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 #include <linux/btf_ids.h>
  65
  66 #ifdef CONFIG_SYSCTL
  67 #include <linux/sysctl.h>
  68 #endif
  69
  70 static int ip6_rt_type_to_error(u8 fib6_type);
  71
  72 #define CREATE_TRACE_POINTS
  73 #include <trace/events/fib6.h>
  74 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  75 #undef CREATE_TRACE_POINTS
  76
  77 enum rt6_nud_state {
  78         RT6_NUD_FAIL_HARD = -3,
  79         RT6_NUD_FAIL_PROBE = -2,
  80         RT6_NUD_FAIL_DO_RR = -1,
  81         RT6_NUD_SUCCEED = 1
  82 };
  83
  84 INDIRECT_CALLABLE_SCOPE
  85 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  86 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  87 INDIRECT_CALLABLE_SCOPE
  88 unsigned int            ip6_mtu(const struct dst_entry *dst);
  89 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  90 static void             ip6_dst_destroy(struct dst_entry *);
  91 static void             ip6_dst_ifdown(struct dst_entry *,
  92                                        struct net_device *dev, int how);
  93 static int               ip6_dst_gc(struct dst_ops *ops);
  94
  95 static int              ip6_pkt_discard(struct sk_buff *skb);
  96 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  97 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  99 static void             ip6_link_failure(struct sk_buff *skb);
 100 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 101                                            struct sk_buff *skb, u32 mtu,
 102                                            bool confirm_neigh);
 103 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 104                                         struct sk_buff *skb);
 105 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 106                            int strict);
 107 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 108 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 109                          struct fib6_info *rt, struct dst_entry *dst,
 110                          struct in6_addr *dest, struct in6_addr *src,
 111                          int iif, int type, u32 portid, u32 seq,
 112                          unsigned int flags);
 113 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 114                                            const struct in6_addr *daddr,
 115                                            const struct in6_addr *saddr);
 116
 117 #ifdef CONFIG_IPV6_ROUTE_INFO
 118 static struct fib6_info *rt6_add_route_info(struct net *net,
 119                                            const struct in6_addr *prefix, int prefixlen,
 120                                            const struct in6_addr *gwaddr,
 121                                            struct net_device *dev,
 122                                            unsigned int pref);
 123 static struct fib6_info *rt6_get_route_info(struct net *net,
 124                                            const struct in6_addr *prefix, int prefixlen,
 125                                            const struct in6_addr *gwaddr,
 126                                            struct net_device *dev);
 127 #endif
 128
 129 struct uncached_list {
 130         spinlock_t              lock;
 131         struct list_head        head;
 132 };
 133
 134 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 135
 136 void rt6_uncached_list_add(struct rt6_info *rt)
 137 {
 138         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 139
 140         rt->rt6i_uncached_list = ul;
 141
 142         spin_lock_bh(&ul->lock);
 143         list_add_tail(&rt->rt6i_uncached, &ul->head);
 144         spin_unlock_bh(&ul->lock);
 145 }
 146
 147 void rt6_uncached_list_del(struct rt6_info *rt)
 148 {
 149         if (!list_empty(&rt->rt6i_uncached)) {
 150                 struct uncached_list *ul = rt->rt6i_uncached_list;
 151                 struct net *net = dev_net(rt->dst.dev);
 152
 153                 spin_lock_bh(&ul->lock);
 154                 list_del(&rt->rt6i_uncached);
 155                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 156                 spin_unlock_bh(&ul->lock);
 157         }
 158 }
 159
 160 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 161 {
 162         struct net_device *loopback_dev = net->loopback_dev;
 163         int cpu;
 164
 165         if (dev == loopback_dev)
 166                 return;
 167
 168         for_each_possible_cpu(cpu) {
 169                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 170                 struct rt6_info *rt;
 171
 172                 spin_lock_bh(&ul->lock);
 173                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 174                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 175                         struct net_device *rt_dev = rt->dst.dev;
 176
 177                         if (rt_idev->dev == dev) {
 178                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 179                                 in6_dev_put(rt_idev);
 180                         }
 181
 182                         if (rt_dev == dev) {
 183                                 rt->dst.dev = blackhole_netdev;
 184                                 dev_hold(rt->dst.dev);
 185                                 dev_put(rt_dev);
 186                         }
 187                 }
 188                 spin_unlock_bh(&ul->lock);
 189         }
 190 }
 191
 192 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 193                                              struct sk_buff *skb,
 194                                              const void *daddr)
 195 {
 196         if (!ipv6_addr_any(p))
 197                 return (const void *) p;
 198         else if (skb)
 199                 return &ipv6_hdr(skb)->daddr;
 200         return daddr;
 201 }
 202
 203 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 204                                    struct net_device *dev,
 205                                    struct sk_buff *skb,
 206                                    const void *daddr)
 207 {
 208         struct neighbour *n;
 209
 210         daddr = choose_neigh_daddr(gw, skb, daddr);
 211         n = __ipv6_neigh_lookup(dev, daddr);
 212         if (n)
 213                 return n;
 214
 215         n = neigh_create(&nd_tbl, daddr, dev);
 216         return IS_ERR(n) ? NULL : n;
 217 }
 218
 219 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 220                                               struct sk_buff *skb,
 221                                               const void *daddr)
 222 {
 223         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 224
 225         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 226                                 dst->dev, skb, daddr);
 227 }
 228
 229 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 230 {
 231         struct net_device *dev = dst->dev;
 232         struct rt6_info *rt = (struct rt6_info *)dst;
 233
 234         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 235         if (!daddr)
 236                 return;
 237         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 238                 return;
 239         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 240                 return;
 241         __ipv6_confirm_neigh(dev, daddr);
 242 }
 243
 244 static struct dst_ops ip6_dst_ops_template = {
 245         .family                 =       AF_INET6,
 246         .gc                     =       ip6_dst_gc,
 247         .gc_thresh              =       1024,
 248         .check                  =       ip6_dst_check,
 249         .default_advmss         =       ip6_default_advmss,
 250         .mtu                    =       ip6_mtu,
 251         .cow_metrics            =       dst_cow_metrics_generic,
 252         .destroy                =       ip6_dst_destroy,
 253         .ifdown                 =       ip6_dst_ifdown,
 254         .negative_advice        =       ip6_negative_advice,
 255         .link_failure           =       ip6_link_failure,
 256         .update_pmtu            =       ip6_rt_update_pmtu,
 257         .redirect               =       rt6_do_redirect,
 258         .local_out              =       __ip6_local_out,
 259         .neigh_lookup           =       ip6_dst_neigh_lookup,
 260         .confirm_neigh          =       ip6_confirm_neigh,
 261 };
 262
 263 static struct dst_ops ip6_dst_blackhole_ops = {
 264         .family                 = AF_INET6,
 265         .default_advmss         = ip6_default_advmss,
 266         .neigh_lookup           = ip6_dst_neigh_lookup,
 267         .check                  = ip6_dst_check,
 268         .destroy                = ip6_dst_destroy,
 269         .cow_metrics            = dst_cow_metrics_generic,
 270         .update_pmtu            = dst_blackhole_update_pmtu,
 271         .redirect               = dst_blackhole_redirect,
 272         .mtu                    = dst_blackhole_mtu,
 273 };
 274
 275 static const u32 ip6_template_metrics[RTAX_MAX] = {
 276         [RTAX_HOPLIMIT - 1] = 0,
 277 };
 278
 279 static const struct fib6_info fib6_null_entry_template = {
 280         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 281         .fib6_protocol  = RTPROT_KERNEL,
 282         .fib6_metric    = ~(u32)0,
 283         .fib6_ref       = REFCOUNT_INIT(1),
 284         .fib6_type      = RTN_UNREACHABLE,
 285         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 286 };
 287
 288 static const struct rt6_info ip6_null_entry_template = {
 289         .dst = {
 290                 .__refcnt       = ATOMIC_INIT(1),
 291                 .__use          = 1,
 292                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 293                 .error          = -ENETUNREACH,
 294                 .input          = ip6_pkt_discard,
 295                 .output         = ip6_pkt_discard_out,
 296         },
 297         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 298 };
 299
 300 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 301
 302 static const struct rt6_info ip6_prohibit_entry_template = {
 303         .dst = {
 304                 .__refcnt       = ATOMIC_INIT(1),
 305                 .__use          = 1,
 306                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 307                 .error          = -EACCES,
 308                 .input          = ip6_pkt_prohibit,
 309                 .output         = ip6_pkt_prohibit_out,
 310         },
 311         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 312 };
 313
 314 static const struct rt6_info ip6_blk_hole_entry_template = {
 315         .dst = {
 316                 .__refcnt       = ATOMIC_INIT(1),
 317                 .__use          = 1,
 318                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 319                 .error          = -EINVAL,
 320                 .input          = dst_discard,
 321                 .output         = dst_discard_out,
 322         },
 323         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 324 };
 325
 326 #endif
 327
 328 static void rt6_info_init(struct rt6_info *rt)
 329 {
 330         struct dst_entry *dst = &rt->dst;
 331
 332         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 333         INIT_LIST_HEAD(&rt->rt6i_uncached);
 334 }
 335
 336 /* allocate dst with ip6_dst_ops */
 337 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 338                                int flags)
 339 {
 340         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 341                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 342
 343         if (rt) {
 344                 rt6_info_init(rt);
 345                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 346         }
 347
 348         return rt;
 349 }
 350 EXPORT_SYMBOL(ip6_dst_alloc);
 351
 352 static void ip6_dst_destroy(struct dst_entry *dst)
 353 {
 354         struct rt6_info *rt = (struct rt6_info *)dst;
 355         struct fib6_info *from;
 356         struct inet6_dev *idev;
 357
 358         ip_dst_metrics_put(dst);
 359         rt6_uncached_list_del(rt);
 360
 361         idev = rt->rt6i_idev;
 362         if (idev) {
 363                 rt->rt6i_idev = NULL;
 364                 in6_dev_put(idev);
 365         }
 366
 367         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 368         fib6_info_release(from);
 369 }
 370
 371 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 372                            int how)
 373 {
 374         struct rt6_info *rt = (struct rt6_info *)dst;
 375         struct inet6_dev *idev = rt->rt6i_idev;
 376         struct net_device *loopback_dev =
 377                 dev_net(dev)->loopback_dev;
 378
 379         if (idev && idev->dev != loopback_dev) {
 380                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 381                 if (loopback_idev) {
 382                         rt->rt6i_idev = loopback_idev;
 383                         in6_dev_put(idev);
 384                 }
 385         }
 386 }
 387
 388 static bool __rt6_check_expired(const struct rt6_info *rt)
 389 {
 390         if (rt->rt6i_flags & RTF_EXPIRES)
 391                 return time_after(jiffies, rt->dst.expires);
 392         else
 393                 return false;
 394 }
 395
 396 static bool rt6_check_expired(const struct rt6_info *rt)
 397 {
 398         struct fib6_info *from;
 399
 400         from = rcu_dereference(rt->from);
 401
 402         if (rt->rt6i_flags & RTF_EXPIRES) {
 403                 if (time_after(jiffies, rt->dst.expires))
 404                         return true;
 405         } else if (from) {
 406                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 407                         fib6_check_expired(from);
 408         }
 409         return false;
 410 }
 411
 412 void fib6_select_path(const struct net *net, struct fib6_result *res,
 413                       struct flowi6 *fl6, int oif, bool have_oif_match,
 414                       const struct sk_buff *skb, int strict)
 415 {
 416         struct fib6_info *sibling, *next_sibling;
 417         struct fib6_info *match = res->f6i;
 418
 419         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 420                 goto out;
 421
 422         if (match->nh && have_oif_match && res->nh)
 423                 return;
 424
 425         /* We might have already computed the hash for ICMPv6 errors. In such
 426          * case it will always be non-zero. Otherwise now is the time to do it.
 427          */
 428         if (!fl6->mp_hash &&
 429             (!match->nh || nexthop_is_multipath(match->nh)))
 430                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 431
 432         if (unlikely(match->nh)) {
 433                 nexthop_path_fib6_result(res, fl6->mp_hash);
 434                 return;
 435         }
 436
 437         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 438                 goto out;
 439
 440         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 441                                  fib6_siblings) {
 442                 const struct fib6_nh *nh = sibling->fib6_nh;
 443                 int nh_upper_bound;
 444
 445                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 446                 if (fl6->mp_hash > nh_upper_bound)
 447                         continue;
 448                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 449                         break;
 450                 match = sibling;
 451                 break;
 452         }
 453
 454 out:
 455         res->f6i = match;
 456         res->nh = match->fib6_nh;
 457 }
 458
 459 /*
 460  *      Route lookup. rcu_read_lock() should be held.
 461  */
 462
 463 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 464                                const struct in6_addr *saddr, int oif, int flags)
 465 {
 466         const struct net_device *dev;
 467
 468         if (nh->fib_nh_flags & RTNH_F_DEAD)
 469                 return false;
 470
 471         dev = nh->fib_nh_dev;
 472         if (oif) {
 473                 if (dev->ifindex == oif)
 474                         return true;
 475         } else {
 476                 if (ipv6_chk_addr(net, saddr, dev,
 477                                   flags & RT6_LOOKUP_F_IFACE))
 478                         return true;
 479         }
 480
 481         return false;
 482 }
 483
 484 struct fib6_nh_dm_arg {
 485         struct net              *net;
 486         const struct in6_addr   *saddr;
 487         int                     oif;
 488         int                     flags;
 489         struct fib6_nh          *nh;
 490 };
 491
 492 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 493 {
 494         struct fib6_nh_dm_arg *arg = _arg;
 495
 496         arg->nh = nh;
 497         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 498                                   arg->flags);
 499 }
 500
 501 /* returns fib6_nh from nexthop or NULL */
 502 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 503                                         struct fib6_result *res,
 504                                         const struct in6_addr *saddr,
 505                                         int oif, int flags)
 506 {
 507         struct fib6_nh_dm_arg arg = {
 508                 .net   = net,
 509                 .saddr = saddr,
 510                 .oif   = oif,
 511                 .flags = flags,
 512         };
 513
 514         if (nexthop_is_blackhole(nh))
 515                 return NULL;
 516
 517         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 518                 return arg.nh;
 519
 520         return NULL;
 521 }
 522
 523 static void rt6_device_match(struct net *net, struct fib6_result *res,
 524                              const struct in6_addr *saddr, int oif, int flags)
 525 {
 526         struct fib6_info *f6i = res->f6i;
 527         struct fib6_info *spf6i;
 528         struct fib6_nh *nh;
 529
 530         if (!oif && ipv6_addr_any(saddr)) {
 531                 if (unlikely(f6i->nh)) {
 532                         nh = nexthop_fib6_nh(f6i->nh);
 533                         if (nexthop_is_blackhole(f6i->nh))
 534                                 goto out_blackhole;
 535                 } else {
 536                         nh = f6i->fib6_nh;
 537                 }
 538                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 539                         goto out;
 540         }
 541
 542         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 543                 bool matched = false;
 544
 545                 if (unlikely(spf6i->nh)) {
 546                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 547                                               oif, flags);
 548                         if (nh)
 549                                 matched = true;
 550                 } else {
 551                         nh = spf6i->fib6_nh;
 552                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 553                                 matched = true;
 554                 }
 555                 if (matched) {
 556                         res->f6i = spf6i;
 557                         goto out;
 558                 }
 559         }
 560
 561         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 562                 res->f6i = net->ipv6.fib6_null_entry;
 563                 nh = res->f6i->fib6_nh;
 564                 goto out;
 565         }
 566
 567         if (unlikely(f6i->nh)) {
 568                 nh = nexthop_fib6_nh(f6i->nh);
 569                 if (nexthop_is_blackhole(f6i->nh))
 570                         goto out_blackhole;
 571         } else {
 572                 nh = f6i->fib6_nh;
 573         }
 574
 575         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 576                 res->f6i = net->ipv6.fib6_null_entry;
 577                 nh = res->f6i->fib6_nh;
 578         }
 579 out:
 580         res->nh = nh;
 581         res->fib6_type = res->f6i->fib6_type;
 582         res->fib6_flags = res->f6i->fib6_flags;
 583         return;
 584
 585 out_blackhole:
 586         res->fib6_flags |= RTF_REJECT;
 587         res->fib6_type = RTN_BLACKHOLE;
 588         res->nh = nh;
 589 }
 590
 591 #ifdef CONFIG_IPV6_ROUTER_PREF
 592 struct __rt6_probe_work {
 593         struct work_struct work;
 594         struct in6_addr target;
 595         struct net_device *dev;
 596 };
 597
 598 static void rt6_probe_deferred(struct work_struct *w)
 599 {
 600         struct in6_addr mcaddr;
 601         struct __rt6_probe_work *work =
 602                 container_of(w, struct __rt6_probe_work, work);
 603
 604         addrconf_addr_solict_mult(&work->target, &mcaddr);
 605         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 606         dev_put(work->dev);
 607         kfree(work);
 608 }
 609
 610 static void rt6_probe(struct fib6_nh *fib6_nh)
 611 {
 612         struct __rt6_probe_work *work = NULL;
 613         const struct in6_addr *nh_gw;
 614         unsigned long last_probe;
 615         struct neighbour *neigh;
 616         struct net_device *dev;
 617         struct inet6_dev *idev;
 618
 619         /*
 620          * Okay, this does not seem to be appropriate
 621          * for now, however, we need to check if it
 622          * is really so; aka Router Reachability Probing.
 623          *
 624          * Router Reachability Probe MUST be rate-limited
 625          * to no more than one per minute.
 626          */
 627         if (!fib6_nh->fib_nh_gw_family)
 628                 return;
 629
 630         nh_gw = &fib6_nh->fib_nh_gw6;
 631         dev = fib6_nh->fib_nh_dev;
 632         rcu_read_lock_bh();
 633         last_probe = READ_ONCE(fib6_nh->last_probe);
 634         idev = __in6_dev_get(dev);
 635         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 636         if (neigh) {
 637                 if (neigh->nud_state & NUD_VALID)
 638                         goto out;
 639
 640                 write_lock(&neigh->lock);
 641                 if (!(neigh->nud_state & NUD_VALID) &&
 642                     time_after(jiffies,
 643                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 644                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 645                         if (work)
 646                                 __neigh_set_probe_once(neigh);
 647                 }
 648                 write_unlock(&neigh->lock);
 649         } else if (time_after(jiffies, last_probe +
 650                                        idev->cnf.rtr_probe_interval)) {
 651                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 652         }
 653
 654         if (!work || cmpxchg(&fib6_nh->last_probe,
 655                              last_probe, jiffies) != last_probe) {
 656                 kfree(work);
 657         } else {
 658                 INIT_WORK(&work->work, rt6_probe_deferred);
 659                 work->target = *nh_gw;
 660                 dev_hold(dev);
 661                 work->dev = dev;
 662                 schedule_work(&work->work);
 663         }
 664
 665 out:
 666         rcu_read_unlock_bh();
 667 }
 668 #else
 669 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 670 {
 671 }
 672 #endif
 673
 674 /*
 675  * Default Router Selection (RFC 2461 6.3.6)
 676  */
 677 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 678 {
 679         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 680         struct neighbour *neigh;
 681
 682         rcu_read_lock_bh();
 683         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 684                                           &fib6_nh->fib_nh_gw6);
 685         if (neigh) {
 686                 read_lock(&neigh->lock);
 687                 if (neigh->nud_state & NUD_VALID)
 688                         ret = RT6_NUD_SUCCEED;
 689 #ifdef CONFIG_IPV6_ROUTER_PREF
 690                 else if (!(neigh->nud_state & NUD_FAILED))
 691                         ret = RT6_NUD_SUCCEED;
 692                 else
 693                         ret = RT6_NUD_FAIL_PROBE;
 694 #endif
 695                 read_unlock(&neigh->lock);
 696         } else {
 697                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 698                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 699         }
 700         rcu_read_unlock_bh();
 701
 702         return ret;
 703 }
 704
 705 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 706                            int strict)
 707 {
 708         int m = 0;
 709
 710         if (!oif || nh->fib_nh_dev->ifindex == oif)
 711                 m = 2;
 712
 713         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 714                 return RT6_NUD_FAIL_HARD;
 715 #ifdef CONFIG_IPV6_ROUTER_PREF
 716         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 717 #endif
 718         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 719             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 720                 int n = rt6_check_neigh(nh);
 721                 if (n < 0)
 722                         return n;
 723         }
 724         return m;
 725 }
 726
 727 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 728                        int oif, int strict, int *mpri, bool *do_rr)
 729 {
 730         bool match_do_rr = false;
 731         bool rc = false;
 732         int m;
 733
 734         if (nh->fib_nh_flags & RTNH_F_DEAD)
 735                 goto out;
 736
 737         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 738             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 739             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 740                 goto out;
 741
 742         m = rt6_score_route(nh, fib6_flags, oif, strict);
 743         if (m == RT6_NUD_FAIL_DO_RR) {
 744                 match_do_rr = true;
 745                 m = 0; /* lowest valid score */
 746         } else if (m == RT6_NUD_FAIL_HARD) {
 747                 goto out;
 748         }
 749
 750         if (strict & RT6_LOOKUP_F_REACHABLE)
 751                 rt6_probe(nh);
 752
 753         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 754         if (m > *mpri) {
 755                 *do_rr = match_do_rr;
 756                 *mpri = m;
 757                 rc = true;
 758         }
 759 out:
 760         return rc;
 761 }
 762
 763 struct fib6_nh_frl_arg {
 764         u32             flags;
 765         int             oif;
 766         int             strict;
 767         int             *mpri;
 768         bool            *do_rr;
 769         struct fib6_nh  *nh;
 770 };
 771
 772 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 773 {
 774         struct fib6_nh_frl_arg *arg = _arg;
 775
 776         arg->nh = nh;
 777         return find_match(nh, arg->flags, arg->oif, arg->strict,
 778                           arg->mpri, arg->do_rr);
 779 }
 780
 781 static void __find_rr_leaf(struct fib6_info *f6i_start,
 782                            struct fib6_info *nomatch, u32 metric,
 783                            struct fib6_result *res, struct fib6_info **cont,
 784                            int oif, int strict, bool *do_rr, int *mpri)
 785 {
 786         struct fib6_info *f6i;
 787
 788         for (f6i = f6i_start;
 789              f6i && f6i != nomatch;
 790              f6i = rcu_dereference(f6i->fib6_next)) {
 791                 bool matched = false;
 792                 struct fib6_nh *nh;
 793
 794                 if (cont && f6i->fib6_metric != metric) {
 795                         *cont = f6i;
 796                         return;
 797                 }
 798
 799                 if (fib6_check_expired(f6i))
 800                         continue;
 801
 802                 if (unlikely(f6i->nh)) {
 803                         struct fib6_nh_frl_arg arg = {
 804                                 .flags  = f6i->fib6_flags,
 805                                 .oif    = oif,
 806                                 .strict = strict,
 807                                 .mpri   = mpri,
 808                                 .do_rr  = do_rr
 809                         };
 810
 811                         if (nexthop_is_blackhole(f6i->nh)) {
 812                                 res->fib6_flags = RTF_REJECT;
 813                                 res->fib6_type = RTN_BLACKHOLE;
 814                                 res->f6i = f6i;
 815                                 res->nh = nexthop_fib6_nh(f6i->nh);
 816                                 return;
 817                         }
 818                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 819                                                      &arg)) {
 820                                 matched = true;
 821                                 nh = arg.nh;
 822                         }
 823                 } else {
 824                         nh = f6i->fib6_nh;
 825                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 826                                        mpri, do_rr))
 827                                 matched = true;
 828                 }
 829                 if (matched) {
 830                         res->f6i = f6i;
 831                         res->nh = nh;
 832                         res->fib6_flags = f6i->fib6_flags;
 833                         res->fib6_type = f6i->fib6_type;
 834                 }
 835         }
 836 }
 837
 838 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 839                          struct fib6_info *rr_head, int oif, int strict,
 840                          bool *do_rr, struct fib6_result *res)
 841 {
 842         u32 metric = rr_head->fib6_metric;
 843         struct fib6_info *cont = NULL;
 844         int mpri = -1;
 845
 846         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 847                        oif, strict, do_rr, &mpri);
 848
 849         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 850                        oif, strict, do_rr, &mpri);
 851
 852         if (res->f6i || !cont)
 853                 return;
 854
 855         __find_rr_leaf(cont, NULL, metric, res, NULL,
 856                        oif, strict, do_rr, &mpri);
 857 }
 858
 859 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 860                        struct fib6_result *res, int strict)
 861 {
 862         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 863         struct fib6_info *rt0;
 864         bool do_rr = false;
 865         int key_plen;
 866
 867         /* make sure this function or its helpers sets f6i */
 868         res->f6i = NULL;
 869
 870         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 871                 goto out;
 872
 873         rt0 = rcu_dereference(fn->rr_ptr);
 874         if (!rt0)
 875                 rt0 = leaf;
 876
 877         /* Double check to make sure fn is not an intermediate node
 878          * and fn->leaf does not points to its child's leaf
 879          * (This might happen if all routes under fn are deleted from
 880          * the tree and fib6_repair_tree() is called on the node.)
 881          */
 882         key_plen = rt0->fib6_dst.plen;
 883 #ifdef CONFIG_IPV6_SUBTREES
 884         if (rt0->fib6_src.plen)
 885                 key_plen = rt0->fib6_src.plen;
 886 #endif
 887         if (fn->fn_bit != key_plen)
 888                 goto out;
 889
 890         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 891         if (do_rr) {
 892                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 893
 894                 /* no entries matched; do round-robin */
 895                 if (!next || next->fib6_metric != rt0->fib6_metric)
 896                         next = leaf;
 897
 898                 if (next != rt0) {
 899                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 900                         /* make sure next is not being deleted from the tree */
 901                         if (next->fib6_node)
 902                                 rcu_assign_pointer(fn->rr_ptr, next);
 903                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 904                 }
 905         }
 906
 907 out:
 908         if (!res->f6i) {
 909                 res->f6i = net->ipv6.fib6_null_entry;
 910                 res->nh = res->f6i->fib6_nh;
 911                 res->fib6_flags = res->f6i->fib6_flags;
 912                 res->fib6_type = res->f6i->fib6_type;
 913         }
 914 }
 915
 916 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 917 {
 918         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 919                res->nh->fib_nh_gw_family;
 920 }
 921
 922 #ifdef CONFIG_IPV6_ROUTE_INFO
 923 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 924                   const struct in6_addr *gwaddr)
 925 {
 926         struct net *net = dev_net(dev);
 927         struct route_info *rinfo = (struct route_info *) opt;
 928         struct in6_addr prefix_buf, *prefix;
 929         unsigned int pref;
 930         unsigned long lifetime;
 931         struct fib6_info *rt;
 932
 933         if (len < sizeof(struct route_info)) {
 934                 return -EINVAL;
 935         }
 936
 937         /* Sanity check for prefix_len and length */
 938         if (rinfo->length > 3) {
 939                 return -EINVAL;
 940         } else if (rinfo->prefix_len > 128) {
 941                 return -EINVAL;
 942         } else if (rinfo->prefix_len > 64) {
 943                 if (rinfo->length < 2) {
 944                         return -EINVAL;
 945                 }
 946         } else if (rinfo->prefix_len > 0) {
 947                 if (rinfo->length < 1) {
 948                         return -EINVAL;
 949                 }
 950         }
 951
 952         pref = rinfo->route_pref;
 953         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 954                 return -EINVAL;
 955
 956         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 957
 958         if (rinfo->length == 3)
 959                 prefix = (struct in6_addr *)rinfo->prefix;
 960         else {
 961                 /* this function is safe */
 962                 ipv6_addr_prefix(&prefix_buf,
 963                                  (struct in6_addr *)rinfo->prefix,
 964                                  rinfo->prefix_len);
 965                 prefix = &prefix_buf;
 966         }
 967
 968         if (rinfo->prefix_len == 0)
 969                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 970         else
 971                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 972                                         gwaddr, dev);
 973
 974         if (rt && !lifetime) {
 975                 ip6_del_rt(net, rt, false);
 976                 rt = NULL;
 977         }
 978
 979         if (!rt && lifetime)
 980                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 981                                         dev, pref);
 982         else if (rt)
 983                 rt->fib6_flags = RTF_ROUTEINFO |
 984                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 985
 986         if (rt) {
 987                 if (!addrconf_finite_timeout(lifetime))
 988                         fib6_clean_expires(rt);
 989                 else
 990                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 991
 992                 fib6_info_release(rt);
 993         }
 994         return 0;
 995 }
 996 #endif
 997
 998 /*
 999  *      Misc support functions
1000  */
1001
1002 /* called with rcu_lock held */
1003 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1004 {
1005         struct net_device *dev = res->nh->fib_nh_dev;
1006
1007         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1008                 /* for copies of local routes, dst->dev needs to be the
1009                  * device if it is a master device, the master device if
1010                  * device is enslaved, and the loopback as the default
1011                  */
1012                 if (netif_is_l3_slave(dev) &&
1013                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1014                         dev = l3mdev_master_dev_rcu(dev);
1015                 else if (!netif_is_l3_master(dev))
1016                         dev = dev_net(dev)->loopback_dev;
1017                 /* last case is netif_is_l3_master(dev) is true in which
1018                  * case we want dev returned to be dev
1019                  */
1020         }
1021
1022         return dev;
1023 }
1024
1025 static const int fib6_prop[RTN_MAX + 1] = {
1026         [RTN_UNSPEC]    = 0,
1027         [RTN_UNICAST]   = 0,
1028         [RTN_LOCAL]     = 0,
1029         [RTN_BROADCAST] = 0,
1030         [RTN_ANYCAST]   = 0,
1031         [RTN_MULTICAST] = 0,
1032         [RTN_BLACKHOLE] = -EINVAL,
1033         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1034         [RTN_PROHIBIT]  = -EACCES,
1035         [RTN_THROW]     = -EAGAIN,
1036         [RTN_NAT]       = -EINVAL,
1037         [RTN_XRESOLVE]  = -EINVAL,
1038 };
1039
1040 static int ip6_rt_type_to_error(u8 fib6_type)
1041 {
1042         return fib6_prop[fib6_type];
1043 }
1044
1045 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1046 {
1047         unsigned short flags = 0;
1048
1049         if (rt->dst_nocount)
1050                 flags |= DST_NOCOUNT;
1051         if (rt->dst_nopolicy)
1052                 flags |= DST_NOPOLICY;
1053
1054         return flags;
1055 }
1056
1057 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1058 {
1059         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1060
1061         switch (fib6_type) {
1062         case RTN_BLACKHOLE:
1063                 rt->dst.output = dst_discard_out;
1064                 rt->dst.input = dst_discard;
1065                 break;
1066         case RTN_PROHIBIT:
1067                 rt->dst.output = ip6_pkt_prohibit_out;
1068                 rt->dst.input = ip6_pkt_prohibit;
1069                 break;
1070         case RTN_THROW:
1071         case RTN_UNREACHABLE:
1072         default:
1073                 rt->dst.output = ip6_pkt_discard_out;
1074                 rt->dst.input = ip6_pkt_discard;
1075                 break;
1076         }
1077 }
1078
1079 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1080 {
1081         struct fib6_info *f6i = res->f6i;
1082
1083         if (res->fib6_flags & RTF_REJECT) {
1084                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1085                 return;
1086         }
1087
1088         rt->dst.error = 0;
1089         rt->dst.output = ip6_output;
1090
1091         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1092                 rt->dst.input = ip6_input;
1093         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1094                 rt->dst.input = ip6_mc_input;
1095         } else {
1096                 rt->dst.input = ip6_forward;
1097         }
1098
1099         if (res->nh->fib_nh_lws) {
1100                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1101                 lwtunnel_set_redirect(&rt->dst);
1102         }
1103
1104         rt->dst.lastuse = jiffies;
1105 }
1106
1107 /* Caller must already hold reference to @from */
1108 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1109 {
1110         rt->rt6i_flags &= ~RTF_EXPIRES;
1111         rcu_assign_pointer(rt->from, from);
1112         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1113 }
1114
1115 /* Caller must already hold reference to f6i in result */
1116 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1117 {
1118         const struct fib6_nh *nh = res->nh;
1119         const struct net_device *dev = nh->fib_nh_dev;
1120         struct fib6_info *f6i = res->f6i;
1121
1122         ip6_rt_init_dst(rt, res);
1123
1124         rt->rt6i_dst = f6i->fib6_dst;
1125         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1126         rt->rt6i_flags = res->fib6_flags;
1127         if (nh->fib_nh_gw_family) {
1128                 rt->rt6i_gateway = nh->fib_nh_gw6;
1129                 rt->rt6i_flags |= RTF_GATEWAY;
1130         }
1131         rt6_set_from(rt, f6i);
1132 #ifdef CONFIG_IPV6_SUBTREES
1133         rt->rt6i_src = f6i->fib6_src;
1134 #endif
1135 }
1136
1137 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1138                                         struct in6_addr *saddr)
1139 {
1140         struct fib6_node *pn, *sn;
1141         while (1) {
1142                 if (fn->fn_flags & RTN_TL_ROOT)
1143                         return NULL;
1144                 pn = rcu_dereference(fn->parent);
1145                 sn = FIB6_SUBTREE(pn);
1146                 if (sn && sn != fn)
1147                         fn = fib6_node_lookup(sn, NULL, saddr);
1148                 else
1149                         fn = pn;
1150                 if (fn->fn_flags & RTN_RTINFO)
1151                         return fn;
1152         }
1153 }
1154
1155 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1156 {
1157         struct rt6_info *rt = *prt;
1158
1159         if (dst_hold_safe(&rt->dst))
1160                 return true;
1161         if (net) {
1162                 rt = net->ipv6.ip6_null_entry;
1163                 dst_hold(&rt->dst);
1164         } else {
1165                 rt = NULL;
1166         }
1167         *prt = rt;
1168         return false;
1169 }
1170
1171 /* called with rcu_lock held */
1172 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1173 {
1174         struct net_device *dev = res->nh->fib_nh_dev;
1175         struct fib6_info *f6i = res->f6i;
1176         unsigned short flags;
1177         struct rt6_info *nrt;
1178
1179         if (!fib6_info_hold_safe(f6i))
1180                 goto fallback;
1181
1182         flags = fib6_info_dst_flags(f6i);
1183         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1184         if (!nrt) {
1185                 fib6_info_release(f6i);
1186                 goto fallback;
1187         }
1188
1189         ip6_rt_copy_init(nrt, res);
1190         return nrt;
1191
1192 fallback:
1193         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1194         dst_hold(&nrt->dst);
1195         return nrt;
1196 }
1197
1198 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1199                                              struct fib6_table *table,
1200                                              struct flowi6 *fl6,
1201                                              const struct sk_buff *skb,
1202                                              int flags)
1203 {
1204         struct fib6_result res = {};
1205         struct fib6_node *fn;
1206         struct rt6_info *rt;
1207
1208         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1209                 flags &= ~RT6_LOOKUP_F_IFACE;
1210
1211         rcu_read_lock();
1212         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1213 restart:
1214         res.f6i = rcu_dereference(fn->leaf);
1215         if (!res.f6i)
1216                 res.f6i = net->ipv6.fib6_null_entry;
1217         else
1218                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1219                                  flags);
1220
1221         if (res.f6i == net->ipv6.fib6_null_entry) {
1222                 fn = fib6_backtrack(fn, &fl6->saddr);
1223                 if (fn)
1224                         goto restart;
1225
1226                 rt = net->ipv6.ip6_null_entry;
1227                 dst_hold(&rt->dst);
1228                 goto out;
1229         } else if (res.fib6_flags & RTF_REJECT) {
1230                 goto do_create;
1231         }
1232
1233         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1234                          fl6->flowi6_oif != 0, skb, flags);
1235
1236         /* Search through exception table */
1237         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1238         if (rt) {
1239                 if (ip6_hold_safe(net, &rt))
1240                         dst_use_noref(&rt->dst, jiffies);
1241         } else {
1242 do_create:
1243                 rt = ip6_create_rt_rcu(&res);
1244         }
1245
1246 out:
1247         trace_fib6_table_lookup(net, &res, table, fl6);
1248
1249         rcu_read_unlock();
1250
1251         return rt;
1252 }
1253
1254 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1255                                    const struct sk_buff *skb, int flags)
1256 {
1257         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1258 }
1259 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1260
1261 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1262                             const struct in6_addr *saddr, int oif,
1263                             const struct sk_buff *skb, int strict)
1264 {
1265         struct flowi6 fl6 = {
1266                 .flowi6_oif = oif,
1267                 .daddr = *daddr,
1268         };
1269         struct dst_entry *dst;
1270         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1271
1272         if (saddr) {
1273                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1274                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1275         }
1276
1277         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1278         if (dst->error == 0)
1279                 return (struct rt6_info *) dst;
1280
1281         dst_release(dst);
1282
1283         return NULL;
1284 }
1285 EXPORT_SYMBOL(rt6_lookup);
1286
1287 /* ip6_ins_rt is called with FREE table->tb6_lock.
1288  * It takes new route entry, the addition fails by any reason the
1289  * route is released.
1290  * Caller must hold dst before calling it.
1291  */
1292
1293 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1294                         struct netlink_ext_ack *extack)
1295 {
1296         int err;
1297         struct fib6_table *table;
1298
1299         table = rt->fib6_table;
1300         spin_lock_bh(&table->tb6_lock);
1301         err = fib6_add(&table->tb6_root, rt, info, extack);
1302         spin_unlock_bh(&table->tb6_lock);
1303
1304         return err;
1305 }
1306
1307 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1308 {
1309         struct nl_info info = { .nl_net = net, };
1310
1311         return __ip6_ins_rt(rt, &info, NULL);
1312 }
1313
1314 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1315                                            const struct in6_addr *daddr,
1316                                            const struct in6_addr *saddr)
1317 {
1318         struct fib6_info *f6i = res->f6i;
1319         struct net_device *dev;
1320         struct rt6_info *rt;
1321
1322         /*
1323          *      Clone the route.
1324          */
1325
1326         if (!fib6_info_hold_safe(f6i))
1327                 return NULL;
1328
1329         dev = ip6_rt_get_dev_rcu(res);
1330         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1331         if (!rt) {
1332                 fib6_info_release(f6i);
1333                 return NULL;
1334         }
1335
1336         ip6_rt_copy_init(rt, res);
1337         rt->rt6i_flags |= RTF_CACHE;
1338         rt->rt6i_dst.addr = *daddr;
1339         rt->rt6i_dst.plen = 128;
1340
1341         if (!rt6_is_gw_or_nonexthop(res)) {
1342                 if (f6i->fib6_dst.plen != 128 &&
1343                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1344                         rt->rt6i_flags |= RTF_ANYCAST;
1345 #ifdef CONFIG_IPV6_SUBTREES
1346                 if (rt->rt6i_src.plen && saddr) {
1347                         rt->rt6i_src.addr = *saddr;
1348                         rt->rt6i_src.plen = 128;
1349                 }
1350 #endif
1351         }
1352
1353         return rt;
1354 }
1355
1356 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1357 {
1358         struct fib6_info *f6i = res->f6i;
1359         unsigned short flags = fib6_info_dst_flags(f6i);
1360         struct net_device *dev;
1361         struct rt6_info *pcpu_rt;
1362
1363         if (!fib6_info_hold_safe(f6i))
1364                 return NULL;
1365
1366         rcu_read_lock();
1367         dev = ip6_rt_get_dev_rcu(res);
1368         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1369         rcu_read_unlock();
1370         if (!pcpu_rt) {
1371                 fib6_info_release(f6i);
1372                 return NULL;
1373         }
1374         ip6_rt_copy_init(pcpu_rt, res);
1375         pcpu_rt->rt6i_flags |= RTF_PCPU;
1376
1377         if (f6i->nh)
1378                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1379
1380         return pcpu_rt;
1381 }
1382
1383 static bool rt6_is_valid(const struct rt6_info *rt6)
1384 {
1385         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
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         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1396                 struct rt6_info *prev, **p;
1397
1398                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1399                 prev = xchg(p, NULL);
1400                 if (prev) {
1401                         dst_dev_put(&prev->dst);
1402                         dst_release(&prev->dst);
1403                 }
1404
1405                 pcpu_rt = NULL;
1406         }
1407
1408         return pcpu_rt;
1409 }
1410
1411 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1412                                             const struct fib6_result *res)
1413 {
1414         struct rt6_info *pcpu_rt, *prev, **p;
1415
1416         pcpu_rt = ip6_rt_pcpu_alloc(res);
1417         if (!pcpu_rt)
1418                 return NULL;
1419
1420         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1421         prev = cmpxchg(p, NULL, pcpu_rt);
1422         BUG_ON(prev);
1423
1424         if (res->f6i->fib6_destroying) {
1425                 struct fib6_info *from;
1426
1427                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1428                 fib6_info_release(from);
1429         }
1430
1431         return pcpu_rt;
1432 }
1433
1434 /* exception hash table implementation
1435  */
1436 static DEFINE_SPINLOCK(rt6_exception_lock);
1437
1438 /* Remove rt6_ex from hash table and free the memory
1439  * Caller must hold rt6_exception_lock
1440  */
1441 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1442                                  struct rt6_exception *rt6_ex)
1443 {
1444         struct fib6_info *from;
1445         struct net *net;
1446
1447         if (!bucket || !rt6_ex)
1448                 return;
1449
1450         net = dev_net(rt6_ex->rt6i->dst.dev);
1451         net->ipv6.rt6_stats->fib_rt_cache--;
1452
1453         /* purge completely the exception to allow releasing the held resources:
1454          * some [sk] cache may keep the dst around for unlimited time
1455          */
1456         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1457         fib6_info_release(from);
1458         dst_dev_put(&rt6_ex->rt6i->dst);
1459
1460         hlist_del_rcu(&rt6_ex->hlist);
1461         dst_release(&rt6_ex->rt6i->dst);
1462         kfree_rcu(rt6_ex, rcu);
1463         WARN_ON_ONCE(!bucket->depth);
1464         bucket->depth--;
1465 }
1466
1467 /* Remove oldest rt6_ex in bucket and free the memory
1468  * Caller must hold rt6_exception_lock
1469  */
1470 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1471 {
1472         struct rt6_exception *rt6_ex, *oldest = NULL;
1473
1474         if (!bucket)
1475                 return;
1476
1477         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1478                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1479                         oldest = rt6_ex;
1480         }
1481         rt6_remove_exception(bucket, oldest);
1482 }
1483
1484 static u32 rt6_exception_hash(const struct in6_addr *dst,
1485                               const struct in6_addr *src)
1486 {
1487         static u32 seed __read_mostly;
1488         u32 val;
1489
1490         net_get_random_once(&seed, sizeof(seed));
1491         val = jhash2((const u32 *)dst, sizeof(*dst)/sizeof(u32), seed);
1492
1493 #ifdef CONFIG_IPV6_SUBTREES
1494         if (src)
1495                 val = jhash2((const u32 *)src, sizeof(*src)/sizeof(u32), val);
1496 #endif
1497         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1498 }
1499
1500 /* Helper function to find the cached rt in the hash table
1501  * and update bucket pointer to point to the bucket for this
1502  * (daddr, saddr) pair
1503  * Caller must hold rt6_exception_lock
1504  */
1505 static struct rt6_exception *
1506 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1507                               const struct in6_addr *daddr,
1508                               const struct in6_addr *saddr)
1509 {
1510         struct rt6_exception *rt6_ex;
1511         u32 hval;
1512
1513         if (!(*bucket) || !daddr)
1514                 return NULL;
1515
1516         hval = rt6_exception_hash(daddr, saddr);
1517         *bucket += hval;
1518
1519         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1520                 struct rt6_info *rt6 = rt6_ex->rt6i;
1521                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1522
1523 #ifdef CONFIG_IPV6_SUBTREES
1524                 if (matched && saddr)
1525                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1526 #endif
1527                 if (matched)
1528                         return rt6_ex;
1529         }
1530         return NULL;
1531 }
1532
1533 /* Helper function to find the cached rt in the hash table
1534  * and update bucket pointer to point to the bucket for this
1535  * (daddr, saddr) pair
1536  * Caller must hold rcu_read_lock()
1537  */
1538 static struct rt6_exception *
1539 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1540                          const struct in6_addr *daddr,
1541                          const struct in6_addr *saddr)
1542 {
1543         struct rt6_exception *rt6_ex;
1544         u32 hval;
1545
1546         WARN_ON_ONCE(!rcu_read_lock_held());
1547
1548         if (!(*bucket) || !daddr)
1549                 return NULL;
1550
1551         hval = rt6_exception_hash(daddr, saddr);
1552         *bucket += hval;
1553
1554         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1555                 struct rt6_info *rt6 = rt6_ex->rt6i;
1556                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1557
1558 #ifdef CONFIG_IPV6_SUBTREES
1559                 if (matched && saddr)
1560                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1561 #endif
1562                 if (matched)
1563                         return rt6_ex;
1564         }
1565         return NULL;
1566 }
1567
1568 static unsigned int fib6_mtu(const struct fib6_result *res)
1569 {
1570         const struct fib6_nh *nh = res->nh;
1571         unsigned int mtu;
1572
1573         if (res->f6i->fib6_pmtu) {
1574                 mtu = res->f6i->fib6_pmtu;
1575         } else {
1576                 struct net_device *dev = nh->fib_nh_dev;
1577                 struct inet6_dev *idev;
1578
1579                 rcu_read_lock();
1580                 idev = __in6_dev_get(dev);
1581                 mtu = idev->cnf.mtu6;
1582                 rcu_read_unlock();
1583         }
1584
1585         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1586
1587         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1588 }
1589
1590 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1591
1592 /* used when the flushed bit is not relevant, only access to the bucket
1593  * (ie., all bucket users except rt6_insert_exception);
1594  *
1595  * called under rcu lock; sometimes called with rt6_exception_lock held
1596  */
1597 static
1598 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1599                                                        spinlock_t *lock)
1600 {
1601         struct rt6_exception_bucket *bucket;
1602
1603         if (lock)
1604                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1605                                                    lockdep_is_held(lock));
1606         else
1607                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1608
1609         /* remove bucket flushed bit if set */
1610         if (bucket) {
1611                 unsigned long p = (unsigned long)bucket;
1612
1613                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1614                 bucket = (struct rt6_exception_bucket *)p;
1615         }
1616
1617         return bucket;
1618 }
1619
1620 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1621 {
1622         unsigned long p = (unsigned long)bucket;
1623
1624         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1625 }
1626
1627 /* called with rt6_exception_lock held */
1628 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1629                                               spinlock_t *lock)
1630 {
1631         struct rt6_exception_bucket *bucket;
1632         unsigned long p;
1633
1634         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1635                                            lockdep_is_held(lock));
1636
1637         p = (unsigned long)bucket;
1638         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1639         bucket = (struct rt6_exception_bucket *)p;
1640         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1641 }
1642
1643 static int rt6_insert_exception(struct rt6_info *nrt,
1644                                 const struct fib6_result *res)
1645 {
1646         struct net *net = dev_net(nrt->dst.dev);
1647         struct rt6_exception_bucket *bucket;
1648         struct fib6_info *f6i = res->f6i;
1649         struct in6_addr *src_key = NULL;
1650         struct rt6_exception *rt6_ex;
1651         struct fib6_nh *nh = res->nh;
1652         int err = 0;
1653
1654         spin_lock_bh(&rt6_exception_lock);
1655
1656         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1657                                           lockdep_is_held(&rt6_exception_lock));
1658         if (!bucket) {
1659                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1660                                  GFP_ATOMIC);
1661                 if (!bucket) {
1662                         err = -ENOMEM;
1663                         goto out;
1664                 }
1665                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1666         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1667                 err = -EINVAL;
1668                 goto out;
1669         }
1670
1671 #ifdef CONFIG_IPV6_SUBTREES
1672         /* fib6_src.plen != 0 indicates f6i is in subtree
1673          * and exception table is indexed by a hash of
1674          * both fib6_dst and fib6_src.
1675          * Otherwise, the exception table is indexed by
1676          * a hash of only fib6_dst.
1677          */
1678         if (f6i->fib6_src.plen)
1679                 src_key = &nrt->rt6i_src.addr;
1680 #endif
1681         /* rt6_mtu_change() might lower mtu on f6i.
1682          * Only insert this exception route if its mtu
1683          * is less than f6i's mtu value.
1684          */
1685         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1686                 err = -EINVAL;
1687                 goto out;
1688         }
1689
1690         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1691                                                src_key);
1692         if (rt6_ex)
1693                 rt6_remove_exception(bucket, rt6_ex);
1694
1695         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1696         if (!rt6_ex) {
1697                 err = -ENOMEM;
1698                 goto out;
1699         }
1700         rt6_ex->rt6i = nrt;
1701         rt6_ex->stamp = jiffies;
1702         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1703         bucket->depth++;
1704         net->ipv6.rt6_stats->fib_rt_cache++;
1705
1706         if (bucket->depth > FIB6_MAX_DEPTH)
1707                 rt6_exception_remove_oldest(bucket);
1708
1709 out:
1710         spin_unlock_bh(&rt6_exception_lock);
1711
1712         /* Update fn->fn_sernum to invalidate all cached dst */
1713         if (!err) {
1714                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1715                 fib6_update_sernum(net, f6i);
1716                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1717                 fib6_force_start_gc(net);
1718         }
1719
1720         return err;
1721 }
1722
1723 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1724 {
1725         struct rt6_exception_bucket *bucket;
1726         struct rt6_exception *rt6_ex;
1727         struct hlist_node *tmp;
1728         int i;
1729
1730         spin_lock_bh(&rt6_exception_lock);
1731
1732         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1733         if (!bucket)
1734                 goto out;
1735
1736         /* Prevent rt6_insert_exception() to recreate the bucket list */
1737         if (!from)
1738                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1739
1740         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1741                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1742                         if (!from ||
1743                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1744                                 rt6_remove_exception(bucket, rt6_ex);
1745                 }
1746                 WARN_ON_ONCE(!from && bucket->depth);
1747                 bucket++;
1748         }
1749 out:
1750         spin_unlock_bh(&rt6_exception_lock);
1751 }
1752
1753 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1754 {
1755         struct fib6_info *f6i = arg;
1756
1757         fib6_nh_flush_exceptions(nh, f6i);
1758
1759         return 0;
1760 }
1761
1762 void rt6_flush_exceptions(struct fib6_info *f6i)
1763 {
1764         if (f6i->nh)
1765                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1766                                          f6i);
1767         else
1768                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1769 }
1770
1771 /* Find cached rt in the hash table inside passed in rt
1772  * Caller has to hold rcu_read_lock()
1773  */
1774 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1775                                            const struct in6_addr *daddr,
1776                                            const struct in6_addr *saddr)
1777 {
1778         const struct in6_addr *src_key = NULL;
1779         struct rt6_exception_bucket *bucket;
1780         struct rt6_exception *rt6_ex;
1781         struct rt6_info *ret = NULL;
1782
1783 #ifdef CONFIG_IPV6_SUBTREES
1784         /* fib6i_src.plen != 0 indicates f6i is in subtree
1785          * and exception table is indexed by a hash of
1786          * both fib6_dst and fib6_src.
1787          * However, the src addr used to create the hash
1788          * might not be exactly the passed in saddr which
1789          * is a /128 addr from the flow.
1790          * So we need to use f6i->fib6_src to redo lookup
1791          * if the passed in saddr does not find anything.
1792          * (See the logic in ip6_rt_cache_alloc() on how
1793          * rt->rt6i_src is updated.)
1794          */
1795         if (res->f6i->fib6_src.plen)
1796                 src_key = saddr;
1797 find_ex:
1798 #endif
1799         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1800         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1801
1802         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1803                 ret = rt6_ex->rt6i;
1804
1805 #ifdef CONFIG_IPV6_SUBTREES
1806         /* Use fib6_src as src_key and redo lookup */
1807         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1808                 src_key = &res->f6i->fib6_src.addr;
1809                 goto find_ex;
1810         }
1811 #endif
1812
1813         return ret;
1814 }
1815
1816 /* Remove the passed in cached rt from the hash table that contains it */
1817 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1818                                     const struct rt6_info *rt)
1819 {
1820         const struct in6_addr *src_key = NULL;
1821         struct rt6_exception_bucket *bucket;
1822         struct rt6_exception *rt6_ex;
1823         int err;
1824
1825         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1826                 return -ENOENT;
1827
1828         spin_lock_bh(&rt6_exception_lock);
1829         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1830
1831 #ifdef CONFIG_IPV6_SUBTREES
1832         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1833          * and exception table is indexed by a hash of
1834          * both rt6i_dst and rt6i_src.
1835          * Otherwise, the exception table is indexed by
1836          * a hash of only rt6i_dst.
1837          */
1838         if (plen)
1839                 src_key = &rt->rt6i_src.addr;
1840 #endif
1841         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1842                                                &rt->rt6i_dst.addr,
1843                                                src_key);
1844         if (rt6_ex) {
1845                 rt6_remove_exception(bucket, rt6_ex);
1846                 err = 0;
1847         } else {
1848                 err = -ENOENT;
1849         }
1850
1851         spin_unlock_bh(&rt6_exception_lock);
1852         return err;
1853 }
1854
1855 struct fib6_nh_excptn_arg {
1856         struct rt6_info *rt;
1857         int             plen;
1858 };
1859
1860 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1861 {
1862         struct fib6_nh_excptn_arg *arg = _arg;
1863         int err;
1864
1865         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1866         if (err == 0)
1867                 return 1;
1868
1869         return 0;
1870 }
1871
1872 static int rt6_remove_exception_rt(struct rt6_info *rt)
1873 {
1874         struct fib6_info *from;
1875
1876         from = rcu_dereference(rt->from);
1877         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1878                 return -EINVAL;
1879
1880         if (from->nh) {
1881                 struct fib6_nh_excptn_arg arg = {
1882                         .rt = rt,
1883                         .plen = from->fib6_src.plen
1884                 };
1885                 int rc;
1886
1887                 /* rc = 1 means an entry was found */
1888                 rc = nexthop_for_each_fib6_nh(from->nh,
1889                                               rt6_nh_remove_exception_rt,
1890                                               &arg);
1891                 return rc ? 0 : -ENOENT;
1892         }
1893
1894         return fib6_nh_remove_exception(from->fib6_nh,
1895                                         from->fib6_src.plen, rt);
1896 }
1897
1898 /* Find rt6_ex which contains the passed in rt cache and
1899  * refresh its stamp
1900  */
1901 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1902                                      const struct rt6_info *rt)
1903 {
1904         const struct in6_addr *src_key = NULL;
1905         struct rt6_exception_bucket *bucket;
1906         struct rt6_exception *rt6_ex;
1907
1908         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1909 #ifdef CONFIG_IPV6_SUBTREES
1910         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1911          * and exception table is indexed by a hash of
1912          * both rt6i_dst and rt6i_src.
1913          * Otherwise, the exception table is indexed by
1914          * a hash of only rt6i_dst.
1915          */
1916         if (plen)
1917                 src_key = &rt->rt6i_src.addr;
1918 #endif
1919         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1920         if (rt6_ex)
1921                 rt6_ex->stamp = jiffies;
1922 }
1923
1924 struct fib6_nh_match_arg {
1925         const struct net_device *dev;
1926         const struct in6_addr   *gw;
1927         struct fib6_nh          *match;
1928 };
1929
1930 /* determine if fib6_nh has given device and gateway */
1931 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1932 {
1933         struct fib6_nh_match_arg *arg = _arg;
1934
1935         if (arg->dev != nh->fib_nh_dev ||
1936             (arg->gw && !nh->fib_nh_gw_family) ||
1937             (!arg->gw && nh->fib_nh_gw_family) ||
1938             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1939                 return 0;
1940
1941         arg->match = nh;
1942
1943         /* found a match, break the loop */
1944         return 1;
1945 }
1946
1947 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1948 {
1949         struct fib6_info *from;
1950         struct fib6_nh *fib6_nh;
1951
1952         rcu_read_lock();
1953
1954         from = rcu_dereference(rt->from);
1955         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1956                 goto unlock;
1957
1958         if (from->nh) {
1959                 struct fib6_nh_match_arg arg = {
1960                         .dev = rt->dst.dev,
1961                         .gw = &rt->rt6i_gateway,
1962                 };
1963
1964                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1965
1966                 if (!arg.match)
1967                         goto unlock;
1968                 fib6_nh = arg.match;
1969         } else {
1970                 fib6_nh = from->fib6_nh;
1971         }
1972         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1973 unlock:
1974         rcu_read_unlock();
1975 }
1976
1977 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1978                                          struct rt6_info *rt, int mtu)
1979 {
1980         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1981          * lowest MTU in the path: always allow updating the route PMTU to
1982          * reflect PMTU decreases.
1983          *
1984          * If the new MTU is higher, and the route PMTU is equal to the local
1985          * MTU, this means the old MTU is the lowest in the path, so allow
1986          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1987          * handle this.
1988          */
1989
1990         if (dst_mtu(&rt->dst) >= mtu)
1991                 return true;
1992
1993         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1994                 return true;
1995
1996         return false;
1997 }
1998
1999 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2000                                        const struct fib6_nh *nh, int mtu)
2001 {
2002         struct rt6_exception_bucket *bucket;
2003         struct rt6_exception *rt6_ex;
2004         int i;
2005
2006         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2007         if (!bucket)
2008                 return;
2009
2010         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2011                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2012                         struct rt6_info *entry = rt6_ex->rt6i;
2013
2014                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2015                          * route), the metrics of its rt->from have already
2016                          * been updated.
2017                          */
2018                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2019                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2020                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2021                 }
2022                 bucket++;
2023         }
2024 }
2025
2026 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2027
2028 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2029                                             const struct in6_addr *gateway)
2030 {
2031         struct rt6_exception_bucket *bucket;
2032         struct rt6_exception *rt6_ex;
2033         struct hlist_node *tmp;
2034         int i;
2035
2036         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2037                 return;
2038
2039         spin_lock_bh(&rt6_exception_lock);
2040         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2041         if (bucket) {
2042                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2043                         hlist_for_each_entry_safe(rt6_ex, tmp,
2044                                                   &bucket->chain, hlist) {
2045                                 struct rt6_info *entry = rt6_ex->rt6i;
2046
2047                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2048                                     RTF_CACHE_GATEWAY &&
2049                                     ipv6_addr_equal(gateway,
2050                                                     &entry->rt6i_gateway)) {
2051                                         rt6_remove_exception(bucket, rt6_ex);
2052                                 }
2053                         }
2054                         bucket++;
2055                 }
2056         }
2057
2058         spin_unlock_bh(&rt6_exception_lock);
2059 }
2060
2061 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2062                                       struct rt6_exception *rt6_ex,
2063                                       struct fib6_gc_args *gc_args,
2064                                       unsigned long now)
2065 {
2066         struct rt6_info *rt = rt6_ex->rt6i;
2067
2068         /* we are pruning and obsoleting aged-out and non gateway exceptions
2069          * even if others have still references to them, so that on next
2070          * dst_check() such references can be dropped.
2071          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2072          * expired, independently from their aging, as per RFC 8201 section 4
2073          */
2074         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2075                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2076                         RT6_TRACE("aging clone %p\n", rt);
2077                         rt6_remove_exception(bucket, rt6_ex);
2078                         return;
2079                 }
2080         } else if (time_after(jiffies, rt->dst.expires)) {
2081                 RT6_TRACE("purging expired route %p\n", rt);
2082                 rt6_remove_exception(bucket, rt6_ex);
2083                 return;
2084         }
2085
2086         if (rt->rt6i_flags & RTF_GATEWAY) {
2087                 struct neighbour *neigh;
2088
2089                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2090
2091                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2092                         RT6_TRACE("purging route %p via non-router but gateway\n",
2093                                   rt);
2094                         rt6_remove_exception(bucket, rt6_ex);
2095                         return;
2096                 }
2097         }
2098
2099         gc_args->more++;
2100 }
2101
2102 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2103                                    struct fib6_gc_args *gc_args,
2104                                    unsigned long now)
2105 {
2106         struct rt6_exception_bucket *bucket;
2107         struct rt6_exception *rt6_ex;
2108         struct hlist_node *tmp;
2109         int i;
2110
2111         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2112                 return;
2113
2114         rcu_read_lock_bh();
2115         spin_lock(&rt6_exception_lock);
2116         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2117         if (bucket) {
2118                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2119                         hlist_for_each_entry_safe(rt6_ex, tmp,
2120                                                   &bucket->chain, hlist) {
2121                                 rt6_age_examine_exception(bucket, rt6_ex,
2122                                                           gc_args, now);
2123                         }
2124                         bucket++;
2125                 }
2126         }
2127         spin_unlock(&rt6_exception_lock);
2128         rcu_read_unlock_bh();
2129 }
2130
2131 struct fib6_nh_age_excptn_arg {
2132         struct fib6_gc_args     *gc_args;
2133         unsigned long           now;
2134 };
2135
2136 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2137 {
2138         struct fib6_nh_age_excptn_arg *arg = _arg;
2139
2140         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2141         return 0;
2142 }
2143
2144 void rt6_age_exceptions(struct fib6_info *f6i,
2145                         struct fib6_gc_args *gc_args,
2146                         unsigned long now)
2147 {
2148         if (f6i->nh) {
2149                 struct fib6_nh_age_excptn_arg arg = {
2150                         .gc_args = gc_args,
2151                         .now = now
2152                 };
2153
2154                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2155                                          &arg);
2156         } else {
2157                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2158         }
2159 }
2160
2161 /* must be called with rcu lock held */
2162 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2163                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2164 {
2165         struct fib6_node *fn, *saved_fn;
2166
2167         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2168         saved_fn = fn;
2169
2170         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2171                 oif = 0;
2172
2173 redo_rt6_select:
2174         rt6_select(net, fn, oif, res, strict);
2175         if (res->f6i == net->ipv6.fib6_null_entry) {
2176                 fn = fib6_backtrack(fn, &fl6->saddr);
2177                 if (fn)
2178                         goto redo_rt6_select;
2179                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2180                         /* also consider unreachable route */
2181                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2182                         fn = saved_fn;
2183                         goto redo_rt6_select;
2184                 }
2185         }
2186
2187         trace_fib6_table_lookup(net, res, table, fl6);
2188
2189         return 0;
2190 }
2191
2192 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2193                                int oif, struct flowi6 *fl6,
2194                                const struct sk_buff *skb, int flags)
2195 {
2196         struct fib6_result res = {};
2197         struct rt6_info *rt = NULL;
2198         int strict = 0;
2199
2200         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2201                      !rcu_read_lock_held());
2202
2203         strict |= flags & RT6_LOOKUP_F_IFACE;
2204         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2205         if (net->ipv6.devconf_all->forwarding == 0)
2206                 strict |= RT6_LOOKUP_F_REACHABLE;
2207
2208         rcu_read_lock();
2209
2210         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2211         if (res.f6i == net->ipv6.fib6_null_entry)
2212                 goto out;
2213
2214         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2215
2216         /*Search through exception table */
2217         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2218         if (rt) {
2219                 goto out;
2220         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2221                             !res.nh->fib_nh_gw_family)) {
2222                 /* Create a RTF_CACHE clone which will not be
2223                  * owned by the fib6 tree.  It is for the special case where
2224                  * the daddr in the skb during the neighbor look-up is different
2225                  * from the fl6->daddr used to look-up route here.
2226                  */
2227                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2228
2229                 if (rt) {
2230                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2231                          * As rt6_uncached_list_add() does not consume refcnt,
2232                          * this refcnt is always returned to the caller even
2233                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2234                          */
2235                         rt6_uncached_list_add(rt);
2236                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2237                         rcu_read_unlock();
2238
2239                         return rt;
2240                 }
2241         } else {
2242                 /* Get a percpu copy */
2243                 local_bh_disable();
2244                 rt = rt6_get_pcpu_route(&res);
2245
2246                 if (!rt)
2247                         rt = rt6_make_pcpu_route(net, &res);
2248
2249                 local_bh_enable();
2250         }
2251 out:
2252         if (!rt)
2253                 rt = net->ipv6.ip6_null_entry;
2254         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2255                 ip6_hold_safe(net, &rt);
2256         rcu_read_unlock();
2257
2258         return rt;
2259 }
2260 EXPORT_SYMBOL_GPL(ip6_pol_route);
2261
2262 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2263                                             struct fib6_table *table,
2264                                             struct flowi6 *fl6,
2265                                             const struct sk_buff *skb,
2266                                             int flags)
2267 {
2268         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2269 }
2270
2271 struct dst_entry *ip6_route_input_lookup(struct net *net,
2272                                          struct net_device *dev,
2273                                          struct flowi6 *fl6,
2274                                          const struct sk_buff *skb,
2275                                          int flags)
2276 {
2277         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2278                 flags |= RT6_LOOKUP_F_IFACE;
2279
2280         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2281 }
2282 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2283
2284 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2285                                   struct flow_keys *keys,
2286                                   struct flow_keys *flkeys)
2287 {
2288         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2289         const struct ipv6hdr *key_iph = outer_iph;
2290         struct flow_keys *_flkeys = flkeys;
2291         const struct ipv6hdr *inner_iph;
2292         const struct icmp6hdr *icmph;
2293         struct ipv6hdr _inner_iph;
2294         struct icmp6hdr _icmph;
2295
2296         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2297                 goto out;
2298
2299         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2300                                    sizeof(_icmph), &_icmph);
2301         if (!icmph)
2302                 goto out;
2303
2304         if (!icmpv6_is_err(icmph->icmp6_type))
2305                 goto out;
2306
2307         inner_iph = skb_header_pointer(skb,
2308                                        skb_transport_offset(skb) + sizeof(*icmph),
2309                                        sizeof(_inner_iph), &_inner_iph);
2310         if (!inner_iph)
2311                 goto out;
2312
2313         key_iph = inner_iph;
2314         _flkeys = NULL;
2315 out:
2316         if (_flkeys) {
2317                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2318                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2319                 keys->tags.flow_label = _flkeys->tags.flow_label;
2320                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2321         } else {
2322                 keys->addrs.v6addrs.src = key_iph->saddr;
2323                 keys->addrs.v6addrs.dst = key_iph->daddr;
2324                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2325                 keys->basic.ip_proto = key_iph->nexthdr;
2326         }
2327 }
2328
2329 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2330                                            const struct sk_buff *skb,
2331                                            bool *p_has_inner)
2332 {
2333         u32 hash_fields = ip6_multipath_hash_fields(net);
2334         struct flow_keys keys, hash_keys;
2335
2336         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2337                 return 0;
2338
2339         memset(&hash_keys, 0, sizeof(hash_keys));
2340         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2341
2342         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2343         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2344                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2345         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2346                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2347         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2348                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2349         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2350                 hash_keys.tags.flow_label = keys.tags.flow_label;
2351         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2352                 hash_keys.ports.src = keys.ports.src;
2353         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2354                 hash_keys.ports.dst = keys.ports.dst;
2355
2356         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2357         return flow_hash_from_keys(&hash_keys);
2358 }
2359
2360 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2361                                            const struct sk_buff *skb,
2362                                            bool has_inner)
2363 {
2364         u32 hash_fields = ip6_multipath_hash_fields(net);
2365         struct flow_keys keys, hash_keys;
2366
2367         /* We assume the packet carries an encapsulation, but if none was
2368          * encountered during dissection of the outer flow, then there is no
2369          * point in calling the flow dissector again.
2370          */
2371         if (!has_inner)
2372                 return 0;
2373
2374         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2375                 return 0;
2376
2377         memset(&hash_keys, 0, sizeof(hash_keys));
2378         skb_flow_dissect_flow_keys(skb, &keys, 0);
2379
2380         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2381                 return 0;
2382
2383         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2384                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2385                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2386                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2387                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2388                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2389         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2390                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2391                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2392                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2393                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2394                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2395                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2396                         hash_keys.tags.flow_label = keys.tags.flow_label;
2397         }
2398
2399         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2400                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2401         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2402                 hash_keys.ports.src = keys.ports.src;
2403         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2404                 hash_keys.ports.dst = keys.ports.dst;
2405
2406         return flow_hash_from_keys(&hash_keys);
2407 }
2408
2409 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2410                                          const struct sk_buff *skb)
2411 {
2412         u32 mhash, mhash_inner;
2413         bool has_inner = true;
2414
2415         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2416         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2417
2418         return jhash_2words(mhash, mhash_inner, 0);
2419 }
2420
2421 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2422                                          const struct flowi6 *fl6)
2423 {
2424         u32 hash_fields = ip6_multipath_hash_fields(net);
2425         struct flow_keys hash_keys;
2426
2427         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2428                 return 0;
2429
2430         memset(&hash_keys, 0, sizeof(hash_keys));
2431         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2432         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2433                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2434         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2435                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2436         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2437                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2438         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2439                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2440         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2441                 hash_keys.ports.src = fl6->fl6_sport;
2442         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2443                 hash_keys.ports.dst = fl6->fl6_dport;
2444
2445         return flow_hash_from_keys(&hash_keys);
2446 }
2447
2448 /* if skb is set it will be used and fl6 can be NULL */
2449 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2450                        const struct sk_buff *skb, struct flow_keys *flkeys)
2451 {
2452         struct flow_keys hash_keys;
2453         u32 mhash = 0;
2454
2455         switch (ip6_multipath_hash_policy(net)) {
2456         case 0:
2457                 memset(&hash_keys, 0, sizeof(hash_keys));
2458                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2459                 if (skb) {
2460                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2461                 } else {
2462                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2463                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2464                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2465                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2466                 }
2467                 mhash = flow_hash_from_keys(&hash_keys);
2468                 break;
2469         case 1:
2470                 if (skb) {
2471                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2472                         struct flow_keys keys;
2473
2474                         /* short-circuit if we already have L4 hash present */
2475                         if (skb->l4_hash)
2476                                 return skb_get_hash_raw(skb) >> 1;
2477
2478                         memset(&hash_keys, 0, sizeof(hash_keys));
2479
2480                         if (!flkeys) {
2481                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2482                                 flkeys = &keys;
2483                         }
2484                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2485                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2486                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2487                         hash_keys.ports.src = flkeys->ports.src;
2488                         hash_keys.ports.dst = flkeys->ports.dst;
2489                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2490                 } else {
2491                         memset(&hash_keys, 0, sizeof(hash_keys));
2492                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2493                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2494                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2495                         hash_keys.ports.src = fl6->fl6_sport;
2496                         hash_keys.ports.dst = fl6->fl6_dport;
2497                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2498                 }
2499                 mhash = flow_hash_from_keys(&hash_keys);
2500                 break;
2501         case 2:
2502                 memset(&hash_keys, 0, sizeof(hash_keys));
2503                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2504                 if (skb) {
2505                         struct flow_keys keys;
2506
2507                         if (!flkeys) {
2508                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2509                                 flkeys = &keys;
2510                         }
2511
2512                         /* Inner can be v4 or v6 */
2513                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2514                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2515                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2516                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2517                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2518                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2519                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2520                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2521                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2522                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2523                         } else {
2524                                 /* Same as case 0 */
2525                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2526                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2527                         }
2528                 } else {
2529                         /* Same as case 0 */
2530                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2531                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2532                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2533                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2534                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2535                 }
2536                 mhash = flow_hash_from_keys(&hash_keys);
2537                 break;
2538         case 3:
2539                 if (skb)
2540                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2541                 else
2542                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2543                 break;
2544         }
2545
2546         return mhash >> 1;
2547 }
2548
2549 /* Called with rcu held */
2550 void ip6_route_input(struct sk_buff *skb)
2551 {
2552         const struct ipv6hdr *iph = ipv6_hdr(skb);
2553         struct net *net = dev_net(skb->dev);
2554         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2555         struct ip_tunnel_info *tun_info;
2556         struct flowi6 fl6 = {
2557                 .flowi6_iif = skb->dev->ifindex,
2558                 .daddr = iph->daddr,
2559                 .saddr = iph->saddr,
2560                 .flowlabel = ip6_flowinfo(iph),
2561                 .flowi6_mark = skb->mark,
2562                 .flowi6_proto = iph->nexthdr,
2563         };
2564         struct flow_keys *flkeys = NULL, _flkeys;
2565
2566         tun_info = skb_tunnel_info(skb);
2567         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2568                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2569
2570         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2571                 flkeys = &_flkeys;
2572
2573         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2574                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2575         skb_dst_drop(skb);
2576         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2577                                                       &fl6, skb, flags));
2578 }
2579
2580 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2581                                              struct fib6_table *table,
2582                                              struct flowi6 *fl6,
2583                                              const struct sk_buff *skb,
2584                                              int flags)
2585 {
2586         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2587 }
2588
2589 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2590                                                const struct sock *sk,
2591                                                struct flowi6 *fl6, int flags)
2592 {
2593         bool any_src;
2594
2595         if (ipv6_addr_type(&fl6->daddr) &
2596             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2597                 struct dst_entry *dst;
2598
2599                 /* This function does not take refcnt on the dst */
2600                 dst = l3mdev_link_scope_lookup(net, fl6);
2601                 if (dst)
2602                         return dst;
2603         }
2604
2605         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2606
2607         flags |= RT6_LOOKUP_F_DST_NOREF;
2608         any_src = ipv6_addr_any(&fl6->saddr);
2609         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2610             (fl6->flowi6_oif && any_src))
2611                 flags |= RT6_LOOKUP_F_IFACE;
2612
2613         if (!any_src)
2614                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2615         else if (sk)
2616                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2617
2618         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2619 }
2620 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2621
2622 struct dst_entry *ip6_route_output_flags(struct net *net,
2623                                          const struct sock *sk,
2624                                          struct flowi6 *fl6,
2625                                          int flags)
2626 {
2627         struct dst_entry *dst;
2628         struct rt6_info *rt6;
2629
2630         rcu_read_lock();
2631         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2632         rt6 = (struct rt6_info *)dst;
2633         /* For dst cached in uncached_list, refcnt is already taken. */
2634         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2635                 dst = &net->ipv6.ip6_null_entry->dst;
2636                 dst_hold(dst);
2637         }
2638         rcu_read_unlock();
2639
2640         return dst;
2641 }
2642 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2643
2644 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2645 {
2646         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2647         struct net_device *loopback_dev = net->loopback_dev;
2648         struct dst_entry *new = NULL;
2649
2650         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2651                        DST_OBSOLETE_DEAD, 0);
2652         if (rt) {
2653                 rt6_info_init(rt);
2654                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2655
2656                 new = &rt->dst;
2657                 new->__use = 1;
2658                 new->input = dst_discard;
2659                 new->output = dst_discard_out;
2660
2661                 dst_copy_metrics(new, &ort->dst);
2662
2663                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2664                 rt->rt6i_gateway = ort->rt6i_gateway;
2665                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2666
2667                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2668 #ifdef CONFIG_IPV6_SUBTREES
2669                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2670 #endif
2671         }
2672
2673         dst_release(dst_orig);
2674         return new ? new : ERR_PTR(-ENOMEM);
2675 }
2676
2677 /*
2678  *      Destination cache support functions
2679  */
2680
2681 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2682 {
2683         u32 rt_cookie = 0;
2684
2685         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2686                 return false;
2687
2688         if (fib6_check_expired(f6i))
2689                 return false;
2690
2691         return true;
2692 }
2693
2694 static struct dst_entry *rt6_check(struct rt6_info *rt,
2695                                    struct fib6_info *from,
2696                                    u32 cookie)
2697 {
2698         u32 rt_cookie = 0;
2699
2700         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2701             rt_cookie != cookie)
2702                 return NULL;
2703
2704         if (rt6_check_expired(rt))
2705                 return NULL;
2706
2707         return &rt->dst;
2708 }
2709
2710 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2711                                             struct fib6_info *from,
2712                                             u32 cookie)
2713 {
2714         if (!__rt6_check_expired(rt) &&
2715             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2716             fib6_check(from, cookie))
2717                 return &rt->dst;
2718         else
2719                 return NULL;
2720 }
2721
2722 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2723                                                         u32 cookie)
2724 {
2725         struct dst_entry *dst_ret;
2726         struct fib6_info *from;
2727         struct rt6_info *rt;
2728
2729         rt = container_of(dst, struct rt6_info, dst);
2730
2731         if (rt->sernum)
2732                 return rt6_is_valid(rt) ? dst : NULL;
2733
2734         rcu_read_lock();
2735
2736         /* All IPV6 dsts are created with ->obsolete set to the value
2737          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2738          * into this function always.
2739          */
2740
2741         from = rcu_dereference(rt->from);
2742
2743         if (from && (rt->rt6i_flags & RTF_PCPU ||
2744             unlikely(!list_empty(&rt->rt6i_uncached))))
2745                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2746         else
2747                 dst_ret = rt6_check(rt, from, cookie);
2748
2749         rcu_read_unlock();
2750
2751         return dst_ret;
2752 }
2753 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2754
2755 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2756 {
2757         struct rt6_info *rt = (struct rt6_info *) dst;
2758
2759         if (rt) {
2760                 if (rt->rt6i_flags & RTF_CACHE) {
2761                         rcu_read_lock();
2762                         if (rt6_check_expired(rt)) {
2763                                 rt6_remove_exception_rt(rt);
2764                                 dst = NULL;
2765                         }
2766                         rcu_read_unlock();
2767                 } else {
2768                         dst_release(dst);
2769                         dst = NULL;
2770                 }
2771         }
2772         return dst;
2773 }
2774
2775 static void ip6_link_failure(struct sk_buff *skb)
2776 {
2777         struct rt6_info *rt;
2778
2779         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2780
2781         rt = (struct rt6_info *) skb_dst(skb);
2782         if (rt) {
2783                 rcu_read_lock();
2784                 if (rt->rt6i_flags & RTF_CACHE) {
2785                         rt6_remove_exception_rt(rt);
2786                 } else {
2787                         struct fib6_info *from;
2788                         struct fib6_node *fn;
2789
2790                         from = rcu_dereference(rt->from);
2791                         if (from) {
2792                                 fn = rcu_dereference(from->fib6_node);
2793                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2794                                         fn->fn_sernum = -1;
2795                         }
2796                 }
2797                 rcu_read_unlock();
2798         }
2799 }
2800
2801 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2802 {
2803         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2804                 struct fib6_info *from;
2805
2806                 rcu_read_lock();
2807                 from = rcu_dereference(rt0->from);
2808                 if (from)
2809                         rt0->dst.expires = from->expires;
2810                 rcu_read_unlock();
2811         }
2812
2813         dst_set_expires(&rt0->dst, timeout);
2814         rt0->rt6i_flags |= RTF_EXPIRES;
2815 }
2816
2817 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2818 {
2819         struct net *net = dev_net(rt->dst.dev);
2820
2821         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2822         rt->rt6i_flags |= RTF_MODIFIED;
2823         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2824 }
2825
2826 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2827 {
2828         return !(rt->rt6i_flags & RTF_CACHE) &&
2829                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2830 }
2831
2832 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2833                                  const struct ipv6hdr *iph, u32 mtu,
2834                                  bool confirm_neigh)
2835 {
2836         const struct in6_addr *daddr, *saddr;
2837         struct rt6_info *rt6 = (struct rt6_info *)dst;
2838
2839         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2840          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2841          * [see also comment in rt6_mtu_change_route()]
2842          */
2843
2844         if (iph) {
2845                 daddr = &iph->daddr;
2846                 saddr = &iph->saddr;
2847         } else if (sk) {
2848                 daddr = &sk->sk_v6_daddr;
2849                 saddr = &inet6_sk(sk)->saddr;
2850         } else {
2851                 daddr = NULL;
2852                 saddr = NULL;
2853         }
2854
2855         if (confirm_neigh)
2856                 dst_confirm_neigh(dst, daddr);
2857
2858         if (mtu < IPV6_MIN_MTU)
2859                 return;
2860         if (mtu >= dst_mtu(dst))
2861                 return;
2862
2863         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2864                 rt6_do_update_pmtu(rt6, mtu);
2865                 /* update rt6_ex->stamp for cache */
2866                 if (rt6->rt6i_flags & RTF_CACHE)
2867                         rt6_update_exception_stamp_rt(rt6);
2868         } else if (daddr) {
2869                 struct fib6_result res = {};
2870                 struct rt6_info *nrt6;
2871
2872                 rcu_read_lock();
2873                 res.f6i = rcu_dereference(rt6->from);
2874                 if (!res.f6i)
2875                         goto out_unlock;
2876
2877                 res.fib6_flags = res.f6i->fib6_flags;
2878                 res.fib6_type = res.f6i->fib6_type;
2879
2880                 if (res.f6i->nh) {
2881                         struct fib6_nh_match_arg arg = {
2882                                 .dev = dst->dev,
2883                                 .gw = &rt6->rt6i_gateway,
2884                         };
2885
2886                         nexthop_for_each_fib6_nh(res.f6i->nh,
2887                                                  fib6_nh_find_match, &arg);
2888
2889                         /* fib6_info uses a nexthop that does not have fib6_nh
2890                          * using the dst->dev + gw. Should be impossible.
2891                          */
2892                         if (!arg.match)
2893                                 goto out_unlock;
2894
2895                         res.nh = arg.match;
2896                 } else {
2897                         res.nh = res.f6i->fib6_nh;
2898                 }
2899
2900                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2901                 if (nrt6) {
2902                         rt6_do_update_pmtu(nrt6, mtu);
2903                         if (rt6_insert_exception(nrt6, &res))
2904                                 dst_release_immediate(&nrt6->dst);
2905                 }
2906 out_unlock:
2907                 rcu_read_unlock();
2908         }
2909 }
2910
2911 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2912                                struct sk_buff *skb, u32 mtu,
2913                                bool confirm_neigh)
2914 {
2915         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2916                              confirm_neigh);
2917 }
2918
2919 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2920                      int oif, u32 mark, kuid_t uid)
2921 {
2922         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2923         struct dst_entry *dst;
2924         struct flowi6 fl6 = {
2925                 .flowi6_oif = oif,
2926                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2927                 .daddr = iph->daddr,
2928                 .saddr = iph->saddr,
2929                 .flowlabel = ip6_flowinfo(iph),
2930                 .flowi6_uid = uid,
2931         };
2932
2933         dst = ip6_route_output(net, NULL, &fl6);
2934         if (!dst->error)
2935                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2936         dst_release(dst);
2937 }
2938 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2939
2940 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2941 {
2942         int oif = sk->sk_bound_dev_if;
2943         struct dst_entry *dst;
2944
2945         if (!oif && skb->dev)
2946                 oif = l3mdev_master_ifindex(skb->dev);
2947
2948         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2949
2950         dst = __sk_dst_get(sk);
2951         if (!dst || !dst->obsolete ||
2952             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2953                 return;
2954
2955         bh_lock_sock(sk);
2956         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2957                 ip6_datagram_dst_update(sk, false);
2958         bh_unlock_sock(sk);
2959 }
2960 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2961
2962 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2963                            const struct flowi6 *fl6)
2964 {
2965 #ifdef CONFIG_IPV6_SUBTREES
2966         struct ipv6_pinfo *np = inet6_sk(sk);
2967 #endif
2968
2969         ip6_dst_store(sk, dst,
2970                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2971                       &sk->sk_v6_daddr : NULL,
2972 #ifdef CONFIG_IPV6_SUBTREES
2973                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2974                       &np->saddr :
2975 #endif
2976                       NULL);
2977 }
2978
2979 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2980                                   struct flowi6 *fl6,
2981                                   const struct in6_addr *gw,
2982                                   struct rt6_info **ret)
2983 {
2984         const struct fib6_nh *nh = res->nh;
2985
2986         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2987             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2988                 return false;
2989
2990         /* rt_cache's gateway might be different from its 'parent'
2991          * in the case of an ip redirect.
2992          * So we keep searching in the exception table if the gateway
2993          * is different.
2994          */
2995         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2996                 struct rt6_info *rt_cache;
2997
2998                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2999                 if (rt_cache &&
3000                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3001                         *ret = rt_cache;
3002                         return true;
3003                 }
3004                 return false;
3005         }
3006         return true;
3007 }
3008
3009 struct fib6_nh_rd_arg {
3010         struct fib6_result      *res;
3011         struct flowi6           *fl6;
3012         const struct in6_addr   *gw;
3013         struct rt6_info         **ret;
3014 };
3015
3016 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3017 {
3018         struct fib6_nh_rd_arg *arg = _arg;
3019
3020         arg->res->nh = nh;
3021         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3022 }
3023
3024 /* Handle redirects */
3025 struct ip6rd_flowi {
3026         struct flowi6 fl6;
3027         struct in6_addr gateway;
3028 };
3029
3030 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3031                                              struct fib6_table *table,
3032                                              struct flowi6 *fl6,
3033                                              const struct sk_buff *skb,
3034                                              int flags)
3035 {
3036         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3037         struct rt6_info *ret = NULL;
3038         struct fib6_result res = {};
3039         struct fib6_nh_rd_arg arg = {
3040                 .res = &res,
3041                 .fl6 = fl6,
3042                 .gw  = &rdfl->gateway,
3043                 .ret = &ret
3044         };
3045         struct fib6_info *rt;
3046         struct fib6_node *fn;
3047
3048         /* l3mdev_update_flow overrides oif if the device is enslaved; in
3049          * this case we must match on the real ingress device, so reset it
3050          */
3051         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
3052                 fl6->flowi6_oif = skb->dev->ifindex;
3053
3054         /* Get the "current" route for this destination and
3055          * check if the redirect has come from appropriate router.
3056          *
3057          * RFC 4861 specifies that redirects should only be
3058          * accepted if they come from the nexthop to the target.
3059          * Due to the way the routes are chosen, this notion
3060          * is a bit fuzzy and one might need to check all possible
3061          * routes.
3062          */
3063
3064         rcu_read_lock();
3065         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3066 restart:
3067         for_each_fib6_node_rt_rcu(fn) {
3068                 res.f6i = rt;
3069                 if (fib6_check_expired(rt))
3070                         continue;
3071                 if (rt->fib6_flags & RTF_REJECT)
3072                         break;
3073                 if (unlikely(rt->nh)) {
3074                         if (nexthop_is_blackhole(rt->nh))
3075                                 continue;
3076                         /* on match, res->nh is filled in and potentially ret */
3077                         if (nexthop_for_each_fib6_nh(rt->nh,
3078                                                      fib6_nh_redirect_match,
3079                                                      &arg))
3080                                 goto out;
3081                 } else {
3082                         res.nh = rt->fib6_nh;
3083                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3084                                                   &ret))
3085                                 goto out;
3086                 }
3087         }
3088
3089         if (!rt)
3090                 rt = net->ipv6.fib6_null_entry;
3091         else if (rt->fib6_flags & RTF_REJECT) {
3092                 ret = net->ipv6.ip6_null_entry;
3093                 goto out;
3094         }
3095
3096         if (rt == net->ipv6.fib6_null_entry) {
3097                 fn = fib6_backtrack(fn, &fl6->saddr);
3098                 if (fn)
3099                         goto restart;
3100         }
3101
3102         res.f6i = rt;
3103         res.nh = rt->fib6_nh;
3104 out:
3105         if (ret) {
3106                 ip6_hold_safe(net, &ret);
3107         } else {
3108                 res.fib6_flags = res.f6i->fib6_flags;
3109                 res.fib6_type = res.f6i->fib6_type;
3110                 ret = ip6_create_rt_rcu(&res);
3111         }
3112
3113         rcu_read_unlock();
3114
3115         trace_fib6_table_lookup(net, &res, table, fl6);
3116         return ret;
3117 };
3118
3119 static struct dst_entry *ip6_route_redirect(struct net *net,
3120                                             const struct flowi6 *fl6,
3121                                             const struct sk_buff *skb,
3122                                             const struct in6_addr *gateway)
3123 {
3124         int flags = RT6_LOOKUP_F_HAS_SADDR;
3125         struct ip6rd_flowi rdfl;
3126
3127         rdfl.fl6 = *fl6;
3128         rdfl.gateway = *gateway;
3129
3130         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3131                                 flags, __ip6_route_redirect);
3132 }
3133
3134 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3135                   kuid_t uid)
3136 {
3137         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3138         struct dst_entry *dst;
3139         struct flowi6 fl6 = {
3140                 .flowi6_iif = LOOPBACK_IFINDEX,
3141                 .flowi6_oif = oif,
3142                 .flowi6_mark = mark,
3143                 .daddr = iph->daddr,
3144                 .saddr = iph->saddr,
3145                 .flowlabel = ip6_flowinfo(iph),
3146                 .flowi6_uid = uid,
3147         };
3148
3149         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3150         rt6_do_redirect(dst, NULL, skb);
3151         dst_release(dst);
3152 }
3153 EXPORT_SYMBOL_GPL(ip6_redirect);
3154
3155 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3156 {
3157         const struct ipv6hdr *iph = ipv6_hdr(skb);
3158         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3159         struct dst_entry *dst;
3160         struct flowi6 fl6 = {
3161                 .flowi6_iif = LOOPBACK_IFINDEX,
3162                 .flowi6_oif = oif,
3163                 .daddr = msg->dest,
3164                 .saddr = iph->daddr,
3165                 .flowi6_uid = sock_net_uid(net, NULL),
3166         };
3167
3168         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3169         rt6_do_redirect(dst, NULL, skb);
3170         dst_release(dst);
3171 }
3172
3173 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3174 {
3175         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3176                      sk->sk_uid);
3177 }
3178 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3179
3180 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3181 {
3182         struct net_device *dev = dst->dev;
3183         unsigned int mtu = dst_mtu(dst);
3184         struct net *net = dev_net(dev);
3185
3186         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3187
3188         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3189                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3190
3191         /*
3192          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3193          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3194          * IPV6_MAXPLEN is also valid and means: "any MSS,
3195          * rely only on pmtu discovery"
3196          */
3197         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3198                 mtu = IPV6_MAXPLEN;
3199         return mtu;
3200 }
3201
3202 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3203 {
3204         struct inet6_dev *idev;
3205         unsigned int mtu;
3206
3207         mtu = dst_metric_raw(dst, RTAX_MTU);
3208         if (mtu)
3209                 goto out;
3210
3211         mtu = IPV6_MIN_MTU;
3212
3213         rcu_read_lock();
3214         idev = __in6_dev_get(dst->dev);
3215         if (idev)
3216                 mtu = idev->cnf.mtu6;
3217         rcu_read_unlock();
3218
3219 out:
3220         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3221
3222         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3223 }
3224 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3225
3226 /* MTU selection:
3227  * 1. mtu on route is locked - use it
3228  * 2. mtu from nexthop exception
3229  * 3. mtu from egress device
3230  *
3231  * based on ip6_dst_mtu_forward and exception logic of
3232  * rt6_find_cached_rt; called with rcu_read_lock
3233  */
3234 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3235                       const struct in6_addr *daddr,
3236                       const struct in6_addr *saddr)
3237 {
3238         const struct fib6_nh *nh = res->nh;
3239         struct fib6_info *f6i = res->f6i;
3240         struct inet6_dev *idev;
3241         struct rt6_info *rt;
3242         u32 mtu = 0;
3243
3244         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3245                 mtu = f6i->fib6_pmtu;
3246                 if (mtu)
3247                         goto out;
3248         }
3249
3250         rt = rt6_find_cached_rt(res, daddr, saddr);
3251         if (unlikely(rt)) {
3252                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3253         } else {
3254                 struct net_device *dev = nh->fib_nh_dev;
3255
3256                 mtu = IPV6_MIN_MTU;
3257                 idev = __in6_dev_get(dev);
3258                 if (idev && idev->cnf.mtu6 > mtu)
3259                         mtu = idev->cnf.mtu6;
3260         }
3261
3262         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3263 out:
3264         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3265 }
3266
3267 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3268                                   struct flowi6 *fl6)
3269 {
3270         struct dst_entry *dst;
3271         struct rt6_info *rt;
3272         struct inet6_dev *idev = in6_dev_get(dev);
3273         struct net *net = dev_net(dev);
3274
3275         if (unlikely(!idev))
3276                 return ERR_PTR(-ENODEV);
3277
3278         rt = ip6_dst_alloc(net, dev, 0);
3279         if (unlikely(!rt)) {
3280                 in6_dev_put(idev);
3281                 dst = ERR_PTR(-ENOMEM);
3282                 goto out;
3283         }
3284
3285         rt->dst.input = ip6_input;
3286         rt->dst.output  = ip6_output;
3287         rt->rt6i_gateway  = fl6->daddr;
3288         rt->rt6i_dst.addr = fl6->daddr;
3289         rt->rt6i_dst.plen = 128;
3290         rt->rt6i_idev     = idev;
3291         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3292
3293         /* Add this dst into uncached_list so that rt6_disable_ip() can
3294          * do proper release of the net_device
3295          */
3296         rt6_uncached_list_add(rt);
3297         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3298
3299         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3300
3301 out:
3302         return dst;
3303 }
3304
3305 static int ip6_dst_gc(struct dst_ops *ops)
3306 {
3307         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3308         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3309         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3310         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3311         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3312         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3313         int entries;
3314
3315         entries = dst_entries_get_fast(ops);
3316         if (entries > rt_max_size)
3317                 entries = dst_entries_get_slow(ops);
3318
3319         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3320             entries <= rt_max_size)
3321                 goto out;
3322
3323         net->ipv6.ip6_rt_gc_expire++;
3324         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3325         entries = dst_entries_get_slow(ops);
3326         if (entries < ops->gc_thresh)
3327                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3328 out:
3329         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3330         return entries > rt_max_size;
3331 }
3332
3333 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3334                                const struct in6_addr *gw_addr, u32 tbid,
3335                                int flags, struct fib6_result *res)
3336 {
3337         struct flowi6 fl6 = {
3338                 .flowi6_oif = cfg->fc_ifindex,
3339                 .daddr = *gw_addr,
3340                 .saddr = cfg->fc_prefsrc,
3341         };
3342         struct fib6_table *table;
3343         int err;
3344
3345         table = fib6_get_table(net, tbid);
3346         if (!table)
3347                 return -EINVAL;
3348
3349         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3350                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3351
3352         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3353
3354         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3355         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3356                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3357                                  cfg->fc_ifindex != 0, NULL, flags);
3358
3359         return err;
3360 }
3361
3362 static int ip6_route_check_nh_onlink(struct net *net,
3363                                      struct fib6_config *cfg,
3364                                      const struct net_device *dev,
3365                                      struct netlink_ext_ack *extack)
3366 {
3367         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3368         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3369         struct fib6_result res = {};
3370         int err;
3371
3372         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3373         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3374             /* ignore match if it is the default route */
3375             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3376             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3377                 NL_SET_ERR_MSG(extack,
3378                                "Nexthop has invalid gateway or device mismatch");
3379                 err = -EINVAL;
3380         }
3381
3382         return err;
3383 }
3384
3385 static int ip6_route_check_nh(struct net *net,
3386                               struct fib6_config *cfg,
3387                               struct net_device **_dev,
3388                               struct inet6_dev **idev)
3389 {
3390         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3391         struct net_device *dev = _dev ? *_dev : NULL;
3392         int flags = RT6_LOOKUP_F_IFACE;
3393         struct fib6_result res = {};
3394         int err = -EHOSTUNREACH;
3395
3396         if (cfg->fc_table) {
3397                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3398                                           cfg->fc_table, flags, &res);
3399                 /* gw_addr can not require a gateway or resolve to a reject
3400                  * route. If a device is given, it must match the result.
3401                  */
3402                 if (err || res.fib6_flags & RTF_REJECT ||
3403                     res.nh->fib_nh_gw_family ||
3404                     (dev && dev != res.nh->fib_nh_dev))
3405                         err = -EHOSTUNREACH;
3406         }
3407
3408         if (err < 0) {
3409                 struct flowi6 fl6 = {
3410                         .flowi6_oif = cfg->fc_ifindex,
3411                         .daddr = *gw_addr,
3412                 };
3413
3414                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3415                 if (err || res.fib6_flags & RTF_REJECT ||
3416                     res.nh->fib_nh_gw_family)
3417                         err = -EHOSTUNREACH;
3418
3419                 if (err)
3420                         return err;
3421
3422                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3423                                  cfg->fc_ifindex != 0, NULL, flags);
3424         }
3425
3426         err = 0;
3427         if (dev) {
3428                 if (dev != res.nh->fib_nh_dev)
3429                         err = -EHOSTUNREACH;
3430         } else {
3431                 *_dev = dev = res.nh->fib_nh_dev;
3432                 dev_hold(dev);
3433                 *idev = in6_dev_get(dev);
3434         }
3435
3436         return err;
3437 }
3438
3439 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3440                            struct net_device **_dev, struct inet6_dev **idev,
3441                            struct netlink_ext_ack *extack)
3442 {
3443         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3444         int gwa_type = ipv6_addr_type(gw_addr);
3445         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3446         const struct net_device *dev = *_dev;
3447         bool need_addr_check = !dev;
3448         int err = -EINVAL;
3449
3450         /* if gw_addr is local we will fail to detect this in case
3451          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3452          * will return already-added prefix route via interface that
3453          * prefix route was assigned to, which might be non-loopback.
3454          */
3455         if (dev &&
3456             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3457                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3458                 goto out;
3459         }
3460
3461         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3462                 /* IPv6 strictly inhibits using not link-local
3463                  * addresses as nexthop address.
3464                  * Otherwise, router will not able to send redirects.
3465                  * It is very good, but in some (rare!) circumstances
3466                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3467                  * some exceptions. --ANK
3468                  * We allow IPv4-mapped nexthops to support RFC4798-type
3469                  * addressing
3470                  */
3471                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3472                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3473                         goto out;
3474                 }
3475
3476                 rcu_read_lock();
3477
3478                 if (cfg->fc_flags & RTNH_F_ONLINK)
3479                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3480                 else
3481                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3482
3483                 rcu_read_unlock();
3484
3485                 if (err)
3486                         goto out;
3487         }
3488
3489         /* reload in case device was changed */
3490         dev = *_dev;
3491
3492         err = -EINVAL;
3493         if (!dev) {
3494                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3495                 goto out;
3496         } else if (dev->flags & IFF_LOOPBACK) {
3497                 NL_SET_ERR_MSG(extack,
3498                                "Egress device can not be loopback device for this route");
3499                 goto out;
3500         }
3501
3502         /* if we did not check gw_addr above, do so now that the
3503          * egress device has been resolved.
3504          */
3505         if (need_addr_check &&
3506             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3507                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3508                 goto out;
3509         }
3510
3511         err = 0;
3512 out:
3513         return err;
3514 }
3515
3516 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3517 {
3518         if ((flags & RTF_REJECT) ||
3519             (dev && (dev->flags & IFF_LOOPBACK) &&
3520              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3521              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3522                 return true;
3523
3524         return false;
3525 }
3526
3527 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3528                  struct fib6_config *cfg, gfp_t gfp_flags,
3529                  struct netlink_ext_ack *extack)
3530 {
3531         struct net_device *dev = NULL;
3532         struct inet6_dev *idev = NULL;
3533         int addr_type;
3534         int err;
3535
3536         fib6_nh->fib_nh_family = AF_INET6;
3537 #ifdef CONFIG_IPV6_ROUTER_PREF
3538         fib6_nh->last_probe = jiffies;
3539 #endif
3540         if (cfg->fc_is_fdb) {
3541                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3542                 fib6_nh->fib_nh_gw_family = AF_INET6;
3543                 return 0;
3544         }
3545
3546         err = -ENODEV;
3547         if (cfg->fc_ifindex) {
3548                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3549                 if (!dev)
3550                         goto out;
3551                 idev = in6_dev_get(dev);
3552                 if (!idev)
3553                         goto out;
3554         }
3555
3556         if (cfg->fc_flags & RTNH_F_ONLINK) {
3557                 if (!dev) {
3558                         NL_SET_ERR_MSG(extack,
3559                                        "Nexthop device required for onlink");
3560                         goto out;
3561                 }
3562
3563                 if (!(dev->flags & IFF_UP)) {
3564                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3565                         err = -ENETDOWN;
3566                         goto out;
3567                 }
3568
3569                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3570         }
3571
3572         fib6_nh->fib_nh_weight = 1;
3573
3574         /* We cannot add true routes via loopback here,
3575          * they would result in kernel looping; promote them to reject routes
3576          */
3577         addr_type = ipv6_addr_type(&cfg->fc_dst);
3578         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3579                 /* hold loopback dev/idev if we haven't done so. */
3580                 if (dev != net->loopback_dev) {
3581                         if (dev) {
3582                                 dev_put(dev);
3583                                 in6_dev_put(idev);
3584                         }
3585                         dev = net->loopback_dev;
3586                         dev_hold(dev);
3587                         idev = in6_dev_get(dev);
3588                         if (!idev) {
3589                                 err = -ENODEV;
3590                                 goto out;
3591                         }
3592                 }
3593                 goto pcpu_alloc;
3594         }
3595
3596         if (cfg->fc_flags & RTF_GATEWAY) {
3597                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3598                 if (err)
3599                         goto out;
3600
3601                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3602                 fib6_nh->fib_nh_gw_family = AF_INET6;
3603         }
3604
3605         err = -ENODEV;
3606         if (!dev)
3607                 goto out;
3608
3609         if (idev->cnf.disable_ipv6) {
3610                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3611                 err = -EACCES;
3612                 goto out;
3613         }
3614
3615         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3616                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3617                 err = -ENETDOWN;
3618                 goto out;
3619         }
3620
3621         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3622             !netif_carrier_ok(dev))
3623                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3624
3625         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3626                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3627         if (err)
3628                 goto out;
3629
3630 pcpu_alloc:
3631         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3632         if (!fib6_nh->rt6i_pcpu) {
3633                 err = -ENOMEM;
3634                 goto out;
3635         }
3636
3637         fib6_nh->fib_nh_dev = dev;
3638         fib6_nh->fib_nh_oif = dev->ifindex;
3639         err = 0;
3640 out:
3641         if (idev)
3642                 in6_dev_put(idev);
3643
3644         if (err) {
3645                 lwtstate_put(fib6_nh->fib_nh_lws);
3646                 fib6_nh->fib_nh_lws = NULL;
3647                 if (dev)
3648                         dev_put(dev);
3649         }
3650
3651         return err;
3652 }
3653
3654 void fib6_nh_release(struct fib6_nh *fib6_nh)
3655 {
3656         struct rt6_exception_bucket *bucket;
3657
3658         rcu_read_lock();
3659
3660         fib6_nh_flush_exceptions(fib6_nh, NULL);
3661         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3662         if (bucket) {
3663                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3664                 kfree(bucket);
3665         }
3666
3667         rcu_read_unlock();
3668
3669         if (fib6_nh->rt6i_pcpu) {
3670                 int cpu;
3671
3672                 for_each_possible_cpu(cpu) {
3673                         struct rt6_info **ppcpu_rt;
3674                         struct rt6_info *pcpu_rt;
3675
3676                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3677                         pcpu_rt = *ppcpu_rt;
3678                         if (pcpu_rt) {
3679                                 dst_dev_put(&pcpu_rt->dst);
3680                                 dst_release(&pcpu_rt->dst);
3681                                 *ppcpu_rt = NULL;
3682                         }
3683                 }
3684
3685                 free_percpu(fib6_nh->rt6i_pcpu);
3686         }
3687
3688         fib_nh_common_release(&fib6_nh->nh_common);
3689 }
3690
3691 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3692                                               gfp_t gfp_flags,
3693                                               struct netlink_ext_ack *extack)
3694 {
3695         struct net *net = cfg->fc_nlinfo.nl_net;
3696         struct fib6_info *rt = NULL;
3697         struct nexthop *nh = NULL;
3698         struct fib6_table *table;
3699         struct fib6_nh *fib6_nh;
3700         int err = -EINVAL;
3701         int addr_type;
3702
3703         /* RTF_PCPU is an internal flag; can not be set by userspace */
3704         if (cfg->fc_flags & RTF_PCPU) {
3705                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3706                 goto out;
3707         }
3708
3709         /* RTF_CACHE is an internal flag; can not be set by userspace */
3710         if (cfg->fc_flags & RTF_CACHE) {
3711                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3712                 goto out;
3713         }
3714
3715         if (cfg->fc_type > RTN_MAX) {
3716                 NL_SET_ERR_MSG(extack, "Invalid route type");
3717                 goto out;
3718         }
3719
3720         if (cfg->fc_dst_len > 128) {
3721                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3722                 goto out;
3723         }
3724         if (cfg->fc_src_len > 128) {
3725                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3726                 goto out;
3727         }
3728 #ifndef CONFIG_IPV6_SUBTREES
3729         if (cfg->fc_src_len) {
3730                 NL_SET_ERR_MSG(extack,
3731                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3732                 goto out;
3733         }
3734 #endif
3735         if (cfg->fc_nh_id) {
3736                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3737                 if (!nh) {
3738                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3739                         goto out;
3740                 }
3741                 err = fib6_check_nexthop(nh, cfg, extack);
3742                 if (err)
3743                         goto out;
3744         }
3745
3746         err = -ENOBUFS;
3747         if (cfg->fc_nlinfo.nlh &&
3748             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3749                 table = fib6_get_table(net, cfg->fc_table);
3750                 if (!table) {
3751                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3752                         table = fib6_new_table(net, cfg->fc_table);
3753                 }
3754         } else {
3755                 table = fib6_new_table(net, cfg->fc_table);
3756         }
3757
3758         if (!table)
3759                 goto out;
3760
3761         err = -ENOMEM;
3762         rt = fib6_info_alloc(gfp_flags, !nh);
3763         if (!rt)
3764                 goto out;
3765
3766         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3767                                                extack);
3768         if (IS_ERR(rt->fib6_metrics)) {
3769                 err = PTR_ERR(rt->fib6_metrics);
3770                 /* Do not leave garbage there. */
3771                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3772                 goto out_free;
3773         }
3774
3775         if (cfg->fc_flags & RTF_ADDRCONF)
3776                 rt->dst_nocount = true;
3777
3778         if (cfg->fc_flags & RTF_EXPIRES)
3779                 fib6_set_expires(rt, jiffies +
3780                                 clock_t_to_jiffies(cfg->fc_expires));
3781         else
3782                 fib6_clean_expires(rt);
3783
3784         if (cfg->fc_protocol == RTPROT_UNSPEC)
3785                 cfg->fc_protocol = RTPROT_BOOT;
3786         rt->fib6_protocol = cfg->fc_protocol;
3787
3788         rt->fib6_table = table;
3789         rt->fib6_metric = cfg->fc_metric;
3790         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3791         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3792
3793         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3794         rt->fib6_dst.plen = cfg->fc_dst_len;
3795
3796 #ifdef CONFIG_IPV6_SUBTREES
3797         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3798         rt->fib6_src.plen = cfg->fc_src_len;
3799 #endif
3800         if (nh) {
3801                 if (rt->fib6_src.plen) {
3802                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3803                         goto out_free;
3804                 }
3805                 if (!nexthop_get(nh)) {
3806                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3807                         goto out_free;
3808                 }
3809                 rt->nh = nh;
3810                 fib6_nh = nexthop_fib6_nh(rt->nh);
3811         } else {
3812                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3813                 if (err)
3814                         goto out;
3815
3816                 fib6_nh = rt->fib6_nh;
3817
3818                 /* We cannot add true routes via loopback here, they would
3819                  * result in kernel looping; promote them to reject routes
3820                  */
3821                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3822                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3823                                    addr_type))
3824                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3825         }
3826
3827         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3828                 struct net_device *dev = fib6_nh->fib_nh_dev;
3829
3830                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3831                         NL_SET_ERR_MSG(extack, "Invalid source address");
3832                         err = -EINVAL;
3833                         goto out;
3834                 }
3835                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3836                 rt->fib6_prefsrc.plen = 128;
3837         } else
3838                 rt->fib6_prefsrc.plen = 0;
3839
3840         return rt;
3841 out:
3842         fib6_info_release(rt);
3843         return ERR_PTR(err);
3844 out_free:
3845         ip_fib_metrics_put(rt->fib6_metrics);
3846         kfree(rt);
3847         return ERR_PTR(err);
3848 }
3849
3850 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3851                   struct netlink_ext_ack *extack)
3852 {
3853         struct fib6_info *rt;
3854         int err;
3855
3856         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3857         if (IS_ERR(rt))
3858                 return PTR_ERR(rt);
3859
3860         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3861         fib6_info_release(rt);
3862
3863         return err;
3864 }
3865
3866 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3867 {
3868         struct net *net = info->nl_net;
3869         struct fib6_table *table;
3870         int err;
3871
3872         if (rt == net->ipv6.fib6_null_entry) {
3873                 err = -ENOENT;
3874                 goto out;
3875         }
3876
3877         table = rt->fib6_table;
3878         spin_lock_bh(&table->tb6_lock);
3879         err = fib6_del(rt, info);
3880         spin_unlock_bh(&table->tb6_lock);
3881
3882 out:
3883         fib6_info_release(rt);
3884         return err;
3885 }
3886
3887 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3888 {
3889         struct nl_info info = {
3890                 .nl_net = net,
3891                 .skip_notify = skip_notify
3892         };
3893
3894         return __ip6_del_rt(rt, &info);
3895 }
3896
3897 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3898 {
3899         struct nl_info *info = &cfg->fc_nlinfo;
3900         struct net *net = info->nl_net;
3901         struct sk_buff *skb = NULL;
3902         struct fib6_table *table;
3903         int err = -ENOENT;
3904
3905         if (rt == net->ipv6.fib6_null_entry)
3906                 goto out_put;
3907         table = rt->fib6_table;
3908         spin_lock_bh(&table->tb6_lock);
3909
3910         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3911                 struct fib6_info *sibling, *next_sibling;
3912                 struct fib6_node *fn;
3913
3914                 /* prefer to send a single notification with all hops */
3915                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3916                 if (skb) {
3917                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3918
3919                         if (rt6_fill_node(net, skb, rt, NULL,
3920                                           NULL, NULL, 0, RTM_DELROUTE,
3921                                           info->portid, seq, 0) < 0) {
3922                                 kfree_skb(skb);
3923                                 skb = NULL;
3924                         } else
3925                                 info->skip_notify = 1;
3926                 }
3927
3928                 /* 'rt' points to the first sibling route. If it is not the
3929                  * leaf, then we do not need to send a notification. Otherwise,
3930                  * we need to check if the last sibling has a next route or not
3931                  * and emit a replace or delete notification, respectively.
3932                  */
3933                 info->skip_notify_kernel = 1;
3934                 fn = rcu_dereference_protected(rt->fib6_node,
3935                                             lockdep_is_held(&table->tb6_lock));
3936                 if (rcu_access_pointer(fn->leaf) == rt) {
3937                         struct fib6_info *last_sibling, *replace_rt;
3938
3939                         last_sibling = list_last_entry(&rt->fib6_siblings,
3940                                                        struct fib6_info,
3941                                                        fib6_siblings);
3942                         replace_rt = rcu_dereference_protected(
3943                                             last_sibling->fib6_next,
3944                                             lockdep_is_held(&table->tb6_lock));
3945                         if (replace_rt)
3946                                 call_fib6_entry_notifiers_replace(net,
3947                                                                   replace_rt);
3948                         else
3949                                 call_fib6_multipath_entry_notifiers(net,
3950                                                        FIB_EVENT_ENTRY_DEL,
3951                                                        rt, rt->fib6_nsiblings,
3952                                                        NULL);
3953                 }
3954                 list_for_each_entry_safe(sibling, next_sibling,
3955                                          &rt->fib6_siblings,
3956                                          fib6_siblings) {
3957                         err = fib6_del(sibling, info);
3958                         if (err)
3959                                 goto out_unlock;
3960                 }
3961         }
3962
3963         err = fib6_del(rt, info);
3964 out_unlock:
3965         spin_unlock_bh(&table->tb6_lock);
3966 out_put:
3967         fib6_info_release(rt);
3968
3969         if (skb) {
3970                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3971                             info->nlh, gfp_any());
3972         }
3973         return err;
3974 }
3975
3976 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3977 {
3978         int rc = -ESRCH;
3979
3980         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3981                 goto out;
3982
3983         if (cfg->fc_flags & RTF_GATEWAY &&
3984             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3985                 goto out;
3986
3987         rc = rt6_remove_exception_rt(rt);
3988 out:
3989         return rc;
3990 }
3991
3992 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3993                              struct fib6_nh *nh)
3994 {
3995         struct fib6_result res = {
3996                 .f6i = rt,
3997                 .nh = nh,
3998         };
3999         struct rt6_info *rt_cache;
4000
4001         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
4002         if (rt_cache)
4003                 return __ip6_del_cached_rt(rt_cache, cfg);
4004
4005         return 0;
4006 }
4007
4008 struct fib6_nh_del_cached_rt_arg {
4009         struct fib6_config *cfg;
4010         struct fib6_info *f6i;
4011 };
4012
4013 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4014 {
4015         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4016         int rc;
4017
4018         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4019         return rc != -ESRCH ? rc : 0;
4020 }
4021
4022 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4023 {
4024         struct fib6_nh_del_cached_rt_arg arg = {
4025                 .cfg = cfg,
4026                 .f6i = f6i
4027         };
4028
4029         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4030 }
4031
4032 static int ip6_route_del(struct fib6_config *cfg,
4033                          struct netlink_ext_ack *extack)
4034 {
4035         struct fib6_table *table;
4036         struct fib6_info *rt;
4037         struct fib6_node *fn;
4038         int err = -ESRCH;
4039
4040         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4041         if (!table) {
4042                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4043                 return err;
4044         }
4045
4046         rcu_read_lock();
4047
4048         fn = fib6_locate(&table->tb6_root,
4049                          &cfg->fc_dst, cfg->fc_dst_len,
4050                          &cfg->fc_src, cfg->fc_src_len,
4051                          !(cfg->fc_flags & RTF_CACHE));
4052
4053         if (fn) {
4054                 for_each_fib6_node_rt_rcu(fn) {
4055                         struct fib6_nh *nh;
4056
4057                         if (rt->nh && cfg->fc_nh_id &&
4058                             rt->nh->id != cfg->fc_nh_id)
4059                                 continue;
4060
4061                         if (cfg->fc_flags & RTF_CACHE) {
4062                                 int rc = 0;
4063
4064                                 if (rt->nh) {
4065                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4066                                 } else if (cfg->fc_nh_id) {
4067                                         continue;
4068                                 } else {
4069                                         nh = rt->fib6_nh;
4070                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4071                                 }
4072                                 if (rc != -ESRCH) {
4073                                         rcu_read_unlock();
4074                                         return rc;
4075                                 }
4076                                 continue;
4077                         }
4078
4079                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4080                                 continue;
4081                         if (cfg->fc_protocol &&
4082                             cfg->fc_protocol != rt->fib6_protocol)
4083                                 continue;
4084
4085                         if (rt->nh) {
4086                                 if (!fib6_info_hold_safe(rt))
4087                                         continue;
4088                                 rcu_read_unlock();
4089
4090                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4091                         }
4092                         if (cfg->fc_nh_id)
4093                                 continue;
4094
4095                         nh = rt->fib6_nh;
4096                         if (cfg->fc_ifindex &&
4097                             (!nh->fib_nh_dev ||
4098                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4099                                 continue;
4100                         if (cfg->fc_flags & RTF_GATEWAY &&
4101                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4102                                 continue;
4103                         if (!fib6_info_hold_safe(rt))
4104                                 continue;
4105                         rcu_read_unlock();
4106
4107                         /* if gateway was specified only delete the one hop */
4108                         if (cfg->fc_flags & RTF_GATEWAY)
4109                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4110
4111                         return __ip6_del_rt_siblings(rt, cfg);
4112                 }
4113         }
4114         rcu_read_unlock();
4115
4116         return err;
4117 }
4118
4119 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4120 {
4121         struct netevent_redirect netevent;
4122         struct rt6_info *rt, *nrt = NULL;
4123         struct fib6_result res = {};
4124         struct ndisc_options ndopts;
4125         struct inet6_dev *in6_dev;
4126         struct neighbour *neigh;
4127         struct rd_msg *msg;
4128         int optlen, on_link;
4129         u8 *lladdr;
4130
4131         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4132         optlen -= sizeof(*msg);
4133
4134         if (optlen < 0) {
4135                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4136                 return;
4137         }
4138
4139         msg = (struct rd_msg *)icmp6_hdr(skb);
4140
4141         if (ipv6_addr_is_multicast(&msg->dest)) {
4142                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4143                 return;
4144         }
4145
4146         on_link = 0;
4147         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4148                 on_link = 1;
4149         } else if (ipv6_addr_type(&msg->target) !=
4150                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4151                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4152                 return;
4153         }
4154
4155         in6_dev = __in6_dev_get(skb->dev);
4156         if (!in6_dev)
4157                 return;
4158         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4159                 return;
4160
4161         /* RFC2461 8.1:
4162          *      The IP source address of the Redirect MUST be the same as the current
4163          *      first-hop router for the specified ICMP Destination Address.
4164          */
4165
4166         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4167                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4168                 return;
4169         }
4170
4171         lladdr = NULL;
4172         if (ndopts.nd_opts_tgt_lladdr) {
4173                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4174                                              skb->dev);
4175                 if (!lladdr) {
4176                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4177                         return;
4178                 }
4179         }
4180
4181         rt = (struct rt6_info *) dst;
4182         if (rt->rt6i_flags & RTF_REJECT) {
4183                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4184                 return;
4185         }
4186
4187         /* Redirect received -> path was valid.
4188          * Look, redirects are sent only in response to data packets,
4189          * so that this nexthop apparently is reachable. --ANK
4190          */
4191         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4192
4193         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4194         if (!neigh)
4195                 return;
4196
4197         /*
4198          *      We have finally decided to accept it.
4199          */
4200
4201         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4202                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4203                      NEIGH_UPDATE_F_OVERRIDE|
4204                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4205                                      NEIGH_UPDATE_F_ISROUTER)),
4206                      NDISC_REDIRECT, &ndopts);
4207
4208         rcu_read_lock();
4209         res.f6i = rcu_dereference(rt->from);
4210         if (!res.f6i)
4211                 goto out;
4212
4213         if (res.f6i->nh) {
4214                 struct fib6_nh_match_arg arg = {
4215                         .dev = dst->dev,
4216                         .gw = &rt->rt6i_gateway,
4217                 };
4218
4219                 nexthop_for_each_fib6_nh(res.f6i->nh,
4220                                          fib6_nh_find_match, &arg);
4221
4222                 /* fib6_info uses a nexthop that does not have fib6_nh
4223                  * using the dst->dev. Should be impossible
4224                  */
4225                 if (!arg.match)
4226                         goto out;
4227                 res.nh = arg.match;
4228         } else {
4229                 res.nh = res.f6i->fib6_nh;
4230         }
4231
4232         res.fib6_flags = res.f6i->fib6_flags;
4233         res.fib6_type = res.f6i->fib6_type;
4234         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4235         if (!nrt)
4236                 goto out;
4237
4238         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4239         if (on_link)
4240                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4241
4242         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4243
4244         /* rt6_insert_exception() will take care of duplicated exceptions */
4245         if (rt6_insert_exception(nrt, &res)) {
4246                 dst_release_immediate(&nrt->dst);
4247                 goto out;
4248         }
4249
4250         netevent.old = &rt->dst;
4251         netevent.new = &nrt->dst;
4252         netevent.daddr = &msg->dest;
4253         netevent.neigh = neigh;
4254         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4255
4256 out:
4257         rcu_read_unlock();
4258         neigh_release(neigh);
4259 }
4260
4261 #ifdef CONFIG_IPV6_ROUTE_INFO
4262 static struct fib6_info *rt6_get_route_info(struct net *net,
4263                                            const struct in6_addr *prefix, int prefixlen,
4264                                            const struct in6_addr *gwaddr,
4265                                            struct net_device *dev)
4266 {
4267         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4268         int ifindex = dev->ifindex;
4269         struct fib6_node *fn;
4270         struct fib6_info *rt = NULL;
4271         struct fib6_table *table;
4272
4273         table = fib6_get_table(net, tb_id);
4274         if (!table)
4275                 return NULL;
4276
4277         rcu_read_lock();
4278         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4279         if (!fn)
4280                 goto out;
4281
4282         for_each_fib6_node_rt_rcu(fn) {
4283                 /* these routes do not use nexthops */
4284                 if (rt->nh)
4285                         continue;
4286                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4287                         continue;
4288                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4289                     !rt->fib6_nh->fib_nh_gw_family)
4290                         continue;
4291                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4292                         continue;
4293                 if (!fib6_info_hold_safe(rt))
4294                         continue;
4295                 break;
4296         }
4297 out:
4298         rcu_read_unlock();
4299         return rt;
4300 }
4301
4302 static struct fib6_info *rt6_add_route_info(struct net *net,
4303                                            const struct in6_addr *prefix, int prefixlen,
4304                                            const struct in6_addr *gwaddr,
4305                                            struct net_device *dev,
4306                                            unsigned int pref)
4307 {
4308         struct fib6_config cfg = {
4309                 .fc_metric      = IP6_RT_PRIO_USER,
4310                 .fc_ifindex     = dev->ifindex,
4311                 .fc_dst_len     = prefixlen,
4312                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4313                                   RTF_UP | RTF_PREF(pref),
4314                 .fc_protocol = RTPROT_RA,
4315                 .fc_type = RTN_UNICAST,
4316                 .fc_nlinfo.portid = 0,
4317                 .fc_nlinfo.nlh = NULL,
4318                 .fc_nlinfo.nl_net = net,
4319         };
4320
4321         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4322         cfg.fc_dst = *prefix;
4323         cfg.fc_gateway = *gwaddr;
4324
4325         /* We should treat it as a default route if prefix length is 0. */
4326         if (!prefixlen)
4327                 cfg.fc_flags |= RTF_DEFAULT;
4328
4329         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4330
4331         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4332 }
4333 #endif
4334
4335 struct fib6_info *rt6_get_dflt_router(struct net *net,
4336                                      const struct in6_addr *addr,
4337                                      struct net_device *dev)
4338 {
4339         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4340         struct fib6_info *rt;
4341         struct fib6_table *table;
4342
4343         table = fib6_get_table(net, tb_id);
4344         if (!table)
4345                 return NULL;
4346
4347         rcu_read_lock();
4348         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4349                 struct fib6_nh *nh;
4350
4351                 /* RA routes do not use nexthops */
4352                 if (rt->nh)
4353                         continue;
4354
4355                 nh = rt->fib6_nh;
4356                 if (dev == nh->fib_nh_dev &&
4357                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4358                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4359                         break;
4360         }
4361         if (rt && !fib6_info_hold_safe(rt))
4362                 rt = NULL;
4363         rcu_read_unlock();
4364         return rt;
4365 }
4366
4367 struct fib6_info *rt6_add_dflt_router(struct net *net,
4368                                      const struct in6_addr *gwaddr,
4369                                      struct net_device *dev,
4370                                      unsigned int pref,
4371                                      u32 defrtr_usr_metric)
4372 {
4373         struct fib6_config cfg = {
4374                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4375                 .fc_metric      = defrtr_usr_metric,
4376                 .fc_ifindex     = dev->ifindex,
4377                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4378                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4379                 .fc_protocol = RTPROT_RA,
4380                 .fc_type = RTN_UNICAST,
4381                 .fc_nlinfo.portid = 0,
4382                 .fc_nlinfo.nlh = NULL,
4383                 .fc_nlinfo.nl_net = net,
4384         };
4385
4386         cfg.fc_gateway = *gwaddr;
4387
4388         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4389                 struct fib6_table *table;
4390
4391                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4392                 if (table)
4393                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4394         }
4395
4396         return rt6_get_dflt_router(net, gwaddr, dev);
4397 }
4398
4399 static void __rt6_purge_dflt_routers(struct net *net,
4400                                      struct fib6_table *table)
4401 {
4402         struct fib6_info *rt;
4403
4404 restart:
4405         rcu_read_lock();
4406         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4407                 struct net_device *dev = fib6_info_nh_dev(rt);
4408                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4409
4410                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4411                     (!idev || idev->cnf.accept_ra != 2) &&
4412                     fib6_info_hold_safe(rt)) {
4413                         rcu_read_unlock();
4414                         ip6_del_rt(net, rt, false);
4415                         goto restart;
4416                 }
4417         }
4418         rcu_read_unlock();
4419
4420         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4421 }
4422
4423 void rt6_purge_dflt_routers(struct net *net)
4424 {
4425         struct fib6_table *table;
4426         struct hlist_head *head;
4427         unsigned int h;
4428
4429         rcu_read_lock();
4430
4431         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4432                 head = &net->ipv6.fib_table_hash[h];
4433                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4434                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4435                                 __rt6_purge_dflt_routers(net, table);
4436                 }
4437         }
4438
4439         rcu_read_unlock();
4440 }
4441
4442 static void rtmsg_to_fib6_config(struct net *net,
4443                                  struct in6_rtmsg *rtmsg,
4444                                  struct fib6_config *cfg)
4445 {
4446         *cfg = (struct fib6_config){
4447                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4448                          : RT6_TABLE_MAIN,
4449                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4450                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4451                 .fc_expires = rtmsg->rtmsg_info,
4452                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4453                 .fc_src_len = rtmsg->rtmsg_src_len,
4454                 .fc_flags = rtmsg->rtmsg_flags,
4455                 .fc_type = rtmsg->rtmsg_type,
4456
4457                 .fc_nlinfo.nl_net = net,
4458
4459                 .fc_dst = rtmsg->rtmsg_dst,
4460                 .fc_src = rtmsg->rtmsg_src,
4461                 .fc_gateway = rtmsg->rtmsg_gateway,
4462         };
4463 }
4464
4465 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4466 {
4467         struct fib6_config cfg;
4468         int err;
4469
4470         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4471                 return -EINVAL;
4472         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4473                 return -EPERM;
4474
4475         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4476
4477         rtnl_lock();
4478         switch (cmd) {
4479         case SIOCADDRT:
4480                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4481                 break;
4482         case SIOCDELRT:
4483                 err = ip6_route_del(&cfg, NULL);
4484                 break;
4485         }
4486         rtnl_unlock();
4487         return err;
4488 }
4489
4490 /*
4491  *      Drop the packet on the floor
4492  */
4493
4494 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4495 {
4496         struct dst_entry *dst = skb_dst(skb);
4497         struct net *net = dev_net(dst->dev);
4498         struct inet6_dev *idev;
4499         int type;
4500
4501         if (netif_is_l3_master(skb->dev) &&
4502             dst->dev == net->loopback_dev)
4503                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4504         else
4505                 idev = ip6_dst_idev(dst);
4506
4507         switch (ipstats_mib_noroutes) {
4508         case IPSTATS_MIB_INNOROUTES:
4509                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4510                 if (type == IPV6_ADDR_ANY) {
4511                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4512                         break;
4513                 }
4514                 fallthrough;
4515         case IPSTATS_MIB_OUTNOROUTES:
4516                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4517                 break;
4518         }
4519
4520         /* Start over by dropping the dst for l3mdev case */
4521         if (netif_is_l3_master(skb->dev))
4522                 skb_dst_drop(skb);
4523
4524         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4525         kfree_skb(skb);
4526         return 0;
4527 }
4528
4529 static int ip6_pkt_discard(struct sk_buff *skb)
4530 {
4531         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4532 }
4533
4534 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4535 {
4536         skb->dev = skb_dst(skb)->dev;
4537         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4538 }
4539
4540 static int ip6_pkt_prohibit(struct sk_buff *skb)
4541 {
4542         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4543 }
4544
4545 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4546 {
4547         skb->dev = skb_dst(skb)->dev;
4548         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4549 }
4550
4551 /*
4552  *      Allocate a dst for local (unicast / anycast) address.
4553  */
4554
4555 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4556                                      struct inet6_dev *idev,
4557                                      const struct in6_addr *addr,
4558                                      bool anycast, gfp_t gfp_flags)
4559 {
4560         struct fib6_config cfg = {
4561                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4562                 .fc_ifindex = idev->dev->ifindex,
4563                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4564                 .fc_dst = *addr,
4565                 .fc_dst_len = 128,
4566                 .fc_protocol = RTPROT_KERNEL,
4567                 .fc_nlinfo.nl_net = net,
4568                 .fc_ignore_dev_down = true,
4569         };
4570         struct fib6_info *f6i;
4571
4572         if (anycast) {
4573                 cfg.fc_type = RTN_ANYCAST;
4574                 cfg.fc_flags |= RTF_ANYCAST;
4575         } else {
4576                 cfg.fc_type = RTN_LOCAL;
4577                 cfg.fc_flags |= RTF_LOCAL;
4578         }
4579
4580         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4581         if (!IS_ERR(f6i))
4582                 f6i->dst_nocount = true;
4583         return f6i;
4584 }
4585
4586 /* remove deleted ip from prefsrc entries */
4587 struct arg_dev_net_ip {
4588         struct net_device *dev;
4589         struct net *net;
4590         struct in6_addr *addr;
4591 };
4592
4593 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4594 {
4595         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4596         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4597         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4598
4599         if (!rt->nh &&
4600             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4601             rt != net->ipv6.fib6_null_entry &&
4602             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4603                 spin_lock_bh(&rt6_exception_lock);
4604                 /* remove prefsrc entry */
4605                 rt->fib6_prefsrc.plen = 0;
4606                 spin_unlock_bh(&rt6_exception_lock);
4607         }
4608         return 0;
4609 }
4610
4611 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4612 {
4613         struct net *net = dev_net(ifp->idev->dev);
4614         struct arg_dev_net_ip adni = {
4615                 .dev = ifp->idev->dev,
4616                 .net = net,
4617                 .addr = &ifp->addr,
4618         };
4619         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4620 }
4621
4622 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4623
4624 /* Remove routers and update dst entries when gateway turn into host. */
4625 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4626 {
4627         struct in6_addr *gateway = (struct in6_addr *)arg;
4628         struct fib6_nh *nh;
4629
4630         /* RA routes do not use nexthops */
4631         if (rt->nh)
4632                 return 0;
4633
4634         nh = rt->fib6_nh;
4635         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4636             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4637                 return -1;
4638
4639         /* Further clean up cached routes in exception table.
4640          * This is needed because cached route may have a different
4641          * gateway than its 'parent' in the case of an ip redirect.
4642          */
4643         fib6_nh_exceptions_clean_tohost(nh, gateway);
4644
4645         return 0;
4646 }
4647
4648 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4649 {
4650         fib6_clean_all(net, fib6_clean_tohost, gateway);
4651 }
4652
4653 struct arg_netdev_event {
4654         const struct net_device *dev;
4655         union {
4656                 unsigned char nh_flags;
4657                 unsigned long event;
4658         };
4659 };
4660
4661 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4662 {
4663         struct fib6_info *iter;
4664         struct fib6_node *fn;
4665
4666         fn = rcu_dereference_protected(rt->fib6_node,
4667                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4668         iter = rcu_dereference_protected(fn->leaf,
4669                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4670         while (iter) {
4671                 if (iter->fib6_metric == rt->fib6_metric &&
4672                     rt6_qualify_for_ecmp(iter))
4673                         return iter;
4674                 iter = rcu_dereference_protected(iter->fib6_next,
4675                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4676         }
4677
4678         return NULL;
4679 }
4680
4681 /* only called for fib entries with builtin fib6_nh */
4682 static bool rt6_is_dead(const struct fib6_info *rt)
4683 {
4684         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4685             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4686              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4687                 return true;
4688
4689         return false;
4690 }
4691
4692 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4693 {
4694         struct fib6_info *iter;
4695         int total = 0;
4696
4697         if (!rt6_is_dead(rt))
4698                 total += rt->fib6_nh->fib_nh_weight;
4699
4700         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4701                 if (!rt6_is_dead(iter))
4702                         total += iter->fib6_nh->fib_nh_weight;
4703         }
4704
4705         return total;
4706 }
4707
4708 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4709 {
4710         int upper_bound = -1;
4711
4712         if (!rt6_is_dead(rt)) {
4713                 *weight += rt->fib6_nh->fib_nh_weight;
4714                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4715                                                     total) - 1;
4716         }
4717         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4718 }
4719
4720 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4721 {
4722         struct fib6_info *iter;
4723         int weight = 0;
4724
4725         rt6_upper_bound_set(rt, &weight, total);
4726
4727         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4728                 rt6_upper_bound_set(iter, &weight, total);
4729 }
4730
4731 void rt6_multipath_rebalance(struct fib6_info *rt)
4732 {
4733         struct fib6_info *first;
4734         int total;
4735
4736         /* In case the entire multipath route was marked for flushing,
4737          * then there is no need to rebalance upon the removal of every
4738          * sibling route.
4739          */
4740         if (!rt->fib6_nsiblings || rt->should_flush)
4741                 return;
4742
4743         /* During lookup routes are evaluated in order, so we need to
4744          * make sure upper bounds are assigned from the first sibling
4745          * onwards.
4746          */
4747         first = rt6_multipath_first_sibling(rt);
4748         if (WARN_ON_ONCE(!first))
4749                 return;
4750
4751         total = rt6_multipath_total_weight(first);
4752         rt6_multipath_upper_bound_set(first, total);
4753 }
4754
4755 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4756 {
4757         const struct arg_netdev_event *arg = p_arg;
4758         struct net *net = dev_net(arg->dev);
4759
4760         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4761             rt->fib6_nh->fib_nh_dev == arg->dev) {
4762                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4763                 fib6_update_sernum_upto_root(net, rt);
4764                 rt6_multipath_rebalance(rt);
4765         }
4766
4767         return 0;
4768 }
4769
4770 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4771 {
4772         struct arg_netdev_event arg = {
4773                 .dev = dev,
4774                 {
4775                         .nh_flags = nh_flags,
4776                 },
4777         };
4778
4779         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4780                 arg.nh_flags |= RTNH_F_LINKDOWN;
4781
4782         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4783 }
4784
4785 /* only called for fib entries with inline fib6_nh */
4786 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4787                                    const struct net_device *dev)
4788 {
4789         struct fib6_info *iter;
4790
4791         if (rt->fib6_nh->fib_nh_dev == dev)
4792                 return true;
4793         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4794                 if (iter->fib6_nh->fib_nh_dev == dev)
4795                         return true;
4796
4797         return false;
4798 }
4799
4800 static void rt6_multipath_flush(struct fib6_info *rt)
4801 {
4802         struct fib6_info *iter;
4803
4804         rt->should_flush = 1;
4805         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4806                 iter->should_flush = 1;
4807 }
4808
4809 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4810                                              const struct net_device *down_dev)
4811 {
4812         struct fib6_info *iter;
4813         unsigned int dead = 0;
4814
4815         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4816             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4817                 dead++;
4818         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4819                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4820                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4821                         dead++;
4822
4823         return dead;
4824 }
4825
4826 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4827                                        const struct net_device *dev,
4828                                        unsigned char nh_flags)
4829 {
4830         struct fib6_info *iter;
4831
4832         if (rt->fib6_nh->fib_nh_dev == dev)
4833                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4834         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4835                 if (iter->fib6_nh->fib_nh_dev == dev)
4836                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4837 }
4838
4839 /* called with write lock held for table with rt */
4840 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4841 {
4842         const struct arg_netdev_event *arg = p_arg;
4843         const struct net_device *dev = arg->dev;
4844         struct net *net = dev_net(dev);
4845
4846         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4847                 return 0;
4848
4849         switch (arg->event) {
4850         case NETDEV_UNREGISTER:
4851                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4852         case NETDEV_DOWN:
4853                 if (rt->should_flush)
4854                         return -1;
4855                 if (!rt->fib6_nsiblings)
4856                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4857                 if (rt6_multipath_uses_dev(rt, dev)) {
4858                         unsigned int count;
4859
4860                         count = rt6_multipath_dead_count(rt, dev);
4861                         if (rt->fib6_nsiblings + 1 == count) {
4862                                 rt6_multipath_flush(rt);
4863                                 return -1;
4864                         }
4865                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4866                                                    RTNH_F_LINKDOWN);
4867                         fib6_update_sernum(net, rt);
4868                         rt6_multipath_rebalance(rt);
4869                 }
4870                 return -2;
4871         case NETDEV_CHANGE:
4872                 if (rt->fib6_nh->fib_nh_dev != dev ||
4873                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4874                         break;
4875                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4876                 rt6_multipath_rebalance(rt);
4877                 break;
4878         }
4879
4880         return 0;
4881 }
4882
4883 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4884 {
4885         struct arg_netdev_event arg = {
4886                 .dev = dev,
4887                 {
4888                         .event = event,
4889                 },
4890         };
4891         struct net *net = dev_net(dev);
4892
4893         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4894                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4895         else
4896                 fib6_clean_all(net, fib6_ifdown, &arg);
4897 }
4898
4899 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4900 {
4901         rt6_sync_down_dev(dev, event);
4902         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4903         neigh_ifdown(&nd_tbl, dev);
4904 }
4905
4906 struct rt6_mtu_change_arg {
4907         struct net_device *dev;
4908         unsigned int mtu;
4909         struct fib6_info *f6i;
4910 };
4911
4912 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4913 {
4914         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4915         struct fib6_info *f6i = arg->f6i;
4916
4917         /* For administrative MTU increase, there is no way to discover
4918          * IPv6 PMTU increase, so PMTU increase should be updated here.
4919          * Since RFC 1981 doesn't include administrative MTU increase
4920          * update PMTU increase is a MUST. (i.e. jumbo frame)
4921          */
4922         if (nh->fib_nh_dev == arg->dev) {
4923                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4924                 u32 mtu = f6i->fib6_pmtu;
4925
4926                 if (mtu >= arg->mtu ||
4927                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4928                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4929
4930                 spin_lock_bh(&rt6_exception_lock);
4931                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4932                 spin_unlock_bh(&rt6_exception_lock);
4933         }
4934
4935         return 0;
4936 }
4937
4938 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4939 {
4940         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4941         struct inet6_dev *idev;
4942
4943         /* In IPv6 pmtu discovery is not optional,
4944            so that RTAX_MTU lock cannot disable it.
4945            We still use this lock to block changes
4946            caused by addrconf/ndisc.
4947         */
4948
4949         idev = __in6_dev_get(arg->dev);
4950         if (!idev)
4951                 return 0;
4952
4953         if (fib6_metric_locked(f6i, RTAX_MTU))
4954                 return 0;
4955
4956         arg->f6i = f6i;
4957         if (f6i->nh) {
4958                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4959                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4960                                                 arg);
4961         }
4962
4963         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4964 }
4965
4966 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4967 {
4968         struct rt6_mtu_change_arg arg = {
4969                 .dev = dev,
4970                 .mtu = mtu,
4971         };
4972
4973         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4974 }
4975
4976 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4977         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4978         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4979         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4980         [RTA_OIF]               = { .type = NLA_U32 },
4981         [RTA_IIF]               = { .type = NLA_U32 },
4982         [RTA_PRIORITY]          = { .type = NLA_U32 },
4983         [RTA_METRICS]           = { .type = NLA_NESTED },
4984         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4985         [RTA_PREF]              = { .type = NLA_U8 },
4986         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4987         [RTA_ENCAP]             = { .type = NLA_NESTED },
4988         [RTA_EXPIRES]           = { .type = NLA_U32 },
4989         [RTA_UID]               = { .type = NLA_U32 },
4990         [RTA_MARK]              = { .type = NLA_U32 },
4991         [RTA_TABLE]             = { .type = NLA_U32 },
4992         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4993         [RTA_SPORT]             = { .type = NLA_U16 },
4994         [RTA_DPORT]             = { .type = NLA_U16 },
4995         [RTA_NH_ID]             = { .type = NLA_U32 },
4996 };
4997
4998 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4999                               struct fib6_config *cfg,
5000                               struct netlink_ext_ack *extack)
5001 {
5002         struct rtmsg *rtm;
5003         struct nlattr *tb[RTA_MAX+1];
5004         unsigned int pref;
5005         int err;
5006
5007         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5008                                      rtm_ipv6_policy, extack);
5009         if (err < 0)
5010                 goto errout;
5011
5012         err = -EINVAL;
5013         rtm = nlmsg_data(nlh);
5014
5015         *cfg = (struct fib6_config){
5016                 .fc_table = rtm->rtm_table,
5017                 .fc_dst_len = rtm->rtm_dst_len,
5018                 .fc_src_len = rtm->rtm_src_len,
5019                 .fc_flags = RTF_UP,
5020                 .fc_protocol = rtm->rtm_protocol,
5021                 .fc_type = rtm->rtm_type,
5022
5023                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5024                 .fc_nlinfo.nlh = nlh,
5025                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5026         };
5027
5028         if (rtm->rtm_type == RTN_UNREACHABLE ||
5029             rtm->rtm_type == RTN_BLACKHOLE ||
5030             rtm->rtm_type == RTN_PROHIBIT ||
5031             rtm->rtm_type == RTN_THROW)
5032                 cfg->fc_flags |= RTF_REJECT;
5033
5034         if (rtm->rtm_type == RTN_LOCAL)
5035                 cfg->fc_flags |= RTF_LOCAL;
5036
5037         if (rtm->rtm_flags & RTM_F_CLONED)
5038                 cfg->fc_flags |= RTF_CACHE;
5039
5040         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5041
5042         if (tb[RTA_NH_ID]) {
5043                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5044                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5045                         NL_SET_ERR_MSG(extack,
5046                                        "Nexthop specification and nexthop id are mutually exclusive");
5047                         goto errout;
5048                 }
5049                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5050         }
5051
5052         if (tb[RTA_GATEWAY]) {
5053                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5054                 cfg->fc_flags |= RTF_GATEWAY;
5055         }
5056         if (tb[RTA_VIA]) {
5057                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5058                 goto errout;
5059         }
5060
5061         if (tb[RTA_DST]) {
5062                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5063
5064                 if (nla_len(tb[RTA_DST]) < plen)
5065                         goto errout;
5066
5067                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5068         }
5069
5070         if (tb[RTA_SRC]) {
5071                 int plen = (rtm->rtm_src_len + 7) >> 3;
5072
5073                 if (nla_len(tb[RTA_SRC]) < plen)
5074                         goto errout;
5075
5076                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5077         }
5078
5079         if (tb[RTA_PREFSRC])
5080                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5081
5082         if (tb[RTA_OIF])
5083                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5084
5085         if (tb[RTA_PRIORITY])
5086                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5087
5088         if (tb[RTA_METRICS]) {
5089                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5090                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5091         }
5092
5093         if (tb[RTA_TABLE])
5094                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5095
5096         if (tb[RTA_MULTIPATH]) {
5097                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5098                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5099
5100                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5101                                                      cfg->fc_mp_len, extack);
5102                 if (err < 0)
5103                         goto errout;
5104         }
5105
5106         if (tb[RTA_PREF]) {
5107                 pref = nla_get_u8(tb[RTA_PREF]);
5108                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5109                     pref != ICMPV6_ROUTER_PREF_HIGH)
5110                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5111                 cfg->fc_flags |= RTF_PREF(pref);
5112         }
5113
5114         if (tb[RTA_ENCAP])
5115                 cfg->fc_encap = tb[RTA_ENCAP];
5116
5117         if (tb[RTA_ENCAP_TYPE]) {
5118                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5119
5120                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5121                 if (err < 0)
5122                         goto errout;
5123         }
5124
5125         if (tb[RTA_EXPIRES]) {
5126                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5127
5128                 if (addrconf_finite_timeout(timeout)) {
5129                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5130                         cfg->fc_flags |= RTF_EXPIRES;
5131                 }
5132         }
5133
5134         err = 0;
5135 errout:
5136         return err;
5137 }
5138
5139 struct rt6_nh {
5140         struct fib6_info *fib6_info;
5141         struct fib6_config r_cfg;
5142         struct list_head next;
5143 };
5144
5145 static int ip6_route_info_append(struct net *net,
5146                                  struct list_head *rt6_nh_list,
5147                                  struct fib6_info *rt,
5148                                  struct fib6_config *r_cfg)
5149 {
5150         struct rt6_nh *nh;
5151         int err = -EEXIST;
5152
5153         list_for_each_entry(nh, rt6_nh_list, next) {
5154                 /* check if fib6_info already exists */
5155                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5156                         return err;
5157         }
5158
5159         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5160         if (!nh)
5161                 return -ENOMEM;
5162         nh->fib6_info = rt;
5163         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5164         list_add_tail(&nh->next, rt6_nh_list);
5165
5166         return 0;
5167 }
5168
5169 static void ip6_route_mpath_notify(struct fib6_info *rt,
5170                                    struct fib6_info *rt_last,
5171                                    struct nl_info *info,
5172                                    __u16 nlflags)
5173 {
5174         /* if this is an APPEND route, then rt points to the first route
5175          * inserted and rt_last points to last route inserted. Userspace
5176          * wants a consistent dump of the route which starts at the first
5177          * nexthop. Since sibling routes are always added at the end of
5178          * the list, find the first sibling of the last route appended
5179          */
5180         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5181                 rt = list_first_entry(&rt_last->fib6_siblings,
5182                                       struct fib6_info,
5183                                       fib6_siblings);
5184         }
5185
5186         if (rt)
5187                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5188 }
5189
5190 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5191 {
5192         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5193         bool should_notify = false;
5194         struct fib6_info *leaf;
5195         struct fib6_node *fn;
5196
5197         rcu_read_lock();
5198         fn = rcu_dereference(rt->fib6_node);
5199         if (!fn)
5200                 goto out;
5201
5202         leaf = rcu_dereference(fn->leaf);
5203         if (!leaf)
5204                 goto out;
5205
5206         if (rt == leaf ||
5207             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5208              rt6_qualify_for_ecmp(leaf)))
5209                 should_notify = true;
5210 out:
5211         rcu_read_unlock();
5212
5213         return should_notify;
5214 }
5215
5216 static int ip6_route_multipath_add(struct fib6_config *cfg,
5217                                    struct netlink_ext_ack *extack)
5218 {
5219         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5220         struct nl_info *info = &cfg->fc_nlinfo;
5221         struct fib6_config r_cfg;
5222         struct rtnexthop *rtnh;
5223         struct fib6_info *rt;
5224         struct rt6_nh *err_nh;
5225         struct rt6_nh *nh, *nh_safe;
5226         __u16 nlflags;
5227         int remaining;
5228         int attrlen;
5229         int err = 1;
5230         int nhn = 0;
5231         int replace = (cfg->fc_nlinfo.nlh &&
5232                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5233         LIST_HEAD(rt6_nh_list);
5234
5235         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5236         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5237                 nlflags |= NLM_F_APPEND;
5238
5239         remaining = cfg->fc_mp_len;
5240         rtnh = (struct rtnexthop *)cfg->fc_mp;
5241
5242         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5243          * fib6_info structs per nexthop
5244          */
5245         while (rtnh_ok(rtnh, remaining)) {
5246                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5247                 if (rtnh->rtnh_ifindex)
5248                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5249
5250                 attrlen = rtnh_attrlen(rtnh);
5251                 if (attrlen > 0) {
5252                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5253
5254                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5255                         if (nla) {
5256                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5257                                 r_cfg.fc_flags |= RTF_GATEWAY;
5258                         }
5259                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5260                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5261                         if (nla)
5262                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5263                 }
5264
5265                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5266                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5267                 if (IS_ERR(rt)) {
5268                         err = PTR_ERR(rt);
5269                         rt = NULL;
5270                         goto cleanup;
5271                 }
5272                 if (!rt6_qualify_for_ecmp(rt)) {
5273                         err = -EINVAL;
5274                         NL_SET_ERR_MSG(extack,
5275                                        "Device only routes can not be added for IPv6 using the multipath API.");
5276                         fib6_info_release(rt);
5277                         goto cleanup;
5278                 }
5279
5280                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5281
5282                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5283                                             rt, &r_cfg);
5284                 if (err) {
5285                         fib6_info_release(rt);
5286                         goto cleanup;
5287                 }
5288
5289                 rtnh = rtnh_next(rtnh, &remaining);
5290         }
5291
5292         if (list_empty(&rt6_nh_list)) {
5293                 NL_SET_ERR_MSG(extack,
5294                                "Invalid nexthop configuration - no valid nexthops");
5295                 return -EINVAL;
5296         }
5297
5298         /* for add and replace send one notification with all nexthops.
5299          * Skip the notification in fib6_add_rt2node and send one with
5300          * the full route when done
5301          */
5302         info->skip_notify = 1;
5303
5304         /* For add and replace, send one notification with all nexthops. For
5305          * append, send one notification with all appended nexthops.
5306          */
5307         info->skip_notify_kernel = 1;
5308
5309         err_nh = NULL;
5310         list_for_each_entry(nh, &rt6_nh_list, next) {
5311                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5312                 fib6_info_release(nh->fib6_info);
5313
5314                 if (!err) {
5315                         /* save reference to last route successfully inserted */
5316                         rt_last = nh->fib6_info;
5317
5318                         /* save reference to first route for notification */
5319                         if (!rt_notif)
5320                                 rt_notif = nh->fib6_info;
5321                 }
5322
5323                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5324                 nh->fib6_info = NULL;
5325                 if (err) {
5326                         if (replace && nhn)
5327                                 NL_SET_ERR_MSG_MOD(extack,
5328                                                    "multipath route replace failed (check consistency of installed routes)");
5329                         err_nh = nh;
5330                         goto add_errout;
5331                 }
5332
5333                 /* Because each route is added like a single route we remove
5334                  * these flags after the first nexthop: if there is a collision,
5335                  * we have already failed to add the first nexthop:
5336                  * fib6_add_rt2node() has rejected it; when replacing, old
5337                  * nexthops have been replaced by first new, the rest should
5338                  * be added to it.
5339                  */
5340                 if (cfg->fc_nlinfo.nlh) {
5341                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5342                                                              NLM_F_REPLACE);
5343                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5344                 }
5345                 nhn++;
5346         }
5347
5348         /* An in-kernel notification should only be sent in case the new
5349          * multipath route is added as the first route in the node, or if
5350          * it was appended to it. We pass 'rt_notif' since it is the first
5351          * sibling and might allow us to skip some checks in the replace case.
5352          */
5353         if (ip6_route_mpath_should_notify(rt_notif)) {
5354                 enum fib_event_type fib_event;
5355
5356                 if (rt_notif->fib6_nsiblings != nhn - 1)
5357                         fib_event = FIB_EVENT_ENTRY_APPEND;
5358                 else
5359                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5360
5361                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5362                                                           fib_event, rt_notif,
5363                                                           nhn - 1, extack);
5364                 if (err) {
5365                         /* Delete all the siblings that were just added */
5366                         err_nh = NULL;
5367                         goto add_errout;
5368                 }
5369         }
5370
5371         /* success ... tell user about new route */
5372         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5373         goto cleanup;
5374
5375 add_errout:
5376         /* send notification for routes that were added so that
5377          * the delete notifications sent by ip6_route_del are
5378          * coherent
5379          */
5380         if (rt_notif)
5381                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5382
5383         /* Delete routes that were already added */
5384         list_for_each_entry(nh, &rt6_nh_list, next) {
5385                 if (err_nh == nh)
5386                         break;
5387                 ip6_route_del(&nh->r_cfg, extack);
5388         }
5389
5390 cleanup:
5391         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5392                 if (nh->fib6_info)
5393                         fib6_info_release(nh->fib6_info);
5394                 list_del(&nh->next);
5395                 kfree(nh);
5396         }
5397
5398         return err;
5399 }
5400
5401 static int ip6_route_multipath_del(struct fib6_config *cfg,
5402                                    struct netlink_ext_ack *extack)
5403 {
5404         struct fib6_config r_cfg;
5405         struct rtnexthop *rtnh;
5406         int last_err = 0;
5407         int remaining;
5408         int attrlen;
5409         int err;
5410
5411         remaining = cfg->fc_mp_len;
5412         rtnh = (struct rtnexthop *)cfg->fc_mp;
5413
5414         /* Parse a Multipath Entry */
5415         while (rtnh_ok(rtnh, remaining)) {
5416                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5417                 if (rtnh->rtnh_ifindex)
5418                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5419
5420                 attrlen = rtnh_attrlen(rtnh);
5421                 if (attrlen > 0) {
5422                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5423
5424                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5425                         if (nla) {
5426                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5427                                 r_cfg.fc_flags |= RTF_GATEWAY;
5428                         }
5429                 }
5430                 err = ip6_route_del(&r_cfg, extack);
5431                 if (err)
5432                         last_err = err;
5433
5434                 rtnh = rtnh_next(rtnh, &remaining);
5435         }
5436
5437         return last_err;
5438 }
5439
5440 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5441                               struct netlink_ext_ack *extack)
5442 {
5443         struct fib6_config cfg;
5444         int err;
5445
5446         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5447         if (err < 0)
5448                 return err;
5449
5450         if (cfg.fc_nh_id &&
5451             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5452                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5453                 return -EINVAL;
5454         }
5455
5456         if (cfg.fc_mp)
5457                 return ip6_route_multipath_del(&cfg, extack);
5458         else {
5459                 cfg.fc_delete_all_nh = 1;
5460                 return ip6_route_del(&cfg, extack);
5461         }
5462 }
5463
5464 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5465                               struct netlink_ext_ack *extack)
5466 {
5467         struct fib6_config cfg;
5468         int err;
5469
5470         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5471         if (err < 0)
5472                 return err;
5473
5474         if (cfg.fc_metric == 0)
5475                 cfg.fc_metric = IP6_RT_PRIO_USER;
5476
5477         if (cfg.fc_mp)
5478                 return ip6_route_multipath_add(&cfg, extack);
5479         else
5480                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5481 }
5482
5483 /* add the overhead of this fib6_nh to nexthop_len */
5484 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5485 {
5486         int *nexthop_len = arg;
5487
5488         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5489                      + NLA_ALIGN(sizeof(struct rtnexthop))
5490                      + nla_total_size(16); /* RTA_GATEWAY */
5491
5492         if (nh->fib_nh_lws) {
5493                 /* RTA_ENCAP_TYPE */
5494                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5495                 /* RTA_ENCAP */
5496                 *nexthop_len += nla_total_size(2);
5497         }
5498
5499         return 0;
5500 }
5501
5502 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5503 {
5504         int nexthop_len;
5505
5506         if (f6i->nh) {
5507                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5508                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5509                                          &nexthop_len);
5510         } else {
5511                 struct fib6_nh *nh = f6i->fib6_nh;
5512
5513                 nexthop_len = 0;
5514                 if (f6i->fib6_nsiblings) {
5515                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5516                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5517                                     + nla_total_size(16) /* RTA_GATEWAY */
5518                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5519
5520                         nexthop_len *= f6i->fib6_nsiblings;
5521                 }
5522                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5523         }
5524
5525         return NLMSG_ALIGN(sizeof(struct rtmsg))
5526                + nla_total_size(16) /* RTA_SRC */
5527                + nla_total_size(16) /* RTA_DST */
5528                + nla_total_size(16) /* RTA_GATEWAY */
5529                + nla_total_size(16) /* RTA_PREFSRC */
5530                + nla_total_size(4) /* RTA_TABLE */
5531                + nla_total_size(4) /* RTA_IIF */
5532                + nla_total_size(4) /* RTA_OIF */
5533                + nla_total_size(4) /* RTA_PRIORITY */
5534                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5535                + nla_total_size(sizeof(struct rta_cacheinfo))
5536                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5537                + nla_total_size(1) /* RTA_PREF */
5538                + nexthop_len;
5539 }
5540
5541 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5542                                  unsigned char *flags)
5543 {
5544         if (nexthop_is_multipath(nh)) {
5545                 struct nlattr *mp;
5546
5547                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5548                 if (!mp)
5549                         goto nla_put_failure;
5550
5551                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5552                         goto nla_put_failure;
5553
5554                 nla_nest_end(skb, mp);
5555         } else {
5556                 struct fib6_nh *fib6_nh;
5557
5558                 fib6_nh = nexthop_fib6_nh(nh);
5559                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5560                                      flags, false) < 0)
5561                         goto nla_put_failure;
5562         }
5563
5564         return 0;
5565
5566 nla_put_failure:
5567         return -EMSGSIZE;
5568 }
5569
5570 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5571                          struct fib6_info *rt, struct dst_entry *dst,
5572                          struct in6_addr *dest, struct in6_addr *src,
5573                          int iif, int type, u32 portid, u32 seq,
5574                          unsigned int flags)
5575 {
5576         struct rt6_info *rt6 = (struct rt6_info *)dst;
5577         struct rt6key *rt6_dst, *rt6_src;
5578         u32 *pmetrics, table, rt6_flags;
5579         unsigned char nh_flags = 0;
5580         struct nlmsghdr *nlh;
5581         struct rtmsg *rtm;
5582         long expires = 0;
5583
5584         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5585         if (!nlh)
5586                 return -EMSGSIZE;
5587
5588         if (rt6) {
5589                 rt6_dst = &rt6->rt6i_dst;
5590                 rt6_src = &rt6->rt6i_src;
5591                 rt6_flags = rt6->rt6i_flags;
5592         } else {
5593                 rt6_dst = &rt->fib6_dst;
5594                 rt6_src = &rt->fib6_src;
5595                 rt6_flags = rt->fib6_flags;
5596         }
5597
5598         rtm = nlmsg_data(nlh);
5599         rtm->rtm_family = AF_INET6;
5600         rtm->rtm_dst_len = rt6_dst->plen;
5601         rtm->rtm_src_len = rt6_src->plen;
5602         rtm->rtm_tos = 0;
5603         if (rt->fib6_table)
5604                 table = rt->fib6_table->tb6_id;
5605         else
5606                 table = RT6_TABLE_UNSPEC;
5607         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5608         if (nla_put_u32(skb, RTA_TABLE, table))
5609                 goto nla_put_failure;
5610
5611         rtm->rtm_type = rt->fib6_type;
5612         rtm->rtm_flags = 0;
5613         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5614         rtm->rtm_protocol = rt->fib6_protocol;
5615
5616         if (rt6_flags & RTF_CACHE)
5617                 rtm->rtm_flags |= RTM_F_CLONED;
5618
5619         if (dest) {
5620                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5621                         goto nla_put_failure;
5622                 rtm->rtm_dst_len = 128;
5623         } else if (rtm->rtm_dst_len)
5624                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5625                         goto nla_put_failure;
5626 #ifdef CONFIG_IPV6_SUBTREES
5627         if (src) {
5628                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5629                         goto nla_put_failure;
5630                 rtm->rtm_src_len = 128;
5631         } else if (rtm->rtm_src_len &&
5632                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5633                 goto nla_put_failure;
5634 #endif
5635         if (iif) {
5636 #ifdef CONFIG_IPV6_MROUTE
5637                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5638                         int err = ip6mr_get_route(net, skb, rtm, portid);
5639
5640                         if (err == 0)
5641                                 return 0;
5642                         if (err < 0)
5643                                 goto nla_put_failure;
5644                 } else
5645 #endif
5646                         if (nla_put_u32(skb, RTA_IIF, iif))
5647                                 goto nla_put_failure;
5648         } else if (dest) {
5649                 struct in6_addr saddr_buf;
5650                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5651                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5652                         goto nla_put_failure;
5653         }
5654
5655         if (rt->fib6_prefsrc.plen) {
5656                 struct in6_addr saddr_buf;
5657                 saddr_buf = rt->fib6_prefsrc.addr;
5658                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5659                         goto nla_put_failure;
5660         }
5661
5662         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5663         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5664                 goto nla_put_failure;
5665
5666         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5667                 goto nla_put_failure;
5668
5669         /* For multipath routes, walk the siblings list and add
5670          * each as a nexthop within RTA_MULTIPATH.
5671          */
5672         if (rt6) {
5673                 if (rt6_flags & RTF_GATEWAY &&
5674                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5675                         goto nla_put_failure;
5676
5677                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5678                         goto nla_put_failure;
5679
5680                 if (dst->lwtstate &&
5681                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5682                         goto nla_put_failure;
5683         } else if (rt->fib6_nsiblings) {
5684                 struct fib6_info *sibling, *next_sibling;
5685                 struct nlattr *mp;
5686
5687                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5688                 if (!mp)
5689                         goto nla_put_failure;
5690
5691                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5692                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5693                         goto nla_put_failure;
5694
5695                 list_for_each_entry_safe(sibling, next_sibling,
5696                                          &rt->fib6_siblings, fib6_siblings) {
5697                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5698                                             sibling->fib6_nh->fib_nh_weight,
5699                                             AF_INET6) < 0)
5700                                 goto nla_put_failure;
5701                 }
5702
5703                 nla_nest_end(skb, mp);
5704         } else if (rt->nh) {
5705                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5706                         goto nla_put_failure;
5707
5708                 if (nexthop_is_blackhole(rt->nh))
5709                         rtm->rtm_type = RTN_BLACKHOLE;
5710
5711                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5712                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5713                         goto nla_put_failure;
5714
5715                 rtm->rtm_flags |= nh_flags;
5716         } else {
5717                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5718                                      &nh_flags, false) < 0)
5719                         goto nla_put_failure;
5720
5721                 rtm->rtm_flags |= nh_flags;
5722         }
5723
5724         if (rt6_flags & RTF_EXPIRES) {
5725                 expires = dst ? dst->expires : rt->expires;
5726                 expires -= jiffies;
5727         }
5728
5729         if (!dst) {
5730                 if (rt->offload)
5731                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5732                 if (rt->trap)
5733                         rtm->rtm_flags |= RTM_F_TRAP;
5734                 if (rt->offload_failed)
5735                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5736         }
5737
5738         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5739                 goto nla_put_failure;
5740
5741         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5742                 goto nla_put_failure;
5743
5744
5745         nlmsg_end(skb, nlh);
5746         return 0;
5747
5748 nla_put_failure:
5749         nlmsg_cancel(skb, nlh);
5750         return -EMSGSIZE;
5751 }
5752
5753 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5754 {
5755         const struct net_device *dev = arg;
5756
5757         if (nh->fib_nh_dev == dev)
5758                 return 1;
5759
5760         return 0;
5761 }
5762
5763 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5764                                const struct net_device *dev)
5765 {
5766         if (f6i->nh) {
5767                 struct net_device *_dev = (struct net_device *)dev;
5768
5769                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5770                                                   fib6_info_nh_uses_dev,
5771                                                   _dev);
5772         }
5773
5774         if (f6i->fib6_nh->fib_nh_dev == dev)
5775                 return true;
5776
5777         if (f6i->fib6_nsiblings) {
5778                 struct fib6_info *sibling, *next_sibling;
5779
5780                 list_for_each_entry_safe(sibling, next_sibling,
5781                                          &f6i->fib6_siblings, fib6_siblings) {
5782                         if (sibling->fib6_nh->fib_nh_dev == dev)
5783                                 return true;
5784                 }
5785         }
5786
5787         return false;
5788 }
5789
5790 struct fib6_nh_exception_dump_walker {
5791         struct rt6_rtnl_dump_arg *dump;
5792         struct fib6_info *rt;
5793         unsigned int flags;
5794         unsigned int skip;
5795         unsigned int count;
5796 };
5797
5798 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5799 {
5800         struct fib6_nh_exception_dump_walker *w = arg;
5801         struct rt6_rtnl_dump_arg *dump = w->dump;
5802         struct rt6_exception_bucket *bucket;
5803         struct rt6_exception *rt6_ex;
5804         int i, err;
5805
5806         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5807         if (!bucket)
5808                 return 0;
5809
5810         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5811                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5812                         if (w->skip) {
5813                                 w->skip--;
5814                                 continue;
5815                         }
5816
5817                         /* Expiration of entries doesn't bump sernum, insertion
5818                          * does. Removal is triggered by insertion, so we can
5819                          * rely on the fact that if entries change between two
5820                          * partial dumps, this node is scanned again completely,
5821                          * see rt6_insert_exception() and fib6_dump_table().
5822                          *
5823                          * Count expired entries we go through as handled
5824                          * entries that we'll skip next time, in case of partial
5825                          * node dump. Otherwise, if entries expire meanwhile,
5826                          * we'll skip the wrong amount.
5827                          */
5828                         if (rt6_check_expired(rt6_ex->rt6i)) {
5829                                 w->count++;
5830                                 continue;
5831                         }
5832
5833                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5834                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5835                                             RTM_NEWROUTE,
5836                                             NETLINK_CB(dump->cb->skb).portid,
5837                                             dump->cb->nlh->nlmsg_seq, w->flags);
5838                         if (err)
5839                                 return err;
5840
5841                         w->count++;
5842                 }
5843                 bucket++;
5844         }
5845
5846         return 0;
5847 }
5848
5849 /* Return -1 if done with node, number of handled routes on partial dump */
5850 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5851 {
5852         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5853         struct fib_dump_filter *filter = &arg->filter;
5854         unsigned int flags = NLM_F_MULTI;
5855         struct net *net = arg->net;
5856         int count = 0;
5857
5858         if (rt == net->ipv6.fib6_null_entry)
5859                 return -1;
5860
5861         if ((filter->flags & RTM_F_PREFIX) &&
5862             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5863                 /* success since this is not a prefix route */
5864                 return -1;
5865         }
5866         if (filter->filter_set &&
5867             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5868              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5869              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5870                 return -1;
5871         }
5872
5873         if (filter->filter_set ||
5874             !filter->dump_routes || !filter->dump_exceptions) {
5875                 flags |= NLM_F_DUMP_FILTERED;
5876         }
5877
5878         if (filter->dump_routes) {
5879                 if (skip) {
5880                         skip--;
5881                 } else {
5882                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5883                                           0, RTM_NEWROUTE,
5884                                           NETLINK_CB(arg->cb->skb).portid,
5885                                           arg->cb->nlh->nlmsg_seq, flags)) {
5886                                 return 0;
5887                         }
5888                         count++;
5889                 }
5890         }
5891
5892         if (filter->dump_exceptions) {
5893                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5894                                                            .rt = rt,
5895                                                            .flags = flags,
5896                                                            .skip = skip,
5897                                                            .count = 0 };
5898                 int err;
5899
5900                 rcu_read_lock();
5901                 if (rt->nh) {
5902                         err = nexthop_for_each_fib6_nh(rt->nh,
5903                                                        rt6_nh_dump_exceptions,
5904                                                        &w);
5905                 } else {
5906                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5907                 }
5908                 rcu_read_unlock();
5909
5910                 if (err)
5911                         return count += w.count;
5912         }
5913
5914         return -1;
5915 }
5916
5917 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5918                                         const struct nlmsghdr *nlh,
5919                                         struct nlattr **tb,
5920                                         struct netlink_ext_ack *extack)
5921 {
5922         struct rtmsg *rtm;
5923         int i, err;
5924
5925         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5926                 NL_SET_ERR_MSG_MOD(extack,
5927                                    "Invalid header for get route request");
5928                 return -EINVAL;
5929         }
5930
5931         if (!netlink_strict_get_check(skb))
5932                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5933                                               rtm_ipv6_policy, extack);
5934
5935         rtm = nlmsg_data(nlh);
5936         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5937             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5938             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5939             rtm->rtm_type) {
5940                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5941                 return -EINVAL;
5942         }
5943         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5944                 NL_SET_ERR_MSG_MOD(extack,
5945                                    "Invalid flags for get route request");
5946                 return -EINVAL;
5947         }
5948
5949         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5950                                             rtm_ipv6_policy, extack);
5951         if (err)
5952                 return err;
5953
5954         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5955             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5956                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5957                 return -EINVAL;
5958         }
5959
5960         for (i = 0; i <= RTA_MAX; i++) {
5961                 if (!tb[i])
5962                         continue;
5963
5964                 switch (i) {
5965                 case RTA_SRC:
5966                 case RTA_DST:
5967                 case RTA_IIF:
5968                 case RTA_OIF:
5969                 case RTA_MARK:
5970                 case RTA_UID:
5971                 case RTA_SPORT:
5972                 case RTA_DPORT:
5973                 case RTA_IP_PROTO:
5974                         break;
5975                 default:
5976                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5977                         return -EINVAL;
5978                 }
5979         }
5980
5981         return 0;
5982 }
5983
5984 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5985                               struct netlink_ext_ack *extack)
5986 {
5987         struct net *net = sock_net(in_skb->sk);
5988         struct nlattr *tb[RTA_MAX+1];
5989         int err, iif = 0, oif = 0;
5990         struct fib6_info *from;
5991         struct dst_entry *dst;
5992         struct rt6_info *rt;
5993         struct sk_buff *skb;
5994         struct rtmsg *rtm;
5995         struct flowi6 fl6 = {};
5996         bool fibmatch;
5997
5998         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5999         if (err < 0)
6000                 goto errout;
6001
6002         err = -EINVAL;
6003         rtm = nlmsg_data(nlh);
6004         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
6005         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
6006
6007         if (tb[RTA_SRC]) {
6008                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6009                         goto errout;
6010
6011                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6012         }
6013
6014         if (tb[RTA_DST]) {
6015                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6016                         goto errout;
6017
6018                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6019         }
6020
6021         if (tb[RTA_IIF])
6022                 iif = nla_get_u32(tb[RTA_IIF]);
6023
6024         if (tb[RTA_OIF])
6025                 oif = nla_get_u32(tb[RTA_OIF]);
6026
6027         if (tb[RTA_MARK])
6028                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6029
6030         if (tb[RTA_UID])
6031                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6032                                            nla_get_u32(tb[RTA_UID]));
6033         else
6034                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6035
6036         if (tb[RTA_SPORT])
6037                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6038
6039         if (tb[RTA_DPORT])
6040                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6041
6042         if (tb[RTA_IP_PROTO]) {
6043                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6044                                                   &fl6.flowi6_proto, AF_INET6,
6045                                                   extack);
6046                 if (err)
6047                         goto errout;
6048         }
6049
6050         if (iif) {
6051                 struct net_device *dev;
6052                 int flags = 0;
6053
6054                 rcu_read_lock();
6055
6056                 dev = dev_get_by_index_rcu(net, iif);
6057                 if (!dev) {
6058                         rcu_read_unlock();
6059                         err = -ENODEV;
6060                         goto errout;
6061                 }
6062
6063                 fl6.flowi6_iif = iif;
6064
6065                 if (!ipv6_addr_any(&fl6.saddr))
6066                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6067
6068                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6069
6070                 rcu_read_unlock();
6071         } else {
6072                 fl6.flowi6_oif = oif;
6073
6074                 dst = ip6_route_output(net, NULL, &fl6);
6075         }
6076
6077
6078         rt = container_of(dst, struct rt6_info, dst);
6079         if (rt->dst.error) {
6080                 err = rt->dst.error;
6081                 ip6_rt_put(rt);
6082                 goto errout;
6083         }
6084
6085         if (rt == net->ipv6.ip6_null_entry) {
6086                 err = rt->dst.error;
6087                 ip6_rt_put(rt);
6088                 goto errout;
6089         }
6090
6091         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6092         if (!skb) {
6093                 ip6_rt_put(rt);
6094                 err = -ENOBUFS;
6095                 goto errout;
6096         }
6097
6098         skb_dst_set(skb, &rt->dst);
6099
6100         rcu_read_lock();
6101         from = rcu_dereference(rt->from);
6102         if (from) {
6103                 if (fibmatch)
6104                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6105                                             iif, RTM_NEWROUTE,
6106                                             NETLINK_CB(in_skb).portid,
6107                                             nlh->nlmsg_seq, 0);
6108                 else
6109                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6110                                             &fl6.saddr, iif, RTM_NEWROUTE,
6111                                             NETLINK_CB(in_skb).portid,
6112                                             nlh->nlmsg_seq, 0);
6113         } else {
6114                 err = -ENETUNREACH;
6115         }
6116         rcu_read_unlock();
6117
6118         if (err < 0) {
6119                 kfree_skb(skb);
6120                 goto errout;
6121         }
6122
6123         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6124 errout:
6125         return err;
6126 }
6127
6128 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6129                      unsigned int nlm_flags)
6130 {
6131         struct sk_buff *skb;
6132         struct net *net = info->nl_net;
6133         u32 seq;
6134         int err;
6135
6136         err = -ENOBUFS;
6137         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6138
6139         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6140         if (!skb)
6141                 goto errout;
6142
6143         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6144                             event, info->portid, seq, nlm_flags);
6145         if (err < 0) {
6146                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6147                 WARN_ON(err == -EMSGSIZE);
6148                 kfree_skb(skb);
6149                 goto errout;
6150         }
6151         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6152                     info->nlh, gfp_any());
6153         return;
6154 errout:
6155         if (err < 0)
6156                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6157 }
6158
6159 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6160                     struct nl_info *info)
6161 {
6162         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6163         struct sk_buff *skb;
6164         int err = -ENOBUFS;
6165
6166         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6167         if (!skb)
6168                 goto errout;
6169
6170         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6171                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6172         if (err < 0) {
6173                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6174                 WARN_ON(err == -EMSGSIZE);
6175                 kfree_skb(skb);
6176                 goto errout;
6177         }
6178         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6179                     info->nlh, gfp_any());
6180         return;
6181 errout:
6182         if (err < 0)
6183                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6184 }
6185
6186 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6187                             bool offload, bool trap, bool offload_failed)
6188 {
6189         struct sk_buff *skb;
6190         int err;
6191
6192         if (f6i->offload == offload && f6i->trap == trap &&
6193             f6i->offload_failed == offload_failed)
6194                 return;
6195
6196         f6i->offload = offload;
6197         f6i->trap = trap;
6198
6199         /* 2 means send notifications only if offload_failed was changed. */
6200         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6201             f6i->offload_failed == offload_failed)
6202                 return;
6203
6204         f6i->offload_failed = offload_failed;
6205
6206         if (!rcu_access_pointer(f6i->fib6_node))
6207                 /* The route was removed from the tree, do not send
6208                  * notification.
6209                  */
6210                 return;
6211
6212         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6213                 return;
6214
6215         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6216         if (!skb) {
6217                 err = -ENOBUFS;
6218                 goto errout;
6219         }
6220
6221         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6222                             0, 0);
6223         if (err < 0) {
6224                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6225                 WARN_ON(err == -EMSGSIZE);
6226                 kfree_skb(skb);
6227                 goto errout;
6228         }
6229
6230         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6231         return;
6232
6233 errout:
6234         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6235 }
6236 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6237
6238 static int ip6_route_dev_notify(struct notifier_block *this,
6239                                 unsigned long event, void *ptr)
6240 {
6241         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6242         struct net *net = dev_net(dev);
6243
6244         if (!(dev->flags & IFF_LOOPBACK))
6245                 return NOTIFY_OK;
6246
6247         if (event == NETDEV_REGISTER) {
6248                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6249                 net->ipv6.ip6_null_entry->dst.dev = dev;
6250                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6251 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6252                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6253                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6254                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6255                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6256 #endif
6257          } else if (event == NETDEV_UNREGISTER &&
6258                     dev->reg_state != NETREG_UNREGISTERED) {
6259                 /* NETDEV_UNREGISTER could be fired for multiple times by
6260                  * netdev_wait_allrefs(). Make sure we only call this once.
6261                  */
6262                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6263 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6264                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6265                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6266 #endif
6267         }
6268
6269         return NOTIFY_OK;
6270 }
6271
6272 /*
6273  *      /proc
6274  */
6275
6276 #ifdef CONFIG_PROC_FS
6277 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6278 {
6279         struct net *net = (struct net *)seq->private;
6280         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6281                    net->ipv6.rt6_stats->fib_nodes,
6282                    net->ipv6.rt6_stats->fib_route_nodes,
6283                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6284                    net->ipv6.rt6_stats->fib_rt_entries,
6285                    net->ipv6.rt6_stats->fib_rt_cache,
6286                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6287                    net->ipv6.rt6_stats->fib_discarded_routes);
6288
6289         return 0;
6290 }
6291 #endif  /* CONFIG_PROC_FS */
6292
6293 #ifdef CONFIG_SYSCTL
6294
6295 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6296                               void *buffer, size_t *lenp, loff_t *ppos)
6297 {
6298         struct net *net;
6299         int delay;
6300         int ret;
6301         if (!write)
6302                 return -EINVAL;
6303
6304         net = (struct net *)ctl->extra1;
6305         delay = net->ipv6.sysctl.flush_delay;
6306         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6307         if (ret)
6308                 return ret;
6309
6310         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6311         return 0;
6312 }
6313
6314 static struct ctl_table ipv6_route_table_template[] = {
6315         {
6316                 .procname       =       "flush",
6317                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6318                 .maxlen         =       sizeof(int),
6319                 .mode           =       0200,
6320                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6321         },
6322         {
6323                 .procname       =       "gc_thresh",
6324                 .data           =       &ip6_dst_ops_template.gc_thresh,
6325                 .maxlen         =       sizeof(int),
6326                 .mode           =       0644,
6327                 .proc_handler   =       proc_dointvec,
6328         },
6329         {
6330                 .procname       =       "max_size",
6331                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6332                 .maxlen         =       sizeof(int),
6333                 .mode           =       0644,
6334                 .proc_handler   =       proc_dointvec,
6335         },
6336         {
6337                 .procname       =       "gc_min_interval",
6338                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6339                 .maxlen         =       sizeof(int),
6340                 .mode           =       0644,
6341                 .proc_handler   =       proc_dointvec_jiffies,
6342         },
6343         {
6344                 .procname       =       "gc_timeout",
6345                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6346                 .maxlen         =       sizeof(int),
6347                 .mode           =       0644,
6348                 .proc_handler   =       proc_dointvec_jiffies,
6349         },
6350         {
6351                 .procname       =       "gc_interval",
6352                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6353                 .maxlen         =       sizeof(int),
6354                 .mode           =       0644,
6355                 .proc_handler   =       proc_dointvec_jiffies,
6356         },
6357         {
6358                 .procname       =       "gc_elasticity",
6359                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6360                 .maxlen         =       sizeof(int),
6361                 .mode           =       0644,
6362                 .proc_handler   =       proc_dointvec,
6363         },
6364         {
6365                 .procname       =       "mtu_expires",
6366                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6367                 .maxlen         =       sizeof(int),
6368                 .mode           =       0644,
6369                 .proc_handler   =       proc_dointvec_jiffies,
6370         },
6371         {
6372                 .procname       =       "min_adv_mss",
6373                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6374                 .maxlen         =       sizeof(int),
6375                 .mode           =       0644,
6376                 .proc_handler   =       proc_dointvec,
6377         },
6378         {
6379                 .procname       =       "gc_min_interval_ms",
6380                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6381                 .maxlen         =       sizeof(int),
6382                 .mode           =       0644,
6383                 .proc_handler   =       proc_dointvec_ms_jiffies,
6384         },
6385         {
6386                 .procname       =       "skip_notify_on_dev_down",
6387                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6388                 .maxlen         =       sizeof(int),
6389                 .mode           =       0644,
6390                 .proc_handler   =       proc_dointvec_minmax,
6391                 .extra1         =       SYSCTL_ZERO,
6392                 .extra2         =       SYSCTL_ONE,
6393         },
6394         { }
6395 };
6396
6397 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6398 {
6399         struct ctl_table *table;
6400
6401         table = kmemdup(ipv6_route_table_template,
6402                         sizeof(ipv6_route_table_template),
6403                         GFP_KERNEL);
6404
6405         if (table) {
6406                 table[0].data = &net->ipv6.sysctl.flush_delay;
6407                 table[0].extra1 = net;
6408                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6409                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6410                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6411                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6412                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6413                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6414                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6415                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6416                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6417                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6418
6419                 /* Don't export sysctls to unprivileged users */
6420                 if (net->user_ns != &init_user_ns)
6421                         table[0].procname = NULL;
6422         }
6423
6424         return table;
6425 }
6426 #endif
6427
6428 static int __net_init ip6_route_net_init(struct net *net)
6429 {
6430         int ret = -ENOMEM;
6431
6432         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6433                sizeof(net->ipv6.ip6_dst_ops));
6434
6435         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6436                 goto out_ip6_dst_ops;
6437
6438         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6439         if (!net->ipv6.fib6_null_entry)
6440                 goto out_ip6_dst_entries;
6441         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6442                sizeof(*net->ipv6.fib6_null_entry));
6443
6444         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6445                                            sizeof(*net->ipv6.ip6_null_entry),
6446                                            GFP_KERNEL);
6447         if (!net->ipv6.ip6_null_entry)
6448                 goto out_fib6_null_entry;
6449         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6450         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6451                          ip6_template_metrics, true);
6452         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6453
6454 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6455         net->ipv6.fib6_has_custom_rules = false;
6456         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6457                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6458                                                GFP_KERNEL);
6459         if (!net->ipv6.ip6_prohibit_entry)
6460                 goto out_ip6_null_entry;
6461         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6462         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6463                          ip6_template_metrics, true);
6464         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6465
6466         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6467                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6468                                                GFP_KERNEL);
6469         if (!net->ipv6.ip6_blk_hole_entry)
6470                 goto out_ip6_prohibit_entry;
6471         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6472         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6473                          ip6_template_metrics, true);
6474         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6475 #ifdef CONFIG_IPV6_SUBTREES
6476         net->ipv6.fib6_routes_require_src = 0;
6477 #endif
6478 #endif
6479
6480         net->ipv6.sysctl.flush_delay = 0;
6481         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6482         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6483         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6484         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6485         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6486         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6487         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6488         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6489
6490         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6491
6492         ret = 0;
6493 out:
6494         return ret;
6495
6496 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6497 out_ip6_prohibit_entry:
6498         kfree(net->ipv6.ip6_prohibit_entry);
6499 out_ip6_null_entry:
6500         kfree(net->ipv6.ip6_null_entry);
6501 #endif
6502 out_fib6_null_entry:
6503         kfree(net->ipv6.fib6_null_entry);
6504 out_ip6_dst_entries:
6505         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6506 out_ip6_dst_ops:
6507         goto out;
6508 }
6509
6510 static void __net_exit ip6_route_net_exit(struct net *net)
6511 {
6512         kfree(net->ipv6.fib6_null_entry);
6513         kfree(net->ipv6.ip6_null_entry);
6514 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6515         kfree(net->ipv6.ip6_prohibit_entry);
6516         kfree(net->ipv6.ip6_blk_hole_entry);
6517 #endif
6518         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6519 }
6520
6521 static int __net_init ip6_route_net_init_late(struct net *net)
6522 {
6523 #ifdef CONFIG_PROC_FS
6524         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6525                         sizeof(struct ipv6_route_iter));
6526         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6527                         rt6_stats_seq_show, NULL);
6528 #endif
6529         return 0;
6530 }
6531
6532 static void __net_exit ip6_route_net_exit_late(struct net *net)
6533 {
6534 #ifdef CONFIG_PROC_FS
6535         remove_proc_entry("ipv6_route", net->proc_net);
6536         remove_proc_entry("rt6_stats", net->proc_net);
6537 #endif
6538 }
6539
6540 static struct pernet_operations ip6_route_net_ops = {
6541         .init = ip6_route_net_init,
6542         .exit = ip6_route_net_exit,
6543 };
6544
6545 static int __net_init ipv6_inetpeer_init(struct net *net)
6546 {
6547         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6548
6549         if (!bp)
6550                 return -ENOMEM;
6551         inet_peer_base_init(bp);
6552         net->ipv6.peers = bp;
6553         return 0;
6554 }
6555
6556 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6557 {
6558         struct inet_peer_base *bp = net->ipv6.peers;
6559
6560         net->ipv6.peers = NULL;
6561         inetpeer_invalidate_tree(bp);
6562         kfree(bp);
6563 }
6564
6565 static struct pernet_operations ipv6_inetpeer_ops = {
6566         .init   =       ipv6_inetpeer_init,
6567         .exit   =       ipv6_inetpeer_exit,
6568 };
6569
6570 static struct pernet_operations ip6_route_net_late_ops = {
6571         .init = ip6_route_net_init_late,
6572         .exit = ip6_route_net_exit_late,
6573 };
6574
6575 static struct notifier_block ip6_route_dev_notifier = {
6576         .notifier_call = ip6_route_dev_notify,
6577         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6578 };
6579
6580 void __init ip6_route_init_special_entries(void)
6581 {
6582         /* Registering of the loopback is done before this portion of code,
6583          * the loopback reference in rt6_info will not be taken, do it
6584          * manually for init_net */
6585         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6586         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6587         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6588   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6589         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6590         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6591         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6592         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6593   #endif
6594 }
6595
6596 #if IS_BUILTIN(CONFIG_IPV6)
6597 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6598 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6599
6600 BTF_ID_LIST(btf_fib6_info_id)
6601 BTF_ID(struct, fib6_info)
6602
6603 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6604         .seq_ops                = &ipv6_route_seq_ops,
6605         .init_seq_private       = bpf_iter_init_seq_net,
6606         .fini_seq_private       = bpf_iter_fini_seq_net,
6607         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6608 };
6609
6610 static struct bpf_iter_reg ipv6_route_reg_info = {
6611         .target                 = "ipv6_route",
6612         .ctx_arg_info_size      = 1,
6613         .ctx_arg_info           = {
6614                 { offsetof(struct bpf_iter__ipv6_route, rt),
6615                   PTR_TO_BTF_ID_OR_NULL },
6616         },
6617         .seq_info               = &ipv6_route_seq_info,
6618 };
6619
6620 static int __init bpf_iter_register(void)
6621 {
6622         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6623         return bpf_iter_reg_target(&ipv6_route_reg_info);
6624 }
6625
6626 static void bpf_iter_unregister(void)
6627 {
6628         bpf_iter_unreg_target(&ipv6_route_reg_info);
6629 }
6630 #endif
6631 #endif
6632
6633 int __init ip6_route_init(void)
6634 {
6635         int ret;
6636         int cpu;
6637
6638         ret = -ENOMEM;
6639         ip6_dst_ops_template.kmem_cachep =
6640                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6641                                   SLAB_HWCACHE_ALIGN, NULL);
6642         if (!ip6_dst_ops_template.kmem_cachep)
6643                 goto out;
6644
6645         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6646         if (ret)
6647                 goto out_kmem_cache;
6648
6649         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6650         if (ret)
6651                 goto out_dst_entries;
6652
6653         ret = register_pernet_subsys(&ip6_route_net_ops);
6654         if (ret)
6655                 goto out_register_inetpeer;
6656
6657         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6658
6659         ret = fib6_init();
6660         if (ret)
6661                 goto out_register_subsys;
6662
6663         ret = xfrm6_init();
6664         if (ret)
6665                 goto out_fib6_init;
6666
6667         ret = fib6_rules_init();
6668         if (ret)
6669                 goto xfrm6_init;
6670
6671         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6672         if (ret)
6673                 goto fib6_rules_init;
6674
6675         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6676                                    inet6_rtm_newroute, NULL, 0);
6677         if (ret < 0)
6678                 goto out_register_late_subsys;
6679
6680         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6681                                    inet6_rtm_delroute, NULL, 0);
6682         if (ret < 0)
6683                 goto out_register_late_subsys;
6684
6685         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6686                                    inet6_rtm_getroute, NULL,
6687                                    RTNL_FLAG_DOIT_UNLOCKED);
6688         if (ret < 0)
6689                 goto out_register_late_subsys;
6690
6691         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6692         if (ret)
6693                 goto out_register_late_subsys;
6694
6695 #if IS_BUILTIN(CONFIG_IPV6)
6696 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6697         ret = bpf_iter_register();
6698         if (ret)
6699                 goto out_register_late_subsys;
6700 #endif
6701 #endif
6702
6703         for_each_possible_cpu(cpu) {
6704                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6705
6706                 INIT_LIST_HEAD(&ul->head);
6707                 spin_lock_init(&ul->lock);
6708         }
6709
6710 out:
6711         return ret;
6712
6713 out_register_late_subsys:
6714         rtnl_unregister_all(PF_INET6);
6715         unregister_pernet_subsys(&ip6_route_net_late_ops);
6716 fib6_rules_init:
6717         fib6_rules_cleanup();
6718 xfrm6_init:
6719         xfrm6_fini();
6720 out_fib6_init:
6721         fib6_gc_cleanup();
6722 out_register_subsys:
6723         unregister_pernet_subsys(&ip6_route_net_ops);
6724 out_register_inetpeer:
6725         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6726 out_dst_entries:
6727         dst_entries_destroy(&ip6_dst_blackhole_ops);
6728 out_kmem_cache:
6729         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6730         goto out;
6731 }
6732
6733 void ip6_route_cleanup(void)
6734 {
6735 #if IS_BUILTIN(CONFIG_IPV6)
6736 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6737         bpf_iter_unregister();
6738 #endif
6739 #endif
6740         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6741         unregister_pernet_subsys(&ip6_route_net_late_ops);
6742         fib6_rules_cleanup();
6743         xfrm6_fini();
6744         fib6_gc_cleanup();
6745         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6746         unregister_pernet_subsys(&ip6_route_net_ops);
6747         dst_entries_destroy(&ip6_dst_blackhole_ops);
6748         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6749 }