *dev)
4330 {
4331         struct fib6_info *iter;
4332
4333         if (rt->fib6_nh->fib_nh_dev == dev)
4334                 return true;
4335         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4336                 if (iter->fib6_nh->fib_nh_dev == dev)
4337                         return true;
4338
4339         return false;
4340 }
4341
4342 static void rt6_multipath_flush(struct fib6_info *rt)
4343 {
4344         struct fib6_info *iter;
4345
4346         rt->should_flush = 1;
4347         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4348                 iter->should_flush = 1;
4349 }
4350
4351 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4352                                              const struct net_device *down_dev)
4353 {
4354         struct fib6_info *iter;
4355         unsigned int dead = 0;
4356
4357         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4358             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4359                 dead++;
4360         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4361                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4362                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4363                         dead++;
4364
4365         return dead;
4366 }
4367
4368 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4369                                        const struct net_device *dev,
4370                                        unsigned char nh_flags)
4371 {
4372         struct fib6_info *iter;
4373
4374         if (rt->fib6_nh->fib_nh_dev == dev)
4375                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4376         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4377                 if (iter->fib6_nh->fib_nh_dev == dev)
4378                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4379 }
4380
4381 /* called with write lock held for table with rt */
4382 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4383 {
4384         const struct arg_netdev_event *arg = p_arg;
4385         const struct net_device *dev = arg->dev;
4386         struct net *net = dev_net(dev);
4387
4388         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4389                 return 0;
4390
4391         switch (arg->event) {
4392         case NETDEV_UNREGISTER:
4393                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4394         case NETDEV_DOWN:
4395                 if (rt->should_flush)
4396                         return -1;
4397                 if (!rt->fib6_nsiblings)
4398                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4399                 if (rt6_multipath_uses_dev(rt, dev)) {
4400                         unsigned int count;
4401
4402                         count = rt6_multipath_dead_count(rt, dev);
4403                         if (rt->fib6_nsiblings + 1 == count) {
4404                                 rt6_multipath_flush(rt);
4405                                 return -1;
4406                         }
4407                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4408                                                    RTNH_F_LINKDOWN);
4409                         fib6_update_sernum(net, rt);
4410                         rt6_multipath_rebalance(rt);
4411                 }
4412                 return -2;
4413         case NETDEV_CHANGE:
4414                 if (rt->fib6_nh->fib_nh_dev != dev ||
4415                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4416                         break;
4417                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4418                 rt6_multipath_rebalance(rt);
4419                 break;
4420         }
4421
4422         return 0;
4423 }
4424
4425 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4426 {
4427         struct arg_netdev_event arg = {
4428                 .dev = dev,
4429                 {
4430                         .event = event,
4431                 },
4432         };
4433         struct net *net = dev_net(dev);
4434
4435         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4436                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4437         else
4438                 fib6_clean_all(net, fib6_ifdown, &arg);
4439 }
4440
4441 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4442 {
4443         rt6_sync_down_dev(dev, event);
4444         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4445         neigh_ifdown(&nd_tbl, dev);
4446 }
4447
4448 struct rt6_mtu_change_arg {
4449         struct net_device *dev;
4450         unsigned int mtu;
4451         struct fib6_info *f6i;
4452 };
4453
4454 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4455 {
4456         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4457         struct fib6_info *f6i = arg->f6i;
4458
4459         /* For administrative MTU increase, there is no way to discover
4460          * IPv6 PMTU increase, so PMTU increase should be updated here.
4461          * Since RFC 1981 doesn't include administrative MTU increase
4462          * update PMTU increase is a MUST. (i.e. jumbo frame)
4463          */
4464         if (nh->fib_nh_dev == arg->dev) {
4465                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4466                 u32 mtu = f6i->fib6_pmtu;
4467
4468                 if (mtu >= arg->mtu ||
4469                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4470                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4471
4472                 spin_lock_bh(&rt6_exception_lock);
4473                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4474                 spin_unlock_bh(&rt6_exception_lock);
4475         }
4476
4477         return 0;
4478 }
4479
4480 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4481 {
4482         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4483         struct inet6_dev *idev;
4484
4485         /* In IPv6 pmtu discovery is not optional,
4486            so that RTAX_MTU lock cannot disable it.
4487            We still use this lock to block changes
4488            caused by addrconf/ndisc.
4489         */
4490
4491         idev = __in6_dev_get(arg->dev);
4492         if (!idev)
4493                 return 0;
4494
4495         if (fib6_metric_locked(f6i, RTAX_MTU))
4496                 return 0;
4497
4498         arg->f6i = f6i;
4499         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4500 }
4501
4502 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4503 {
4504         struct rt6_mtu_change_arg arg = {
4505                 .dev = dev,
4506                 .mtu = mtu,
4507         };
4508
4509         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4510 }
4511
4512 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4513         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4514         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4515         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4516         [RTA_OIF]               = { .type = NLA_U32 },
4517         [RTA_IIF]               = { .type = NLA_U32 },
4518         [RTA_PRIORITY]          = { .type = NLA_U32 },
4519         [RTA_METRICS]           = { .type = NLA_NESTED },
4520         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4521         [RTA_PREF]              = { .type = NLA_U8 },
4522         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4523         [RTA_ENCAP]             = { .type = NLA_NESTED },
4524         [RTA_EXPIRES]           = { .type = NLA_U32 },
4525         [RTA_UID]               = { .type = NLA_U32 },
4526         [RTA_MARK]              = { .type = NLA_U32 },
4527         [RTA_TABLE]             = { .type = NLA_U32 },
4528         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4529         [RTA_SPORT]             = { .type = NLA_U16 },
4530         [RTA_DPORT]             = { .type = NLA_U16 },
4531 };
4532
4533 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4534                               struct fib6_config *cfg,
4535                               struct netlink_ext_ack *extack)
4536 {
4537         struct rtmsg *rtm;
4538         struct nlattr *tb[RTA_MAX+1];
4539         unsigned int pref;
4540         int err;
4541
4542         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4543                                      rtm_ipv6_policy, extack);
4544         if (err < 0)
4545                 goto errout;
4546
4547         err = -EINVAL;
4548         rtm = nlmsg_data(nlh);
4549
4550         *cfg = (struct fib6_config){
4551                 .fc_table = rtm->rtm_table,
4552                 .fc_dst_len = rtm->rtm_dst_len,
4553                 .fc_src_len = rtm->rtm_src_len,
4554                 .fc_flags = RTF_UP,
4555                 .fc_protocol = rtm->rtm_protocol,
4556                 .fc_type = rtm->rtm_type,
4557
4558                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4559                 .fc_nlinfo.nlh = nlh,
4560                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4561         };
4562
4563         if (rtm->rtm_type == RTN_UNREACHABLE ||
4564             rtm->rtm_type == RTN_BLACKHOLE ||
4565             rtm->rtm_type == RTN_PROHIBIT ||
4566             rtm->rtm_type == RTN_THROW)
4567                 cfg->fc_flags |= RTF_REJECT;
4568
4569         if (rtm->rtm_type == RTN_LOCAL)
4570                 cfg->fc_flags |= RTF_LOCAL;
4571
4572         if (rtm->rtm_flags & RTM_F_CLONED)
4573                 cfg->fc_flags |= RTF_CACHE;
4574
4575         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4576
4577         if (tb[RTA_GATEWAY]) {
4578                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4579                 cfg->fc_flags |= RTF_GATEWAY;
4580         }
4581         if (tb[RTA_VIA]) {
4582                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4583                 goto errout;
4584         }
4585
4586         if (tb[RTA_DST]) {
4587                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4588
4589                 if (nla_len(tb[RTA_DST]) < plen)
4590                         goto errout;
4591
4592                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4593         }
4594
4595         if (tb[RTA_SRC]) {
4596                 int plen = (rtm->rtm_src_len + 7) >> 3;
4597
4598                 if (nla_len(tb[RTA_SRC]) < plen)
4599                         goto errout;
4600
4601                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4602         }
4603
4604         if (tb[RTA_PREFSRC])
4605                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4606
4607         if (tb[RTA_OIF])
4608                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4609
4610         if (tb[RTA_PRIORITY])
4611                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4612
4613         if (tb[RTA_METRICS]) {
4614                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4615                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4616         }
4617
4618         if (tb[RTA_TABLE])
4619                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4620
4621         if (tb[RTA_MULTIPATH]) {
4622                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4623                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4624
4625                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4626                                                      cfg->fc_mp_len, extack);
4627                 if (err < 0)
4628                         goto errout;
4629         }
4630
4631         if (tb[RTA_PREF]) {
4632                 pref = nla_get_u8(tb[RTA_PREF]);
4633                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4634                     pref != ICMPV6_ROUTER_PREF_HIGH)
4635                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4636                 cfg->fc_flags |= RTF_PREF(pref);
4637         }
4638
4639         if (tb[RTA_ENCAP])
4640                 cfg->fc_encap = tb[RTA_ENCAP];
4641
4642         if (tb[RTA_ENCAP_TYPE]) {
4643                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4644
4645                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4646                 if (err < 0)
4647                         goto errout;
4648         }
4649
4650         if (tb[RTA_EXPIRES]) {
4651                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4652
4653                 if (addrconf_finite_timeout(timeout)) {
4654                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
4655                         cfg->fc_flags |= RTF_EXPIRES;
4656                 }
4657         }
4658
4659         err = 0;
4660 errout:
4661         return err;
4662 }
4663
4664 struct rt6_nh {
4665         struct fib6_info *fib6_info;
4666         struct fib6_config r_cfg;
4667         struct list_head next;
4668 };
4669
4670 static int ip6_route_info_append(struct net *net,
4671                                  struct list_head *rt6_nh_list,
4672                                  struct fib6_info *rt,
4673                                  struct fib6_config *r_cfg)
4674 {
4675         struct rt6_nh *nh;
4676         int err = -EEXIST;
4677
4678         list_for_each_entry(nh, rt6_nh_list, next) {
4679                 /* check if fib6_info already exists */
4680                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
4681                         return err;
4682         }
4683
4684         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
4685         if (!nh)
4686                 return -ENOMEM;
4687         nh->fib6_info = rt;
4688         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
4689         list_add_tail(&nh->next, rt6_nh_list);
4690
4691         return 0;
4692 }
4693
4694 static void ip6_route_mpath_notify(struct fib6_info *rt,
4695                                    struct fib6_info *rt_last,
4696                                    struct nl_info *info,
4697                                    __u16 nlflags)
4698 {
4699         /* if this is an APPEND route, then rt points to the first route
4700          * inserted and rt_last points to last route inserted. Userspace
4701          * wants a consistent dump of the route which starts at the first
4702          * nexthop. Since sibling routes are always added at the end of
4703          * the list, find the first sibling of the last route appended
4704          */
4705         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
4706                 rt = list_first_entry(&rt_last->fib6_siblings,
4707                                       struct fib6_info,
4708                                       fib6_siblings);
4709         }
4710
4711         if (rt)
4712                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
4713 }
4714
4715 static int ip6_route_multipath_add(struct fib6_config *cfg,
4716                                    struct netlink_ext_ack *extack)
4717 {
4718         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
4719         struct nl_info *info = &cfg->fc_nlinfo;
4720         struct fib6_config r_cfg;
4721         struct rtnexthop *rtnh;
4722         struct fib6_info *rt;
4723         struct rt6_nh *err_nh;
4724         struct rt6_nh *nh, *nh_safe;
4725         __u16 nlflags;
4726         int remaining;
4727         int attrlen;
4728         int err = 1;
4729         int nhn = 0;
4730         int replace = (cfg->fc_nlinfo.nlh &&
4731                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
4732         LIST_HEAD(rt6_nh_list);
4733
4734         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
4735         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
4736                 nlflags |= NLM_F_APPEND;
4737
4738         remaining = cfg->fc_mp_len;
4739         rtnh = (struct rtnexthop *)cfg->fc_mp;
4740
4741         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
4742          * fib6_info structs per nexthop
4743          */
4744         while (rtnh_ok(rtnh, remaining)) {
4745                 memcpy(&r_cfg, cfg, sizeof(*cfg));
4746                 if (rtnh->rtnh_ifindex)
4747                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
4748
4749                 attrlen = rtnh_attrlen(rtnh);
4750                 if (attrlen > 0) {
4751                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
4752
4753                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
4754                         if (nla) {
4755                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
4756                                 r_cfg.fc_flags |= RTF_GATEWAY;
4757                         }
4758                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
4759                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
4760                         if (nla)
4761                                 r_cfg.fc_encap_type = nla_get_u16(nla);
4762                 }
4763
4764                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
4765                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
4766                 if (IS_ERR(rt)) {
4767                         err = PTR_ERR(rt);
4768                         rt = NULL;
4769                         goto cleanup;
4770                 }
4771                 if (!rt6_qualify_for_ecmp(rt)) {
4772                         err = -EINVAL;
4773                         NL_SET_ERR_MSG(extack,
4774                                        "Device only routes can not be added for IPv6 using the multipath API.");
4775                         fib6_info_release(rt);
4776                         goto cleanup;
4777                 }
4778
4779                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
4780
4781                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
4782                                             rt, &r_cfg);
4783                 if (err) {
4784                         fib6_info_release(rt);
4785                         goto cleanup;
4786                 }
4787
4788                 rtnh = rtnh_next(rtnh, &remaining);
4789         }
4790
4791         /* for add and replace send one notification with all nexthops.
4792          * Skip the notification in fib6_add_rt2node and send one with
4793          * the full route when done
4794          */
4795         info->skip_notify = 1;
4796
4797         err_nh = NULL;
4798         list_for_each_entry(nh, &rt6_nh_list, next) {
4799                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
4800                 fib6_info_release(nh->fib6_info);
4801
4802                 if (!err) {
4803                         /* save reference to last route successfully inserted */
4804                         rt_last = nh->fib6_info;
4805
4806                         /* save reference to first route for notification */
4807                         if (!rt_notif)
4808                                 rt_notif = nh->fib6_info;
4809                 }
4810
4811                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
4812                 nh->fib6_info = NULL;
4813                 if (err) {
4814                         if (replace && nhn)
4815                                 NL_SET_ERR_MSG_MOD(extack,
4816                                                    "multipath route replace failed (check consistency of installed routes)");
4817                         err_nh = nh;
4818                         goto add_errout;
4819                 }
4820
4821                 /* Because each route is added like a single route we remove
4822                  * these flags after the first nexthop: if there is a collision,
4823                  * we have already failed to add the first nexthop:
4824                  * fib6_add_rt2node() has rejected it; when replacing, old
4825                  * nexthops have been replaced by first new, the rest should
4826                  * be added to it.
4827                  */
4828                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
4829                                                      NLM_F_REPLACE);
4830                 nhn++;
4831         }
4832
4833         /* success ... tell user about new route */
4834         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
4835         goto cleanup;
4836
4837 add_errout:
4838         /* send notification for routes that were added so that
4839          * the delete notifications sent by ip6_route_del are
4840          * coherent
4841          */
4842         if (rt_notif)
4843                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
4844
4845         /* Delete routes that were already added */
4846         list_for_each_entry(nh, &rt6_nh_list, next) {
4847                 if (err_nh == nh)
4848                         break;
4849                 ip6_route_del(&nh->r_cfg, extack);
4850         }
4851
4852 cleanup:
4853         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
4854                 if (nh->fib6_info)
4855                         fib6_info_release(nh->fib6_info);
4856                 list_del(&nh->next);
4857                 kfree(nh);
4858         }
4859
4860         return err;
4861 }
4862
4863 static int ip6_route_multipath_del(struct fib6_config *cfg,
4864                                    struct netlink_ext_ack *extack)
4865 {
4866         struct fib6_config r_cfg;
4867         struct rtnexthop *rtnh;
4868         int remaining;
4869         int attrlen;
4870         int err = 1, last_err = 0;
4871
4872         remaining = cfg->fc_mp_len;
4873         rtnh = (struct rtnexthop *)cfg->fc_mp;
4874
4875         /* Parse a Multipath Entry */
4876         while (rtnh_ok(rtnh, remaining)) {
4877                 memcpy(&r_cfg, cfg, sizeof(*cfg));
4878                 if (rtnh->rtnh_ifindex)
4879                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
4880
4881                 attrlen = rtnh_attrlen(rtnh);
4882                 if (attrlen > 0) {
4883                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
4884
4885                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
4886                         if (nla) {
4887                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
4888                                 r_cfg.fc_flags |= RTF_GATEWAY;
4889                         }
4890                 }
4891                 err = ip6_route_del(&r_cfg, extack);
4892                 if (err)
4893                         last_err = err;
4894
4895                 rtnh = rtnh_next(rtnh, &remaining);
4896         }
4897
4898         return last_err;
4899 }
4900
4901 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
4902                               struct netlink_ext_ack *extack)
4903 {
4904         struct fib6_config cfg;
4905         int err;
4906
4907         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
4908         if (err < 0)
4909                 return err;
4910
4911         if (cfg.fc_mp)
4912                 return ip6_route_multipath_del(&cfg, extack);
4913         else {
4914                 cfg.fc_delete_all_nh = 1;
4915                 return ip6_route_del(&cfg, extack);
4916         }
4917 }
4918
4919 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
4920                               struct netlink_ext_ack *extack)
4921 {
4922         struct fib6_config cfg;
4923         int err;
4924
4925         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
4926         if (err < 0)
4927                 return err;
4928
4929         if (cfg.fc_metric == 0)
4930                 cfg.fc_metric = IP6_RT_PRIO_USER;
4931
4932         if (cfg.fc_mp)
4933                 return ip6_route_multipath_add(&cfg, extack);
4934         else
4935                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
4936 }
4937
4938 static size_t rt6_nlmsg_size(struct fib6_info *rt)
4939 {
4940         int nexthop_len = 0;
4941
4942         if (rt->nh)
4943                 nexthop_len += nla_total_size(4); /* RTA_NH_ID */
4944
4945         if (rt->fib6_nsiblings) {
4946                 nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
4947                             + NLA_ALIGN(sizeof(struct rtnexthop))
4948                             + nla_total_size(16) /* RTA_GATEWAY */
4949                             + lwtunnel_get_encap_size(rt->fib6_nh->fib_nh_lws);
4950
4951                 nexthop_len *= rt->fib6_nsiblings;
4952         }
4953
4954         return NLMSG_ALIGN(sizeof(struct rtmsg))
4955                + nla_total_size(16) /* RTA_SRC */
4956                + nla_total_size(16) /* RTA_DST */
4957                + nla_total_size(16) /* RTA_GATEWAY */
4958                + nla_total_size(16) /* RTA_PREFSRC */
4959                + nla_total_size(4) /* RTA_TABLE */
4960                + nla_total_size(4) /* RTA_IIF */
4961                + nla_total_size(4) /* RTA_OIF */
4962                + nla_total_size(4) /* RTA_PRIORITY */
4963                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
4964                + nla_total_size(sizeof(struct rta_cacheinfo))
4965                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
4966                + nla_total_size(1) /* RTA_PREF */
4967                + lwtunnel_get_encap_size(rt->fib6_nh->fib_nh_lws)
4968                + nexthop_len;
4969 }
4970
4971 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
4972                                  unsigned char *flags)
4973 {
4974         if (nexthop_is_multipath(nh)) {
4975                 struct nlattr *mp;
4976
4977                 mp = nla_nest_start(skb, RTA_MULTIPATH);
4978                 if (!mp)
4979                         goto nla_put_failure;
4980
4981                 if (nexthop_mpath_fill_node(skb, nh))
4982                         goto nla_put_failure;
4983
4984                 nla_nest_end(skb, mp);
4985         } else {
4986                 struct fib6_nh *fib6_nh;
4987
4988                 fib6_nh = nexthop_fib6_nh(nh);
4989                 if (fib_nexthop_info(skb, &fib6_nh->nh_common,
4990                                      flags, false) < 0)
4991                         goto nla_put_failure;
4992         }
4993
4994         return 0;
4995
4996 nla_put_failure:
4997         return -EMSGSIZE;
4998 }
4999
5000 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5001                          struct fib6_info *rt, struct dst_entry *dst,
5002                          struct in6_addr *dest, struct in6_addr *src,
5003                          int iif, int type, u32 portid, u32 seq,
5004                          unsigned int flags)
5005 {
5006         struct rt6_info *rt6 = (struct rt6_info *)dst;
5007         struct rt6key *rt6_dst, *rt6_src;
5008         u32 *pmetrics, table, rt6_flags;
5009         unsigned char nh_flags = 0;
5010         struct nlmsghdr *nlh;
5011         struct rtmsg *rtm;
5012         long expires = 0;
5013
5014         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5015         if (!nlh)
5016                 return -EMSGSIZE;
5017
5018         if (rt6) {
5019                 rt6_dst = &rt6->rt6i_dst;
5020                 rt6_src = &rt6->rt6i_src;
5021                 rt6_flags = rt6->rt6i_flags;
5022         } else {
5023                 rt6_dst = &rt->fib6_dst;
5024                 rt6_src = &rt->fib6_src;
5025                 rt6_flags = rt->fib6_flags;
5026         }
5027
5028         rtm = nlmsg_data(nlh);
5029         rtm->rtm_family = AF_INET6;
5030         rtm->rtm_dst_len = rt6_dst->plen;
5031         rtm->rtm_src_len = rt6_src->plen;
5032         rtm->rtm_tos = 0;
5033         if (rt->fib6_table)
5034                 table = rt->fib6_table->tb6_id;
5035         else
5036                 table = RT6_TABLE_UNSPEC;
5037         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5038         if (nla_put_u32(skb, RTA_TABLE, table))
5039                 goto nla_put_failure;
5040
5041         rtm->rtm_type = rt->fib6_type;
5042         rtm->rtm_flags = 0;
5043         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5044         rtm->rtm_protocol = rt->fib6_protocol;
5045
5046         if (rt6_flags & RTF_CACHE)
5047                 rtm->rtm_flags |= RTM_F_CLONED;
5048
5049         if (dest) {
5050                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5051                         goto nla_put_failure;
5052                 rtm->rtm_dst_len = 128;
5053         } else if (rtm->rtm_dst_len)
5054                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5055                         goto nla_put_failure;
5056 #ifdef CONFIG_IPV6_SUBTREES
5057         if (src) {
5058                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5059                         goto nla_put_failure;
5060                 rtm->rtm_src_len = 128;
5061         } else if (rtm->rtm_src_len &&
5062                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5063                 goto nla_put_failure;
5064 #endif
5065         if (iif) {
5066 #ifdef CONFIG_IPV6_MROUTE
5067                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5068                         int err = ip6mr_get_route(net, skb, rtm, portid);
5069
5070                         if (err == 0)
5071                                 return 0;
5072                         if (err < 0)
5073                                 goto nla_put_failure;
5074                 } else
5075 #endif
5076                         if (nla_put_u32(skb, RTA_IIF, iif))
5077                                 goto nla_put_failure;
5078         } else if (dest) {
5079                 struct in6_addr saddr_buf;
5080                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5081                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5082                         goto nla_put_failure;
5083         }
5084
5085         if (rt->fib6_prefsrc.plen) {
5086                 struct in6_addr saddr_buf;
5087                 saddr_buf = rt->fib6_prefsrc.addr;
5088                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5089                         goto nla_put_failure;
5090         }
5091
5092         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5093         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5094                 goto nla_put_failure;
5095
5096         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5097                 goto nla_put_failure;
5098
5099         /* For multipath routes, walk the siblings list and add
5100          * each as a nexthop within RTA_MULTIPATH.
5101          */
5102         if (rt6) {
5103                 if (rt6_flags & RTF_GATEWAY &&
5104                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5105                         goto nla_put_failure;
5106
5107                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5108                         goto nla_put_failure;
5109         } else if (rt->fib6_nsiblings) {
5110                 struct fib6_info *sibling, *next_sibling;
5111                 struct nlattr *mp;
5112
5113                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5114                 if (!mp)
5115                         goto nla_put_failure;
5116
5117                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5118                                     rt->fib6_nh->fib_nh_weight) < 0)
5119                         goto nla_put_failure;
5120
5121                 list_for_each_entry_safe(sibling, next_sibling,
5122                                          &rt->fib6_siblings, fib6_siblings) {
5123                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5124                                             sibling->fib6_nh->fib_nh_weight) < 0)
5125                                 goto nla_put_failure;
5126                 }
5127
5128                 nla_nest_end(skb, mp);
5129         } else if (rt->nh) {
5130                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5131                         goto nla_put_failure;
5132
5133                 if (nexthop_is_blackhole(rt->nh))
5134                         rtm->rtm_type = RTN_BLACKHOLE;
5135
5136                 if (rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5137                         goto nla_put_failure;
5138
5139                 rtm->rtm_flags |= nh_flags;
5140         } else {
5141                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common,
5142                                      &nh_flags, false) < 0)
5143                         goto nla_put_failure;
5144
5145                 rtm->rtm_flags |= nh_flags;
5146         }
5147
5148         if (rt6_flags & RTF_EXPIRES) {
5149                 expires = dst ? dst->expires : rt->expires;
5150                 expires -= jiffies;
5151         }
5152
5153         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5154                 goto nla_put_failure;
5155
5156         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5157                 goto nla_put_failure;
5158
5159
5160         nlmsg_end(skb, nlh);
5161         return 0;
5162
5163 nla_put_failure:
5164         nlmsg_cancel(skb, nlh);
5165         return -EMSGSIZE;
5166 }
5167
5168 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5169                                const struct net_device *dev)
5170 {
5171         if (f6i->fib6_nh->fib_nh_dev == dev)
5172                 return true;
5173
5174         if (f6i->fib6_nsiblings) {
5175                 struct fib6_info *sibling, *next_sibling;
5176
5177                 list_for_each_entry_safe(sibling, next_sibling,
5178                                          &f6i->fib6_siblings, fib6_siblings) {
5179                         if (sibling->fib6_nh->fib_nh_dev == dev)
5180                                 return true;
5181                 }
5182         }
5183
5184         return false;
5185 }
5186
5187 int rt6_dump_route(struct fib6_info *rt, void *p_arg)
5188 {
5189         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5190         struct fib_dump_filter *filter = &arg->filter;
5191         unsigned int flags = NLM_F_MULTI;
5192         struct net *net = arg->net;
5193
5194         if (rt == net->ipv6.fib6_null_entry)
5195                 return 0;
5196
5197         if ((filter->flags & RTM_F_PREFIX) &&
5198             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5199                 /* success since this is not a prefix route */
5200                 return 1;
5201         }
5202         if (filter->filter_set) {
5203                 if ((filter->rt_type && rt->fib6_type != filter->rt_type) ||
5204                     (filter->dev && !fib6_info_uses_dev(rt, filter->dev)) ||
5205                     (filter->protocol && rt->fib6_protocol != filter->protocol)) {
5206                         return 1;
5207                 }
5208                 flags |= NLM_F_DUMP_FILTERED;
5209         }
5210
5211         return rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL, 0,
5212                              RTM_NEWROUTE, NETLINK_CB(arg->cb->skb).portid,
5213                              arg->cb->nlh->nlmsg_seq, flags);
5214 }
5215
5216 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5217                                         const struct nlmsghdr *nlh,
5218                                         struct nlattr **tb,
5219                                         struct netlink_ext_ack *extack)
5220 {
5221         struct rtmsg *rtm;
5222         int i, err;
5223
5224         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5225                 NL_SET_ERR_MSG_MOD(extack,
5226                                    "Invalid header for get route request");
5227                 return -EINVAL;
5228         }
5229
5230         if (!netlink_strict_get_check(skb))
5231                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5232                                               rtm_ipv6_policy, extack);
5233
5234         rtm = nlmsg_data(nlh);
5235         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5236             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5237             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5238             rtm->rtm_type) {
5239                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5240                 return -EINVAL;
5241         }
5242         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5243                 NL_SET_ERR_MSG_MOD(extack,
5244                                    "Invalid flags for get route request");
5245                 return -EINVAL;
5246         }
5247
5248         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5249                                             rtm_ipv6_policy, extack);
5250         if (err)
5251                 return err;
5252
5253         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5254             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5255                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5256                 return -EINVAL;
5257         }
5258
5259         for (i = 0; i <= RTA_MAX; i++) {
5260                 if (!tb[i])
5261                         continue;
5262
5263                 switch (i) {
5264                 case RTA_SRC:
5265                 case RTA_DST:
5266                 case RTA_IIF:
5267                 case RTA_OIF:
5268                 case RTA_MARK:
5269                 case RTA_UID:
5270                 case RTA_SPORT:
5271                 case RTA_DPORT:
5272                 case RTA_IP_PROTO:
5273                         break;
5274                 default:
5275                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5276                         return -EINVAL;
5277                 }
5278         }
5279
5280         return 0;
5281 }
5282
5283 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5284                               struct netlink_ext_ack *extack)
5285 {
5286         struct net *net = sock_net(in_skb->sk);
5287         struct nlattr *tb[RTA_MAX+1];
5288         int err, iif = 0, oif = 0;
5289         struct fib6_info *from;
5290         struct dst_entry *dst;
5291         struct rt6_info *rt;
5292         struct sk_buff *skb;
5293         struct rtmsg *rtm;
5294         struct flowi6 fl6 = {};
5295         bool fibmatch;
5296
5297         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5298         if (err < 0)
5299                 goto errout;
5300
5301         err = -EINVAL;
5302         rtm = nlmsg_data(nlh);
5303         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5304         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5305
5306         if (tb[RTA_SRC]) {
5307                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5308                         goto errout;
5309
5310                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5311         }
5312
5313         if (tb[RTA_DST]) {
5314                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5315                         goto errout;
5316
5317                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5318         }
5319
5320         if (tb[RTA_IIF])
5321                 iif = nla_get_u32(tb[RTA_IIF]);
5322
5323         if (tb[RTA_OIF])
5324                 oif = nla_get_u32(tb[RTA_OIF]);
5325
5326         if (tb[RTA_MARK])
5327                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5328
5329         if (tb[RTA_UID])
5330                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5331                                            nla_get_u32(tb[RTA_UID]));
5332         else
5333                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5334
5335         if (tb[RTA_SPORT])
5336                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5337
5338         if (tb[RTA_DPORT])
5339                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5340
5341         if (tb[RTA_IP_PROTO]) {
5342                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5343                                                   &fl6.flowi6_proto, AF_INET6,
5344                                                   extack);
5345                 if (err)
5346                         goto errout;
5347         }
5348
5349         if (iif) {
5350                 struct net_device *dev;
5351                 int flags = 0;
5352
5353                 rcu_read_lock();
5354
5355                 dev = dev_get_by_index_rcu(net, iif);
5356                 if (!dev) {
5357                         rcu_read_unlock();
5358                         err = -ENODEV;
5359                         goto errout;
5360                 }
5361
5362                 fl6.flowi6_iif = iif;
5363
5364                 if (!ipv6_addr_any(&fl6.saddr))
5365                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5366
5367                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5368
5369                 rcu_read_unlock();
5370         } else {
5371                 fl6.flowi6_oif = oif;
5372
5373                 dst = ip6_route_output(net, NULL, &fl6);
5374         }
5375
5376
5377         rt = container_of(dst, struct rt6_info, dst);
5378         if (rt->dst.error) {
5379                 err = rt->dst.error;
5380                 ip6_rt_put(rt);
5381                 goto errout;
5382         }
5383
5384         if (rt == net->ipv6.ip6_null_entry) {
5385                 err = rt->dst.error;
5386                 ip6_rt_put(rt);
5387                 goto errout;
5388         }
5389
5390         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5391         if (!skb) {
5392                 ip6_rt_put(rt);
5393                 err = -ENOBUFS;
5394                 goto errout;
5395         }
5396
5397         skb_dst_set(skb, &rt->dst);
5398
5399         rcu_read_lock();
5400         from = rcu_dereference(rt->from);
5401         if (from) {
5402                 if (fibmatch)
5403                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5404                                             iif, RTM_NEWROUTE,
5405                                             NETLINK_CB(in_skb).portid,
5406                                             nlh->nlmsg_seq, 0);
5407                 else
5408                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5409                                             &fl6.saddr, iif, RTM_NEWROUTE,
5410                                             NETLINK_CB(in_skb).portid,
5411                                             nlh->nlmsg_seq, 0);
5412         } else {
5413                 err = -ENETUNREACH;
5414         }
5415         rcu_read_unlock();
5416
5417         if (err < 0) {
5418                 kfree_skb(skb);
5419                 goto errout;
5420         }
5421
5422         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5423 errout:
5424         return err;
5425 }
5426
5427 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
5428                      unsigned int nlm_flags)
5429 {
5430         struct sk_buff *skb;
5431         struct net *net = info->nl_net;
5432         u32 seq;
5433         int err;
5434
5435         err = -ENOBUFS;
5436         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5437
5438         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5439         if (!skb)
5440                 goto errout;
5441
5442         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5443                             event, info->portid, seq, nlm_flags);
5444         if (err < 0) {
5445                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5446                 WARN_ON(err == -EMSGSIZE);
5447                 kfree_skb(skb);
5448                 goto errout;
5449         }
5450         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5451                     info->nlh, gfp_any());
5452         return;
5453 errout:
5454         if (err < 0)
5455                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5456 }
5457
5458 void fib6_rt_update(struct net *net, struct fib6_info *rt,
5459                     struct nl_info *info)
5460 {
5461         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5462         struct sk_buff *skb;
5463         int err = -ENOBUFS;
5464
5465         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
5466          * is implemented and supported for nexthop objects
5467          */
5468         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
5469
5470         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5471         if (!skb)
5472                 goto errout;
5473
5474         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5475                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
5476         if (err < 0) {
5477                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5478                 WARN_ON(err == -EMSGSIZE);
5479                 kfree_skb(skb);
5480                 goto errout;
5481         }
5482         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5483                     info->nlh, gfp_any());
5484         return;
5485 errout:
5486         if (err < 0)
5487                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5488 }
5489
5490 static int ip6_route_dev_notify(struct notifier_block *this,
5491                                 unsigned long event, void *ptr)
5492 {
5493         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
5494         struct net *net = dev_net(dev);
5495
5496         if (!(dev->flags & IFF_LOOPBACK))
5497                 return NOTIFY_OK;
5498
5499         if (event == NETDEV_REGISTER) {
5500                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
5501                 net->ipv6.ip6_null_entry->dst.dev = dev;
5502                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
5503 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5504                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
5505                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
5506                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
5507                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
5508 #endif
5509          } else if (event == NETDEV_UNREGISTER &&
5510                     dev->reg_state != NETREG_UNREGISTERED) {
5511                 /* NETDEV_UNREGISTER could be fired for multiple times by
5512                  * netdev_wait_allrefs(). Make sure we only call this once.
5513                  */
5514                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
5515 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5516                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
5517                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
5518 #endif
5519         }
5520
5521         return NOTIFY_OK;
5522 }
5523
5524 /*
5525  *      /proc
5526  */
5527
5528 #ifdef CONFIG_PROC_FS
5529 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
5530 {
5531         struct net *net = (struct net *)seq->private;
5532         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
5533                    net->ipv6.rt6_stats->fib_nodes,
5534                    net->ipv6.rt6_stats->fib_route_nodes,
5535                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
5536                    net->ipv6.rt6_stats->fib_rt_entries,
5537                    net->ipv6.rt6_stats->fib_rt_cache,
5538                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
5539                    net->ipv6.rt6_stats->fib_discarded_routes);
5540
5541         return 0;
5542 }
5543 #endif  /* CONFIG_PROC_FS */
5544
5545 #ifdef CONFIG_SYSCTL
5546
5547 static
5548 int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
5549                               void __user *buffer, size_t *lenp, loff_t *ppos)
5550 {
5551         struct net *net;
5552         int delay;
5553         int ret;
5554         if (!write)
5555                 return -EINVAL;
5556
5557         net = (struct net *)ctl->extra1;
5558         delay = net->ipv6.sysctl.flush_delay;
5559         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
5560         if (ret)
5561                 return ret;
5562
5563         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
5564         return 0;
5565 }
5566
5567 static int zero;
5568 static int one = 1;
5569
5570 static struct ctl_table ipv6_route_table_template[] = {
5571         {
5572                 .procname       =       "flush",
5573                 .data           =       &init_net.ipv6.sysctl.flush_delay,
5574                 .maxlen         =       sizeof(int),
5575                 .mode           =       0200,
5576                 .proc_handler   =       ipv6_sysctl_rtcache_flush
5577         },
5578         {
5579                 .procname       =       "gc_thresh",
5580                 .data           =       &ip6_dst_ops_template.gc_thresh,
5581                 .maxlen         =       sizeof(int),
5582                 .mode           =       0644,
5583                 .proc_handler   =       proc_dointvec,
5584         },
5585         {
5586                 .procname       =       "max_size",
5587                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
5588                 .maxlen         =       sizeof(int),
5589                 .mode           =       0644,
5590                 .proc_handler   =       proc_dointvec,
5591         },
5592         {
5593                 .procname       =       "gc_min_interval",
5594                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
5595                 .maxlen         =       sizeof(int),
5596                 .mode           =       0644,
5597                 .proc_handler   =       proc_dointvec_jiffies,
5598         },
5599         {
5600                 .procname       =       "gc_timeout",
5601                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
5602                 .maxlen         =       sizeof(int),
5603                 .mode           =       0644,
5604                 .proc_handler   =       proc_dointvec_jiffies,
5605         },
5606         {
5607                 .procname       =       "gc_interval",
5608                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
5609                 .maxlen         =       sizeof(int),
5610                 .mode           =       0644,
5611                 .proc_handler   =       proc_dointvec_jiffies,
5612         },
5613         {
5614                 .procname       =       "gc_elasticity",
5615                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
5616                 .maxlen         =       sizeof(int),
5617                 .mode           =       0644,
5618                 .proc_handler   =       proc_dointvec,
5619         },
5620         {
5621                 .procname       =       "mtu_expires",
5622                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
5623                 .maxlen         =       sizeof(int),
5624                 .mode           =       0644,
5625                 .proc_handler   =       proc_dointvec_jiffies,
5626         },
5627         {
5628                 .procname       =       "min_adv_mss",
5629                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
5630                 .maxlen         =       sizeof(int),
5631                 .mode           =       0644,
5632                 .proc_handler   =       proc_dointvec,
5633         },
5634         {
5635                 .procname       =       "gc_min_interval_ms",
5636                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
5637                 .maxlen         =       sizeof(int),
5638                 .mode           =       0644,
5639                 .proc_handler   =       proc_dointvec_ms_jiffies,
5640         },
5641         {
5642                 .procname       =       "skip_notify_on_dev_down",
5643                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
5644                 .maxlen         =       sizeof(int),
5645                 .mode           =       0644,
5646                 .proc_handler   =       proc_dointvec,
5647                 .extra1         =       &zero,
5648                 .extra2         =       &one,
5649         },
5650         { }
5651 };
5652
5653 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
5654 {
5655         struct ctl_table *table;
5656
5657         table = kmemdup(ipv6_route_table_template,
5658                         sizeof(ipv6_route_table_template),
5659                         GFP_KERNEL);
5660
5661         if (table) {
5662                 table[0].data = &net->ipv6.sysctl.flush_delay;
5663                 table[0].extra1 = net;
5664                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
5665                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
5666                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
5667                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
5668                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
5669                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
5670                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
5671                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
5672                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
5673                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
5674
5675                 /* Don't export sysctls to unprivileged users */
5676                 if (net->user_ns != &init_user_ns)
5677                         table[0].procname = NULL;
5678         }
5679
5680         return table;
5681 }
5682 #endif
5683
5684 static int __net_init ip6_route_net_init(struct net *net)
5685 {
5686         int ret = -ENOMEM;
5687
5688         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
5689                sizeof(net->ipv6.ip6_dst_ops));
5690
5691         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
5692                 goto out_ip6_dst_ops;
5693
5694         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
5695         if (!net->ipv6.fib6_null_entry)
5696                 goto out_ip6_dst_entries;
5697         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
5698                sizeof(*net->ipv6.fib6_null_entry));
5699
5700         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
5701                                            sizeof(*net->ipv6.ip6_null_entry),
5702                                            GFP_KERNEL);
5703         if (!net->ipv6.ip6_null_entry)
5704                 goto out_fib6_null_entry;
5705         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5706         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
5707                          ip6_template_metrics, true);
5708
5709 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5710         net->ipv6.fib6_has_custom_rules = false;
5711         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
5712                                                sizeof(*net->ipv6.ip6_prohibit_entry),
5713                                                GFP_KERNEL);
5714         if (!net->ipv6.ip6_prohibit_entry)
5715                 goto out_ip6_null_entry;
5716         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5717         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
5718                          ip6_template_metrics, true);
5719
5720         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
5721                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
5722                                                GFP_KERNEL);
5723         if (!net->ipv6.ip6_blk_hole_entry)
5724                 goto out_ip6_prohibit_entry;
5725         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5726         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
5727                          ip6_template_metrics, true);
5728 #endif
5729
5730         net->ipv6.sysctl.flush_delay = 0;
5731         net->ipv6.sysctl.ip6_rt_max_size = 4096;
5732         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
5733         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
5734         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
5735         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
5736         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
5737         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
5738         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
5739
5740         net->ipv6.ip6_rt_gc_expire = 30*HZ;
5741
5742         ret = 0;
5743 out:
5744         return ret;
5745
5746 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5747 out_ip6_prohibit_entry:
5748         kfree(net->ipv6.ip6_prohibit_entry);
5749 out_ip6_null_entry:
5750         kfree(net->ipv6.ip6_null_entry);
5751 #endif
5752 out_fib6_null_entry:
5753         kfree(net->ipv6.fib6_null_entry);
5754 out_ip6_dst_entries:
5755         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
5756 out_ip6_dst_ops:
5757         goto out;
5758 }
5759
5760 static void __net_exit ip6_route_net_exit(struct net *net)
5761 {
5762         kfree(net->ipv6.fib6_null_entry);
5763         kfree(net->ipv6.ip6_null_entry);
5764 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5765         kfree(net->ipv6.ip6_prohibit_entry);
5766         kfree(net->ipv6.ip6_blk_hole_entry);
5767 #endif
5768         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
5769 }
5770
5771 static int __net_init ip6_route_net_init_late(struct net *net)
5772 {
5773 #ifdef CONFIG_PROC_FS
5774         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
5775                         sizeof(struct ipv6_route_iter));
5776         proc_create_net_single("rt6_stats", 0444, net->proc_net,
5777                         rt6_stats_seq_show, NULL);
5778 #endif
5779         return 0;
5780 }
5781
5782 static void __net_exit ip6_route_net_exit_late(struct net *net)
5783 {
5784 #ifdef CONFIG_PROC_FS
5785         remove_proc_entry("ipv6_route", net->proc_net);
5786         remove_proc_entry("rt6_stats", net->proc_net);
5787 #endif
5788 }
5789
5790 static struct pernet_operations ip6_route_net_ops = {
5791         .init = ip6_route_net_init,
5792         .exit = ip6_route_net_exit,
5793 };
5794
5795 static int __net_init ipv6_inetpeer_init(struct net *net)
5796 {
5797         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
5798
5799         if (!bp)
5800                 return -ENOMEM;
5801         inet_peer_base_init(bp);
5802         net->ipv6.peers = bp;
5803         return 0;
5804 }
5805
5806 static void __net_exit ipv6_inetpeer_exit(struct net *net)
5807 {
5808         struct inet_peer_base *bp = net->ipv6.peers;
5809
5810         net->ipv6.peers = NULL;
5811         inetpeer_invalidate_tree(bp);
5812         kfree(bp);
5813 }
5814
5815 static struct pernet_operations ipv6_inetpeer_ops = {
5816         .init   =       ipv6_inetpeer_init,
5817         .exit   =       ipv6_inetpeer_exit,
5818 };
5819
5820 static struct pernet_operations ip6_route_net_late_ops = {
5821         .init = ip6_route_net_init_late,
5822         .exit = ip6_route_net_exit_late,
5823 };
5824
5825 static struct notifier_block ip6_route_dev_notifier = {
5826         .notifier_call = ip6_route_dev_notify,
5827         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
5828 };
5829
5830 void __init ip6_route_init_special_entries(void)
5831 {
5832         /* Registering of the loopback is done before this portion of code,
5833          * the loopback reference in rt6_info will not be taken, do it
5834          * manually for init_net */
5835         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
5836         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
5837         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5838   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5839         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
5840         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5841         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
5842         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5843   #endif
5844 }
5845
5846 int __init ip6_route_init(void)
5847 {
5848         int ret;
5849         int cpu;
5850
5851         ret = -ENOMEM;
5852         ip6_dst_ops_template.kmem_cachep =
5853                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
5854                                   SLAB_HWCACHE_ALIGN, NULL);
5855         if (!ip6_dst_ops_template.kmem_cachep)
5856                 goto out;
5857
5858         ret = dst_entries_init(&ip6_dst_blackhole_ops);
5859         if (ret)
5860                 goto out_kmem_cache;
5861
5862         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
5863         if (ret)
5864                 goto out_dst_entries;
5865
5866         ret = register_pernet_subsys(&ip6_route_net_ops);
5867         if (ret)
5868                 goto out_register_inetpeer;
5869
5870         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
5871
5872         ret = fib6_init();
5873         if (ret)
5874                 goto out_register_subsys;
5875
5876         ret = xfrm6_init();
5877         if (ret)
5878                 goto out_fib6_init;
5879
5880         ret = fib6_rules_init();
5881         if (ret)
5882                 goto xfrm6_init;
5883
5884         ret = register_pernet_subsys(&ip6_route_net_late_ops);
5885         if (ret)
5886                 goto fib6_rules_init;
5887
5888         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
5889                                    inet6_rtm_newroute, NULL, 0);
5890         if (ret < 0)
5891                 goto out_register_late_subsys;
5892
5893         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
5894                                    inet6_rtm_delroute, NULL, 0);
5895         if (ret < 0)
5896                 goto out_register_late_subsys;
5897
5898         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
5899                                    inet6_rtm_getroute, NULL,
5900                                    RTNL_FLAG_DOIT_UNLOCKED);
5901         if (ret < 0)
5902                 goto out_register_late_subsys;
5903
5904         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
5905         if (ret)
5906                 goto out_register_late_subsys;
5907
5908         for_each_possible_cpu(cpu) {
5909                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
5910
5911                 INIT_LIST_HEAD(&ul->head);
5912                 spin_lock_init(&ul->lock);
5913         }
5914
5915 out:
5916         return ret;
5917
5918 out_register_late_subsys:
5919         rtnl_unregister_all(PF_INET6);
5920         unregister_pernet_subsys(&ip6_route_net_late_ops);
5921 fib6_rules_init:
5922         fib6_rules_cleanup();
5923 xfrm6_init:
5924         xfrm6_fini();
5925 out_fib6_init:
5926         fib6_gc_cleanup();
5927 out_register_subsys:
5928         unregister_pernet_subsys(&ip6_route_net_ops);
5929 out_register_inetpeer:
5930         unregister_pernet_subsys(&ipv6_inetpeer_ops);
5931 out_dst_entries:
5932         dst_entries_destroy(&ip6_dst_blackhole_ops);
5933 out_kmem_cache:
5934         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
5935         goto out;
5936 }
5937
5938 void ip6_route_cleanup(void)
5939 {
5940         unregister_netdevice_notifier(&ip6_route_dev_notifier);
5941         unregister_pernet_subsys(&ip6_route_net_late_ops);
5942         fib6_rules_cleanup();
5943         xfrm6_fini();
5944         fib6_gc_cleanup();
5945         unregister_pernet_subsys(&ipv6_inetpeer_ops);
5946         unregister_pernet_subsys(&ip6_route_net_ops);
5947         dst_entries_destroy(&ip6_dst_blackhole_ops);
5948         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
5949 }