net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <net/net_namespace.h>
  45 #include <net/snmp.h>
  46 #include <net/ipv6.h>
  47 #include <net/ip6_fib.h>
  48 #include <net/ip6_route.h>
  49 #include <net/ndisc.h>
  50 #include <net/addrconf.h>
  51 #include <net/tcp.h>
  52 #include <linux/rtnetlink.h>
  53 #include <net/dst.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/xfrm.h>
  56 #include <net/netevent.h>
  57 #include <net/netlink.h>
  58 #include <net/rtnh.h>
  59 #include <net/lwtunnel.h>
  60 #include <net/ip_tunnels.h>
  61 #include <net/l3mdev.h>
  62 #include <net/ip.h>
  63 #include <linux/uaccess.h>
  64
  65 #ifdef CONFIG_SYSCTL
  66 #include <linux/sysctl.h>
  67 #endif
  68
  69 static int ip6_rt_type_to_error(u8 fib6_type);
  70
  71 #define CREATE_TRACE_POINTS
  72 #include <trace/events/fib6.h>
  73 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  74 #undef CREATE_TRACE_POINTS
  75
  76 enum rt6_nud_state {
  77         RT6_NUD_FAIL_HARD = -3,
  78         RT6_NUD_FAIL_PROBE = -2,
  79         RT6_NUD_FAIL_DO_RR = -1,
  80         RT6_NUD_SUCCEED = 1
  81 };
  82
  83 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  84 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  85 static unsigned int      ip6_mtu(const struct dst_entry *dst);
  86 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  87 static void             ip6_dst_destroy(struct dst_entry *);
  88 static void             ip6_dst_ifdown(struct dst_entry *,
  89                                        struct net_device *dev, int how);
  90 static int               ip6_dst_gc(struct dst_ops *ops);
  91
  92 static int              ip6_pkt_discard(struct sk_buff *skb);
  93 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  94 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  95 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  96 static void             ip6_link_failure(struct sk_buff *skb);
  97 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
  98                                            struct sk_buff *skb, u32 mtu);
  99 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 100                                         struct sk_buff *skb);
 101 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 102                            int strict);
 103 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 104 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 105                          struct fib6_info *rt, struct dst_entry *dst,
 106                          struct in6_addr *dest, struct in6_addr *src,
 107                          int iif, int type, u32 portid, u32 seq,
 108                          unsigned int flags);
 109 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 110                                            const struct in6_addr *daddr,
 111                                            const struct in6_addr *saddr);
 112
 113 #ifdef CONFIG_IPV6_ROUTE_INFO
 114 static struct fib6_info *rt6_add_route_info(struct net *net,
 115                                            const struct in6_addr *prefix, int prefixlen,
 116                                            const struct in6_addr *gwaddr,
 117                                            struct net_device *dev,
 118                                            unsigned int pref);
 119 static struct fib6_info *rt6_get_route_info(struct net *net,
 120                                            const struct in6_addr *prefix, int prefixlen,
 121                                            const struct in6_addr *gwaddr,
 122                                            struct net_device *dev);
 123 #endif
 124
 125 struct uncached_list {
 126         spinlock_t              lock;
 127         struct list_head        head;
 128 };
 129
 130 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 131
 132 void rt6_uncached_list_add(struct rt6_info *rt)
 133 {
 134         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 135
 136         rt->rt6i_uncached_list = ul;
 137
 138         spin_lock_bh(&ul->lock);
 139         list_add_tail(&rt->rt6i_uncached, &ul->head);
 140         spin_unlock_bh(&ul->lock);
 141 }
 142
 143 void rt6_uncached_list_del(struct rt6_info *rt)
 144 {
 145         if (!list_empty(&rt->rt6i_uncached)) {
 146                 struct uncached_list *ul = rt->rt6i_uncached_list;
 147                 struct net *net = dev_net(rt->dst.dev);
 148
 149                 spin_lock_bh(&ul->lock);
 150                 list_del(&rt->rt6i_uncached);
 151                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 152                 spin_unlock_bh(&ul->lock);
 153         }
 154 }
 155
 156 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 157 {
 158         struct net_device *loopback_dev = net->loopback_dev;
 159         int cpu;
 160
 161         if (dev == loopback_dev)
 162                 return;
 163
 164         for_each_possible_cpu(cpu) {
 165                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 166                 struct rt6_info *rt;
 167
 168                 spin_lock_bh(&ul->lock);
 169                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 170                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 171                         struct net_device *rt_dev = rt->dst.dev;
 172
 173                         if (rt_idev->dev == dev) {
 174                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 175                                 in6_dev_put(rt_idev);
 176                         }
 177
 178                         if (rt_dev == dev) {
 179                                 rt->dst.dev = blackhole_netdev;
 180                                 dev_hold(rt->dst.dev);
 181                                 dev_put(rt_dev);
 182                         }
 183                 }
 184                 spin_unlock_bh(&ul->lock);
 185         }
 186 }
 187
 188 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 189                                              struct sk_buff *skb,
 190                                              const void *daddr)
 191 {
 192         if (!ipv6_addr_any(p))
 193                 return (const void *) p;
 194         else if (skb)
 195                 return &ipv6_hdr(skb)->daddr;
 196         return daddr;
 197 }
 198
 199 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 200                                    struct net_device *dev,
 201                                    struct sk_buff *skb,
 202                                    const void *daddr)
 203 {
 204         struct neighbour *n;
 205
 206         daddr = choose_neigh_daddr(gw, skb, daddr);
 207         n = __ipv6_neigh_lookup(dev, daddr);
 208         if (n)
 209                 return n;
 210
 211         n = neigh_create(&nd_tbl, daddr, dev);
 212         return IS_ERR(n) ? NULL : n;
 213 }
 214
 215 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 216                                               struct sk_buff *skb,
 217                                               const void *daddr)
 218 {
 219         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 220
 221         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 222                                 dst->dev, skb, daddr);
 223 }
 224
 225 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 226 {
 227         struct net_device *dev = dst->dev;
 228         struct rt6_info *rt = (struct rt6_info *)dst;
 229
 230         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 231         if (!daddr)
 232                 return;
 233         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 234                 return;
 235         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 236                 return;
 237         __ipv6_confirm_neigh(dev, daddr);
 238 }
 239
 240 static struct dst_ops ip6_dst_ops_template = {
 241         .family                 =       AF_INET6,
 242         .gc                     =       ip6_dst_gc,
 243         .gc_thresh              =       1024,
 244         .check                  =       ip6_dst_check,
 245         .default_advmss         =       ip6_default_advmss,
 246         .mtu                    =       ip6_mtu,
 247         .cow_metrics            =       dst_cow_metrics_generic,
 248         .destroy                =       ip6_dst_destroy,
 249         .ifdown                 =       ip6_dst_ifdown,
 250         .negative_advice        =       ip6_negative_advice,
 251         .link_failure           =       ip6_link_failure,
 252         .update_pmtu            =       ip6_rt_update_pmtu,
 253         .redirect               =       rt6_do_redirect,
 254         .local_out              =       __ip6_local_out,
 255         .neigh_lookup           =       ip6_dst_neigh_lookup,
 256         .confirm_neigh          =       ip6_confirm_neigh,
 257 };
 258
 259 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 260 {
 261         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 262
 263         return mtu ? : dst->dev->mtu;
 264 }
 265
 266 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 267                                          struct sk_buff *skb, u32 mtu)
 268 {
 269 }
 270
 271 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 272                                       struct sk_buff *skb)
 273 {
 274 }
 275
 276 static struct dst_ops ip6_dst_blackhole_ops = {
 277         .family                 =       AF_INET6,
 278         .destroy                =       ip6_dst_destroy,
 279         .check                  =       ip6_dst_check,
 280         .mtu                    =       ip6_blackhole_mtu,
 281         .default_advmss         =       ip6_default_advmss,
 282         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 283         .redirect               =       ip6_rt_blackhole_redirect,
 284         .cow_metrics            =       dst_cow_metrics_generic,
 285         .neigh_lookup           =       ip6_dst_neigh_lookup,
 286 };
 287
 288 static const u32 ip6_template_metrics[RTAX_MAX] = {
 289         [RTAX_HOPLIMIT - 1] = 0,
 290 };
 291
 292 static const struct fib6_info fib6_null_entry_template = {
 293         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 294         .fib6_protocol  = RTPROT_KERNEL,
 295         .fib6_metric    = ~(u32)0,
 296         .fib6_ref       = REFCOUNT_INIT(1),
 297         .fib6_type      = RTN_UNREACHABLE,
 298         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 299 };
 300
 301 static const struct rt6_info ip6_null_entry_template = {
 302         .dst = {
 303                 .__refcnt       = ATOMIC_INIT(1),
 304                 .__use          = 1,
 305                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 306                 .error          = -ENETUNREACH,
 307                 .input          = ip6_pkt_discard,
 308                 .output         = ip6_pkt_discard_out,
 309         },
 310         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 311 };
 312
 313 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 314
 315 static const struct rt6_info ip6_prohibit_entry_template = {
 316         .dst = {
 317                 .__refcnt       = ATOMIC_INIT(1),
 318                 .__use          = 1,
 319                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 320                 .error          = -EACCES,
 321                 .input          = ip6_pkt_prohibit,
 322                 .output         = ip6_pkt_prohibit_out,
 323         },
 324         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 325 };
 326
 327 static const struct rt6_info ip6_blk_hole_entry_template = {
 328         .dst = {
 329                 .__refcnt       = ATOMIC_INIT(1),
 330                 .__use          = 1,
 331                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 332                 .error          = -EINVAL,
 333                 .input          = dst_discard,
 334                 .output         = dst_discard_out,
 335         },
 336         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 337 };
 338
 339 #endif
 340
 341 static void rt6_info_init(struct rt6_info *rt)
 342 {
 343         struct dst_entry *dst = &rt->dst;
 344
 345         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 346         INIT_LIST_HEAD(&rt->rt6i_uncached);
 347 }
 348
 349 /* allocate dst with ip6_dst_ops */
 350 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 351                                int flags)
 352 {
 353         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 354                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 355
 356         if (rt) {
 357                 rt6_info_init(rt);
 358                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 359         }
 360
 361         return rt;
 362 }
 363 EXPORT_SYMBOL(ip6_dst_alloc);
 364
 365 static void ip6_dst_destroy(struct dst_entry *dst)
 366 {
 367         struct rt6_info *rt = (struct rt6_info *)dst;
 368         struct fib6_info *from;
 369         struct inet6_dev *idev;
 370
 371         ip_dst_metrics_put(dst);
 372         rt6_uncached_list_del(rt);
 373
 374         idev = rt->rt6i_idev;
 375         if (idev) {
 376                 rt->rt6i_idev = NULL;
 377                 in6_dev_put(idev);
 378         }
 379
 380         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 381         fib6_info_release(from);
 382 }
 383
 384 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 385                            int how)
 386 {
 387         struct rt6_info *rt = (struct rt6_info *)dst;
 388         struct inet6_dev *idev = rt->rt6i_idev;
 389         struct net_device *loopback_dev =
 390                 dev_net(dev)->loopback_dev;
 391
 392         if (idev && idev->dev != loopback_dev) {
 393                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 394                 if (loopback_idev) {
 395                         rt->rt6i_idev = loopback_idev;
 396                         in6_dev_put(idev);
 397                 }
 398         }
 399 }
 400
 401 static bool __rt6_check_expired(const struct rt6_info *rt)
 402 {
 403         if (rt->rt6i_flags & RTF_EXPIRES)
 404                 return time_after(jiffies, rt->dst.expires);
 405         else
 406                 return false;
 407 }
 408
 409 static bool rt6_check_expired(const struct rt6_info *rt)
 410 {
 411         struct fib6_info *from;
 412
 413         from = rcu_dereference(rt->from);
 414
 415         if (rt->rt6i_flags & RTF_EXPIRES) {
 416                 if (time_after(jiffies, rt->dst.expires))
 417                         return true;
 418         } else if (from) {
 419                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 420                         fib6_check_expired(from);
 421         }
 422         return false;
 423 }
 424
 425 void fib6_select_path(const struct net *net, struct fib6_result *res,
 426                       struct flowi6 *fl6, int oif, bool have_oif_match,
 427                       const struct sk_buff *skb, int strict)
 428 {
 429         struct fib6_info *sibling, *next_sibling;
 430         struct fib6_info *match = res->f6i;
 431
 432         if ((!match->fib6_nsiblings && !match->nh) || have_oif_match)
 433                 goto out;
 434
 435         /* We might have already computed the hash for ICMPv6 errors. In such
 436          * case it will always be non-zero. Otherwise now is the time to do it.
 437          */
 438         if (!fl6->mp_hash &&
 439             (!match->nh || nexthop_is_multipath(match->nh)))
 440                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 441
 442         if (unlikely(match->nh)) {
 443                 nexthop_path_fib6_result(res, fl6->mp_hash);
 444                 return;
 445         }
 446
 447         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 448                 goto out;
 449
 450         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 451                                  fib6_siblings) {
 452                 const struct fib6_nh *nh = sibling->fib6_nh;
 453                 int nh_upper_bound;
 454
 455                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 456                 if (fl6->mp_hash > nh_upper_bound)
 457                         continue;
 458                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 459                         break;
 460                 match = sibling;
 461                 break;
 462         }
 463
 464 out:
 465         res->f6i = match;
 466         res->nh = match->fib6_nh;
 467 }
 468
 469 /*
 470  *      Route lookup. rcu_read_lock() should be held.
 471  */
 472
 473 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 474                                const struct in6_addr *saddr, int oif, int flags)
 475 {
 476         const struct net_device *dev;
 477
 478         if (nh->fib_nh_flags & RTNH_F_DEAD)
 479                 return false;
 480
 481         dev = nh->fib_nh_dev;
 482         if (oif) {
 483                 if (dev->ifindex == oif)
 484                         return true;
 485         } else {
 486                 if (ipv6_chk_addr(net, saddr, dev,
 487                                   flags & RT6_LOOKUP_F_IFACE))
 488                         return true;
 489         }
 490
 491         return false;
 492 }
 493
 494 struct fib6_nh_dm_arg {
 495         struct net              *net;
 496         const struct in6_addr   *saddr;
 497         int                     oif;
 498         int                     flags;
 499         struct fib6_nh          *nh;
 500 };
 501
 502 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 503 {
 504         struct fib6_nh_dm_arg *arg = _arg;
 505
 506         arg->nh = nh;
 507         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 508                                   arg->flags);
 509 }
 510
 511 /* returns fib6_nh from nexthop or NULL */
 512 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 513                                         struct fib6_result *res,
 514                                         const struct in6_addr *saddr,
 515                                         int oif, int flags)
 516 {
 517         struct fib6_nh_dm_arg arg = {
 518                 .net   = net,
 519                 .saddr = saddr,
 520                 .oif   = oif,
 521                 .flags = flags,
 522         };
 523
 524         if (nexthop_is_blackhole(nh))
 525                 return NULL;
 526
 527         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 528                 return arg.nh;
 529
 530         return NULL;
 531 }
 532
 533 static void rt6_device_match(struct net *net, struct fib6_result *res,
 534                              const struct in6_addr *saddr, int oif, int flags)
 535 {
 536         struct fib6_info *f6i = res->f6i;
 537         struct fib6_info *spf6i;
 538         struct fib6_nh *nh;
 539
 540         if (!oif && ipv6_addr_any(saddr)) {
 541                 if (unlikely(f6i->nh)) {
 542                         nh = nexthop_fib6_nh(f6i->nh);
 543                         if (nexthop_is_blackhole(f6i->nh))
 544                                 goto out_blackhole;
 545                 } else {
 546                         nh = f6i->fib6_nh;
 547                 }
 548                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 549                         goto out;
 550         }
 551
 552         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 553                 bool matched = false;
 554
 555                 if (unlikely(spf6i->nh)) {
 556                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 557                                               oif, flags);
 558                         if (nh)
 559                                 matched = true;
 560                 } else {
 561                         nh = spf6i->fib6_nh;
 562                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 563                                 matched = true;
 564                 }
 565                 if (matched) {
 566                         res->f6i = spf6i;
 567                         goto out;
 568                 }
 569         }
 570
 571         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 572                 res->f6i = net->ipv6.fib6_null_entry;
 573                 nh = res->f6i->fib6_nh;
 574                 goto out;
 575         }
 576
 577         if (unlikely(f6i->nh)) {
 578                 nh = nexthop_fib6_nh(f6i->nh);
 579                 if (nexthop_is_blackhole(f6i->nh))
 580                         goto out_blackhole;
 581         } else {
 582                 nh = f6i->fib6_nh;
 583         }
 584
 585         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 586                 res->f6i = net->ipv6.fib6_null_entry;
 587                 nh = res->f6i->fib6_nh;
 588         }
 589 out:
 590         res->nh = nh;
 591         res->fib6_type = res->f6i->fib6_type;
 592         res->fib6_flags = res->f6i->fib6_flags;
 593         return;
 594
 595 out_blackhole:
 596         res->fib6_flags |= RTF_REJECT;
 597         res->fib6_type = RTN_BLACKHOLE;
 598         res->nh = nh;
 599 }
 600
 601 #ifdef CONFIG_IPV6_ROUTER_PREF
 602 struct __rt6_probe_work {
 603         struct work_struct work;
 604         struct in6_addr target;
 605         struct net_device *dev;
 606 };
 607
 608 static void rt6_probe_deferred(struct work_struct *w)
 609 {
 610         struct in6_addr mcaddr;
 611         struct __rt6_probe_work *work =
 612                 container_of(w, struct __rt6_probe_work, work);
 613
 614         addrconf_addr_solict_mult(&work->target, &mcaddr);
 615         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 616         dev_put(work->dev);
 617         kfree(work);
 618 }
 619
 620 static void rt6_probe(struct fib6_nh *fib6_nh)
 621 {
 622         struct __rt6_probe_work *work = NULL;
 623         const struct in6_addr *nh_gw;
 624         struct neighbour *neigh;
 625         struct net_device *dev;
 626         struct inet6_dev *idev;
 627
 628         /*
 629          * Okay, this does not seem to be appropriate
 630          * for now, however, we need to check if it
 631          * is really so; aka Router Reachability Probing.
 632          *
 633          * Router Reachability Probe MUST be rate-limited
 634          * to no more than one per minute.
 635          */
 636         if (fib6_nh->fib_nh_gw_family)
 637                 return;
 638
 639         nh_gw = &fib6_nh->fib_nh_gw6;
 640         dev = fib6_nh->fib_nh_dev;
 641         rcu_read_lock_bh();
 642         idev = __in6_dev_get(dev);
 643         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 644         if (neigh) {
 645                 if (neigh->nud_state & NUD_VALID)
 646                         goto out;
 647
 648                 write_lock(&neigh->lock);
 649                 if (!(neigh->nud_state & NUD_VALID) &&
 650                     time_after(jiffies,
 651                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 652                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 653                         if (work)
 654                                 __neigh_set_probe_once(neigh);
 655                 }
 656                 write_unlock(&neigh->lock);
 657         } else if (time_after(jiffies, fib6_nh->last_probe +
 658                                        idev->cnf.rtr_probe_interval)) {
 659                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 660         }
 661
 662         if (work) {
 663                 fib6_nh->last_probe = jiffies;
 664                 INIT_WORK(&work->work, rt6_probe_deferred);
 665                 work->target = *nh_gw;
 666                 dev_hold(dev);
 667                 work->dev = dev;
 668                 schedule_work(&work->work);
 669         }
 670
 671 out:
 672         rcu_read_unlock_bh();
 673 }
 674 #else
 675 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 676 {
 677 }
 678 #endif
 679
 680 /*
 681  * Default Router Selection (RFC 2461 6.3.6)
 682  */
 683 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 684 {
 685         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 686         struct neighbour *neigh;
 687
 688         rcu_read_lock_bh();
 689         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 690                                           &fib6_nh->fib_nh_gw6);
 691         if (neigh) {
 692                 read_lock(&neigh->lock);
 693                 if (neigh->nud_state & NUD_VALID)
 694                         ret = RT6_NUD_SUCCEED;
 695 #ifdef CONFIG_IPV6_ROUTER_PREF
 696                 else if (!(neigh->nud_state & NUD_FAILED))
 697                         ret = RT6_NUD_SUCCEED;
 698                 else
 699                         ret = RT6_NUD_FAIL_PROBE;
 700 #endif
 701                 read_unlock(&neigh->lock);
 702         } else {
 703                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 704                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 705         }
 706         rcu_read_unlock_bh();
 707
 708         return ret;
 709 }
 710
 711 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 712                            int strict)
 713 {
 714         int m = 0;
 715
 716         if (!oif || nh->fib_nh_dev->ifindex == oif)
 717                 m = 2;
 718
 719         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 720                 return RT6_NUD_FAIL_HARD;
 721 #ifdef CONFIG_IPV6_ROUTER_PREF
 722         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 723 #endif
 724         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 725             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 726                 int n = rt6_check_neigh(nh);
 727                 if (n < 0)
 728                         return n;
 729         }
 730         return m;
 731 }
 732
 733 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 734                        int oif, int strict, int *mpri, bool *do_rr)
 735 {
 736         bool match_do_rr = false;
 737         bool rc = false;
 738         int m;
 739
 740         if (nh->fib_nh_flags & RTNH_F_DEAD)
 741                 goto out;
 742
 743         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 744             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 745             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 746                 goto out;
 747
 748         m = rt6_score_route(nh, fib6_flags, oif, strict);
 749         if (m == RT6_NUD_FAIL_DO_RR) {
 750                 match_do_rr = true;
 751                 m = 0; /* lowest valid score */
 752         } else if (m == RT6_NUD_FAIL_HARD) {
 753                 goto out;
 754         }
 755
 756         if (strict & RT6_LOOKUP_F_REACHABLE)
 757                 rt6_probe(nh);
 758
 759         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 760         if (m > *mpri) {
 761                 *do_rr = match_do_rr;
 762                 *mpri = m;
 763                 rc = true;
 764         }
 765 out:
 766         return rc;
 767 }
 768
 769 struct fib6_nh_frl_arg {
 770         u32             flags;
 771         int             oif;
 772         int             strict;
 773         int             *mpri;
 774         bool            *do_rr;
 775         struct fib6_nh  *nh;
 776 };
 777
 778 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 779 {
 780         struct fib6_nh_frl_arg *arg = _arg;
 781
 782         arg->nh = nh;
 783         return find_match(nh, arg->flags, arg->oif, arg->strict,
 784                           arg->mpri, arg->do_rr);
 785 }
 786
 787 static void __find_rr_leaf(struct fib6_info *f6i_start,
 788                            struct fib6_info *nomatch, u32 metric,
 789                            struct fib6_result *res, struct fib6_info **cont,
 790                            int oif, int strict, bool *do_rr, int *mpri)
 791 {
 792         struct fib6_info *f6i;
 793
 794         for (f6i = f6i_start;
 795              f6i && f6i != nomatch;
 796              f6i = rcu_dereference(f6i->fib6_next)) {
 797                 bool matched = false;
 798                 struct fib6_nh *nh;
 799
 800                 if (cont && f6i->fib6_metric != metric) {
 801                         *cont = f6i;
 802                         return;
 803                 }
 804
 805                 if (fib6_check_expired(f6i))
 806                         continue;
 807
 808                 if (unlikely(f6i->nh)) {
 809                         struct fib6_nh_frl_arg arg = {
 810                                 .flags  = f6i->fib6_flags,
 811                                 .oif    = oif,
 812                                 .strict = strict,
 813                                 .mpri   = mpri,
 814                                 .do_rr  = do_rr
 815                         };
 816
 817                         if (nexthop_is_blackhole(f6i->nh)) {
 818                                 res->fib6_flags = RTF_REJECT;
 819                                 res->fib6_type = RTN_BLACKHOLE;
 820                                 res->f6i = f6i;
 821                                 res->nh = nexthop_fib6_nh(f6i->nh);
 822                                 return;
 823                         }
 824                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 825                                                      &arg)) {
 826                                 matched = true;
 827                                 nh = arg.nh;
 828                         }
 829                 } else {
 830                         nh = f6i->fib6_nh;
 831                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 832                                        mpri, do_rr))
 833                                 matched = true;
 834                 }
 835                 if (matched) {
 836                         res->f6i = f6i;
 837                         res->nh = nh;
 838                         res->fib6_flags = f6i->fib6_flags;
 839                         res->fib6_type = f6i->fib6_type;
 840                 }
 841         }
 842 }
 843
 844 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 845                          struct fib6_info *rr_head, int oif, int strict,
 846                          bool *do_rr, struct fib6_result *res)
 847 {
 848         u32 metric = rr_head->fib6_metric;
 849         struct fib6_info *cont = NULL;
 850         int mpri = -1;
 851
 852         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 853                        oif, strict, do_rr, &mpri);
 854
 855         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 856                        oif, strict, do_rr, &mpri);
 857
 858         if (res->f6i || !cont)
 859                 return;
 860
 861         __find_rr_leaf(cont, NULL, metric, res, NULL,
 862                        oif, strict, do_rr, &mpri);
 863 }
 864
 865 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 866                        struct fib6_result *res, int strict)
 867 {
 868         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 869         struct fib6_info *rt0;
 870         bool do_rr = false;
 871         int key_plen;
 872
 873         /* make sure this function or its helpers sets f6i */
 874         res->f6i = NULL;
 875
 876         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 877                 goto out;
 878
 879         rt0 = rcu_dereference(fn->rr_ptr);
 880         if (!rt0)
 881                 rt0 = leaf;
 882
 883         /* Double check to make sure fn is not an intermediate node
 884          * and fn->leaf does not points to its child's leaf
 885          * (This might happen if all routes under fn are deleted from
 886          * the tree and fib6_repair_tree() is called on the node.)
 887          */
 888         key_plen = rt0->fib6_dst.plen;
 889 #ifdef CONFIG_IPV6_SUBTREES
 890         if (rt0->fib6_src.plen)
 891                 key_plen = rt0->fib6_src.plen;
 892 #endif
 893         if (fn->fn_bit != key_plen)
 894                 goto out;
 895
 896         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 897         if (do_rr) {
 898                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 899
 900                 /* no entries matched; do round-robin */
 901                 if (!next || next->fib6_metric != rt0->fib6_metric)
 902                         next = leaf;
 903
 904                 if (next != rt0) {
 905                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 906                         /* make sure next is not being deleted from the tree */
 907                         if (next->fib6_node)
 908                                 rcu_assign_pointer(fn->rr_ptr, next);
 909                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 910                 }
 911         }
 912
 913 out:
 914         if (!res->f6i) {
 915                 res->f6i = net->ipv6.fib6_null_entry;
 916                 res->nh = res->f6i->fib6_nh;
 917                 res->fib6_flags = res->f6i->fib6_flags;
 918                 res->fib6_type = res->f6i->fib6_type;
 919         }
 920 }
 921
 922 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 923 {
 924         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 925                res->nh->fib_nh_gw_family;
 926 }
 927
 928 #ifdef CONFIG_IPV6_ROUTE_INFO
 929 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 930                   const struct in6_addr *gwaddr)
 931 {
 932         struct net *net = dev_net(dev);
 933         struct route_info *rinfo = (struct route_info *) opt;
 934         struct in6_addr prefix_buf, *prefix;
 935         unsigned int pref;
 936         unsigned long lifetime;
 937         struct fib6_info *rt;
 938
 939         if (len < sizeof(struct route_info)) {
 940                 return -EINVAL;
 941         }
 942
 943         /* Sanity check for prefix_len and length */
 944         if (rinfo->length > 3) {
 945                 return -EINVAL;
 946         } else if (rinfo->prefix_len > 128) {
 947                 return -EINVAL;
 948         } else if (rinfo->prefix_len > 64) {
 949                 if (rinfo->length < 2) {
 950                         return -EINVAL;
 951                 }
 952         } else if (rinfo->prefix_len > 0) {
 953                 if (rinfo->length < 1) {
 954                         return -EINVAL;
 955                 }
 956         }
 957
 958         pref = rinfo->route_pref;
 959         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 960                 return -EINVAL;
 961
 962         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 963
 964         if (rinfo->length == 3)
 965                 prefix = (struct in6_addr *)rinfo->prefix;
 966         else {
 967                 /* this function is safe */
 968                 ipv6_addr_prefix(&prefix_buf,
 969                                  (struct in6_addr *)rinfo->prefix,
 970                                  rinfo->prefix_len);
 971                 prefix = &prefix_buf;
 972         }
 973
 974         if (rinfo->prefix_len == 0)
 975                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 976         else
 977                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 978                                         gwaddr, dev);
 979
 980         if (rt && !lifetime) {
 981                 ip6_del_rt(net, rt);
 982                 rt = NULL;
 983         }
 984
 985         if (!rt && lifetime)
 986                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 987                                         dev, pref);
 988         else if (rt)
 989                 rt->fib6_flags = RTF_ROUTEINFO |
 990                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 991
 992         if (rt) {
 993                 if (!addrconf_finite_timeout(lifetime))
 994                         fib6_clean_expires(rt);
 995                 else
 996                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 997
 998                 fib6_info_release(rt);
 999         }
1000         return 0;
1001 }
1002 #endif
1003
1004 /*
1005  *      Misc support functions
1006  */
1007
1008 /* called with rcu_lock held */
1009 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1010 {
1011         struct net_device *dev = res->nh->fib_nh_dev;
1012
1013         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1014                 /* for copies of local routes, dst->dev needs to be the
1015                  * device if it is a master device, the master device if
1016                  * device is enslaved, and the loopback as the default
1017                  */
1018                 if (netif_is_l3_slave(dev) &&
1019                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1020                         dev = l3mdev_master_dev_rcu(dev);
1021                 else if (!netif_is_l3_master(dev))
1022                         dev = dev_net(dev)->loopback_dev;
1023                 /* last case is netif_is_l3_master(dev) is true in which
1024                  * case we want dev returned to be dev
1025                  */
1026         }
1027
1028         return dev;
1029 }
1030
1031 static const int fib6_prop[RTN_MAX + 1] = {
1032         [RTN_UNSPEC]    = 0,
1033         [RTN_UNICAST]   = 0,
1034         [RTN_LOCAL]     = 0,
1035         [RTN_BROADCAST] = 0,
1036         [RTN_ANYCAST]   = 0,
1037         [RTN_MULTICAST] = 0,
1038         [RTN_BLACKHOLE] = -EINVAL,
1039         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1040         [RTN_PROHIBIT]  = -EACCES,
1041         [RTN_THROW]     = -EAGAIN,
1042         [RTN_NAT]       = -EINVAL,
1043         [RTN_XRESOLVE]  = -EINVAL,
1044 };
1045
1046 static int ip6_rt_type_to_error(u8 fib6_type)
1047 {
1048         return fib6_prop[fib6_type];
1049 }
1050
1051 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1052 {
1053         unsigned short flags = 0;
1054
1055         if (rt->dst_nocount)
1056                 flags |= DST_NOCOUNT;
1057         if (rt->dst_nopolicy)
1058                 flags |= DST_NOPOLICY;
1059         if (rt->dst_host)
1060                 flags |= DST_HOST;
1061
1062         return flags;
1063 }
1064
1065 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1066 {
1067         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1068
1069         switch (fib6_type) {
1070         case RTN_BLACKHOLE:
1071                 rt->dst.output = dst_discard_out;
1072                 rt->dst.input = dst_discard;
1073                 break;
1074         case RTN_PROHIBIT:
1075                 rt->dst.output = ip6_pkt_prohibit_out;
1076                 rt->dst.input = ip6_pkt_prohibit;
1077                 break;
1078         case RTN_THROW:
1079         case RTN_UNREACHABLE:
1080         default:
1081                 rt->dst.output = ip6_pkt_discard_out;
1082                 rt->dst.input = ip6_pkt_discard;
1083                 break;
1084         }
1085 }
1086
1087 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1088 {
1089         struct fib6_info *f6i = res->f6i;
1090
1091         if (res->fib6_flags & RTF_REJECT) {
1092                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1093                 return;
1094         }
1095
1096         rt->dst.error = 0;
1097         rt->dst.output = ip6_output;
1098
1099         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1100                 rt->dst.input = ip6_input;
1101         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1102                 rt->dst.input = ip6_mc_input;
1103         } else {
1104                 rt->dst.input = ip6_forward;
1105         }
1106
1107         if (res->nh->fib_nh_lws) {
1108                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1109                 lwtunnel_set_redirect(&rt->dst);
1110         }
1111
1112         rt->dst.lastuse = jiffies;
1113 }
1114
1115 /* Caller must already hold reference to @from */
1116 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1117 {
1118         rt->rt6i_flags &= ~RTF_EXPIRES;
1119         rcu_assign_pointer(rt->from, from);
1120         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1121 }
1122
1123 /* Caller must already hold reference to f6i in result */
1124 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1125 {
1126         const struct fib6_nh *nh = res->nh;
1127         const struct net_device *dev = nh->fib_nh_dev;
1128         struct fib6_info *f6i = res->f6i;
1129
1130         ip6_rt_init_dst(rt, res);
1131
1132         rt->rt6i_dst = f6i->fib6_dst;
1133         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1134         rt->rt6i_flags = res->fib6_flags;
1135         if (nh->fib_nh_gw_family) {
1136                 rt->rt6i_gateway = nh->fib_nh_gw6;
1137                 rt->rt6i_flags |= RTF_GATEWAY;
1138         }
1139         rt6_set_from(rt, f6i);
1140 #ifdef CONFIG_IPV6_SUBTREES
1141         rt->rt6i_src = f6i->fib6_src;
1142 #endif
1143 }
1144
1145 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1146                                         struct in6_addr *saddr)
1147 {
1148         struct fib6_node *pn, *sn;
1149         while (1) {
1150                 if (fn->fn_flags & RTN_TL_ROOT)
1151                         return NULL;
1152                 pn = rcu_dereference(fn->parent);
1153                 sn = FIB6_SUBTREE(pn);
1154                 if (sn && sn != fn)
1155                         fn = fib6_node_lookup(sn, NULL, saddr);
1156                 else
1157                         fn = pn;
1158                 if (fn->fn_flags & RTN_RTINFO)
1159                         return fn;
1160         }
1161 }
1162
1163 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1164 {
1165         struct rt6_info *rt = *prt;
1166
1167         if (dst_hold_safe(&rt->dst))
1168                 return true;
1169         if (net) {
1170                 rt = net->ipv6.ip6_null_entry;
1171                 dst_hold(&rt->dst);
1172         } else {
1173                 rt = NULL;
1174         }
1175         *prt = rt;
1176         return false;
1177 }
1178
1179 /* called with rcu_lock held */
1180 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1181 {
1182         struct net_device *dev = res->nh->fib_nh_dev;
1183         struct fib6_info *f6i = res->f6i;
1184         unsigned short flags;
1185         struct rt6_info *nrt;
1186
1187         if (!fib6_info_hold_safe(f6i))
1188                 goto fallback;
1189
1190         flags = fib6_info_dst_flags(f6i);
1191         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1192         if (!nrt) {
1193                 fib6_info_release(f6i);
1194                 goto fallback;
1195         }
1196
1197         ip6_rt_copy_init(nrt, res);
1198         return nrt;
1199
1200 fallback:
1201         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1202         dst_hold(&nrt->dst);
1203         return nrt;
1204 }
1205
1206 static struct rt6_info *ip6_pol_route_lookup(struct net *net,
1207                                              struct fib6_table *table,
1208                                              struct flowi6 *fl6,
1209                                              const struct sk_buff *skb,
1210                                              int flags)
1211 {
1212         struct fib6_result res = {};
1213         struct fib6_node *fn;
1214         struct rt6_info *rt;
1215
1216         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1217                 flags &= ~RT6_LOOKUP_F_IFACE;
1218
1219         rcu_read_lock();
1220         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1221 restart:
1222         res.f6i = rcu_dereference(fn->leaf);
1223         if (!res.f6i)
1224                 res.f6i = net->ipv6.fib6_null_entry;
1225         else
1226                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1227                                  flags);
1228
1229         if (res.f6i == net->ipv6.fib6_null_entry) {
1230                 fn = fib6_backtrack(fn, &fl6->saddr);
1231                 if (fn)
1232                         goto restart;
1233
1234                 rt = net->ipv6.ip6_null_entry;
1235                 dst_hold(&rt->dst);
1236                 goto out;
1237         } else if (res.fib6_flags & RTF_REJECT) {
1238                 goto do_create;
1239         }
1240
1241         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1242                          fl6->flowi6_oif != 0, skb, flags);
1243
1244         /* Search through exception table */
1245         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1246         if (rt) {
1247                 if (ip6_hold_safe(net, &rt))
1248                         dst_use_noref(&rt->dst, jiffies);
1249         } else {
1250 do_create:
1251                 rt = ip6_create_rt_rcu(&res);
1252         }
1253
1254 out:
1255         trace_fib6_table_lookup(net, &res, table, fl6);
1256
1257         rcu_read_unlock();
1258
1259         return rt;
1260 }
1261
1262 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1263                                    const struct sk_buff *skb, int flags)
1264 {
1265         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1266 }
1267 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1268
1269 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1270                             const struct in6_addr *saddr, int oif,
1271                             const struct sk_buff *skb, int strict)
1272 {
1273         struct flowi6 fl6 = {
1274                 .flowi6_oif = oif,
1275                 .daddr = *daddr,
1276         };
1277         struct dst_entry *dst;
1278         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1279
1280         if (saddr) {
1281                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1282                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1283         }
1284
1285         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1286         if (dst->error == 0)
1287                 return (struct rt6_info *) dst;
1288
1289         dst_release(dst);
1290
1291         return NULL;
1292 }
1293 EXPORT_SYMBOL(rt6_lookup);
1294
1295 /* ip6_ins_rt is called with FREE table->tb6_lock.
1296  * It takes new route entry, the addition fails by any reason the
1297  * route is released.
1298  * Caller must hold dst before calling it.
1299  */
1300
1301 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1302                         struct netlink_ext_ack *extack)
1303 {
1304         int err;
1305         struct fib6_table *table;
1306
1307         table = rt->fib6_table;
1308         spin_lock_bh(&table->tb6_lock);
1309         err = fib6_add(&table->tb6_root, rt, info, extack);
1310         spin_unlock_bh(&table->tb6_lock);
1311
1312         return err;
1313 }
1314
1315 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1316 {
1317         struct nl_info info = { .nl_net = net, };
1318
1319         return __ip6_ins_rt(rt, &info, NULL);
1320 }
1321
1322 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1323                                            const struct in6_addr *daddr,
1324                                            const struct in6_addr *saddr)
1325 {
1326         struct fib6_info *f6i = res->f6i;
1327         struct net_device *dev;
1328         struct rt6_info *rt;
1329
1330         /*
1331          *      Clone the route.
1332          */
1333
1334         if (!fib6_info_hold_safe(f6i))
1335                 return NULL;
1336
1337         dev = ip6_rt_get_dev_rcu(res);
1338         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1339         if (!rt) {
1340                 fib6_info_release(f6i);
1341                 return NULL;
1342         }
1343
1344         ip6_rt_copy_init(rt, res);
1345         rt->rt6i_flags |= RTF_CACHE;
1346         rt->dst.flags |= DST_HOST;
1347         rt->rt6i_dst.addr = *daddr;
1348         rt->rt6i_dst.plen = 128;
1349
1350         if (!rt6_is_gw_or_nonexthop(res)) {
1351                 if (f6i->fib6_dst.plen != 128 &&
1352                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1353                         rt->rt6i_flags |= RTF_ANYCAST;
1354 #ifdef CONFIG_IPV6_SUBTREES
1355                 if (rt->rt6i_src.plen && saddr) {
1356                         rt->rt6i_src.addr = *saddr;
1357                         rt->rt6i_src.plen = 128;
1358                 }
1359 #endif
1360         }
1361
1362         return rt;
1363 }
1364
1365 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1366 {
1367         struct fib6_info *f6i = res->f6i;
1368         unsigned short flags = fib6_info_dst_flags(f6i);
1369         struct net_device *dev;
1370         struct rt6_info *pcpu_rt;
1371
1372         if (!fib6_info_hold_safe(f6i))
1373                 return NULL;
1374
1375         rcu_read_lock();
1376         dev = ip6_rt_get_dev_rcu(res);
1377         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags);
1378         rcu_read_unlock();
1379         if (!pcpu_rt) {
1380                 fib6_info_release(f6i);
1381                 return NULL;
1382         }
1383         ip6_rt_copy_init(pcpu_rt, res);
1384         pcpu_rt->rt6i_flags |= RTF_PCPU;
1385         return pcpu_rt;
1386 }
1387
1388 /* It should be called with rcu_read_lock() acquired */
1389 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1390 {
1391         struct rt6_info *pcpu_rt;
1392
1393         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1394
1395         return pcpu_rt;
1396 }
1397
1398 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1399                                             const struct fib6_result *res)
1400 {
1401         struct rt6_info *pcpu_rt, *prev, **p;
1402
1403         pcpu_rt = ip6_rt_pcpu_alloc(res);
1404         if (!pcpu_rt)
1405                 return NULL;
1406
1407         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1408         prev = cmpxchg(p, NULL, pcpu_rt);
1409         BUG_ON(prev);
1410
1411         if (res->f6i->fib6_destroying) {
1412                 struct fib6_info *from;
1413
1414                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1415                 fib6_info_release(from);
1416         }
1417
1418         return pcpu_rt;
1419 }
1420
1421 /* exception hash table implementation
1422  */
1423 static DEFINE_SPINLOCK(rt6_exception_lock);
1424
1425 /* Remove rt6_ex from hash table and free the memory
1426  * Caller must hold rt6_exception_lock
1427  */
1428 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1429                                  struct rt6_exception *rt6_ex)
1430 {
1431         struct fib6_info *from;
1432         struct net *net;
1433
1434         if (!bucket || !rt6_ex)
1435                 return;
1436
1437         net = dev_net(rt6_ex->rt6i->dst.dev);
1438         net->ipv6.rt6_stats->fib_rt_cache--;
1439
1440         /* purge completely the exception to allow releasing the held resources:
1441          * some [sk] cache may keep the dst around for unlimited time
1442          */
1443         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1444         fib6_info_release(from);
1445         dst_dev_put(&rt6_ex->rt6i->dst);
1446
1447         hlist_del_rcu(&rt6_ex->hlist);
1448         dst_release(&rt6_ex->rt6i->dst);
1449         kfree_rcu(rt6_ex, rcu);
1450         WARN_ON_ONCE(!bucket->depth);
1451         bucket->depth--;
1452 }
1453
1454 /* Remove oldest rt6_ex in bucket and free the memory
1455  * Caller must hold rt6_exception_lock
1456  */
1457 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1458 {
1459         struct rt6_exception *rt6_ex, *oldest = NULL;
1460
1461         if (!bucket)
1462                 return;
1463
1464         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1465                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1466                         oldest = rt6_ex;
1467         }
1468         rt6_remove_exception(bucket, oldest);
1469 }
1470
1471 static u32 rt6_exception_hash(const struct in6_addr *dst,
1472                               const struct in6_addr *src)
1473 {
1474         static u32 seed __read_mostly;
1475         u32 val;
1476
1477         net_get_random_once(&seed, sizeof(seed));
1478         val = jhash2((const u32 *)dst, sizeof(*dst)/sizeof(u32), seed);
1479
1480 #ifdef CONFIG_IPV6_SUBTREES
1481         if (src)
1482                 val = jhash2((const u32 *)src, sizeof(*src)/sizeof(u32), val);
1483 #endif
1484         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1485 }
1486
1487 /* Helper function to find the cached rt in the hash table
1488  * and update bucket pointer to point to the bucket for this
1489  * (daddr, saddr) pair
1490  * Caller must hold rt6_exception_lock
1491  */
1492 static struct rt6_exception *
1493 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1494                               const struct in6_addr *daddr,
1495                               const struct in6_addr *saddr)
1496 {
1497         struct rt6_exception *rt6_ex;
1498         u32 hval;
1499
1500         if (!(*bucket) || !daddr)
1501                 return NULL;
1502
1503         hval = rt6_exception_hash(daddr, saddr);
1504         *bucket += hval;
1505
1506         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1507                 struct rt6_info *rt6 = rt6_ex->rt6i;
1508                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1509
1510 #ifdef CONFIG_IPV6_SUBTREES
1511                 if (matched && saddr)
1512                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1513 #endif
1514                 if (matched)
1515                         return rt6_ex;
1516         }
1517         return NULL;
1518 }
1519
1520 /* Helper function to find the cached rt in the hash table
1521  * and update bucket pointer to point to the bucket for this
1522  * (daddr, saddr) pair
1523  * Caller must hold rcu_read_lock()
1524  */
1525 static struct rt6_exception *
1526 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1527                          const struct in6_addr *daddr,
1528                          const struct in6_addr *saddr)
1529 {
1530         struct rt6_exception *rt6_ex;
1531         u32 hval;
1532
1533         WARN_ON_ONCE(!rcu_read_lock_held());
1534
1535         if (!(*bucket) || !daddr)
1536                 return NULL;
1537
1538         hval = rt6_exception_hash(daddr, saddr);
1539         *bucket += hval;
1540
1541         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1542                 struct rt6_info *rt6 = rt6_ex->rt6i;
1543                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1544
1545 #ifdef CONFIG_IPV6_SUBTREES
1546                 if (matched && saddr)
1547                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1548 #endif
1549                 if (matched)
1550                         return rt6_ex;
1551         }
1552         return NULL;
1553 }
1554
1555 static unsigned int fib6_mtu(const struct fib6_result *res)
1556 {
1557         const struct fib6_nh *nh = res->nh;
1558         unsigned int mtu;
1559
1560         if (res->f6i->fib6_pmtu) {
1561                 mtu = res->f6i->fib6_pmtu;
1562         } else {
1563                 struct net_device *dev = nh->fib_nh_dev;
1564                 struct inet6_dev *idev;
1565
1566                 rcu_read_lock();
1567                 idev = __in6_dev_get(dev);
1568                 mtu = idev->cnf.mtu6;
1569                 rcu_read_unlock();
1570         }
1571
1572         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1573
1574         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1575 }
1576
1577 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1578
1579 /* used when the flushed bit is not relevant, only access to the bucket
1580  * (ie., all bucket users except rt6_insert_exception);
1581  *
1582  * called under rcu lock; sometimes called with rt6_exception_lock held
1583  */
1584 static
1585 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1586                                                        spinlock_t *lock)
1587 {
1588         struct rt6_exception_bucket *bucket;
1589
1590         if (lock)
1591                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1592                                                    lockdep_is_held(lock));
1593         else
1594                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1595
1596         /* remove bucket flushed bit if set */
1597         if (bucket) {
1598                 unsigned long p = (unsigned long)bucket;
1599
1600                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1601                 bucket = (struct rt6_exception_bucket *)p;
1602         }
1603
1604         return bucket;
1605 }
1606
1607 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1608 {
1609         unsigned long p = (unsigned long)bucket;
1610
1611         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1612 }
1613
1614 /* called with rt6_exception_lock held */
1615 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1616                                               spinlock_t *lock)
1617 {
1618         struct rt6_exception_bucket *bucket;
1619         unsigned long p;
1620
1621         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1622                                            lockdep_is_held(lock));
1623
1624         p = (unsigned long)bucket;
1625         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1626         bucket = (struct rt6_exception_bucket *)p;
1627         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1628 }
1629
1630 static int rt6_insert_exception(struct rt6_info *nrt,
1631                                 const struct fib6_result *res)
1632 {
1633         struct net *net = dev_net(nrt->dst.dev);
1634         struct rt6_exception_bucket *bucket;
1635         struct fib6_info *f6i = res->f6i;
1636         struct in6_addr *src_key = NULL;
1637         struct rt6_exception *rt6_ex;
1638         struct fib6_nh *nh = res->nh;
1639         int err = 0;
1640
1641         spin_lock_bh(&rt6_exception_lock);
1642
1643         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1644                                           lockdep_is_held(&rt6_exception_lock));
1645         if (!bucket) {
1646                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1647                                  GFP_ATOMIC);
1648                 if (!bucket) {
1649                         err = -ENOMEM;
1650                         goto out;
1651                 }
1652                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1653         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1654                 err = -EINVAL;
1655                 goto out;
1656         }
1657
1658 #ifdef CONFIG_IPV6_SUBTREES
1659         /* fib6_src.plen != 0 indicates f6i is in subtree
1660          * and exception table is indexed by a hash of
1661          * both fib6_dst and fib6_src.
1662          * Otherwise, the exception table is indexed by
1663          * a hash of only fib6_dst.
1664          */
1665         if (f6i->fib6_src.plen)
1666                 src_key = &nrt->rt6i_src.addr;
1667 #endif
1668         /* rt6_mtu_change() might lower mtu on f6i.
1669          * Only insert this exception route if its mtu
1670          * is less than f6i's mtu value.
1671          */
1672         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1673                 err = -EINVAL;
1674                 goto out;
1675         }
1676
1677         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1678                                                src_key);
1679         if (rt6_ex)
1680                 rt6_remove_exception(bucket, rt6_ex);
1681
1682         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1683         if (!rt6_ex) {
1684                 err = -ENOMEM;
1685                 goto out;
1686         }
1687         rt6_ex->rt6i = nrt;
1688         rt6_ex->stamp = jiffies;
1689         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1690         bucket->depth++;
1691         net->ipv6.rt6_stats->fib_rt_cache++;
1692
1693         if (bucket->depth > FIB6_MAX_DEPTH)
1694                 rt6_exception_remove_oldest(bucket);
1695
1696 out:
1697         spin_unlock_bh(&rt6_exception_lock);
1698
1699         /* Update fn->fn_sernum to invalidate all cached dst */
1700         if (!err) {
1701                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1702                 fib6_update_sernum(net, f6i);
1703                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1704                 fib6_force_start_gc(net);
1705         }
1706
1707         return err;
1708 }
1709
1710 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1711 {
1712         struct rt6_exception_bucket *bucket;
1713         struct rt6_exception *rt6_ex;
1714         struct hlist_node *tmp;
1715         int i;
1716
1717         spin_lock_bh(&rt6_exception_lock);
1718
1719         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1720         if (!bucket)
1721                 goto out;
1722
1723         /* Prevent rt6_insert_exception() to recreate the bucket list */
1724         if (!from)
1725                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1726
1727         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1728                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1729                         if (!from ||
1730                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1731                                 rt6_remove_exception(bucket, rt6_ex);
1732                 }
1733                 WARN_ON_ONCE(!from && bucket->depth);
1734                 bucket++;
1735         }
1736 out:
1737         spin_unlock_bh(&rt6_exception_lock);
1738 }
1739
1740 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1741 {
1742         struct fib6_info *f6i = arg;
1743
1744         fib6_nh_flush_exceptions(nh, f6i);
1745
1746         return 0;
1747 }
1748
1749 void rt6_flush_exceptions(struct fib6_info *f6i)
1750 {
1751         if (f6i->nh)
1752                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1753                                          f6i);
1754         else
1755                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1756 }
1757
1758 /* Find cached rt in the hash table inside passed in rt
1759  * Caller has to hold rcu_read_lock()
1760  */
1761 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1762                                            const struct in6_addr *daddr,
1763                                            const struct in6_addr *saddr)
1764 {
1765         const struct in6_addr *src_key = NULL;
1766         struct rt6_exception_bucket *bucket;
1767         struct rt6_exception *rt6_ex;
1768         struct rt6_info *ret = NULL;
1769
1770 #ifdef CONFIG_IPV6_SUBTREES
1771         /* fib6i_src.plen != 0 indicates f6i is in subtree
1772          * and exception table is indexed by a hash of
1773          * both fib6_dst and fib6_src.
1774          * However, the src addr used to create the hash
1775          * might not be exactly the passed in saddr which
1776          * is a /128 addr from the flow.
1777          * So we need to use f6i->fib6_src to redo lookup
1778          * if the passed in saddr does not find anything.
1779          * (See the logic in ip6_rt_cache_alloc() on how
1780          * rt->rt6i_src is updated.)
1781          */
1782         if (res->f6i->fib6_src.plen)
1783                 src_key = saddr;
1784 find_ex:
1785 #endif
1786         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1787         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1788
1789         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1790                 ret = rt6_ex->rt6i;
1791
1792 #ifdef CONFIG_IPV6_SUBTREES
1793         /* Use fib6_src as src_key and redo lookup */
1794         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1795                 src_key = &res->f6i->fib6_src.addr;
1796                 goto find_ex;
1797         }
1798 #endif
1799
1800         return ret;
1801 }
1802
1803 /* Remove the passed in cached rt from the hash table that contains it */
1804 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1805                                     const struct rt6_info *rt)
1806 {
1807         const struct in6_addr *src_key = NULL;
1808         struct rt6_exception_bucket *bucket;
1809         struct rt6_exception *rt6_ex;
1810         int err;
1811
1812         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1813                 return -ENOENT;
1814
1815         spin_lock_bh(&rt6_exception_lock);
1816         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1817
1818 #ifdef CONFIG_IPV6_SUBTREES
1819         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1820          * and exception table is indexed by a hash of
1821          * both rt6i_dst and rt6i_src.
1822          * Otherwise, the exception table is indexed by
1823          * a hash of only rt6i_dst.
1824          */
1825         if (plen)
1826                 src_key = &rt->rt6i_src.addr;
1827 #endif
1828         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1829                                                &rt->rt6i_dst.addr,
1830                                                src_key);
1831         if (rt6_ex) {
1832                 rt6_remove_exception(bucket, rt6_ex);
1833                 err = 0;
1834         } else {
1835                 err = -ENOENT;
1836         }
1837
1838         spin_unlock_bh(&rt6_exception_lock);
1839         return err;
1840 }
1841
1842 struct fib6_nh_excptn_arg {
1843         struct rt6_info *rt;
1844         int             plen;
1845 };
1846
1847 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1848 {
1849         struct fib6_nh_excptn_arg *arg = _arg;
1850         int err;
1851
1852         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1853         if (err == 0)
1854                 return 1;
1855
1856         return 0;
1857 }
1858
1859 static int rt6_remove_exception_rt(struct rt6_info *rt)
1860 {
1861         struct fib6_info *from;
1862
1863         from = rcu_dereference(rt->from);
1864         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1865                 return -EINVAL;
1866
1867         if (from->nh) {
1868                 struct fib6_nh_excptn_arg arg = {
1869                         .rt = rt,
1870                         .plen = from->fib6_src.plen
1871                 };
1872                 int rc;
1873
1874                 /* rc = 1 means an entry was found */
1875                 rc = nexthop_for_each_fib6_nh(from->nh,
1876                                               rt6_nh_remove_exception_rt,
1877                                               &arg);
1878                 return rc ? 0 : -ENOENT;
1879         }
1880
1881         return fib6_nh_remove_exception(from->fib6_nh,
1882                                         from->fib6_src.plen, rt);
1883 }
1884
1885 /* Find rt6_ex which contains the passed in rt cache and
1886  * refresh its stamp
1887  */
1888 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1889                                      const struct rt6_info *rt)
1890 {
1891         const struct in6_addr *src_key = NULL;
1892         struct rt6_exception_bucket *bucket;
1893         struct rt6_exception *rt6_ex;
1894
1895         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1896 #ifdef CONFIG_IPV6_SUBTREES
1897         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1898          * and exception table is indexed by a hash of
1899          * both rt6i_dst and rt6i_src.
1900          * Otherwise, the exception table is indexed by
1901          * a hash of only rt6i_dst.
1902          */
1903         if (plen)
1904                 src_key = &rt->rt6i_src.addr;
1905 #endif
1906         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1907         if (rt6_ex)
1908                 rt6_ex->stamp = jiffies;
1909 }
1910
1911 struct fib6_nh_match_arg {
1912         const struct net_device *dev;
1913         const struct in6_addr   *gw;
1914         struct fib6_nh          *match;
1915 };
1916
1917 /* determine if fib6_nh has given device and gateway */
1918 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1919 {
1920         struct fib6_nh_match_arg *arg = _arg;
1921
1922         if (arg->dev != nh->fib_nh_dev ||
1923             (arg->gw && !nh->fib_nh_gw_family) ||
1924             (!arg->gw && nh->fib_nh_gw_family) ||
1925             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1926                 return 0;
1927
1928         arg->match = nh;
1929
1930         /* found a match, break the loop */
1931         return 1;
1932 }
1933
1934 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1935 {
1936         struct fib6_info *from;
1937         struct fib6_nh *fib6_nh;
1938
1939         rcu_read_lock();
1940
1941         from = rcu_dereference(rt->from);
1942         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1943                 goto unlock;
1944
1945         if (from->nh) {
1946                 struct fib6_nh_match_arg arg = {
1947                         .dev = rt->dst.dev,
1948                         .gw = &rt->rt6i_gateway,
1949                 };
1950
1951                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1952
1953                 if (!arg.match)
1954                         goto unlock;
1955                 fib6_nh = arg.match;
1956         } else {
1957                 fib6_nh = from->fib6_nh;
1958         }
1959         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1960 unlock:
1961         rcu_read_unlock();
1962 }
1963
1964 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1965                                          struct rt6_info *rt, int mtu)
1966 {
1967         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1968          * lowest MTU in the path: always allow updating the route PMTU to
1969          * reflect PMTU decreases.
1970          *
1971          * If the new MTU is higher, and the route PMTU is equal to the local
1972          * MTU, this means the old MTU is the lowest in the path, so allow
1973          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1974          * handle this.
1975          */
1976
1977         if (dst_mtu(&rt->dst) >= mtu)
1978                 return true;
1979
1980         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1981                 return true;
1982
1983         return false;
1984 }
1985
1986 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
1987                                        const struct fib6_nh *nh, int mtu)
1988 {
1989         struct rt6_exception_bucket *bucket;
1990         struct rt6_exception *rt6_ex;
1991         int i;
1992
1993         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1994         if (!bucket)
1995                 return;
1996
1997         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1998                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1999                         struct rt6_info *entry = rt6_ex->rt6i;
2000
2001                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2002                          * route), the metrics of its rt->from have already
2003                          * been updated.
2004                          */
2005                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2006                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2007                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2008                 }
2009                 bucket++;
2010         }
2011 }
2012
2013 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2014
2015 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2016                                             const struct in6_addr *gateway)
2017 {
2018         struct rt6_exception_bucket *bucket;
2019         struct rt6_exception *rt6_ex;
2020         struct hlist_node *tmp;
2021         int i;
2022
2023         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2024                 return;
2025
2026         spin_lock_bh(&rt6_exception_lock);
2027         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2028         if (bucket) {
2029                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2030                         hlist_for_each_entry_safe(rt6_ex, tmp,
2031                                                   &bucket->chain, hlist) {
2032                                 struct rt6_info *entry = rt6_ex->rt6i;
2033
2034                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2035                                     RTF_CACHE_GATEWAY &&
2036                                     ipv6_addr_equal(gateway,
2037                                                     &entry->rt6i_gateway)) {
2038                                         rt6_remove_exception(bucket, rt6_ex);
2039                                 }
2040                         }
2041                         bucket++;
2042                 }
2043         }
2044
2045         spin_unlock_bh(&rt6_exception_lock);
2046 }
2047
2048 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2049                                       struct rt6_exception *rt6_ex,
2050                                       struct fib6_gc_args *gc_args,
2051                                       unsigned long now)
2052 {
2053         struct rt6_info *rt = rt6_ex->rt6i;
2054
2055         /* we are pruning and obsoleting aged-out and non gateway exceptions
2056          * even if others have still references to them, so that on next
2057          * dst_check() such references can be dropped.
2058          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2059          * expired, independently from their aging, as per RFC 8201 section 4
2060          */
2061         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2062                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2063                         RT6_TRACE("aging clone %p\n", rt);
2064                         rt6_remove_exception(bucket, rt6_ex);
2065                         return;
2066                 }
2067         } else if (time_after(jiffies, rt->dst.expires)) {
2068                 RT6_TRACE("purging expired route %p\n", rt);
2069                 rt6_remove_exception(bucket, rt6_ex);
2070                 return;
2071         }
2072
2073         if (rt->rt6i_flags & RTF_GATEWAY) {
2074                 struct neighbour *neigh;
2075                 __u8 neigh_flags = 0;
2076
2077                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2078                 if (neigh)
2079                         neigh_flags = neigh->flags;
2080
2081                 if (!(neigh_flags & NTF_ROUTER)) {
2082                         RT6_TRACE("purging route %p via non-router but gateway\n",
2083                                   rt);
2084                         rt6_remove_exception(bucket, rt6_ex);
2085                         return;
2086                 }
2087         }
2088
2089         gc_args->more++;
2090 }
2091
2092 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2093                                    struct fib6_gc_args *gc_args,
2094                                    unsigned long now)
2095 {
2096         struct rt6_exception_bucket *bucket;
2097         struct rt6_exception *rt6_ex;
2098         struct hlist_node *tmp;
2099         int i;
2100
2101         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2102                 return;
2103
2104         rcu_read_lock_bh();
2105         spin_lock(&rt6_exception_lock);
2106         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2107         if (bucket) {
2108                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2109                         hlist_for_each_entry_safe(rt6_ex, tmp,
2110                                                   &bucket->chain, hlist) {
2111                                 rt6_age_examine_exception(bucket, rt6_ex,
2112                                                           gc_args, now);
2113                         }
2114                         bucket++;
2115                 }
2116         }
2117         spin_unlock(&rt6_exception_lock);
2118         rcu_read_unlock_bh();
2119 }
2120
2121 struct fib6_nh_age_excptn_arg {
2122         struct fib6_gc_args     *gc_args;
2123         unsigned long           now;
2124 };
2125
2126 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2127 {
2128         struct fib6_nh_age_excptn_arg *arg = _arg;
2129
2130         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2131         return 0;
2132 }
2133
2134 void rt6_age_exceptions(struct fib6_info *f6i,
2135                         struct fib6_gc_args *gc_args,
2136                         unsigned long now)
2137 {
2138         if (f6i->nh) {
2139                 struct fib6_nh_age_excptn_arg arg = {
2140                         .gc_args = gc_args,
2141                         .now = now
2142                 };
2143
2144                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2145                                          &arg);
2146         } else {
2147                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2148         }
2149 }
2150
2151 /* must be called with rcu lock held */
2152 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2153                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2154 {
2155         struct fib6_node *fn, *saved_fn;
2156
2157         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2158         saved_fn = fn;
2159
2160         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2161                 oif = 0;
2162
2163 redo_rt6_select:
2164         rt6_select(net, fn, oif, res, strict);
2165         if (res->f6i == net->ipv6.fib6_null_entry) {
2166                 fn = fib6_backtrack(fn, &fl6->saddr);
2167                 if (fn)
2168                         goto redo_rt6_select;
2169                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2170                         /* also consider unreachable route */
2171                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2172                         fn = saved_fn;
2173                         goto redo_rt6_select;
2174                 }
2175         }
2176
2177         trace_fib6_table_lookup(net, res, table, fl6);
2178
2179         return 0;
2180 }
2181
2182 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2183                                int oif, struct flowi6 *fl6,
2184                                const struct sk_buff *skb, int flags)
2185 {
2186         struct fib6_result res = {};
2187         struct rt6_info *rt = NULL;
2188         int strict = 0;
2189
2190         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2191                      !rcu_read_lock_held());
2192
2193         strict |= flags & RT6_LOOKUP_F_IFACE;
2194         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2195         if (net->ipv6.devconf_all->forwarding == 0)
2196                 strict |= RT6_LOOKUP_F_REACHABLE;
2197
2198         rcu_read_lock();
2199
2200         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2201         if (res.f6i == net->ipv6.fib6_null_entry)
2202                 goto out;
2203
2204         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2205
2206         /*Search through exception table */
2207         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2208         if (rt) {
2209                 goto out;
2210         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2211                             !res.nh->fib_nh_gw_family)) {
2212                 /* Create a RTF_CACHE clone which will not be
2213                  * owned by the fib6 tree.  It is for the special case where
2214                  * the daddr in the skb during the neighbor look-up is different
2215                  * from the fl6->daddr used to look-up route here.
2216                  */
2217                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2218
2219                 if (rt) {
2220                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2221                          * As rt6_uncached_list_add() does not consume refcnt,
2222                          * this refcnt is always returned to the caller even
2223                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2224                          */
2225                         rt6_uncached_list_add(rt);
2226                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2227                         rcu_read_unlock();
2228
2229                         return rt;
2230                 }
2231         } else {
2232                 /* Get a percpu copy */
2233                 local_bh_disable();
2234                 rt = rt6_get_pcpu_route(&res);
2235
2236                 if (!rt)
2237                         rt = rt6_make_pcpu_route(net, &res);
2238
2239                 local_bh_enable();
2240         }
2241 out:
2242         if (!rt)
2243                 rt = net->ipv6.ip6_null_entry;
2244         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2245                 ip6_hold_safe(net, &rt);
2246         rcu_read_unlock();
2247
2248         return rt;
2249 }
2250 EXPORT_SYMBOL_GPL(ip6_pol_route);
2251
2252 static struct rt6_info *ip6_pol_route_input(struct net *net,
2253                                             struct fib6_table *table,
2254                                             struct flowi6 *fl6,
2255                                             const struct sk_buff *skb,
2256                                             int flags)
2257 {
2258         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2259 }
2260
2261 struct dst_entry *ip6_route_input_lookup(struct net *net,
2262                                          struct net_device *dev,
2263                                          struct flowi6 *fl6,
2264                                          const struct sk_buff *skb,
2265                                          int flags)
2266 {
2267         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2268                 flags |= RT6_LOOKUP_F_IFACE;
2269
2270         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2271 }
2272 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2273
2274 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2275                                   struct flow_keys *keys,
2276                                   struct flow_keys *flkeys)
2277 {
2278         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2279         const struct ipv6hdr *key_iph = outer_iph;
2280         struct flow_keys *_flkeys = flkeys;
2281         const struct ipv6hdr *inner_iph;
2282         const struct icmp6hdr *icmph;
2283         struct ipv6hdr _inner_iph;
2284         struct icmp6hdr _icmph;
2285
2286         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2287                 goto out;
2288
2289         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2290                                    sizeof(_icmph), &_icmph);
2291         if (!icmph)
2292                 goto out;
2293
2294         if (!icmpv6_is_err(icmph->icmp6_type))
2295                 goto out;
2296
2297         inner_iph = skb_header_pointer(skb,
2298                                        skb_transport_offset(skb) + sizeof(*icmph),
2299                                        sizeof(_inner_iph), &_inner_iph);
2300         if (!inner_iph)
2301                 goto out;
2302
2303         key_iph = inner_iph;
2304         _flkeys = NULL;
2305 out:
2306         if (_flkeys) {
2307                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2308                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2309                 keys->tags.flow_label = _flkeys->tags.flow_label;
2310                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2311         } else {
2312                 keys->addrs.v6addrs.src = key_iph->saddr;
2313                 keys->addrs.v6addrs.dst = key_iph->daddr;
2314                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2315                 keys->basic.ip_proto = key_iph->nexthdr;
2316         }
2317 }
2318
2319 /* if skb is set it will be used and fl6 can be NULL */
2320 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2321                        const struct sk_buff *skb, struct flow_keys *flkeys)
2322 {
2323         struct flow_keys hash_keys;
2324         u32 mhash;
2325
2326         switch (ip6_multipath_hash_policy(net)) {
2327         case 0:
2328                 memset(&hash_keys, 0, sizeof(hash_keys));
2329                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2330                 if (skb) {
2331                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2332                 } else {
2333                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2334                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2335                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2336                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2337                 }
2338                 break;
2339         case 1:
2340                 if (skb) {
2341                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2342                         struct flow_keys keys;
2343
2344                         /* short-circuit if we already have L4 hash present */
2345                         if (skb->l4_hash)
2346                                 return skb_get_hash_raw(skb) >> 1;
2347
2348                         memset(&hash_keys, 0, sizeof(hash_keys));
2349
2350                         if (!flkeys) {
2351                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2352                                 flkeys = &keys;
2353                         }
2354                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2355                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2356                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2357                         hash_keys.ports.src = flkeys->ports.src;
2358                         hash_keys.ports.dst = flkeys->ports.dst;
2359                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2360                 } else {
2361                         memset(&hash_keys, 0, sizeof(hash_keys));
2362                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2363                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2364                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2365                         hash_keys.ports.src = fl6->fl6_sport;
2366                         hash_keys.ports.dst = fl6->fl6_dport;
2367                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2368                 }
2369                 break;
2370         case 2:
2371                 memset(&hash_keys, 0, sizeof(hash_keys));
2372                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2373                 if (skb) {
2374                         struct flow_keys keys;
2375
2376                         if (!flkeys) {
2377                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2378                                 flkeys = &keys;
2379                         }
2380
2381                         /* Inner can be v4 or v6 */
2382                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2383                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2384                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2385                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2386                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2387                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2388                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2389                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2390                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2391                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2392                         } else {
2393                                 /* Same as case 0 */
2394                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2395                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2396                         }
2397                 } else {
2398                         /* Same as case 0 */
2399                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2400                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2401                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2402                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2403                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2404                 }
2405                 break;
2406         }
2407         mhash = flow_hash_from_keys(&hash_keys);
2408
2409         return mhash >> 1;
2410 }
2411
2412 /* Called with rcu held */
2413 void ip6_route_input(struct sk_buff *skb)
2414 {
2415         const struct ipv6hdr *iph = ipv6_hdr(skb);
2416         struct net *net = dev_net(skb->dev);
2417         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2418         struct ip_tunnel_info *tun_info;
2419         struct flowi6 fl6 = {
2420                 .flowi6_iif = skb->dev->ifindex,
2421                 .daddr = iph->daddr,
2422                 .saddr = iph->saddr,
2423                 .flowlabel = ip6_flowinfo(iph),
2424                 .flowi6_mark = skb->mark,
2425                 .flowi6_proto = iph->nexthdr,
2426         };
2427         struct flow_keys *flkeys = NULL, _flkeys;
2428
2429         tun_info = skb_tunnel_info(skb);
2430         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2431                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2432
2433         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2434                 flkeys = &_flkeys;
2435
2436         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2437                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2438         skb_dst_drop(skb);
2439         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2440                                                       &fl6, skb, flags));
2441 }
2442
2443 static struct rt6_info *ip6_pol_route_output(struct net *net,
2444                                              struct fib6_table *table,
2445                                              struct flowi6 *fl6,
2446                                              const struct sk_buff *skb,
2447                                              int flags)
2448 {
2449         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2450 }
2451
2452 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2453                                                const struct sock *sk,
2454                                                struct flowi6 *fl6, int flags)
2455 {
2456         bool any_src;
2457
2458         if (ipv6_addr_type(&fl6->daddr) &
2459             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2460                 struct dst_entry *dst;
2461
2462                 /* This function does not take refcnt on the dst */
2463                 dst = l3mdev_link_scope_lookup(net, fl6);
2464                 if (dst)
2465                         return dst;
2466         }
2467
2468         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2469
2470         flags |= RT6_LOOKUP_F_DST_NOREF;
2471         any_src = ipv6_addr_any(&fl6->saddr);
2472         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2473             (fl6->flowi6_oif && any_src))
2474                 flags |= RT6_LOOKUP_F_IFACE;
2475
2476         if (!any_src)
2477                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2478         else if (sk)
2479                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2480
2481         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2482 }
2483 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2484
2485 struct dst_entry *ip6_route_output_flags(struct net *net,
2486                                          const struct sock *sk,
2487                                          struct flowi6 *fl6,
2488                                          int flags)
2489 {
2490         struct dst_entry *dst;
2491         struct rt6_info *rt6;
2492
2493         rcu_read_lock();
2494         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2495         rt6 = (struct rt6_info *)dst;
2496         /* For dst cached in uncached_list, refcnt is already taken. */
2497         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2498                 dst = &net->ipv6.ip6_null_entry->dst;
2499                 dst_hold(dst);
2500         }
2501         rcu_read_unlock();
2502
2503         return dst;
2504 }
2505 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2506
2507 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2508 {
2509         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2510         struct net_device *loopback_dev = net->loopback_dev;
2511         struct dst_entry *new = NULL;
2512
2513         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2514                        DST_OBSOLETE_DEAD, 0);
2515         if (rt) {
2516                 rt6_info_init(rt);
2517                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2518
2519                 new = &rt->dst;
2520                 new->__use = 1;
2521                 new->input = dst_discard;
2522                 new->output = dst_discard_out;
2523
2524                 dst_copy_metrics(new, &ort->dst);
2525
2526                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2527                 rt->rt6i_gateway = ort->rt6i_gateway;
2528                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2529
2530                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2531 #ifdef CONFIG_IPV6_SUBTREES
2532                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2533 #endif
2534         }
2535
2536         dst_release(dst_orig);
2537         return new ? new : ERR_PTR(-ENOMEM);
2538 }
2539
2540 /*
2541  *      Destination cache support functions
2542  */
2543
2544 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2545 {
2546         u32 rt_cookie = 0;
2547
2548         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2549                 return false;
2550
2551         if (fib6_check_expired(f6i))
2552                 return false;
2553
2554         return true;
2555 }
2556
2557 static struct dst_entry *rt6_check(struct rt6_info *rt,
2558                                    struct fib6_info *from,
2559                                    u32 cookie)
2560 {
2561         u32 rt_cookie = 0;
2562
2563         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2564             rt_cookie != cookie)
2565                 return NULL;
2566
2567         if (rt6_check_expired(rt))
2568                 return NULL;
2569
2570         return &rt->dst;
2571 }
2572
2573 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2574                                             struct fib6_info *from,
2575                                             u32 cookie)
2576 {
2577         if (!__rt6_check_expired(rt) &&
2578             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2579             fib6_check(from, cookie))
2580                 return &rt->dst;
2581         else
2582                 return NULL;
2583 }
2584
2585 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2586 {
2587         struct dst_entry *dst_ret;
2588         struct fib6_info *from;
2589         struct rt6_info *rt;
2590
2591         rt = container_of(dst, struct rt6_info, dst);
2592
2593         rcu_read_lock();
2594
2595         /* All IPV6 dsts are created with ->obsolete set to the value
2596          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2597          * into this function always.
2598          */
2599
2600         from = rcu_dereference(rt->from);
2601
2602         if (from && (rt->rt6i_flags & RTF_PCPU ||
2603             unlikely(!list_empty(&rt->rt6i_uncached))))
2604                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2605         else
2606                 dst_ret = rt6_check(rt, from, cookie);
2607
2608         rcu_read_unlock();
2609
2610         return dst_ret;
2611 }
2612
2613 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2614 {
2615         struct rt6_info *rt = (struct rt6_info *) dst;
2616
2617         if (rt) {
2618                 if (rt->rt6i_flags & RTF_CACHE) {
2619                         rcu_read_lock();
2620                         if (rt6_check_expired(rt)) {
2621                                 rt6_remove_exception_rt(rt);
2622                                 dst = NULL;
2623                         }
2624                         rcu_read_unlock();
2625                 } else {
2626                         dst_release(dst);
2627                         dst = NULL;
2628                 }
2629         }
2630         return dst;
2631 }
2632
2633 static void ip6_link_failure(struct sk_buff *skb)
2634 {
2635         struct rt6_info *rt;
2636
2637         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2638
2639         rt = (struct rt6_info *) skb_dst(skb);
2640         if (rt) {
2641                 rcu_read_lock();
2642                 if (rt->rt6i_flags & RTF_CACHE) {
2643                         rt6_remove_exception_rt(rt);
2644                 } else {
2645                         struct fib6_info *from;
2646                         struct fib6_node *fn;
2647
2648                         from = rcu_dereference(rt->from);
2649                         if (from) {
2650                                 fn = rcu_dereference(from->fib6_node);
2651                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2652                                         fn->fn_sernum = -1;
2653                         }
2654                 }
2655                 rcu_read_unlock();
2656         }
2657 }
2658
2659 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2660 {
2661         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2662                 struct fib6_info *from;
2663
2664                 rcu_read_lock();
2665                 from = rcu_dereference(rt0->from);
2666                 if (from)
2667                         rt0->dst.expires = from->expires;
2668                 rcu_read_unlock();
2669         }
2670
2671         dst_set_expires(&rt0->dst, timeout);
2672         rt0->rt6i_flags |= RTF_EXPIRES;
2673 }
2674
2675 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2676 {
2677         struct net *net = dev_net(rt->dst.dev);
2678
2679         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2680         rt->rt6i_flags |= RTF_MODIFIED;
2681         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2682 }
2683
2684 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2685 {
2686         return !(rt->rt6i_flags & RTF_CACHE) &&
2687                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2688 }
2689
2690 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2691                                  const struct ipv6hdr *iph, u32 mtu)
2692 {
2693         const struct in6_addr *daddr, *saddr;
2694         struct rt6_info *rt6 = (struct rt6_info *)dst;
2695
2696         if (dst_metric_locked(dst, RTAX_MTU))
2697                 return;
2698
2699         if (iph) {
2700                 daddr = &iph->daddr;
2701                 saddr = &iph->saddr;
2702         } else if (sk) {
2703                 daddr = &sk->sk_v6_daddr;
2704                 saddr = &inet6_sk(sk)->saddr;
2705         } else {
2706                 daddr = NULL;
2707                 saddr = NULL;
2708         }
2709         dst_confirm_neigh(dst, daddr);
2710         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2711         if (mtu >= dst_mtu(dst))
2712                 return;
2713
2714         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2715                 rt6_do_update_pmtu(rt6, mtu);
2716                 /* update rt6_ex->stamp for cache */
2717                 if (rt6->rt6i_flags & RTF_CACHE)
2718                         rt6_update_exception_stamp_rt(rt6);
2719         } else if (daddr) {
2720                 struct fib6_result res = {};
2721                 struct rt6_info *nrt6;
2722
2723                 rcu_read_lock();
2724                 res.f6i = rcu_dereference(rt6->from);
2725                 if (!res.f6i)
2726                         goto out_unlock;
2727
2728                 res.fib6_flags = res.f6i->fib6_flags;
2729                 res.fib6_type = res.f6i->fib6_type;
2730
2731                 if (res.f6i->nh) {
2732                         struct fib6_nh_match_arg arg = {
2733                                 .dev = dst->dev,
2734                                 .gw = &rt6->rt6i_gateway,
2735                         };
2736
2737                         nexthop_for_each_fib6_nh(res.f6i->nh,
2738                                                  fib6_nh_find_match, &arg);
2739
2740                         /* fib6_info uses a nexthop that does not have fib6_nh
2741                          * using the dst->dev + gw. Should be impossible.
2742                          */
2743                         if (!arg.match)
2744                                 goto out_unlock;
2745
2746                         res.nh = arg.match;
2747                 } else {
2748                         res.nh = res.f6i->fib6_nh;
2749                 }
2750
2751                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2752                 if (nrt6) {
2753                         rt6_do_update_pmtu(nrt6, mtu);
2754                         if (rt6_insert_exception(nrt6, &res))
2755                                 dst_release_immediate(&nrt6->dst);
2756                 }
2757 out_unlock:
2758                 rcu_read_unlock();
2759         }
2760 }
2761
2762 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2763                                struct sk_buff *skb, u32 mtu)
2764 {
2765         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu);
2766 }
2767
2768 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2769                      int oif, u32 mark, kuid_t uid)
2770 {
2771         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2772         struct dst_entry *dst;
2773         struct flowi6 fl6 = {
2774                 .flowi6_oif = oif,
2775                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2776                 .daddr = iph->daddr,
2777                 .saddr = iph->saddr,
2778                 .flowlabel = ip6_flowinfo(iph),
2779                 .flowi6_uid = uid,
2780         };
2781
2782         dst = ip6_route_output(net, NULL, &fl6);
2783         if (!dst->error)
2784                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu));
2785         dst_release(dst);
2786 }
2787 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2788
2789 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2790 {
2791         int oif = sk->sk_bound_dev_if;
2792         struct dst_entry *dst;
2793
2794         if (!oif && skb->dev)
2795                 oif = l3mdev_master_ifindex(skb->dev);
2796
2797         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2798
2799         dst = __sk_dst_get(sk);
2800         if (!dst || !dst->obsolete ||
2801             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2802                 return;
2803
2804         bh_lock_sock(sk);
2805         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2806                 ip6_datagram_dst_update(sk, false);
2807         bh_unlock_sock(sk);
2808 }
2809 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2810
2811 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2812                            const struct flowi6 *fl6)
2813 {
2814 #ifdef CONFIG_IPV6_SUBTREES
2815         struct ipv6_pinfo *np = inet6_sk(sk);
2816 #endif
2817
2818         ip6_dst_store(sk, dst,
2819                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2820                       &sk->sk_v6_daddr : NULL,
2821 #ifdef CONFIG_IPV6_SUBTREES
2822                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2823                       &np->saddr :
2824 #endif
2825                       NULL);
2826 }
2827
2828 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2829                                   struct flowi6 *fl6,
2830                                   const struct in6_addr *gw,
2831                                   struct rt6_info **ret)
2832 {
2833         const struct fib6_nh *nh = res->nh;
2834
2835         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2836             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2837                 return false;
2838
2839         /* rt_cache's gateway might be different from its 'parent'
2840          * in the case of an ip redirect.
2841          * So we keep searching in the exception table if the gateway
2842          * is different.
2843          */
2844         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2845                 struct rt6_info *rt_cache;
2846
2847                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2848                 if (rt_cache &&
2849                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2850                         *ret = rt_cache;
2851                         return true;
2852                 }
2853                 return false;
2854         }
2855         return true;
2856 }
2857
2858 struct fib6_nh_rd_arg {
2859         struct fib6_result      *res;
2860         struct flowi6           *fl6;
2861         const struct in6_addr   *gw;
2862         struct rt6_info         **ret;
2863 };
2864
2865 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2866 {
2867         struct fib6_nh_rd_arg *arg = _arg;
2868
2869         arg->res->nh = nh;
2870         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2871 }
2872
2873 /* Handle redirects */
2874 struct ip6rd_flowi {
2875         struct flowi6 fl6;
2876         struct in6_addr gateway;
2877 };
2878
2879 static struct rt6_info *__ip6_route_redirect(struct net *net,
2880                                              struct fib6_table *table,
2881                                              struct flowi6 *fl6,
2882                                              const struct sk_buff *skb,
2883                                              int flags)
2884 {
2885         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2886         struct rt6_info *ret = NULL;
2887         struct fib6_result res = {};
2888         struct fib6_nh_rd_arg arg = {
2889                 .res = &res,
2890                 .fl6 = fl6,
2891                 .gw  = &rdfl->gateway,
2892                 .ret = &ret
2893         };
2894         struct fib6_info *rt;
2895         struct fib6_node *fn;
2896
2897         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2898          * this case we must match on the real ingress device, so reset it
2899          */
2900         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2901                 fl6->flowi6_oif = skb->dev->ifindex;
2902
2903         /* Get the "current" route for this destination and
2904          * check if the redirect has come from appropriate router.
2905          *
2906          * RFC 4861 specifies that redirects should only be
2907          * accepted if they come from the nexthop to the target.
2908          * Due to the way the routes are chosen, this notion
2909          * is a bit fuzzy and one might need to check all possible
2910          * routes.
2911          */
2912
2913         rcu_read_lock();
2914         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2915 restart:
2916         for_each_fib6_node_rt_rcu(fn) {
2917                 res.f6i = rt;
2918                 if (fib6_check_expired(rt))
2919                         continue;
2920                 if (rt->fib6_flags & RTF_REJECT)
2921                         break;
2922                 if (unlikely(rt->nh)) {
2923                         if (nexthop_is_blackhole(rt->nh))
2924                                 continue;
2925                         /* on match, res->nh is filled in and potentially ret */
2926                         if (nexthop_for_each_fib6_nh(rt->nh,
2927                                                      fib6_nh_redirect_match,
2928                                                      &arg))
2929                                 goto out;
2930                 } else {
2931                         res.nh = rt->fib6_nh;
2932                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2933                                                   &ret))
2934                                 goto out;
2935                 }
2936         }
2937
2938         if (!rt)
2939                 rt = net->ipv6.fib6_null_entry;
2940         else if (rt->fib6_flags & RTF_REJECT) {
2941                 ret = net->ipv6.ip6_null_entry;
2942                 goto out;
2943         }
2944
2945         if (rt == net->ipv6.fib6_null_entry) {
2946                 fn = fib6_backtrack(fn, &fl6->saddr);
2947                 if (fn)
2948                         goto restart;
2949         }
2950
2951         res.f6i = rt;
2952         res.nh = rt->fib6_nh;
2953 out:
2954         if (ret) {
2955                 ip6_hold_safe(net, &ret);
2956         } else {
2957                 res.fib6_flags = res.f6i->fib6_flags;
2958                 res.fib6_type = res.f6i->fib6_type;
2959                 ret = ip6_create_rt_rcu(&res);
2960         }
2961
2962         rcu_read_unlock();
2963
2964         trace_fib6_table_lookup(net, &res, table, fl6);
2965         return ret;
2966 };
2967
2968 static struct dst_entry *ip6_route_redirect(struct net *net,
2969                                             const struct flowi6 *fl6,
2970                                             const struct sk_buff *skb,
2971                                             const struct in6_addr *gateway)
2972 {
2973         int flags = RT6_LOOKUP_F_HAS_SADDR;
2974         struct ip6rd_flowi rdfl;
2975
2976         rdfl.fl6 = *fl6;
2977         rdfl.gateway = *gateway;
2978
2979         return fib6_rule_lookup(net, &rdfl.fl6, skb,
2980                                 flags, __ip6_route_redirect);
2981 }
2982
2983 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
2984                   kuid_t uid)
2985 {
2986         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2987         struct dst_entry *dst;
2988         struct flowi6 fl6 = {
2989                 .flowi6_iif = LOOPBACK_IFINDEX,
2990                 .flowi6_oif = oif,
2991                 .flowi6_mark = mark,
2992                 .daddr = iph->daddr,
2993                 .saddr = iph->saddr,
2994                 .flowlabel = ip6_flowinfo(iph),
2995                 .flowi6_uid = uid,
2996         };
2997
2998         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
2999         rt6_do_redirect(dst, NULL, skb);
3000         dst_release(dst);
3001 }
3002 EXPORT_SYMBOL_GPL(ip6_redirect);
3003
3004 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3005 {
3006         const struct ipv6hdr *iph = ipv6_hdr(skb);
3007         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3008         struct dst_entry *dst;
3009         struct flowi6 fl6 = {
3010                 .flowi6_iif = LOOPBACK_IFINDEX,
3011                 .flowi6_oif = oif,
3012                 .daddr = msg->dest,
3013                 .saddr = iph->daddr,
3014                 .flowi6_uid = sock_net_uid(net, NULL),
3015         };
3016
3017         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3018         rt6_do_redirect(dst, NULL, skb);
3019         dst_release(dst);
3020 }
3021
3022 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3023 {
3024         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3025                      sk->sk_uid);
3026 }
3027 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3028
3029 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3030 {
3031         struct net_device *dev = dst->dev;
3032         unsigned int mtu = dst_mtu(dst);
3033         struct net *net = dev_net(dev);
3034
3035         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3036
3037         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3038                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3039
3040         /*
3041          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3042          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3043          * IPV6_MAXPLEN is also valid and means: "any MSS,
3044          * rely only on pmtu discovery"
3045          */
3046         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3047                 mtu = IPV6_MAXPLEN;
3048         return mtu;
3049 }
3050
3051 static unsigned int ip6_mtu(const struct dst_entry *dst)
3052 {
3053         struct inet6_dev *idev;
3054         unsigned int mtu;
3055
3056         mtu = dst_metric_raw(dst, RTAX_MTU);
3057         if (mtu)
3058                 goto out;
3059
3060         mtu = IPV6_MIN_MTU;
3061
3062         rcu_read_lock();
3063         idev = __in6_dev_get(dst->dev);
3064         if (idev)
3065                 mtu = idev->cnf.mtu6;
3066         rcu_read_unlock();
3067
3068 out:
3069         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3070
3071         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3072 }
3073
3074 /* MTU selection:
3075  * 1. mtu on route is locked - use it
3076  * 2. mtu from nexthop exception
3077  * 3. mtu from egress device
3078  *
3079  * based on ip6_dst_mtu_forward and exception logic of
3080  * rt6_find_cached_rt; called with rcu_read_lock
3081  */
3082 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3083                       const struct in6_addr *daddr,
3084                       const struct in6_addr *saddr)
3085 {
3086         const struct fib6_nh *nh = res->nh;
3087         struct fib6_info *f6i = res->f6i;
3088         struct inet6_dev *idev;
3089         struct rt6_info *rt;
3090         u32 mtu = 0;
3091
3092         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3093                 mtu = f6i->fib6_pmtu;
3094                 if (mtu)
3095                         goto out;
3096         }
3097
3098         rt = rt6_find_cached_rt(res, daddr, saddr);
3099         if (unlikely(rt)) {
3100                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3101         } else {
3102                 struct net_device *dev = nh->fib_nh_dev;
3103
3104                 mtu = IPV6_MIN_MTU;
3105                 idev = __in6_dev_get(dev);
3106                 if (idev && idev->cnf.mtu6 > mtu)
3107                         mtu = idev->cnf.mtu6;
3108         }
3109
3110         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3111 out:
3112         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3113 }
3114
3115 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3116                                   struct flowi6 *fl6)
3117 {
3118         struct dst_entry *dst;
3119         struct rt6_info *rt;
3120         struct inet6_dev *idev = in6_dev_get(dev);
3121         struct net *net = dev_net(dev);
3122
3123         if (unlikely(!idev))
3124                 return ERR_PTR(-ENODEV);
3125
3126         rt = ip6_dst_alloc(net, dev, 0);
3127         if (unlikely(!rt)) {
3128                 in6_dev_put(idev);
3129                 dst = ERR_PTR(-ENOMEM);
3130                 goto out;
3131         }
3132
3133         rt->dst.flags |= DST_HOST;
3134         rt->dst.input = ip6_input;
3135         rt->dst.output  = ip6_output;
3136         rt->rt6i_gateway  = fl6->daddr;
3137         rt->rt6i_dst.addr = fl6->daddr;
3138         rt->rt6i_dst.plen = 128;
3139         rt->rt6i_idev     = idev;
3140         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3141
3142         /* Add this dst into uncached_list so that rt6_disable_ip() can
3143          * do proper release of the net_device
3144          */
3145         rt6_uncached_list_add(rt);
3146         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3147
3148         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3149
3150 out:
3151         return dst;
3152 }
3153
3154 static int ip6_dst_gc(struct dst_ops *ops)
3155 {
3156         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3157         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3158         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3159         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3160         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3161         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3162         int entries;
3163
3164         entries = dst_entries_get_fast(ops);
3165         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3166             entries <= rt_max_size)
3167                 goto out;
3168
3169         net->ipv6.ip6_rt_gc_expire++;
3170         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3171         entries = dst_entries_get_slow(ops);
3172         if (entries < ops->gc_thresh)
3173                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3174 out:
3175         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3176         return entries > rt_max_size;
3177 }
3178
3179 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3180                                const struct in6_addr *gw_addr, u32 tbid,
3181                                int flags, struct fib6_result *res)
3182 {
3183         struct flowi6 fl6 = {
3184                 .flowi6_oif = cfg->fc_ifindex,
3185                 .daddr = *gw_addr,
3186                 .saddr = cfg->fc_prefsrc,
3187         };
3188         struct fib6_table *table;
3189         int err;
3190
3191         table = fib6_get_table(net, tbid);
3192         if (!table)
3193                 return -EINVAL;
3194
3195         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3196                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3197
3198         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3199
3200         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3201         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3202                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3203                                  cfg->fc_ifindex != 0, NULL, flags);
3204
3205         return err;
3206 }
3207
3208 static int ip6_route_check_nh_onlink(struct net *net,
3209                                      struct fib6_config *cfg,
3210                                      const struct net_device *dev,
3211                                      struct netlink_ext_ack *extack)
3212 {
3213         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3214         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3215         struct fib6_result res = {};
3216         int err;
3217
3218         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3219         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3220             /* ignore match if it is the default route */
3221             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3222             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3223                 NL_SET_ERR_MSG(extack,
3224                                "Nexthop has invalid gateway or device mismatch");
3225                 err = -EINVAL;
3226         }
3227
3228         return err;
3229 }
3230
3231 static int ip6_route_check_nh(struct net *net,
3232                               struct fib6_config *cfg,
3233                               struct net_device **_dev,
3234                               struct inet6_dev **idev)
3235 {
3236         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3237         struct net_device *dev = _dev ? *_dev : NULL;
3238         int flags = RT6_LOOKUP_F_IFACE;
3239         struct fib6_result res = {};
3240         int err = -EHOSTUNREACH;
3241
3242         if (cfg->fc_table) {
3243                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3244                                           cfg->fc_table, flags, &res);
3245                 /* gw_addr can not require a gateway or resolve to a reject
3246                  * route. If a device is given, it must match the result.
3247                  */
3248                 if (err || res.fib6_flags & RTF_REJECT ||
3249                     res.nh->fib_nh_gw_family ||
3250                     (dev && dev != res.nh->fib_nh_dev))
3251                         err = -EHOSTUNREACH;
3252         }
3253
3254         if (err < 0) {
3255                 struct flowi6 fl6 = {
3256                         .flowi6_oif = cfg->fc_ifindex,
3257                         .daddr = *gw_addr,
3258                 };
3259
3260                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3261                 if (err || res.fib6_flags & RTF_REJECT ||
3262                     res.nh->fib_nh_gw_family)
3263                         err = -EHOSTUNREACH;
3264
3265                 if (err)
3266                         return err;
3267
3268                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3269                                  cfg->fc_ifindex != 0, NULL, flags);
3270         }
3271
3272         err = 0;
3273         if (dev) {
3274                 if (dev != res.nh->fib_nh_dev)
3275                         err = -EHOSTUNREACH;
3276         } else {
3277                 *_dev = dev = res.nh->fib_nh_dev;
3278                 dev_hold(dev);
3279                 *idev = in6_dev_get(dev);
3280         }
3281
3282         return err;
3283 }
3284
3285 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3286                            struct net_device **_dev, struct inet6_dev **idev,
3287                            struct netlink_ext_ack *extack)
3288 {
3289         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3290         int gwa_type = ipv6_addr_type(gw_addr);
3291         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3292         const struct net_device *dev = *_dev;
3293         bool need_addr_check = !dev;
3294         int err = -EINVAL;
3295
3296         /* if gw_addr is local we will fail to detect this in case
3297          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3298          * will return already-added prefix route via interface that
3299          * prefix route was assigned to, which might be non-loopback.
3300          */
3301         if (dev &&
3302             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3303                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3304                 goto out;
3305         }
3306
3307         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3308                 /* IPv6 strictly inhibits using not link-local
3309                  * addresses as nexthop address.
3310                  * Otherwise, router will not able to send redirects.
3311                  * It is very good, but in some (rare!) circumstances
3312                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3313                  * some exceptions. --ANK
3314                  * We allow IPv4-mapped nexthops to support RFC4798-type
3315                  * addressing
3316                  */
3317                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3318                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3319                         goto out;
3320                 }
3321
3322                 rcu_read_lock();
3323
3324                 if (cfg->fc_flags & RTNH_F_ONLINK)
3325                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3326                 else
3327                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3328
3329                 rcu_read_unlock();
3330
3331                 if (err)
3332                         goto out;
3333         }
3334
3335         /* reload in case device was changed */
3336         dev = *_dev;
3337
3338         err = -EINVAL;
3339         if (!dev) {
3340                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3341                 goto out;
3342         } else if (dev->flags & IFF_LOOPBACK) {
3343                 NL_SET_ERR_MSG(extack,
3344                                "Egress device can not be loopback device for this route");
3345                 goto out;
3346         }
3347
3348         /* if we did not check gw_addr above, do so now that the
3349          * egress device has been resolved.
3350          */
3351         if (need_addr_check &&
3352             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3353                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3354                 goto out;
3355         }
3356
3357         err = 0;
3358 out:
3359         return err;
3360 }
3361
3362 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3363 {
3364         if ((flags & RTF_REJECT) ||
3365             (dev && (dev->flags & IFF_LOOPBACK) &&
3366              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3367              !(flags & RTF_LOCAL)))
3368                 return true;
3369
3370         return false;
3371 }
3372
3373 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3374                  struct fib6_config *cfg, gfp_t gfp_flags,
3375                  struct netlink_ext_ack *extack)
3376 {
3377         struct net_device *dev = NULL;
3378         struct inet6_dev *idev = NULL;
3379         int addr_type;
3380         int err;
3381
3382         fib6_nh->fib_nh_family = AF_INET6;
3383
3384         err = -ENODEV;
3385         if (cfg->fc_ifindex) {
3386                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3387                 if (!dev)
3388                         goto out;
3389                 idev = in6_dev_get(dev);
3390                 if (!idev)
3391                         goto out;
3392         }
3393
3394         if (cfg->fc_flags & RTNH_F_ONLINK) {
3395                 if (!dev) {
3396                         NL_SET_ERR_MSG(extack,
3397                                        "Nexthop device required for onlink");
3398                         goto out;
3399                 }
3400
3401                 if (!(dev->flags & IFF_UP)) {
3402                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3403                         err = -ENETDOWN;
3404                         goto out;
3405                 }
3406
3407                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3408         }
3409
3410         fib6_nh->fib_nh_weight = 1;
3411
3412         /* We cannot add true routes via loopback here,
3413          * they would result in kernel looping; promote them to reject routes
3414          */
3415         addr_type = ipv6_addr_type(&cfg->fc_dst);
3416         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3417                 /* hold loopback dev/idev if we haven't done so. */
3418                 if (dev != net->loopback_dev) {
3419                         if (dev) {
3420                                 dev_put(dev);
3421                                 in6_dev_put(idev);
3422                         }
3423                         dev = net->loopback_dev;
3424                         dev_hold(dev);
3425                         idev = in6_dev_get(dev);
3426                         if (!idev) {
3427                                 err = -ENODEV;
3428                                 goto out;
3429                         }
3430                 }
3431                 goto pcpu_alloc;
3432         }
3433
3434         if (cfg->fc_flags & RTF_GATEWAY) {
3435                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3436                 if (err)
3437                         goto out;
3438
3439                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3440                 fib6_nh->fib_nh_gw_family = AF_INET6;
3441         }
3442
3443         err = -ENODEV;
3444         if (!dev)
3445                 goto out;
3446
3447         if (idev->cnf.disable_ipv6) {
3448                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3449                 err = -EACCES;
3450                 goto out;
3451         }
3452
3453         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3454                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3455                 err = -ENETDOWN;
3456                 goto out;
3457         }
3458
3459         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3460             !netif_carrier_ok(dev))
3461                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3462
3463         err = fib_nh_common_init(&fib6_nh->nh_common, cfg->fc_encap,
3464                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3465         if (err)
3466                 goto out;
3467
3468 pcpu_alloc:
3469         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3470         if (!fib6_nh->rt6i_pcpu) {
3471                 err = -ENOMEM;
3472                 goto out;
3473         }
3474
3475         fib6_nh->fib_nh_dev = dev;
3476         fib6_nh->fib_nh_oif = dev->ifindex;
3477         err = 0;
3478 out:
3479         if (idev)
3480                 in6_dev_put(idev);
3481
3482         if (err) {
3483                 lwtstate_put(fib6_nh->fib_nh_lws);
3484                 fib6_nh->fib_nh_lws = NULL;
3485                 if (dev)
3486                         dev_put(dev);
3487         }
3488
3489         return err;
3490 }
3491
3492 void fib6_nh_release(struct fib6_nh *fib6_nh)
3493 {
3494         struct rt6_exception_bucket *bucket;
3495
3496         rcu_read_lock();
3497
3498         fib6_nh_flush_exceptions(fib6_nh, NULL);
3499         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3500         if (bucket) {
3501                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3502                 kfree(bucket);
3503         }
3504
3505         rcu_read_unlock();
3506
3507         if (fib6_nh->rt6i_pcpu) {
3508                 int cpu;
3509
3510                 for_each_possible_cpu(cpu) {
3511                         struct rt6_info **ppcpu_rt;
3512                         struct rt6_info *pcpu_rt;
3513
3514                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3515                         pcpu_rt = *ppcpu_rt;
3516                         if (pcpu_rt) {
3517                                 dst_dev_put(&pcpu_rt->dst);
3518                                 dst_release(&pcpu_rt->dst);
3519                                 *ppcpu_rt = NULL;
3520                         }
3521                 }
3522
3523                 free_percpu(fib6_nh->rt6i_pcpu);
3524         }
3525
3526         fib_nh_common_release(&fib6_nh->nh_common);
3527 }
3528
3529 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3530                                               gfp_t gfp_flags,
3531                                               struct netlink_ext_ack *extack)
3532 {
3533         struct net *net = cfg->fc_nlinfo.nl_net;
3534         struct fib6_info *rt = NULL;
3535         struct nexthop *nh = NULL;
3536         struct fib6_table *table;
3537         struct fib6_nh *fib6_nh;
3538         int err = -EINVAL;
3539         int addr_type;
3540
3541         /* RTF_PCPU is an internal flag; can not be set by userspace */
3542         if (cfg->fc_flags & RTF_PCPU) {
3543                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3544                 goto out;
3545         }
3546
3547         /* RTF_CACHE is an internal flag; can not be set by userspace */
3548         if (cfg->fc_flags & RTF_CACHE) {
3549                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3550                 goto out;
3551         }
3552
3553         if (cfg->fc_type > RTN_MAX) {
3554                 NL_SET_ERR_MSG(extack, "Invalid route type");
3555                 goto out;
3556         }
3557
3558         if (cfg->fc_dst_len > 128) {
3559                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3560                 goto out;
3561         }
3562         if (cfg->fc_src_len > 128) {
3563                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3564                 goto out;
3565         }
3566 #ifndef CONFIG_IPV6_SUBTREES
3567         if (cfg->fc_src_len) {
3568                 NL_SET_ERR_MSG(extack,
3569                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3570                 goto out;
3571         }
3572 #endif
3573         if (cfg->fc_nh_id) {
3574                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3575                 if (!nh) {
3576                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3577                         goto out;
3578                 }
3579                 err = fib6_check_nexthop(nh, cfg, extack);
3580                 if (err)
3581                         goto out;
3582         }
3583
3584         err = -ENOBUFS;
3585         if (cfg->fc_nlinfo.nlh &&
3586             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3587                 table = fib6_get_table(net, cfg->fc_table);
3588                 if (!table) {
3589                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3590                         table = fib6_new_table(net, cfg->fc_table);
3591                 }
3592         } else {
3593                 table = fib6_new_table(net, cfg->fc_table);
3594         }
3595
3596         if (!table)
3597                 goto out;
3598
3599         err = -ENOMEM;
3600         rt = fib6_info_alloc(gfp_flags, !nh);
3601         if (!rt)
3602                 goto out;
3603
3604         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3605                                                extack);
3606         if (IS_ERR(rt->fib6_metrics)) {
3607                 err = PTR_ERR(rt->fib6_metrics);
3608                 /* Do not leave garbage there. */
3609                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3610                 goto out;
3611         }
3612
3613         if (cfg->fc_flags & RTF_ADDRCONF)
3614                 rt->dst_nocount = true;
3615
3616         if (cfg->fc_flags & RTF_EXPIRES)
3617                 fib6_set_expires(rt, jiffies +
3618                                 clock_t_to_jiffies(cfg->fc_expires));
3619         else
3620                 fib6_clean_expires(rt);
3621
3622         if (cfg->fc_protocol == RTPROT_UNSPEC)
3623                 cfg->fc_protocol = RTPROT_BOOT;
3624         rt->fib6_protocol = cfg->fc_protocol;
3625
3626         rt->fib6_table = table;
3627         rt->fib6_metric = cfg->fc_metric;
3628         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3629         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3630
3631         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3632         rt->fib6_dst.plen = cfg->fc_dst_len;
3633         if (rt->fib6_dst.plen == 128)
3634                 rt->dst_host = true;
3635
3636 #ifdef CONFIG_IPV6_SUBTREES
3637         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3638         rt->fib6_src.plen = cfg->fc_src_len;
3639 #endif
3640         if (nh) {
3641                 if (!nexthop_get(nh)) {
3642                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3643                         goto out;
3644                 }
3645                 if (rt->fib6_src.plen) {
3646                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3647                         goto out;
3648                 }
3649                 rt->nh = nh;
3650                 fib6_nh = nexthop_fib6_nh(rt->nh);
3651         } else {
3652                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3653                 if (err)
3654                         goto out;
3655
3656                 fib6_nh = rt->fib6_nh;
3657
3658                 /* We cannot add true routes via loopback here, they would
3659                  * result in kernel looping; promote them to reject routes
3660                  */
3661                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3662                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3663                                    addr_type))
3664                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3665         }
3666
3667         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3668                 struct net_device *dev = fib6_nh->fib_nh_dev;
3669
3670                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3671                         NL_SET_ERR_MSG(extack, "Invalid source address");
3672                         err = -EINVAL;
3673                         goto out;
3674                 }
3675                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3676                 rt->fib6_prefsrc.plen = 128;
3677         } else
3678                 rt->fib6_prefsrc.plen = 0;
3679
3680         return rt;
3681 out:
3682         fib6_info_release(rt);
3683         return ERR_PTR(err);
3684 }
3685
3686 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3687                   struct netlink_ext_ack *extack)
3688 {
3689         struct fib6_info *rt;
3690         int err;
3691
3692         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3693         if (IS_ERR(rt))
3694                 return PTR_ERR(rt);
3695
3696         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3697         fib6_info_release(rt);
3698
3699         return err;
3700 }
3701
3702 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3703 {
3704         struct net *net = info->nl_net;
3705         struct fib6_table *table;
3706         int err;
3707
3708         if (rt == net->ipv6.fib6_null_entry) {
3709                 err = -ENOENT;
3710                 goto out;
3711         }
3712
3713         table = rt->fib6_table;
3714         spin_lock_bh(&table->tb6_lock);
3715         err = fib6_del(rt, info);
3716         spin_unlock_bh(&table->tb6_lock);
3717
3718 out:
3719         fib6_info_release(rt);
3720         return err;
3721 }
3722
3723 int ip6_del_rt(struct net *net, struct fib6_info *rt)
3724 {
3725         struct nl_info info = { .nl_net = net };
3726
3727         return __ip6_del_rt(rt, &info);
3728 }
3729
3730 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3731 {
3732         struct nl_info *info = &cfg->fc_nlinfo;
3733         struct net *net = info->nl_net;
3734         struct sk_buff *skb = NULL;
3735         struct fib6_table *table;
3736         int err = -ENOENT;
3737
3738         if (rt == net->ipv6.fib6_null_entry)
3739                 goto out_put;
3740         table = rt->fib6_table;
3741         spin_lock_bh(&table->tb6_lock);
3742
3743         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3744                 struct fib6_info *sibling, *next_sibling;
3745
3746                 /* prefer to send a single notification with all hops */
3747                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3748                 if (skb) {
3749                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3750
3751                         if (rt6_fill_node(net, skb, rt, NULL,
3752                                           NULL, NULL, 0, RTM_DELROUTE,
3753                                           info->portid, seq, 0) < 0) {
3754                                 kfree_skb(skb);
3755                                 skb = NULL;
3756                         } else
3757                                 info->skip_notify = 1;
3758                 }
3759
3760                 info->skip_notify_kernel = 1;
3761                 call_fib6_multipath_entry_notifiers(net,
3762                                                     FIB_EVENT_ENTRY_DEL,
3763                                                     rt,
3764                                                     rt->fib6_nsiblings,
3765                                                     NULL);
3766                 list_for_each_entry_safe(sibling, next_sibling,
3767                                          &rt->fib6_siblings,
3768                                          fib6_siblings) {
3769                         err = fib6_del(sibling, info);
3770                         if (err)
3771                                 goto out_unlock;
3772                 }
3773         }
3774
3775         err = fib6_del(rt, info);
3776 out_unlock:
3777         spin_unlock_bh(&table->tb6_lock);
3778 out_put:
3779         fib6_info_release(rt);
3780
3781         if (skb) {
3782                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3783                             info->nlh, gfp_any());
3784         }
3785         return err;
3786 }
3787
3788 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3789 {
3790         int rc = -ESRCH;
3791
3792         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3793                 goto out;
3794
3795         if (cfg->fc_flags & RTF_GATEWAY &&
3796             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3797                 goto out;
3798
3799         rc = rt6_remove_exception_rt(rt);
3800 out:
3801         return rc;
3802 }
3803
3804 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3805                              struct fib6_nh *nh)
3806 {
3807         struct fib6_result res = {
3808                 .f6i = rt,
3809                 .nh = nh,
3810         };
3811         struct rt6_info *rt_cache;
3812
3813         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3814         if (rt_cache)
3815                 return __ip6_del_cached_rt(rt_cache, cfg);
3816
3817         return 0;
3818 }
3819
3820 struct fib6_nh_del_cached_rt_arg {
3821         struct fib6_config *cfg;
3822         struct fib6_info *f6i;
3823 };
3824
3825 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3826 {
3827         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3828         int rc;
3829
3830         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3831         return rc != -ESRCH ? rc : 0;
3832 }
3833
3834 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3835 {
3836         struct fib6_nh_del_cached_rt_arg arg = {
3837                 .cfg = cfg,
3838                 .f6i = f6i
3839         };
3840
3841         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3842 }
3843
3844 static int ip6_route_del(struct fib6_config *cfg,
3845                          struct netlink_ext_ack *extack)
3846 {
3847         struct fib6_table *table;
3848         struct fib6_info *rt;
3849         struct fib6_node *fn;
3850         int err = -ESRCH;
3851
3852         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3853         if (!table) {
3854                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3855                 return err;
3856         }
3857
3858         rcu_read_lock();
3859
3860         fn = fib6_locate(&table->tb6_root,
3861                          &cfg->fc_dst, cfg->fc_dst_len,
3862                          &cfg->fc_src, cfg->fc_src_len,
3863                          !(cfg->fc_flags & RTF_CACHE));
3864
3865         if (fn) {
3866                 for_each_fib6_node_rt_rcu(fn) {
3867                         struct fib6_nh *nh;
3868
3869                         if (rt->nh && cfg->fc_nh_id &&
3870                             rt->nh->id != cfg->fc_nh_id)
3871                                 continue;
3872
3873                         if (cfg->fc_flags & RTF_CACHE) {
3874                                 int rc = 0;
3875
3876                                 if (rt->nh) {
3877                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3878                                 } else if (cfg->fc_nh_id) {
3879                                         continue;
3880                                 } else {
3881                                         nh = rt->fib6_nh;
3882                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3883                                 }
3884                                 if (rc != -ESRCH) {
3885                                         rcu_read_unlock();
3886                                         return rc;
3887                                 }
3888                                 continue;
3889                         }
3890
3891                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3892                                 continue;
3893                         if (cfg->fc_protocol &&
3894                             cfg->fc_protocol != rt->fib6_protocol)
3895                                 continue;
3896
3897                         if (rt->nh) {
3898                                 if (!fib6_info_hold_safe(rt))
3899                                         continue;
3900                                 rcu_read_unlock();
3901
3902                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3903                         }
3904                         if (cfg->fc_nh_id)
3905                                 continue;
3906
3907                         nh = rt->fib6_nh;
3908                         if (cfg->fc_ifindex &&
3909                             (!nh->fib_nh_dev ||
3910                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3911                                 continue;
3912                         if (cfg->fc_flags & RTF_GATEWAY &&
3913                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3914                                 continue;
3915                         if (!fib6_info_hold_safe(rt))
3916                                 continue;
3917                         rcu_read_unlock();
3918
3919                         /* if gateway was specified only delete the one hop */
3920                         if (cfg->fc_flags & RTF_GATEWAY)
3921                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3922
3923                         return __ip6_del_rt_siblings(rt, cfg);
3924                 }
3925         }
3926         rcu_read_unlock();
3927
3928         return err;
3929 }
3930
3931 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3932 {
3933         struct netevent_redirect netevent;
3934         struct rt6_info *rt, *nrt = NULL;
3935         struct fib6_result res = {};
3936         struct ndisc_options ndopts;
3937         struct inet6_dev *in6_dev;
3938         struct neighbour *neigh;
3939         struct rd_msg *msg;
3940         int optlen, on_link;
3941         u8 *lladdr;
3942
3943         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
3944         optlen -= sizeof(*msg);
3945
3946         if (optlen < 0) {
3947                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
3948                 return;
3949         }
3950
3951         msg = (struct rd_msg *)icmp6_hdr(skb);
3952
3953         if (ipv6_addr_is_multicast(&msg->dest)) {
3954                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
3955                 return;
3956         }
3957
3958         on_link = 0;
3959         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
3960                 on_link = 1;
3961         } else if (ipv6_addr_type(&msg->target) !=
3962                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
3963                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
3964                 return;
3965         }
3966
3967         in6_dev = __in6_dev_get(skb->dev);
3968         if (!in6_dev)
3969                 return;
3970         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
3971                 return;
3972
3973         /* RFC2461 8.1:
3974          *      The IP source address of the Redirect MUST be the same as the current
3975          *      first-hop router for the specified ICMP Destination Address.
3976          */
3977
3978         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
3979                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
3980                 return;
3981         }
3982
3983         lladdr = NULL;
3984         if (ndopts.nd_opts_tgt_lladdr) {
3985                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
3986                                              skb->dev);
3987                 if (!lladdr) {
3988                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
3989                         return;
3990                 }
3991         }
3992
3993         rt = (struct rt6_info *) dst;
3994         if (rt->rt6i_flags & RTF_REJECT) {
3995                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
3996                 return;
3997         }
3998
3999         /* Redirect received -> path was valid.
4000          * Look, redirects are sent only in response to data packets,
4001          * so that this nexthop apparently is reachable. --ANK
4002          */
4003         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4004
4005         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4006         if (!neigh)
4007                 return;
4008
4009         /*
4010          *      We have finally decided to accept it.
4011          */
4012
4013         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4014                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4015                      NEIGH_UPDATE_F_OVERRIDE|
4016                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4017                                      NEIGH_UPDATE_F_ISROUTER)),
4018                      NDISC_REDIRECT, &ndopts);
4019
4020         rcu_read_lock();
4021         res.f6i = rcu_dereference(rt->from);
4022         if (!res.f6i)
4023                 goto out;
4024
4025         if (res.f6i->nh) {
4026                 struct fib6_nh_match_arg arg = {
4027                         .dev = dst->dev,
4028                         .gw = &rt->rt6i_gateway,
4029                 };
4030
4031                 nexthop_for_each_fib6_nh(res.f6i->nh,
4032                                          fib6_nh_find_match, &arg);
4033
4034                 /* fib6_info uses a nexthop that does not have fib6_nh
4035                  * using the dst->dev. Should be impossible
4036                  */
4037                 if (!arg.match)
4038                         goto out;
4039                 res.nh = arg.match;
4040         } else {
4041                 res.nh = res.f6i->fib6_nh;
4042         }
4043
4044         res.fib6_flags = res.f6i->fib6_flags;
4045         res.fib6_type = res.f6i->fib6_type;
4046         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4047         if (!nrt)
4048                 goto out;
4049
4050         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4051         if (on_link)
4052                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4053
4054         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4055
4056         /* rt6_insert_exception() will take care of duplicated exceptions */
4057         if (rt6_insert_exception(nrt, &res)) {
4058                 dst_release_immediate(&nrt->dst);
4059                 goto out;
4060         }
4061
4062         netevent.old = &rt->dst;
4063         netevent.new = &nrt->dst;
4064         netevent.daddr = &msg->dest;
4065         netevent.neigh = neigh;
4066         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4067
4068 out:
4069         rcu_read_unlock();
4070         neigh_release(neigh);
4071 }
4072
4073 #ifdef CONFIG_IPV6_ROUTE_INFO
4074 static struct fib6_info *rt6_get_route_info(struct net *net,
4075                                            const struct in6_addr *prefix, int prefixlen,
4076                                            const struct in6_addr *gwaddr,
4077                                            struct net_device *dev)
4078 {
4079         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4080         int ifindex = dev->ifindex;
4081         struct fib6_node *fn;
4082         struct fib6_info *rt = NULL;
4083         struct fib6_table *table;
4084
4085         table = fib6_get_table(net, tb_id);
4086         if (!table)
4087                 return NULL;
4088
4089         rcu_read_lock();
4090         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4091         if (!fn)
4092                 goto out;
4093
4094         for_each_fib6_node_rt_rcu(fn) {
4095                 /* these routes do not use nexthops */
4096                 if (rt->nh)
4097                         continue;
4098                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4099                         continue;
4100                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4101                     !rt->fib6_nh->fib_nh_gw_family)
4102                         continue;
4103                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4104                         continue;
4105                 if (!fib6_info_hold_safe(rt))
4106                         continue;
4107                 break;
4108         }
4109 out:
4110         rcu_read_unlock();
4111         return rt;
4112 }
4113
4114 static struct fib6_info *rt6_add_route_info(struct net *net,
4115                                            const struct in6_addr *prefix, int prefixlen,
4116                                            const struct in6_addr *gwaddr,
4117                                            struct net_device *dev,
4118                                            unsigned int pref)
4119 {
4120         struct fib6_config cfg = {
4121                 .fc_metric      = IP6_RT_PRIO_USER,
4122                 .fc_ifindex     = dev->ifindex,
4123                 .fc_dst_len     = prefixlen,
4124                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4125                                   RTF_UP | RTF_PREF(pref),
4126                 .fc_protocol = RTPROT_RA,
4127                 .fc_type = RTN_UNICAST,
4128                 .fc_nlinfo.portid = 0,
4129                 .fc_nlinfo.nlh = NULL,
4130                 .fc_nlinfo.nl_net = net,
4131         };
4132
4133         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
4134         cfg.fc_dst = *prefix;
4135         cfg.fc_gateway = *gwaddr;
4136
4137         /* We should treat it as a default route if prefix length is 0. */
4138         if (!prefixlen)
4139                 cfg.fc_flags |= RTF_DEFAULT;
4140
4141         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4142
4143         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4144 }
4145 #endif
4146
4147 struct fib6_info *rt6_get_dflt_router(struct net *net,
4148                                      const struct in6_addr *addr,
4149                                      struct net_device *dev)
4150 {
4151         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4152         struct fib6_info *rt;
4153         struct fib6_table *table;
4154
4155         table = fib6_get_table(net, tb_id);
4156         if (!table)
4157                 return NULL;
4158
4159         rcu_read_lock();
4160         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4161                 struct fib6_nh *nh;
4162
4163                 /* RA routes do not use nexthops */
4164                 if (rt->nh)
4165                         continue;
4166
4167                 nh = rt->fib6_nh;
4168                 if (dev == nh->fib_nh_dev &&
4169                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4170                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4171                         break;
4172         }
4173         if (rt && !fib6_info_hold_safe(rt))
4174                 rt = NULL;
4175         rcu_read_unlock();
4176         return rt;
4177 }
4178
4179 struct fib6_info *rt6_add_dflt_router(struct net *net,
4180                                      const struct in6_addr *gwaddr,
4181                                      struct net_device *dev,
4182                                      unsigned int pref)
4183 {
4184         struct fib6_config cfg = {
4185                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4186                 .fc_metric      = IP6_RT_PRIO_USER,
4187                 .fc_ifindex     = dev->ifindex,
4188                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4189                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4190                 .fc_protocol = RTPROT_RA,
4191                 .fc_type = RTN_UNICAST,
4192                 .fc_nlinfo.portid = 0,
4193                 .fc_nlinfo.nlh = NULL,
4194                 .fc_nlinfo.nl_net = net,
4195         };
4196
4197         cfg.fc_gateway = *gwaddr;
4198
4199         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4200                 struct fib6_table *table;
4201
4202                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4203                 if (table)
4204                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4205         }
4206
4207         return rt6_get_dflt_router(net, gwaddr, dev);
4208 }
4209
4210 static void __rt6_purge_dflt_routers(struct net *net,
4211                                      struct fib6_table *table)
4212 {
4213         struct fib6_info *rt;
4214
4215 restart:
4216         rcu_read_lock();
4217         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4218                 struct net_device *dev = fib6_info_nh_dev(rt);
4219                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4220
4221                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4222                     (!idev || idev->cnf.accept_ra != 2) &&
4223                     fib6_info_hold_safe(rt)) {
4224                         rcu_read_unlock();
4225                         ip6_del_rt(net, rt);
4226                         goto restart;
4227                 }
4228         }
4229         rcu_read_unlock();
4230
4231         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4232 }
4233
4234 void rt6_purge_dflt_routers(struct net *net)
4235 {
4236         struct fib6_table *table;
4237         struct hlist_head *head;
4238         unsigned int h;
4239
4240         rcu_read_lock();
4241
4242         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4243                 head = &net->ipv6.fib_table_hash[h];
4244                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4245                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4246                                 __rt6_purge_dflt_routers(net, table);
4247                 }
4248         }
4249
4250         rcu_read_unlock();
4251 }
4252
4253 static void rtmsg_to_fib6_config(struct net *net,
4254                                  struct in6_rtmsg *rtmsg,
4255                                  struct fib6_config *cfg)
4256 {
4257         *cfg = (struct fib6_config){
4258                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4259                          : RT6_TABLE_MAIN,
4260                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4261                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4262                 .fc_expires = rtmsg->rtmsg_info,
4263                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4264                 .fc_src_len = rtmsg->rtmsg_src_len,
4265                 .fc_flags = rtmsg->rtmsg_flags,
4266                 .fc_type = rtmsg->rtmsg_type,
4267
4268                 .fc_nlinfo.nl_net = net,
4269
4270                 .fc_dst = rtmsg->rtmsg_dst,
4271                 .fc_src = rtmsg->rtmsg_src,
4272                 .fc_gateway = rtmsg->rtmsg_gateway,
4273         };
4274 }
4275
4276 int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4277 {
4278         struct fib6_config cfg;
4279         struct in6_rtmsg rtmsg;
4280         int err;
4281
4282         switch (cmd) {
4283         case SIOCADDRT:         /* Add a route */
4284         case SIOCDELRT:         /* Delete a route */
4285                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4286                         return -EPERM;
4287                 err = copy_from_user(&rtmsg, arg,
4288                                      sizeof(struct in6_rtmsg));
4289                 if (err)
4290                         return -EFAULT;
4291
4292                 rtmsg_to_fib6_config(net, &rtmsg, &cfg);
4293
4294                 rtnl_lock();
4295                 switch (cmd) {
4296                 case SIOCADDRT:
4297                         err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4298                         break;
4299                 case SIOCDELRT:
4300                         err = ip6_route_del(&cfg, NULL);
4301                         break;
4302                 default:
4303                         err = -EINVAL;
4304                 }
4305                 rtnl_unlock();
4306
4307                 return err;
4308         }
4309
4310         return -EINVAL;
4311 }
4312
4313 /*
4314  *      Drop the packet on the floor
4315  */
4316
4317 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4318 {
4319         struct dst_entry *dst = skb_dst(skb);
4320         struct net *net = dev_net(dst->dev);
4321         struct inet6_dev *idev;
4322         int type;
4323
4324         if (netif_is_l3_master(skb->dev) &&
4325             dst->dev == net->loopback_dev)
4326                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4327         else
4328                 idev = ip6_dst_idev(dst);
4329
4330         switch (ipstats_mib_noroutes) {
4331         case IPSTATS_MIB_INNOROUTES:
4332                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4333                 if (type == IPV6_ADDR_ANY) {
4334                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4335                         break;
4336                 }
4337                 /* FALLTHROUGH */
4338         case IPSTATS_MIB_OUTNOROUTES:
4339                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4340                 break;
4341         }
4342
4343         /* Start over by dropping the dst for l3mdev case */
4344         if (netif_is_l3_master(skb->dev))
4345                 skb_dst_drop(skb);
4346
4347         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4348         kfree_skb(skb);
4349         return 0;
4350 }
4351
4352 static int ip6_pkt_discard(struct sk_buff *skb)
4353 {
4354         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4355 }
4356
4357 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4358 {
4359         skb->dev = skb_dst(skb)->dev;
4360         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4361 }
4362
4363 static int ip6_pkt_prohibit(struct sk_buff *skb)
4364 {
4365         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4366 }
4367
4368 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4369 {
4370         skb->dev = skb_dst(skb)->dev;
4371         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4372 }
4373
4374 /*
4375  *      Allocate a dst for local (unicast / anycast) address.
4376  */
4377
4378 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4379                                      struct inet6_dev *idev,
4380                                      const struct in6_addr *addr,
4381                                      bool anycast, gfp_t gfp_flags)
4382 {
4383         struct fib6_config cfg = {
4384                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4385                 .fc_ifindex = idev->dev->ifindex,
4386                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4387                 .fc_dst = *addr,
4388                 .fc_dst_len = 128,
4389                 .fc_protocol = RTPROT_KERNEL,
4390                 .fc_nlinfo.nl_net = net,
4391                 .fc_ignore_dev_down = true,
4392         };
4393         struct fib6_info *f6i;
4394
4395         if (anycast) {
4396                 cfg.fc_type = RTN_ANYCAST;
4397                 cfg.fc_flags |= RTF_ANYCAST;
4398         } else {
4399                 cfg.fc_type = RTN_LOCAL;
4400                 cfg.fc_flags |= RTF_LOCAL;
4401         }
4402
4403         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4404         if (!IS_ERR(f6i))
4405                 f6i->dst_nocount = true;
4406         return f6i;
4407 }
4408
4409 /* remove deleted ip from prefsrc entries */
4410 struct arg_dev_net_ip {
4411         struct net_device *dev;
4412         struct net *net;
4413         struct in6_addr *addr;
4414 };
4415
4416 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4417 {
4418         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4419         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4420         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4421
4422         if (!rt->nh &&
4423             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4424             rt != net->ipv6.fib6_null_entry &&
4425             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4426                 spin_lock_bh(&rt6_exception_lock);
4427                 /* remove prefsrc entry */
4428                 rt->fib6_prefsrc.plen = 0;
4429                 spin_unlock_bh(&rt6_exception_lock);
4430         }
4431         return 0;
4432 }
4433
4434 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4435 {
4436         struct net *net = dev_net(ifp->idev->dev);
4437         struct arg_dev_net_ip adni = {
4438                 .dev = ifp->idev->dev,
4439                 .net = net,
4440                 .addr = &ifp->addr,
4441         };
4442         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4443 }
4444
4445 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4446
4447 /* Remove routers and update dst entries when gateway turn into host. */
4448 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4449 {
4450         struct in6_addr *gateway = (struct in6_addr *)arg;
4451         struct fib6_nh *nh;
4452
4453         /* RA routes do not use nexthops */
4454         if (rt->nh)
4455                 return 0;
4456
4457         nh = rt->fib6_nh;
4458         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4459             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4460                 return -1;
4461
4462         /* Further clean up cached routes in exception table.
4463          * This is needed because cached route may have a different
4464          * gateway than its 'parent' in the case of an ip redirect.
4465          */
4466         fib6_nh_exceptions_clean_tohost(nh, gateway);
4467
4468         return 0;
4469 }
4470
4471 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4472 {
4473         fib6_clean_all(net, fib6_clean_tohost, gateway);
4474 }
4475
4476 struct arg_netdev_event {
4477         const struct net_device *dev;
4478         union {
4479                 unsigned char nh_flags;
4480                 unsigned long event;
4481         };
4482 };
4483
4484 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4485 {
4486         struct fib6_info *iter;
4487         struct fib6_node *fn;
4488
4489         fn = rcu_dereference_protected(rt->fib6_node,
4490                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4491         iter = rcu_dereference_protected(fn->leaf,
4492                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4493         while (iter) {
4494                 if (iter->fib6_metric == rt->fib6_metric &&
4495                     rt6_qualify_for_ecmp(iter))
4496                         return iter;
4497                 iter = rcu_dereference_protected(iter->fib6_next,
4498                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4499         }
4500
4501         return NULL;
4502 }
4503
4504 /* only called for fib entries with builtin fib6_nh */
4505 static bool rt6_is_dead(const struct fib6_info *rt)
4506 {
4507         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4508             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4509              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4510                 return true;
4511
4512         return false;
4513 }
4514
4515 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4516 {
4517         struct fib6_info *iter;
4518         int total = 0;
4519
4520         if (!rt6_is_dead(rt))
4521                 total += rt->fib6_nh->fib_nh_weight;
4522
4523         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4524                 if (!rt6_is_dead(iter))
4525                         total += iter->fib6_nh->fib_nh_weight;
4526         }
4527
4528         return total;
4529 }
4530
4531 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4532 {
4533         int upper_bound = -1;
4534
4535         if (!rt6_is_dead(rt)) {
4536                 *weight += rt->fib6_nh->fib_nh_weight;
4537                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4538                                                     total) - 1;
4539         }
4540         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4541 }
4542
4543 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4544 {
4545         struct fib6_info *iter;
4546         int weight = 0;
4547
4548         rt6_upper_bound_set(rt, &weight, total);
4549
4550         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4551                 rt6_upper_bound_set(iter, &weight, total);
4552 }
4553
4554 void rt6_multipath_rebalance(struct fib6_info *rt)
4555 {
4556         struct fib6_info *first;
4557         int total;
4558
4559         /* In case the entire multipath route was marked for flushing,
4560          * then there is no need to rebalance upon the removal of every
4561          * sibling route.
4562          */
4563         if (!rt->fib6_nsiblings || rt->should_flush)
4564                 return;
4565
4566         /* During lookup routes are evaluated in order, so we need to
4567          * make sure upper bounds are assigned from the first sibling
4568          * onwards.
4569          */
4570         first = rt6_multipath_first_sibling(rt);
4571         if (WARN_ON_ONCE(!first))
4572                 return;
4573
4574         total = rt6_multipath_total_weight(first);
4575         rt6_multipath_upper_bound_set(first, total);
4576 }
4577
4578 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4579 {
4580         const struct arg_netdev_event *arg = p_arg;
4581         struct net *net = dev_net(arg->dev);
4582
4583         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4584             rt->fib6_nh->fib_nh_dev == arg->dev) {
4585                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4586                 fib6_update_sernum_upto_root(net, rt);
4587                 rt6_multipath_rebalance(rt);
4588         }
4589
4590         return 0;
4591 }
4592
4593 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4594 {
4595         struct arg_netdev_event arg = {
4596                 .dev = dev,
4597                 {
4598                         .nh_flags = nh_flags,
4599                 },
4600         };
4601
4602         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4603                 arg.nh_flags |= RTNH_F_LINKDOWN;
4604
4605         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4606 }
4607
4608 /* only called for fib entries with inline fib6_nh */
4609 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4610                                    const struct net_device *dev)
4611 {
4612         struct fib6_info *iter;
4613
4614         if (rt->fib6_nh->fib_nh_dev == dev)
4615                 return true;
4616         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4617                 if (iter->fib6_nh->fib_nh_dev == dev)
4618                         return true;
4619
4620         return false;
4621 }
4622
4623 static void rt6_multipath_flush(struct fib6_info *rt)
4624 {
4625         struct fib6_info *iter;
4626
4627         rt->should_flush = 1;
4628         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4629                 iter->should_flush = 1;
4630 }
4631
4632 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4633                                              const struct net_device *down_dev)
4634 {
4635         struct fib6_info *iter;
4636         unsigned int dead = 0;
4637
4638         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4639             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4640                 dead++;
4641         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4642                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4643                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4644                         dead++;
4645
4646         return dead;
4647 }
4648
4649 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4650                                        const struct net_device *dev,
4651                                        unsigned char nh_flags)
4652 {
4653         struct fib6_info *iter;
4654
4655         if (rt->fib6_nh->fib_nh_dev == dev)
4656                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4657         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4658                 if (iter->fib6_nh->fib_nh_dev == dev)
4659                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4660 }
4661
4662 /* called with write lock held for table with rt */
4663 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4664 {
4665         const struct arg_netdev_event *arg = p_arg;
4666         const struct net_device *dev = arg->dev;
4667         struct net *net = dev_net(dev);
4668
4669         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4670                 return 0;
4671
4672         switch (arg->event) {
4673         case NETDEV_UNREGISTER:
4674                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4675         case NETDEV_DOWN:
4676                 if (rt->should_flush)
4677                         return -1;
4678                 if (!rt->fib6_nsiblings)
4679                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4680                 if (rt6_multipath_uses_dev(rt, dev)) {
4681                         unsigned int count;
4682
4683                         count = rt6_multipath_dead_count(rt, dev);
4684                         if (rt->fib6_nsiblings + 1 == count) {
4685                                 rt6_multipath_flush(rt);
4686                                 return -1;
4687                         }
4688                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4689                                                    RTNH_F_LINKDOWN);
4690                         fib6_update_sernum(net, rt);
4691                         rt6_multipath_rebalance(rt);
4692                 }
4693                 return -2;
4694         case NETDEV_CHANGE:
4695                 if (rt->fib6_nh->fib_nh_dev != dev ||
4696                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4697                         break;
4698                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4699                 rt6_multipath_rebalance(rt);
4700                 break;
4701         }
4702
4703         return 0;
4704 }
4705
4706 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4707 {
4708         struct arg_netdev_event arg = {
4709                 .dev = dev,
4710                 {
4711                         .event = event,
4712                 },
4713         };
4714         struct net *net = dev_net(dev);
4715
4716         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4717                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4718         else
4719                 fib6_clean_all(net, fib6_ifdown, &arg);
4720 }
4721
4722 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4723 {
4724         rt6_sync_down_dev(dev, event);
4725         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4726         neigh_ifdown(&nd_tbl, dev);
4727 }
4728
4729 struct rt6_mtu_change_arg {
4730         struct net_device *dev;
4731         unsigned int mtu;
4732         struct fib6_info *f6i;
4733 };
4734
4735 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4736 {
4737         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4738         struct fib6_info *f6i = arg->f6i;
4739
4740         /* For administrative MTU increase, there is no way to discover
4741          * IPv6 PMTU increase, so PMTU increase should be updated here.
4742          * Since RFC 1981 doesn't include administrative MTU increase
4743          * update PMTU increase is a MUST. (i.e. jumbo frame)
4744          */
4745         if (nh->fib_nh_dev == arg->dev) {
4746                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4747                 u32 mtu = f6i->fib6_pmtu;
4748
4749                 if (mtu >= arg->mtu ||
4750                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4751                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4752
4753                 spin_lock_bh(&rt6_exception_lock);
4754                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4755                 spin_unlock_bh(&rt6_exception_lock);
4756         }
4757
4758         return 0;
4759 }
4760
4761 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4762 {
4763         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4764         struct inet6_dev *idev;
4765
4766         /* In IPv6 pmtu discovery is not optional,
4767            so that RTAX_MTU lock cannot disable it.
4768            We still use this lock to block changes
4769            caused by addrconf/ndisc.
4770         */
4771
4772         idev = __in6_dev_get(arg->dev);
4773         if (!idev)
4774                 return 0;
4775
4776         if (fib6_metric_locked(f6i, RTAX_MTU))
4777                 return 0;
4778
4779         arg->f6i = f6i;
4780         if (f6i->nh) {
4781                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4782                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4783                                                 arg);
4784         }
4785
4786         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4787 }
4788
4789 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4790 {
4791         struct rt6_mtu_change_arg arg = {
4792                 .dev = dev,
4793                 .mtu = mtu,
4794         };
4795
4796         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4797 }
4798
4799 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4800         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4801         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4802         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4803         [RTA_OIF]               = { .type = NLA_U32 },
4804         [RTA_IIF]               = { .type = NLA_U32 },
4805         [RTA_PRIORITY]          = { .type = NLA_U32 },
4806         [RTA_METRICS]           = { .type = NLA_NESTED },
4807         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4808         [RTA_PREF]              = { .type = NLA_U8 },
4809         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4810         [RTA_ENCAP]             = { .type = NLA_NESTED },
4811         [RTA_EXPIRES]           = { .type = NLA_U32 },
4812         [RTA_UID]               = { .type = NLA_U32 },
4813         [RTA_MARK]              = { .type = NLA_U32 },
4814         [RTA_TABLE]             = { .type = NLA_U32 },
4815         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4816         [RTA_SPORT]             = { .type = NLA_U16 },
4817         [RTA_DPORT]             = { .type = NLA_U16 },
4818         [RTA_NH_ID]             = { .type = NLA_U32 },
4819 };
4820
4821 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4822                               struct fib6_config *cfg,
4823                               struct netlink_ext_ack *extack)
4824 {
4825         struct rtmsg *rtm;
4826         struct nlattr *tb[RTA_MAX+1];
4827         unsigned int pref;
4828         int err;
4829
4830         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4831                                      rtm_ipv6_policy, extack);
4832         if (err < 0)
4833                 goto errout;
4834
4835         err = -EINVAL;
4836         rtm = nlmsg_data(nlh);
4837
4838         *cfg = (struct fib6_config){
4839                 .fc_table = rtm->rtm_table,
4840                 .fc_dst_len = rtm->rtm_dst_len,
4841                 .fc_src_len = rtm->rtm_src_len,
4842                 .fc_flags = RTF_UP,
4843                 .fc_protocol = rtm->rtm_protocol,
4844                 .fc_type = rtm->rtm_type,
4845
4846                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4847                 .fc_nlinfo.nlh = nlh,
4848                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4849         };
4850
4851         if (rtm->rtm_type == RTN_UNREACHABLE ||
4852             rtm->rtm_type == RTN_BLACKHOLE ||
4853             rtm->rtm_type == RTN_PROHIBIT ||
4854             rtm->rtm_type == RTN_THROW)
4855                 cfg->fc_flags |= RTF_REJECT;
4856
4857         if (rtm->rtm_type == RTN_LOCAL)
4858                 cfg->fc_flags |= RTF_LOCAL;
4859
4860         if (rtm->rtm_flags & RTM_F_CLONED)
4861                 cfg->fc_flags |= RTF_CACHE;
4862
4863         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4864
4865         if (tb[RTA_NH_ID]) {
4866                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4867                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4868                         NL_SET_ERR_MSG(extack,
4869                                        "Nexthop specification and nexthop id are mutually exclusive");
4870                         goto errout;
4871                 }
4872                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4873         }
4874
4875         if (tb[RTA_GATEWAY]) {
4876                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4877                 cfg->fc_flags |= RTF_GATEWAY;
4878         }
4879         if (tb[RTA_VIA]) {
4880                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4881                 goto errout;
4882         }
4883
4884         if (tb[RTA_DST]) {
4885                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4886
4887                 if (nla_len(tb[RTA_DST]) < plen)
4888                         goto errout;
4889
4890                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4891         }
4892
4893         if (tb[RTA_SRC]) {
4894                 int plen = (rtm->rtm_src_len + 7) >> 3;
4895
4896                 if (nla_len(tb[RTA_SRC]) < plen)
4897                         goto errout;
4898
4899                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4900         }
4901
4902         if (tb[RTA_PREFSRC])
4903                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4904
4905         if (tb[RTA_OIF])
4906                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4907
4908         if (tb[RTA_PRIORITY])
4909                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4910
4911         if (tb[RTA_METRICS]) {
4912                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4913                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4914         }
4915
4916         if (tb[RTA_TABLE])
4917                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4918
4919         if (tb[RTA_MULTIPATH]) {
4920                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4921                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4922
4923                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4924                                                      cfg->fc_mp_len, extack);
4925                 if (err < 0)
4926                         goto errout;
4927         }
4928
4929         if (tb[RTA_PREF]) {
4930                 pref = nla_get_u8(tb[RTA_PREF]);
4931                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4932                     pref != ICMPV6_ROUTER_PREF_HIGH)
4933                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4934                 cfg->fc_flags |= RTF_PREF(pref);
4935         }
4936
4937         if (tb[RTA_ENCAP])
4938                 cfg->fc_encap = tb[RTA_ENCAP];
4939
4940         if (tb[RTA_ENCAP_TYPE]) {
4941                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4942
4943                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4944                 if (err < 0)
4945                         goto errout;
4946         }
4947
4948         if (tb[RTA_EXPIRES]) {
4949                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4950
4951                 if (addrconf_finite_timeout(timeout)) {
4952                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
4953                         cfg->fc_flags |= RTF_EXPIRES;
4954                 }
4955         }
4956
4957         err = 0;
4958 errout:
4959         return err;
4960 }
4961
4962 struct rt6_nh {
4963         struct fib6_info *fib6_info;
4964         struct fib6_config r_cfg;
4965         struct list_head next;
4966 };
4967
4968 static int ip6_route_info_append(struct net *net,
4969                                  struct list_head *rt6_nh_list,
4970                                  struct fib6_info *rt,
4971                                  struct fib6_config *r_cfg)
4972 {
4973         struct rt6_nh *nh;
4974         int err = -EEXIST;
4975
4976         list_for_each_entry(nh, rt6_nh_list, next) {
4977                 /* check if fib6_info already exists */
4978                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
4979                         return err;
4980         }
4981
4982         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
4983         if (!nh)
4984                 return -ENOMEM;
4985         nh->fib6_info = rt;
4986         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
4987         list_add_tail(&nh->next, rt6_nh_list);
4988
4989         return 0;
4990 }
4991
4992 static void ip6_route_mpath_notify(struct fib6_info *rt,
4993                                    struct fib6_info *rt_last,
4994                                    struct nl_info *info,
4995                                    __u16 nlflags)
4996 {
4997         /* if this is an APPEND route, then rt points to the first route
4998          * inserted and rt_last points to last route inserted. Userspace
4999          * wants a consistent dump of the route which starts at the first
5000          * nexthop. Since sibling routes are always added at the end of
5001          * the list, find the first sibling of the last route appended
5002          */
5003         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5004                 rt = list_first_entry(&rt_last->fib6_siblings,
5005                                       struct fib6_info,
5006                                       fib6_siblings);
5007         }
5008
5009         if (rt)
5010                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5011 }
5012
5013 static int ip6_route_multipath_add(struct fib6_config *cfg,
5014                                    struct netlink_ext_ack *extack)
5015 {
5016         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5017         struct nl_info *info = &cfg->fc_nlinfo;
5018         enum fib_event_type event_type;
5019         struct fib6_config r_cfg;
5020         struct rtnexthop *rtnh;
5021         struct fib6_info *rt;
5022         struct rt6_nh *err_nh;
5023         struct rt6_nh *nh, *nh_safe;
5024         __u16 nlflags;
5025         int remaining;
5026         int attrlen;
5027         int err = 1;
5028         int nhn = 0;
5029         int replace = (cfg->fc_nlinfo.nlh &&
5030                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5031         LIST_HEAD(rt6_nh_list);
5032
5033         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5034         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5035                 nlflags |= NLM_F_APPEND;
5036
5037         remaining = cfg->fc_mp_len;
5038         rtnh = (struct rtnexthop *)cfg->fc_mp;
5039
5040         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5041          * fib6_info structs per nexthop
5042          */
5043         while (rtnh_ok(rtnh, remaining)) {
5044                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5045                 if (rtnh->rtnh_ifindex)
5046                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5047
5048                 attrlen = rtnh_attrlen(rtnh);
5049                 if (attrlen > 0) {
5050                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5051
5052                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5053                         if (nla) {
5054                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5055                                 r_cfg.fc_flags |= RTF_GATEWAY;
5056                         }
5057                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5058                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5059                         if (nla)
5060                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5061                 }
5062
5063                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5064                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5065                 if (IS_ERR(rt)) {
5066                         err = PTR_ERR(rt);
5067                         rt = NULL;
5068                         goto cleanup;
5069                 }
5070                 if (!rt6_qualify_for_ecmp(rt)) {
5071                         err = -EINVAL;
5072                         NL_SET_ERR_MSG(extack,
5073                                        "Device only routes can not be added for IPv6 using the multipath API.");
5074                         fib6_info_release(rt);
5075                         goto cleanup;
5076                 }
5077
5078                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5079
5080                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5081                                             rt, &r_cfg);
5082                 if (err) {
5083                         fib6_info_release(rt);
5084                         goto cleanup;
5085                 }
5086
5087                 rtnh = rtnh_next(rtnh, &remaining);
5088         }
5089
5090         if (list_empty(&rt6_nh_list)) {
5091                 NL_SET_ERR_MSG(extack,
5092                                "Invalid nexthop configuration - no valid nexthops");
5093                 return -EINVAL;
5094         }
5095
5096         /* for add and replace send one notification with all nexthops.
5097          * Skip the notification in fib6_add_rt2node and send one with
5098          * the full route when done
5099          */
5100         info->skip_notify = 1;
5101
5102         /* For add and replace, send one notification with all nexthops. For
5103          * append, send one notification with all appended nexthops.
5104          */
5105         info->skip_notify_kernel = 1;
5106
5107         err_nh = NULL;
5108         list_for_each_entry(nh, &rt6_nh_list, next) {
5109                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5110                 fib6_info_release(nh->fib6_info);
5111
5112                 if (!err) {
5113                         /* save reference to last route successfully inserted */
5114                         rt_last = nh->fib6_info;
5115
5116                         /* save reference to first route for notification */
5117                         if (!rt_notif)
5118                                 rt_notif = nh->fib6_info;
5119                 }
5120
5121                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5122                 nh->fib6_info = NULL;
5123                 if (err) {
5124                         if (replace && nhn)
5125                                 NL_SET_ERR_MSG_MOD(extack,
5126                                                    "multipath route replace failed (check consistency of installed routes)");
5127                         err_nh = nh;
5128                         goto add_errout;
5129                 }
5130
5131                 /* Because each route is added like a single route we remove
5132                  * these flags after the first nexthop: if there is a collision,
5133                  * we have already failed to add the first nexthop:
5134                  * fib6_add_rt2node() has rejected it; when replacing, old
5135                  * nexthops have been replaced by first new, the rest should
5136                  * be added to it.
5137                  */
5138                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5139                                                      NLM_F_REPLACE);
5140                 nhn++;
5141         }
5142
5143         event_type = replace ? FIB_EVENT_ENTRY_REPLACE : FIB_EVENT_ENTRY_ADD;
5144         err = call_fib6_multipath_entry_notifiers(info->nl_net, event_type,
5145                                                   rt_notif, nhn - 1, extack);
5146         if (err) {
5147                 /* Delete all the siblings that were just added */
5148                 err_nh = NULL;
5149                 goto add_errout;
5150         }
5151
5152         /* success ... tell user about new route */
5153         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5154         goto cleanup;
5155
5156 add_errout:
5157         /* send notification for routes that were added so that
5158          * the delete notifications sent by ip6_route_del are
5159          * coherent
5160          */
5161         if (rt_notif)
5162                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5163
5164         /* Delete routes that were already added */
5165         list_for_each_entry(nh, &rt6_nh_list, next) {
5166                 if (err_nh == nh)
5167                         break;
5168                 ip6_route_del(&nh->r_cfg, extack);
5169         }
5170
5171 cleanup:
5172         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5173                 if (nh->fib6_info)
5174                         fib6_info_release(nh->fib6_info);
5175                 list_del(&nh->next);
5176                 kfree(nh);
5177         }
5178
5179         return err;
5180 }
5181
5182 static int ip6_route_multipath_del(struct fib6_config *cfg,
5183                                    struct netlink_ext_ack *extack)
5184 {
5185         struct fib6_config r_cfg;
5186         struct rtnexthop *rtnh;
5187         int remaining;
5188         int attrlen;
5189         int err = 1, last_err = 0;
5190
5191         remaining = cfg->fc_mp_len;
5192         rtnh = (struct rtnexthop *)cfg->fc_mp;
5193
5194         /* Parse a Multipath Entry */
5195         while (rtnh_ok(rtnh, remaining)) {
5196                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5197                 if (rtnh->rtnh_ifindex)
5198                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5199
5200                 attrlen = rtnh_attrlen(rtnh);
5201                 if (attrlen > 0) {
5202                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5203
5204                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5205                         if (nla) {
5206                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5207                                 r_cfg.fc_flags |= RTF_GATEWAY;
5208                         }
5209                 }
5210                 err = ip6_route_del(&r_cfg, extack);
5211                 if (err)
5212                         last_err = err;
5213
5214                 rtnh = rtnh_next(rtnh, &remaining);
5215         }
5216
5217         return last_err;
5218 }
5219
5220 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5221                               struct netlink_ext_ack *extack)
5222 {
5223         struct fib6_config cfg;
5224         int err;
5225
5226         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5227         if (err < 0)
5228                 return err;
5229
5230         if (cfg.fc_nh_id &&
5231             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5232                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5233                 return -EINVAL;
5234         }
5235
5236         if (cfg.fc_mp)
5237                 return ip6_route_multipath_del(&cfg, extack);
5238         else {
5239                 cfg.fc_delete_all_nh = 1;
5240                 return ip6_route_del(&cfg, extack);
5241         }
5242 }
5243
5244 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5245                               struct netlink_ext_ack *extack)
5246 {
5247         struct fib6_config cfg;
5248         int err;
5249
5250         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5251         if (err < 0)
5252                 return err;
5253
5254         if (cfg.fc_metric == 0)
5255                 cfg.fc_metric = IP6_RT_PRIO_USER;
5256
5257         if (cfg.fc_mp)
5258                 return ip6_route_multipath_add(&cfg, extack);
5259         else
5260                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5261 }
5262
5263 /* add the overhead of this fib6_nh to nexthop_len */
5264 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5265 {
5266         int *nexthop_len = arg;
5267
5268         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5269                      + NLA_ALIGN(sizeof(struct rtnexthop))
5270                      + nla_total_size(16); /* RTA_GATEWAY */
5271
5272         if (nh->fib_nh_lws) {
5273                 /* RTA_ENCAP_TYPE */
5274                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5275                 /* RTA_ENCAP */
5276                 *nexthop_len += nla_total_size(2);
5277         }
5278
5279         return 0;
5280 }
5281
5282 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5283 {
5284         int nexthop_len;
5285
5286         if (f6i->nh) {
5287                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5288                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5289                                          &nexthop_len);
5290         } else {
5291                 struct fib6_nh *nh = f6i->fib6_nh;
5292
5293                 nexthop_len = 0;
5294                 if (f6i->fib6_nsiblings) {
5295                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5296                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5297                                     + nla_total_size(16) /* RTA_GATEWAY */
5298                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5299
5300                         nexthop_len *= f6i->fib6_nsiblings;
5301                 }
5302                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5303         }
5304
5305         return NLMSG_ALIGN(sizeof(struct rtmsg))
5306                + nla_total_size(16) /* RTA_SRC */
5307                + nla_total_size(16) /* RTA_DST */
5308                + nla_total_size(16) /* RTA_GATEWAY */
5309                + nla_total_size(16) /* RTA_PREFSRC */
5310                + nla_total_size(4) /* RTA_TABLE */
5311                + nla_total_size(4) /* RTA_IIF */
5312                + nla_total_size(4) /* RTA_OIF */
5313                + nla_total_size(4) /* RTA_PRIORITY */
5314                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5315                + nla_total_size(sizeof(struct rta_cacheinfo))
5316                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5317                + nla_total_size(1) /* RTA_PREF */
5318                + nexthop_len;
5319 }
5320
5321 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5322                                  unsigned char *flags)
5323 {
5324         if (nexthop_is_multipath(nh)) {
5325                 struct nlattr *mp;
5326
5327                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5328                 if (!mp)
5329                         goto nla_put_failure;
5330
5331                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5332                         goto nla_put_failure;
5333
5334                 nla_nest_end(skb, mp);
5335         } else {
5336                 struct fib6_nh *fib6_nh;
5337
5338                 fib6_nh = nexthop_fib6_nh(nh);
5339                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5340                                      flags, false) < 0)
5341                         goto nla_put_failure;
5342         }
5343
5344         return 0;
5345
5346 nla_put_failure:
5347         return -EMSGSIZE;
5348 }
5349
5350 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5351                          struct fib6_info *rt, struct dst_entry *dst,
5352                          struct in6_addr *dest, struct in6_addr *src,
5353                          int iif, int type, u32 portid, u32 seq,
5354                          unsigned int flags)
5355 {
5356         struct rt6_info *rt6 = (struct rt6_info *)dst;
5357         struct rt6key *rt6_dst, *rt6_src;
5358         u32 *pmetrics, table, rt6_flags;
5359         unsigned char nh_flags = 0;
5360         struct nlmsghdr *nlh;
5361         struct rtmsg *rtm;
5362         long expires = 0;
5363
5364         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5365         if (!nlh)
5366                 return -EMSGSIZE;
5367
5368         if (rt6) {
5369                 rt6_dst = &rt6->rt6i_dst;
5370                 rt6_src = &rt6->rt6i_src;
5371                 rt6_flags = rt6->rt6i_flags;
5372         } else {
5373                 rt6_dst = &rt->fib6_dst;
5374                 rt6_src = &rt->fib6_src;
5375                 rt6_flags = rt->fib6_flags;
5376         }
5377
5378         rtm = nlmsg_data(nlh);
5379         rtm->rtm_family = AF_INET6;
5380         rtm->rtm_dst_len = rt6_dst->plen;
5381         rtm->rtm_src_len = rt6_src->plen;
5382         rtm->rtm_tos = 0;
5383         if (rt->fib6_table)
5384                 table = rt->fib6_table->tb6_id;
5385         else
5386                 table = RT6_TABLE_UNSPEC;
5387         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5388         if (nla_put_u32(skb, RTA_TABLE, table))
5389                 goto nla_put_failure;
5390
5391         rtm->rtm_type = rt->fib6_type;
5392         rtm->rtm_flags = 0;
5393         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5394         rtm->rtm_protocol = rt->fib6_protocol;
5395
5396         if (rt6_flags & RTF_CACHE)
5397                 rtm->rtm_flags |= RTM_F_CLONED;
5398
5399         if (dest) {
5400                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5401                         goto nla_put_failure;
5402                 rtm->rtm_dst_len = 128;
5403         } else if (rtm->rtm_dst_len)
5404                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5405                         goto nla_put_failure;
5406 #ifdef CONFIG_IPV6_SUBTREES
5407         if (src) {
5408                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5409                         goto nla_put_failure;
5410                 rtm->rtm_src_len = 128;
5411         } else if (rtm->rtm_src_len &&
5412                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5413                 goto nla_put_failure;
5414 #endif
5415         if (iif) {
5416 #ifdef CONFIG_IPV6_MROUTE
5417                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5418                         int err = ip6mr_get_route(net, skb, rtm, portid);
5419
5420                         if (err == 0)
5421                                 return 0;
5422                         if (err < 0)
5423                                 goto nla_put_failure;
5424                 } else
5425 #endif
5426                         if (nla_put_u32(skb, RTA_IIF, iif))
5427                                 goto nla_put_failure;
5428         } else if (dest) {
5429                 struct in6_addr saddr_buf;
5430                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5431                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5432                         goto nla_put_failure;
5433         }
5434
5435         if (rt->fib6_prefsrc.plen) {
5436                 struct in6_addr saddr_buf;
5437                 saddr_buf = rt->fib6_prefsrc.addr;
5438                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5439                         goto nla_put_failure;
5440         }
5441
5442         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5443         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5444                 goto nla_put_failure;
5445
5446         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5447                 goto nla_put_failure;
5448
5449         /* For multipath routes, walk the siblings list and add
5450          * each as a nexthop within RTA_MULTIPATH.
5451          */
5452         if (rt6) {
5453                 if (rt6_flags & RTF_GATEWAY &&
5454                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5455                         goto nla_put_failure;
5456
5457                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5458                         goto nla_put_failure;
5459         } else if (rt->fib6_nsiblings) {
5460                 struct fib6_info *sibling, *next_sibling;
5461                 struct nlattr *mp;
5462
5463                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5464                 if (!mp)
5465                         goto nla_put_failure;
5466
5467                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5468                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5469                         goto nla_put_failure;
5470
5471                 list_for_each_entry_safe(sibling, next_sibling,
5472                                          &rt->fib6_siblings, fib6_siblings) {
5473                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5474                                             sibling->fib6_nh->fib_nh_weight,
5475                                             AF_INET6) < 0)
5476                                 goto nla_put_failure;
5477                 }
5478
5479                 nla_nest_end(skb, mp);
5480         } else if (rt->nh) {
5481                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5482                         goto nla_put_failure;
5483
5484                 if (nexthop_is_blackhole(rt->nh))
5485                         rtm->rtm_type = RTN_BLACKHOLE;
5486
5487                 if (rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5488                         goto nla_put_failure;
5489
5490                 rtm->rtm_flags |= nh_flags;
5491         } else {
5492                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5493                                      &nh_flags, false) < 0)
5494                         goto nla_put_failure;
5495
5496                 rtm->rtm_flags |= nh_flags;
5497         }
5498
5499         if (rt6_flags & RTF_EXPIRES) {
5500                 expires = dst ? dst->expires : rt->expires;
5501                 expires -= jiffies;
5502         }
5503
5504         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5505                 goto nla_put_failure;
5506
5507         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5508                 goto nla_put_failure;
5509
5510
5511         nlmsg_end(skb, nlh);
5512         return 0;
5513
5514 nla_put_failure:
5515         nlmsg_cancel(skb, nlh);
5516         return -EMSGSIZE;
5517 }
5518
5519 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5520 {
5521         const struct net_device *dev = arg;
5522
5523         if (nh->fib_nh_dev == dev)
5524                 return 1;
5525
5526         return 0;
5527 }
5528
5529 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5530                                const struct net_device *dev)
5531 {
5532         if (f6i->nh) {
5533                 struct net_device *_dev = (struct net_device *)dev;
5534
5535                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5536                                                   fib6_info_nh_uses_dev,
5537                                                   _dev);
5538         }
5539
5540         if (f6i->fib6_nh->fib_nh_dev == dev)
5541                 return true;
5542
5543         if (f6i->fib6_nsiblings) {
5544                 struct fib6_info *sibling, *next_sibling;
5545
5546                 list_for_each_entry_safe(sibling, next_sibling,
5547                                          &f6i->fib6_siblings, fib6_siblings) {
5548                         if (sibling->fib6_nh->fib_nh_dev == dev)
5549                                 return true;
5550                 }
5551         }
5552
5553         return false;
5554 }
5555
5556 struct fib6_nh_exception_dump_walker {
5557         struct rt6_rtnl_dump_arg *dump;
5558         struct fib6_info *rt;
5559         unsigned int flags;
5560         unsigned int skip;
5561         unsigned int count;
5562 };
5563
5564 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5565 {
5566         struct fib6_nh_exception_dump_walker *w = arg;
5567         struct rt6_rtnl_dump_arg *dump = w->dump;
5568         struct rt6_exception_bucket *bucket;
5569         struct rt6_exception *rt6_ex;
5570         int i, err;
5571
5572         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5573         if (!bucket)
5574                 return 0;
5575
5576         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5577                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5578                         if (w->skip) {
5579                                 w->skip--;
5580                                 continue;
5581                         }
5582
5583                         /* Expiration of entries doesn't bump sernum, insertion
5584                          * does. Removal is triggered by insertion, so we can
5585                          * rely on the fact that if entries change between two
5586                          * partial dumps, this node is scanned again completely,
5587                          * see rt6_insert_exception() and fib6_dump_table().
5588                          *
5589                          * Count expired entries we go through as handled
5590                          * entries that we'll skip next time, in case of partial
5591                          * node dump. Otherwise, if entries expire meanwhile,
5592                          * we'll skip the wrong amount.
5593                          */
5594                         if (rt6_check_expired(rt6_ex->rt6i)) {
5595                                 w->count++;
5596                                 continue;
5597                         }
5598
5599                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5600                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5601                                             RTM_NEWROUTE,
5602                                             NETLINK_CB(dump->cb->skb).portid,
5603                                             dump->cb->nlh->nlmsg_seq, w->flags);
5604                         if (err)
5605                                 return err;
5606
5607                         w->count++;
5608                 }
5609                 bucket++;
5610         }
5611
5612         return 0;
5613 }
5614
5615 /* Return -1 if done with node, number of handled routes on partial dump */
5616 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5617 {
5618         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5619         struct fib_dump_filter *filter = &arg->filter;
5620         unsigned int flags = NLM_F_MULTI;
5621         struct net *net = arg->net;
5622         int count = 0;
5623
5624         if (rt == net->ipv6.fib6_null_entry)
5625                 return -1;
5626
5627         if ((filter->flags & RTM_F_PREFIX) &&
5628             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5629                 /* success since this is not a prefix route */
5630                 return -1;
5631         }
5632         if (filter->filter_set &&
5633             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5634              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5635              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5636                 return -1;
5637         }
5638
5639         if (filter->filter_set ||
5640             !filter->dump_routes || !filter->dump_exceptions) {
5641                 flags |= NLM_F_DUMP_FILTERED;
5642         }
5643
5644         if (filter->dump_routes) {
5645                 if (skip) {
5646                         skip--;
5647                 } else {
5648                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5649                                           0, RTM_NEWROUTE,
5650                                           NETLINK_CB(arg->cb->skb).portid,
5651                                           arg->cb->nlh->nlmsg_seq, flags)) {
5652                                 return 0;
5653                         }
5654                         count++;
5655                 }
5656         }
5657
5658         if (filter->dump_exceptions) {
5659                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5660                                                            .rt = rt,
5661                                                            .flags = flags,
5662                                                            .skip = skip,
5663                                                            .count = 0 };
5664                 int err;
5665
5666                 rcu_read_lock();
5667                 if (rt->nh) {
5668                         err = nexthop_for_each_fib6_nh(rt->nh,
5669                                                        rt6_nh_dump_exceptions,
5670                                                        &w);
5671                 } else {
5672                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5673                 }
5674                 rcu_read_unlock();
5675
5676                 if (err)
5677                         return count += w.count;
5678         }
5679
5680         return -1;
5681 }
5682
5683 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5684                                         const struct nlmsghdr *nlh,
5685                                         struct nlattr **tb,
5686                                         struct netlink_ext_ack *extack)
5687 {
5688         struct rtmsg *rtm;
5689         int i, err;
5690
5691         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5692                 NL_SET_ERR_MSG_MOD(extack,
5693                                    "Invalid header for get route request");
5694                 return -EINVAL;
5695         }
5696
5697         if (!netlink_strict_get_check(skb))
5698                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5699                                               rtm_ipv6_policy, extack);
5700
5701         rtm = nlmsg_data(nlh);
5702         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5703             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5704             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5705             rtm->rtm_type) {
5706                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5707                 return -EINVAL;
5708         }
5709         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5710                 NL_SET_ERR_MSG_MOD(extack,
5711                                    "Invalid flags for get route request");
5712                 return -EINVAL;
5713         }
5714
5715         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5716                                             rtm_ipv6_policy, extack);
5717         if (err)
5718                 return err;
5719
5720         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5721             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5722                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5723                 return -EINVAL;
5724         }
5725
5726         for (i = 0; i <= RTA_MAX; i++) {
5727                 if (!tb[i])
5728                         continue;
5729
5730                 switch (i) {
5731                 case RTA_SRC:
5732                 case RTA_DST:
5733                 case RTA_IIF:
5734                 case RTA_OIF:
5735                 case RTA_MARK:
5736                 case RTA_UID:
5737                 case RTA_SPORT:
5738                 case RTA_DPORT:
5739                 case RTA_IP_PROTO:
5740                         break;
5741                 default:
5742                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5743                         return -EINVAL;
5744                 }
5745         }
5746
5747         return 0;
5748 }
5749
5750 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5751                               struct netlink_ext_ack *extack)
5752 {
5753         struct net *net = sock_net(in_skb->sk);
5754         struct nlattr *tb[RTA_MAX+1];
5755         int err, iif = 0, oif = 0;
5756         struct fib6_info *from;
5757         struct dst_entry *dst;
5758         struct rt6_info *rt;
5759         struct sk_buff *skb;
5760         struct rtmsg *rtm;
5761         struct flowi6 fl6 = {};
5762         bool fibmatch;
5763
5764         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5765         if (err < 0)
5766                 goto errout;
5767
5768         err = -EINVAL;
5769         rtm = nlmsg_data(nlh);
5770         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5771         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5772
5773         if (tb[RTA_SRC]) {
5774                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5775                         goto errout;
5776
5777                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5778         }
5779
5780         if (tb[RTA_DST]) {
5781                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5782                         goto errout;
5783
5784                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5785         }
5786
5787         if (tb[RTA_IIF])
5788                 iif = nla_get_u32(tb[RTA_IIF]);
5789
5790         if (tb[RTA_OIF])
5791                 oif = nla_get_u32(tb[RTA_OIF]);
5792
5793         if (tb[RTA_MARK])
5794                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5795
5796         if (tb[RTA_UID])
5797                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5798                                            nla_get_u32(tb[RTA_UID]));
5799         else
5800                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5801
5802         if (tb[RTA_SPORT])
5803                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5804
5805         if (tb[RTA_DPORT])
5806                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5807
5808         if (tb[RTA_IP_PROTO]) {
5809                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5810                                                   &fl6.flowi6_proto, AF_INET6,
5811                                                   extack);
5812                 if (err)
5813                         goto errout;
5814         }
5815
5816         if (iif) {
5817                 struct net_device *dev;
5818                 int flags = 0;
5819
5820                 rcu_read_lock();
5821
5822                 dev = dev_get_by_index_rcu(net, iif);
5823                 if (!dev) {
5824                         rcu_read_unlock();
5825                         err = -ENODEV;
5826                         goto errout;
5827                 }
5828
5829                 fl6.flowi6_iif = iif;
5830
5831                 if (!ipv6_addr_any(&fl6.saddr))
5832                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5833
5834                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5835
5836                 rcu_read_unlock();
5837         } else {
5838                 fl6.flowi6_oif = oif;
5839
5840                 dst = ip6_route_output(net, NULL, &fl6);
5841         }
5842
5843
5844         rt = container_of(dst, struct rt6_info, dst);
5845         if (rt->dst.error) {
5846                 err = rt->dst.error;
5847                 ip6_rt_put(rt);
5848                 goto errout;
5849         }
5850
5851         if (rt == net->ipv6.ip6_null_entry) {
5852                 err = rt->dst.error;
5853                 ip6_rt_put(rt);
5854                 goto errout;
5855         }
5856
5857         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5858         if (!skb) {
5859                 ip6_rt_put(rt);
5860                 err = -ENOBUFS;
5861                 goto errout;
5862         }
5863
5864         skb_dst_set(skb, &rt->dst);
5865
5866         rcu_read_lock();
5867         from = rcu_dereference(rt->from);
5868         if (from) {
5869                 if (fibmatch)
5870                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5871                                             iif, RTM_NEWROUTE,
5872                                             NETLINK_CB(in_skb).portid,
5873                                             nlh->nlmsg_seq, 0);
5874                 else
5875                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5876                                             &fl6.saddr, iif, RTM_NEWROUTE,
5877                                             NETLINK_CB(in_skb).portid,
5878                                             nlh->nlmsg_seq, 0);
5879         } else {
5880                 err = -ENETUNREACH;
5881         }
5882         rcu_read_unlock();
5883
5884         if (err < 0) {
5885                 kfree_skb(skb);
5886                 goto errout;
5887         }
5888
5889         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5890 errout:
5891         return err;
5892 }
5893
5894 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
5895                      unsigned int nlm_flags)
5896 {
5897         struct sk_buff *skb;
5898         struct net *net = info->nl_net;
5899         u32 seq;
5900         int err;
5901
5902         err = -ENOBUFS;
5903         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5904
5905         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5906         if (!skb)
5907                 goto errout;
5908
5909         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5910                             event, info->portid, seq, nlm_flags);
5911         if (err < 0) {
5912                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5913                 WARN_ON(err == -EMSGSIZE);
5914                 kfree_skb(skb);
5915                 goto errout;
5916         }
5917         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5918                     info->nlh, gfp_any());
5919         return;
5920 errout:
5921         if (err < 0)
5922                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5923 }
5924
5925 void fib6_rt_update(struct net *net, struct fib6_info *rt,
5926                     struct nl_info *info)
5927 {
5928         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5929         struct sk_buff *skb;
5930         int err = -ENOBUFS;
5931
5932         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
5933          * is implemented and supported for nexthop objects
5934          */
5935         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
5936
5937         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5938         if (!skb)
5939                 goto errout;
5940
5941         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5942                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
5943         if (err < 0) {
5944                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5945                 WARN_ON(err == -EMSGSIZE);
5946                 kfree_skb(skb);
5947                 goto errout;
5948         }
5949         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5950                     info->nlh, gfp_any());
5951         return;
5952 errout:
5953         if (err < 0)
5954                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5955 }
5956
5957 static int ip6_route_dev_notify(struct notifier_block *this,
5958                                 unsigned long event, void *ptr)
5959 {
5960         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
5961         struct net *net = dev_net(dev);
5962
5963         if (!(dev->flags & IFF_LOOPBACK))
5964                 return NOTIFY_OK;
5965
5966         if (event == NETDEV_REGISTER) {
5967                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
5968                 net->ipv6.ip6_null_entry->dst.dev = dev;
5969                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
5970 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5971                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
5972                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
5973                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
5974                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
5975 #endif
5976          } else if (event == NETDEV_UNREGISTER &&
5977                     dev->reg_state != NETREG_UNREGISTERED) {
5978                 /* NETDEV_UNREGISTER could be fired for multiple times by
5979                  * netdev_wait_allrefs(). Make sure we only call this once.
5980                  */
5981                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
5982 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5983                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
5984                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
5985 #endif
5986         }
5987
5988         return NOTIFY_OK;
5989 }
5990
5991 /*
5992  *      /proc
5993  */
5994
5995 #ifdef CONFIG_PROC_FS
5996 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
5997 {
5998         struct net *net = (struct net *)seq->private;
5999         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6000                    net->ipv6.rt6_stats->fib_nodes,
6001                    net->ipv6.rt6_stats->fib_route_nodes,
6002                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6003                    net->ipv6.rt6_stats->fib_rt_entries,
6004                    net->ipv6.rt6_stats->fib_rt_cache,
6005                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6006                    net->ipv6.rt6_stats->fib_discarded_routes);
6007
6008         return 0;
6009 }
6010 #endif  /* CONFIG_PROC_FS */
6011
6012 #ifdef CONFIG_SYSCTL
6013
6014 static
6015 int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6016                               void __user *buffer, size_t *lenp, loff_t *ppos)
6017 {
6018         struct net *net;
6019         int delay;
6020         int ret;
6021         if (!write)
6022                 return -EINVAL;
6023
6024         net = (struct net *)ctl->extra1;
6025         delay = net->ipv6.sysctl.flush_delay;
6026         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6027         if (ret)
6028                 return ret;
6029
6030         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6031         return 0;
6032 }
6033
6034 static struct ctl_table ipv6_route_table_template[] = {
6035         {
6036                 .procname       =       "flush",
6037                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6038                 .maxlen         =       sizeof(int),
6039                 .mode           =       0200,
6040                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6041         },
6042         {
6043                 .procname       =       "gc_thresh",
6044                 .data           =       &ip6_dst_ops_template.gc_thresh,
6045                 .maxlen         =       sizeof(int),
6046                 .mode           =       0644,
6047                 .proc_handler   =       proc_dointvec,
6048         },
6049         {
6050                 .procname       =       "max_size",
6051                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6052                 .maxlen         =       sizeof(int),
6053                 .mode           =       0644,
6054                 .proc_handler   =       proc_dointvec,
6055         },
6056         {
6057                 .procname       =       "gc_min_interval",
6058                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6059                 .maxlen         =       sizeof(int),
6060                 .mode           =       0644,
6061                 .proc_handler   =       proc_dointvec_jiffies,
6062         },
6063         {
6064                 .procname       =       "gc_timeout",
6065                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6066                 .maxlen         =       sizeof(int),
6067                 .mode           =       0644,
6068                 .proc_handler   =       proc_dointvec_jiffies,
6069         },
6070         {
6071                 .procname       =       "gc_interval",
6072                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6073                 .maxlen         =       sizeof(int),
6074                 .mode           =       0644,
6075                 .proc_handler   =       proc_dointvec_jiffies,
6076         },
6077         {
6078                 .procname       =       "gc_elasticity",
6079                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6080                 .maxlen         =       sizeof(int),
6081                 .mode           =       0644,
6082                 .proc_handler   =       proc_dointvec,
6083         },
6084         {
6085                 .procname       =       "mtu_expires",
6086                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6087                 .maxlen         =       sizeof(int),
6088                 .mode           =       0644,
6089                 .proc_handler   =       proc_dointvec_jiffies,
6090         },
6091         {
6092                 .procname       =       "min_adv_mss",
6093                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6094                 .maxlen         =       sizeof(int),
6095                 .mode           =       0644,
6096                 .proc_handler   =       proc_dointvec,
6097         },
6098         {
6099                 .procname       =       "gc_min_interval_ms",
6100                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6101                 .maxlen         =       sizeof(int),
6102                 .mode           =       0644,
6103                 .proc_handler   =       proc_dointvec_ms_jiffies,
6104         },
6105         {
6106                 .procname       =       "skip_notify_on_dev_down",
6107                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6108                 .maxlen         =       sizeof(int),
6109                 .mode           =       0644,
6110                 .proc_handler   =       proc_dointvec_minmax,
6111                 .extra1         =       SYSCTL_ZERO,
6112                 .extra2         =       SYSCTL_ONE,
6113         },
6114         { }
6115 };
6116
6117 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6118 {
6119         struct ctl_table *table;
6120
6121         table = kmemdup(ipv6_route_table_template,
6122                         sizeof(ipv6_route_table_template),
6123                         GFP_KERNEL);
6124
6125         if (table) {
6126                 table[0].data = &net->ipv6.sysctl.flush_delay;
6127                 table[0].extra1 = net;
6128                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6129                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6130                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6131                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6132                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6133                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6134                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6135                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6136                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6137                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6138
6139                 /* Don't export sysctls to unprivileged users */
6140                 if (net->user_ns != &init_user_ns)
6141                         table[0].procname = NULL;
6142         }
6143
6144         return table;
6145 }
6146 #endif
6147
6148 static int __net_init ip6_route_net_init(struct net *net)
6149 {
6150         int ret = -ENOMEM;
6151
6152         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6153                sizeof(net->ipv6.ip6_dst_ops));
6154
6155         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6156                 goto out_ip6_dst_ops;
6157
6158         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6159         if (!net->ipv6.fib6_null_entry)
6160                 goto out_ip6_dst_entries;
6161         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6162                sizeof(*net->ipv6.fib6_null_entry));
6163
6164         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6165                                            sizeof(*net->ipv6.ip6_null_entry),
6166                                            GFP_KERNEL);
6167         if (!net->ipv6.ip6_null_entry)
6168                 goto out_fib6_null_entry;
6169         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6170         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6171                          ip6_template_metrics, true);
6172         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6173
6174 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6175         net->ipv6.fib6_has_custom_rules = false;
6176         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6177                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6178                                                GFP_KERNEL);
6179         if (!net->ipv6.ip6_prohibit_entry)
6180                 goto out_ip6_null_entry;
6181         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6182         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6183                          ip6_template_metrics, true);
6184         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6185
6186         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6187                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6188                                                GFP_KERNEL);
6189         if (!net->ipv6.ip6_blk_hole_entry)
6190                 goto out_ip6_prohibit_entry;
6191         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6192         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6193                          ip6_template_metrics, true);
6194         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6195 #endif
6196
6197         net->ipv6.sysctl.flush_delay = 0;
6198         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6199         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6200         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6201         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6202         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6203         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6204         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6205         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6206
6207         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6208
6209         ret = 0;
6210 out:
6211         return ret;
6212
6213 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6214 out_ip6_prohibit_entry:
6215         kfree(net->ipv6.ip6_prohibit_entry);
6216 out_ip6_null_entry:
6217         kfree(net->ipv6.ip6_null_entry);
6218 #endif
6219 out_fib6_null_entry:
6220         kfree(net->ipv6.fib6_null_entry);
6221 out_ip6_dst_entries:
6222         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6223 out_ip6_dst_ops:
6224         goto out;
6225 }
6226
6227 static void __net_exit ip6_route_net_exit(struct net *net)
6228 {
6229         kfree(net->ipv6.fib6_null_entry);
6230         kfree(net->ipv6.ip6_null_entry);
6231 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6232         kfree(net->ipv6.ip6_prohibit_entry);
6233         kfree(net->ipv6.ip6_blk_hole_entry);
6234 #endif
6235         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6236 }
6237
6238 static int __net_init ip6_route_net_init_late(struct net *net)
6239 {
6240 #ifdef CONFIG_PROC_FS
6241         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6242                         sizeof(struct ipv6_route_iter));
6243         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6244                         rt6_stats_seq_show, NULL);
6245 #endif
6246         return 0;
6247 }
6248
6249 static void __net_exit ip6_route_net_exit_late(struct net *net)
6250 {
6251 #ifdef CONFIG_PROC_FS
6252         remove_proc_entry("ipv6_route", net->proc_net);
6253         remove_proc_entry("rt6_stats", net->proc_net);
6254 #endif
6255 }
6256
6257 static struct pernet_operations ip6_route_net_ops = {
6258         .init = ip6_route_net_init,
6259         .exit = ip6_route_net_exit,
6260 };
6261
6262 static int __net_init ipv6_inetpeer_init(struct net *net)
6263 {
6264         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6265
6266         if (!bp)
6267                 return -ENOMEM;
6268         inet_peer_base_init(bp);
6269         net->ipv6.peers = bp;
6270         return 0;
6271 }
6272
6273 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6274 {
6275         struct inet_peer_base *bp = net->ipv6.peers;
6276
6277         net->ipv6.peers = NULL;
6278         inetpeer_invalidate_tree(bp);
6279         kfree(bp);
6280 }
6281
6282 static struct pernet_operations ipv6_inetpeer_ops = {
6283         .init   =       ipv6_inetpeer_init,
6284         .exit   =       ipv6_inetpeer_exit,
6285 };
6286
6287 static struct pernet_operations ip6_route_net_late_ops = {
6288         .init = ip6_route_net_init_late,
6289         .exit = ip6_route_net_exit_late,
6290 };
6291
6292 static struct notifier_block ip6_route_dev_notifier = {
6293         .notifier_call = ip6_route_dev_notify,
6294         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6295 };
6296
6297 void __init ip6_route_init_special_entries(void)
6298 {
6299         /* Registering of the loopback is done before this portion of code,
6300          * the loopback reference in rt6_info will not be taken, do it
6301          * manually for init_net */
6302         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6303         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6304         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6305   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6306         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6307         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6308         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6309         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6310   #endif
6311 }
6312
6313 int __init ip6_route_init(void)
6314 {
6315         int ret;
6316         int cpu;
6317
6318         ret = -ENOMEM;
6319         ip6_dst_ops_template.kmem_cachep =
6320                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6321                                   SLAB_HWCACHE_ALIGN, NULL);
6322         if (!ip6_dst_ops_template.kmem_cachep)
6323                 goto out;
6324
6325         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6326         if (ret)
6327                 goto out_kmem_cache;
6328
6329         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6330         if (ret)
6331                 goto out_dst_entries;
6332
6333         ret = register_pernet_subsys(&ip6_route_net_ops);
6334         if (ret)
6335                 goto out_register_inetpeer;
6336
6337         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6338
6339         ret = fib6_init();
6340         if (ret)
6341                 goto out_register_subsys;
6342
6343         ret = xfrm6_init();
6344         if (ret)
6345                 goto out_fib6_init;
6346
6347         ret = fib6_rules_init();
6348         if (ret)
6349                 goto xfrm6_init;
6350
6351         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6352         if (ret)
6353                 goto fib6_rules_init;
6354
6355         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6356                                    inet6_rtm_newroute, NULL, 0);
6357         if (ret < 0)
6358                 goto out_register_late_subsys;
6359
6360         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6361                                    inet6_rtm_delroute, NULL, 0);
6362         if (ret < 0)
6363                 goto out_register_late_subsys;
6364
6365         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6366                                    inet6_rtm_getroute, NULL,
6367                                    RTNL_FLAG_DOIT_UNLOCKED);
6368         if (ret < 0)
6369                 goto out_register_late_subsys;
6370
6371         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6372         if (ret)
6373                 goto out_register_late_subsys;
6374
6375         for_each_possible_cpu(cpu) {
6376                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6377
6378                 INIT_LIST_HEAD(&ul->head);
6379                 spin_lock_init(&ul->lock);
6380         }
6381
6382 out:
6383         return ret;
6384
6385 out_register_late_subsys:
6386         rtnl_unregister_all(PF_INET6);
6387         unregister_pernet_subsys(&ip6_route_net_late_ops);
6388 fib6_rules_init:
6389         fib6_rules_cleanup();
6390 xfrm6_init:
6391         xfrm6_fini();
6392 out_fib6_init:
6393         fib6_gc_cleanup();
6394 out_register_subsys:
6395         unregister_pernet_subsys(&ip6_route_net_ops);
6396 out_register_inetpeer:
6397         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6398 out_dst_entries:
6399         dst_entries_destroy(&ip6_dst_blackhole_ops);
6400 out_kmem_cache:
6401         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6402         goto out;
6403 }
6404
6405 void ip6_route_cleanup(void)
6406 {
6407         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6408         unregister_pernet_subsys(&ip6_route_net_late_ops);
6409         fib6_rules_cleanup();
6410         xfrm6_fini();
6411         fib6_gc_cleanup();
6412         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6413         unregister_pernet_subsys(&ip6_route_net_ops);
6414         dst_entries_destroy(&ip6_dst_blackhole_ops);
6415         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6416 }