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             <[email protected]>
   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                 __u8 neigh_flags = 0;
2089
2090                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2091                 if (neigh)
2092                         neigh_flags = neigh->flags;
2093
2094                 if (!(neigh_flags & NTF_ROUTER)) {
2095                         RT6_TRACE("purging route %p via non-router but gateway\n",
2096                                   rt);
2097                         rt6_remove_exception(bucket, rt6_ex);
2098                         return;
2099                 }
2100         }
2101
2102         gc_args->more++;
2103 }
2104
2105 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2106                                    struct fib6_gc_args *gc_args,
2107                                    unsigned long now)
2108 {
2109         struct rt6_exception_bucket *bucket;
2110         struct rt6_exception *rt6_ex;
2111         struct hlist_node *tmp;
2112         int i;
2113
2114         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2115                 return;
2116
2117         rcu_read_lock_bh();
2118         spin_lock(&rt6_exception_lock);
2119         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2120         if (bucket) {
2121                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2122                         hlist_for_each_entry_safe(rt6_ex, tmp,
2123                                                   &bucket->chain, hlist) {
2124                                 rt6_age_examine_exception(bucket, rt6_ex,
2125                                                           gc_args, now);
2126                         }
2127                         bucket++;
2128                 }
2129         }
2130         spin_unlock(&rt6_exception_lock);
2131         rcu_read_unlock_bh();
2132 }
2133
2134 struct fib6_nh_age_excptn_arg {
2135         struct fib6_gc_args     *gc_args;
2136         unsigned long           now;
2137 };
2138
2139 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2140 {
2141         struct fib6_nh_age_excptn_arg *arg = _arg;
2142
2143         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2144         return 0;
2145 }
2146
2147 void rt6_age_exceptions(struct fib6_info *f6i,
2148                         struct fib6_gc_args *gc_args,
2149                         unsigned long now)
2150 {
2151         if (f6i->nh) {
2152                 struct fib6_nh_age_excptn_arg arg = {
2153                         .gc_args = gc_args,
2154                         .now = now
2155                 };
2156
2157                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2158                                          &arg);
2159         } else {
2160                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2161         }
2162 }
2163
2164 /* must be called with rcu lock held */
2165 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2166                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2167 {
2168         struct fib6_node *fn, *saved_fn;
2169
2170         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2171         saved_fn = fn;
2172
2173         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2174                 oif = 0;
2175
2176 redo_rt6_select:
2177         rt6_select(net, fn, oif, res, strict);
2178         if (res->f6i == net->ipv6.fib6_null_entry) {
2179                 fn = fib6_backtrack(fn, &fl6->saddr);
2180                 if (fn)
2181                         goto redo_rt6_select;
2182                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2183                         /* also consider unreachable route */
2184                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2185                         fn = saved_fn;
2186                         goto redo_rt6_select;
2187                 }
2188         }
2189
2190         trace_fib6_table_lookup(net, res, table, fl6);
2191
2192         return 0;
2193 }
2194
2195 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2196                                int oif, struct flowi6 *fl6,
2197                                const struct sk_buff *skb, int flags)
2198 {
2199         struct fib6_result res = {};
2200         struct rt6_info *rt = NULL;
2201         int strict = 0;
2202
2203         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2204                      !rcu_read_lock_held());
2205
2206         strict |= flags & RT6_LOOKUP_F_IFACE;
2207         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2208         if (net->ipv6.devconf_all->forwarding == 0)
2209                 strict |= RT6_LOOKUP_F_REACHABLE;
2210
2211         rcu_read_lock();
2212
2213         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2214         if (res.f6i == net->ipv6.fib6_null_entry)
2215                 goto out;
2216
2217         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2218
2219         /*Search through exception table */
2220         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2221         if (rt) {
2222                 goto out;
2223         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2224                             !res.nh->fib_nh_gw_family)) {
2225                 /* Create a RTF_CACHE clone which will not be
2226                  * owned by the fib6 tree.  It is for the special case where
2227                  * the daddr in the skb during the neighbor look-up is different
2228                  * from the fl6->daddr used to look-up route here.
2229                  */
2230                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2231
2232                 if (rt) {
2233                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2234                          * As rt6_uncached_list_add() does not consume refcnt,
2235                          * this refcnt is always returned to the caller even
2236                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2237                          */
2238                         rt6_uncached_list_add(rt);
2239                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2240                         rcu_read_unlock();
2241
2242                         return rt;
2243                 }
2244         } else {
2245                 /* Get a percpu copy */
2246                 local_bh_disable();
2247                 rt = rt6_get_pcpu_route(&res);
2248
2249                 if (!rt)
2250                         rt = rt6_make_pcpu_route(net, &res);
2251
2252                 local_bh_enable();
2253         }
2254 out:
2255         if (!rt)
2256                 rt = net->ipv6.ip6_null_entry;
2257         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2258                 ip6_hold_safe(net, &rt);
2259         rcu_read_unlock();
2260
2261         return rt;
2262 }
2263 EXPORT_SYMBOL_GPL(ip6_pol_route);
2264
2265 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2266                                             struct fib6_table *table,
2267                                             struct flowi6 *fl6,
2268                                             const struct sk_buff *skb,
2269                                             int flags)
2270 {
2271         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2272 }
2273
2274 struct dst_entry *ip6_route_input_lookup(struct net *net,
2275                                          struct net_device *dev,
2276                                          struct flowi6 *fl6,
2277                                          const struct sk_buff *skb,
2278                                          int flags)
2279 {
2280         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2281                 flags |= RT6_LOOKUP_F_IFACE;
2282
2283         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2284 }
2285 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2286
2287 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2288                                   struct flow_keys *keys,
2289                                   struct flow_keys *flkeys)
2290 {
2291         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2292         const struct ipv6hdr *key_iph = outer_iph;
2293         struct flow_keys *_flkeys = flkeys;
2294         const struct ipv6hdr *inner_iph;
2295         const struct icmp6hdr *icmph;
2296         struct ipv6hdr _inner_iph;
2297         struct icmp6hdr _icmph;
2298
2299         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2300                 goto out;
2301
2302         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2303                                    sizeof(_icmph), &_icmph);
2304         if (!icmph)
2305                 goto out;
2306
2307         if (!icmpv6_is_err(icmph->icmp6_type))
2308                 goto out;
2309
2310         inner_iph = skb_header_pointer(skb,
2311                                        skb_transport_offset(skb) + sizeof(*icmph),
2312                                        sizeof(_inner_iph), &_inner_iph);
2313         if (!inner_iph)
2314                 goto out;
2315
2316         key_iph = inner_iph;
2317         _flkeys = NULL;
2318 out:
2319         if (_flkeys) {
2320                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2321                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2322                 keys->tags.flow_label = _flkeys->tags.flow_label;
2323                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2324         } else {
2325                 keys->addrs.v6addrs.src = key_iph->saddr;
2326                 keys->addrs.v6addrs.dst = key_iph->daddr;
2327                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2328                 keys->basic.ip_proto = key_iph->nexthdr;
2329         }
2330 }
2331
2332 /* if skb is set it will be used and fl6 can be NULL */
2333 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2334                        const struct sk_buff *skb, struct flow_keys *flkeys)
2335 {
2336         struct flow_keys hash_keys;
2337         u32 mhash;
2338
2339         switch (ip6_multipath_hash_policy(net)) {
2340         case 0:
2341                 memset(&hash_keys, 0, sizeof(hash_keys));
2342                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2343                 if (skb) {
2344                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2345                 } else {
2346                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2347                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2348                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2349                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2350                 }
2351                 break;
2352         case 1:
2353                 if (skb) {
2354                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2355                         struct flow_keys keys;
2356
2357                         /* short-circuit if we already have L4 hash present */
2358                         if (skb->l4_hash)
2359                                 return skb_get_hash_raw(skb) >> 1;
2360
2361                         memset(&hash_keys, 0, sizeof(hash_keys));
2362
2363                         if (!flkeys) {
2364                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2365                                 flkeys = &keys;
2366                         }
2367                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2368                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2369                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2370                         hash_keys.ports.src = flkeys->ports.src;
2371                         hash_keys.ports.dst = flkeys->ports.dst;
2372                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2373                 } else {
2374                         memset(&hash_keys, 0, sizeof(hash_keys));
2375                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2376                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2377                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2378                         hash_keys.ports.src = fl6->fl6_sport;
2379                         hash_keys.ports.dst = fl6->fl6_dport;
2380                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2381                 }
2382                 break;
2383         case 2:
2384                 memset(&hash_keys, 0, sizeof(hash_keys));
2385                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2386                 if (skb) {
2387                         struct flow_keys keys;
2388
2389                         if (!flkeys) {
2390                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2391                                 flkeys = &keys;
2392                         }
2393
2394                         /* Inner can be v4 or v6 */
2395                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2396                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2397                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2398                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2399                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2400                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2401                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2402                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2403                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2404                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2405                         } else {
2406                                 /* Same as case 0 */
2407                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2408                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2409                         }
2410                 } else {
2411                         /* Same as case 0 */
2412                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2413                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2414                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2415                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2416                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2417                 }
2418                 break;
2419         }
2420         mhash = flow_hash_from_keys(&hash_keys);
2421
2422         return mhash >> 1;
2423 }
2424
2425 /* Called with rcu held */
2426 void ip6_route_input(struct sk_buff *skb)
2427 {
2428         const struct ipv6hdr *iph = ipv6_hdr(skb);
2429         struct net *net = dev_net(skb->dev);
2430         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2431         struct ip_tunnel_info *tun_info;
2432         struct flowi6 fl6 = {
2433                 .flowi6_iif = skb->dev->ifindex,
2434                 .daddr = iph->daddr,
2435                 .saddr = iph->saddr,
2436                 .flowlabel = ip6_flowinfo(iph),
2437                 .flowi6_mark = skb->mark,
2438                 .flowi6_proto = iph->nexthdr,
2439         };
2440         struct flow_keys *flkeys = NULL, _flkeys;
2441
2442         tun_info = skb_tunnel_info(skb);
2443         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2444                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2445
2446         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2447                 flkeys = &_flkeys;
2448
2449         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2450                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2451         skb_dst_drop(skb);
2452         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2453                                                       &fl6, skb, flags));
2454 }
2455
2456 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2457                                              struct fib6_table *table,
2458                                              struct flowi6 *fl6,
2459                                              const struct sk_buff *skb,
2460                                              int flags)
2461 {
2462         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2463 }
2464
2465 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2466                                                const struct sock *sk,
2467                                                struct flowi6 *fl6, int flags)
2468 {
2469         bool any_src;
2470
2471         if (ipv6_addr_type(&fl6->daddr) &
2472             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2473                 struct dst_entry *dst;
2474
2475                 /* This function does not take refcnt on the dst */
2476                 dst = l3mdev_link_scope_lookup(net, fl6);
2477                 if (dst)
2478                         return dst;
2479         }
2480
2481         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2482
2483         flags |= RT6_LOOKUP_F_DST_NOREF;
2484         any_src = ipv6_addr_any(&fl6->saddr);
2485         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2486             (fl6->flowi6_oif && any_src))
2487                 flags |= RT6_LOOKUP_F_IFACE;
2488
2489         if (!any_src)
2490                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2491         else if (sk)
2492                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2493
2494         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2495 }
2496 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2497
2498 struct dst_entry *ip6_route_output_flags(struct net *net,
2499                                          const struct sock *sk,
2500                                          struct flowi6 *fl6,
2501                                          int flags)
2502 {
2503         struct dst_entry *dst;
2504         struct rt6_info *rt6;
2505
2506         rcu_read_lock();
2507         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2508         rt6 = (struct rt6_info *)dst;
2509         /* For dst cached in uncached_list, refcnt is already taken. */
2510         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2511                 dst = &net->ipv6.ip6_null_entry->dst;
2512                 dst_hold(dst);
2513         }
2514         rcu_read_unlock();
2515
2516         return dst;
2517 }
2518 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2519
2520 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2521 {
2522         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2523         struct net_device *loopback_dev = net->loopback_dev;
2524         struct dst_entry *new = NULL;
2525
2526         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2527                        DST_OBSOLETE_DEAD, 0);
2528         if (rt) {
2529                 rt6_info_init(rt);
2530                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2531
2532                 new = &rt->dst;
2533                 new->__use = 1;
2534                 new->input = dst_discard;
2535                 new->output = dst_discard_out;
2536
2537                 dst_copy_metrics(new, &ort->dst);
2538
2539                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2540                 rt->rt6i_gateway = ort->rt6i_gateway;
2541                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2542
2543                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2544 #ifdef CONFIG_IPV6_SUBTREES
2545                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2546 #endif
2547         }
2548
2549         dst_release(dst_orig);
2550         return new ? new : ERR_PTR(-ENOMEM);
2551 }
2552
2553 /*
2554  *      Destination cache support functions
2555  */
2556
2557 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2558 {
2559         u32 rt_cookie = 0;
2560
2561         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2562                 return false;
2563
2564         if (fib6_check_expired(f6i))
2565                 return false;
2566
2567         return true;
2568 }
2569
2570 static struct dst_entry *rt6_check(struct rt6_info *rt,
2571                                    struct fib6_info *from,
2572                                    u32 cookie)
2573 {
2574         u32 rt_cookie = 0;
2575
2576         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2577             rt_cookie != cookie)
2578                 return NULL;
2579
2580         if (rt6_check_expired(rt))
2581                 return NULL;
2582
2583         return &rt->dst;
2584 }
2585
2586 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2587                                             struct fib6_info *from,
2588                                             u32 cookie)
2589 {
2590         if (!__rt6_check_expired(rt) &&
2591             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2592             fib6_check(from, cookie))
2593                 return &rt->dst;
2594         else
2595                 return NULL;
2596 }
2597
2598 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2599                                                         u32 cookie)
2600 {
2601         struct dst_entry *dst_ret;
2602         struct fib6_info *from;
2603         struct rt6_info *rt;
2604
2605         rt = container_of(dst, struct rt6_info, dst);
2606
2607         if (rt->sernum)
2608                 return rt6_is_valid(rt) ? dst : NULL;
2609
2610         rcu_read_lock();
2611
2612         /* All IPV6 dsts are created with ->obsolete set to the value
2613          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2614          * into this function always.
2615          */
2616
2617         from = rcu_dereference(rt->from);
2618
2619         if (from && (rt->rt6i_flags & RTF_PCPU ||
2620             unlikely(!list_empty(&rt->rt6i_uncached))))
2621                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2622         else
2623                 dst_ret = rt6_check(rt, from, cookie);
2624
2625         rcu_read_unlock();
2626
2627         return dst_ret;
2628 }
2629 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2630
2631 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2632 {
2633         struct rt6_info *rt = (struct rt6_info *) dst;
2634
2635         if (rt) {
2636                 if (rt->rt6i_flags & RTF_CACHE) {
2637                         rcu_read_lock();
2638                         if (rt6_check_expired(rt)) {
2639                                 rt6_remove_exception_rt(rt);
2640                                 dst = NULL;
2641                         }
2642                         rcu_read_unlock();
2643                 } else {
2644                         dst_release(dst);
2645                         dst = NULL;
2646                 }
2647         }
2648         return dst;
2649 }
2650
2651 static void ip6_link_failure(struct sk_buff *skb)
2652 {
2653         struct rt6_info *rt;
2654
2655         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2656
2657         rt = (struct rt6_info *) skb_dst(skb);
2658         if (rt) {
2659                 rcu_read_lock();
2660                 if (rt->rt6i_flags & RTF_CACHE) {
2661                         rt6_remove_exception_rt(rt);
2662                 } else {
2663                         struct fib6_info *from;
2664                         struct fib6_node *fn;
2665
2666                         from = rcu_dereference(rt->from);
2667                         if (from) {
2668                                 fn = rcu_dereference(from->fib6_node);
2669                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2670                                         fn->fn_sernum = -1;
2671                         }
2672                 }
2673                 rcu_read_unlock();
2674         }
2675 }
2676
2677 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2678 {
2679         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2680                 struct fib6_info *from;
2681
2682                 rcu_read_lock();
2683                 from = rcu_dereference(rt0->from);
2684                 if (from)
2685                         rt0->dst.expires = from->expires;
2686                 rcu_read_unlock();
2687         }
2688
2689         dst_set_expires(&rt0->dst, timeout);
2690         rt0->rt6i_flags |= RTF_EXPIRES;
2691 }
2692
2693 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2694 {
2695         struct net *net = dev_net(rt->dst.dev);
2696
2697         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2698         rt->rt6i_flags |= RTF_MODIFIED;
2699         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2700 }
2701
2702 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2703 {
2704         return !(rt->rt6i_flags & RTF_CACHE) &&
2705                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2706 }
2707
2708 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2709                                  const struct ipv6hdr *iph, u32 mtu,
2710                                  bool confirm_neigh)
2711 {
2712         const struct in6_addr *daddr, *saddr;
2713         struct rt6_info *rt6 = (struct rt6_info *)dst;
2714
2715         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2716          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2717          * [see also comment in rt6_mtu_change_route()]
2718          */
2719
2720         if (iph) {
2721                 daddr = &iph->daddr;
2722                 saddr = &iph->saddr;
2723         } else if (sk) {
2724                 daddr = &sk->sk_v6_daddr;
2725                 saddr = &inet6_sk(sk)->saddr;
2726         } else {
2727                 daddr = NULL;
2728                 saddr = NULL;
2729         }
2730
2731         if (confirm_neigh)
2732                 dst_confirm_neigh(dst, daddr);
2733
2734         if (mtu < IPV6_MIN_MTU)
2735                 return;
2736         if (mtu >= dst_mtu(dst))
2737                 return;
2738
2739         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2740                 rt6_do_update_pmtu(rt6, mtu);
2741                 /* update rt6_ex->stamp for cache */
2742                 if (rt6->rt6i_flags & RTF_CACHE)
2743                         rt6_update_exception_stamp_rt(rt6);
2744         } else if (daddr) {
2745                 struct fib6_result res = {};
2746                 struct rt6_info *nrt6;
2747
2748                 rcu_read_lock();
2749                 res.f6i = rcu_dereference(rt6->from);
2750                 if (!res.f6i)
2751                         goto out_unlock;
2752
2753                 res.fib6_flags = res.f6i->fib6_flags;
2754                 res.fib6_type = res.f6i->fib6_type;
2755
2756                 if (res.f6i->nh) {
2757                         struct fib6_nh_match_arg arg = {
2758                                 .dev = dst->dev,
2759                                 .gw = &rt6->rt6i_gateway,
2760                         };
2761
2762                         nexthop_for_each_fib6_nh(res.f6i->nh,
2763                                                  fib6_nh_find_match, &arg);
2764
2765                         /* fib6_info uses a nexthop that does not have fib6_nh
2766                          * using the dst->dev + gw. Should be impossible.
2767                          */
2768                         if (!arg.match)
2769                                 goto out_unlock;
2770
2771                         res.nh = arg.match;
2772                 } else {
2773                         res.nh = res.f6i->fib6_nh;
2774                 }
2775
2776                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2777                 if (nrt6) {
2778                         rt6_do_update_pmtu(nrt6, mtu);
2779                         if (rt6_insert_exception(nrt6, &res))
2780                                 dst_release_immediate(&nrt6->dst);
2781                 }
2782 out_unlock:
2783                 rcu_read_unlock();
2784         }
2785 }
2786
2787 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2788                                struct sk_buff *skb, u32 mtu,
2789                                bool confirm_neigh)
2790 {
2791         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2792                              confirm_neigh);
2793 }
2794
2795 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2796                      int oif, u32 mark, kuid_t uid)
2797 {
2798         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2799         struct dst_entry *dst;
2800         struct flowi6 fl6 = {
2801                 .flowi6_oif = oif,
2802                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2803                 .daddr = iph->daddr,
2804                 .saddr = iph->saddr,
2805                 .flowlabel = ip6_flowinfo(iph),
2806                 .flowi6_uid = uid,
2807         };
2808
2809         dst = ip6_route_output(net, NULL, &fl6);
2810         if (!dst->error)
2811                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2812         dst_release(dst);
2813 }
2814 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2815
2816 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2817 {
2818         int oif = sk->sk_bound_dev_if;
2819         struct dst_entry *dst;
2820
2821         if (!oif && skb->dev)
2822                 oif = l3mdev_master_ifindex(skb->dev);
2823
2824         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2825
2826         dst = __sk_dst_get(sk);
2827         if (!dst || !dst->obsolete ||
2828             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2829                 return;
2830
2831         bh_lock_sock(sk);
2832         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2833                 ip6_datagram_dst_update(sk, false);
2834         bh_unlock_sock(sk);
2835 }
2836 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2837
2838 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2839                            const struct flowi6 *fl6)
2840 {
2841 #ifdef CONFIG_IPV6_SUBTREES
2842         struct ipv6_pinfo *np = inet6_sk(sk);
2843 #endif
2844
2845         ip6_dst_store(sk, dst,
2846                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2847                       &sk->sk_v6_daddr : NULL,
2848 #ifdef CONFIG_IPV6_SUBTREES
2849                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2850                       &np->saddr :
2851 #endif
2852                       NULL);
2853 }
2854
2855 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2856                                   struct flowi6 *fl6,
2857                                   const struct in6_addr *gw,
2858                                   struct rt6_info **ret)
2859 {
2860         const struct fib6_nh *nh = res->nh;
2861
2862         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2863             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2864                 return false;
2865
2866         /* rt_cache's gateway might be different from its 'parent'
2867          * in the case of an ip redirect.
2868          * So we keep searching in the exception table if the gateway
2869          * is different.
2870          */
2871         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2872                 struct rt6_info *rt_cache;
2873
2874                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2875                 if (rt_cache &&
2876                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2877                         *ret = rt_cache;
2878                         return true;
2879                 }
2880                 return false;
2881         }
2882         return true;
2883 }
2884
2885 struct fib6_nh_rd_arg {
2886         struct fib6_result      *res;
2887         struct flowi6           *fl6;
2888         const struct in6_addr   *gw;
2889         struct rt6_info         **ret;
2890 };
2891
2892 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2893 {
2894         struct fib6_nh_rd_arg *arg = _arg;
2895
2896         arg->res->nh = nh;
2897         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2898 }
2899
2900 /* Handle redirects */
2901 struct ip6rd_flowi {
2902         struct flowi6 fl6;
2903         struct in6_addr gateway;
2904 };
2905
2906 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
2907                                              struct fib6_table *table,
2908                                              struct flowi6 *fl6,
2909                                              const struct sk_buff *skb,
2910                                              int flags)
2911 {
2912         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2913         struct rt6_info *ret = NULL;
2914         struct fib6_result res = {};
2915         struct fib6_nh_rd_arg arg = {
2916                 .res = &res,
2917                 .fl6 = fl6,
2918                 .gw  = &rdfl->gateway,
2919                 .ret = &ret
2920         };
2921         struct fib6_info *rt;
2922         struct fib6_node *fn;
2923
2924         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2925          * this case we must match on the real ingress device, so reset it
2926          */
2927         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2928                 fl6->flowi6_oif = skb->dev->ifindex;
2929
2930         /* Get the "current" route for this destination and
2931          * check if the redirect has come from appropriate router.
2932          *
2933          * RFC 4861 specifies that redirects should only be
2934          * accepted if they come from the nexthop to the target.
2935          * Due to the way the routes are chosen, this notion
2936          * is a bit fuzzy and one might need to check all possible
2937          * routes.
2938          */
2939
2940         rcu_read_lock();
2941         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2942 restart:
2943         for_each_fib6_node_rt_rcu(fn) {
2944                 res.f6i = rt;
2945                 if (fib6_check_expired(rt))
2946                         continue;
2947                 if (rt->fib6_flags & RTF_REJECT)
2948                         break;
2949                 if (unlikely(rt->nh)) {
2950                         if (nexthop_is_blackhole(rt->nh))
2951                                 continue;
2952                         /* on match, res->nh is filled in and potentially ret */
2953                         if (nexthop_for_each_fib6_nh(rt->nh,
2954                                                      fib6_nh_redirect_match,
2955                                                      &arg))
2956                                 goto out;
2957                 } else {
2958                         res.nh = rt->fib6_nh;
2959                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2960                                                   &ret))
2961                                 goto out;
2962                 }
2963         }
2964
2965         if (!rt)
2966                 rt = net->ipv6.fib6_null_entry;
2967         else if (rt->fib6_flags & RTF_REJECT) {
2968                 ret = net->ipv6.ip6_null_entry;
2969                 goto out;
2970         }
2971
2972         if (rt == net->ipv6.fib6_null_entry) {
2973                 fn = fib6_backtrack(fn, &fl6->saddr);
2974                 if (fn)
2975                         goto restart;
2976         }
2977
2978         res.f6i = rt;
2979         res.nh = rt->fib6_nh;
2980 out:
2981         if (ret) {
2982                 ip6_hold_safe(net, &ret);
2983         } else {
2984                 res.fib6_flags = res.f6i->fib6_flags;
2985                 res.fib6_type = res.f6i->fib6_type;
2986                 ret = ip6_create_rt_rcu(&res);
2987         }
2988
2989         rcu_read_unlock();
2990
2991         trace_fib6_table_lookup(net, &res, table, fl6);
2992         return ret;
2993 };
2994
2995 static struct dst_entry *ip6_route_redirect(struct net *net,
2996                                             const struct flowi6 *fl6,
2997                                             const struct sk_buff *skb,
2998                                             const struct in6_addr *gateway)
2999 {
3000         int flags = RT6_LOOKUP_F_HAS_SADDR;
3001         struct ip6rd_flowi rdfl;
3002
3003         rdfl.fl6 = *fl6;
3004         rdfl.gateway = *gateway;
3005
3006         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3007                                 flags, __ip6_route_redirect);
3008 }
3009
3010 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3011                   kuid_t uid)
3012 {
3013         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3014         struct dst_entry *dst;
3015         struct flowi6 fl6 = {
3016                 .flowi6_iif = LOOPBACK_IFINDEX,
3017                 .flowi6_oif = oif,
3018                 .flowi6_mark = mark,
3019                 .daddr = iph->daddr,
3020                 .saddr = iph->saddr,
3021                 .flowlabel = ip6_flowinfo(iph),
3022                 .flowi6_uid = uid,
3023         };
3024
3025         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3026         rt6_do_redirect(dst, NULL, skb);
3027         dst_release(dst);
3028 }
3029 EXPORT_SYMBOL_GPL(ip6_redirect);
3030
3031 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3032 {
3033         const struct ipv6hdr *iph = ipv6_hdr(skb);
3034         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3035         struct dst_entry *dst;
3036         struct flowi6 fl6 = {
3037                 .flowi6_iif = LOOPBACK_IFINDEX,
3038                 .flowi6_oif = oif,
3039                 .daddr = msg->dest,
3040                 .saddr = iph->daddr,
3041                 .flowi6_uid = sock_net_uid(net, NULL),
3042         };
3043
3044         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3045         rt6_do_redirect(dst, NULL, skb);
3046         dst_release(dst);
3047 }
3048
3049 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3050 {
3051         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3052                      sk->sk_uid);
3053 }
3054 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3055
3056 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3057 {
3058         struct net_device *dev = dst->dev;
3059         unsigned int mtu = dst_mtu(dst);
3060         struct net *net = dev_net(dev);
3061
3062         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3063
3064         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3065                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3066
3067         /*
3068          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3069          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3070          * IPV6_MAXPLEN is also valid and means: "any MSS,
3071          * rely only on pmtu discovery"
3072          */
3073         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3074                 mtu = IPV6_MAXPLEN;
3075         return mtu;
3076 }
3077
3078 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3079 {
3080         struct inet6_dev *idev;
3081         unsigned int mtu;
3082
3083         mtu = dst_metric_raw(dst, RTAX_MTU);
3084         if (mtu)
3085                 goto out;
3086
3087         mtu = IPV6_MIN_MTU;
3088
3089         rcu_read_lock();
3090         idev = __in6_dev_get(dst->dev);
3091         if (idev)
3092                 mtu = idev->cnf.mtu6;
3093         rcu_read_unlock();
3094
3095 out:
3096         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3097
3098         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3099 }
3100 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3101
3102 /* MTU selection:
3103  * 1. mtu on route is locked - use it
3104  * 2. mtu from nexthop exception
3105  * 3. mtu from egress device
3106  *
3107  * based on ip6_dst_mtu_forward and exception logic of
3108  * rt6_find_cached_rt; called with rcu_read_lock
3109  */
3110 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3111                       const struct in6_addr *daddr,
3112                       const struct in6_addr *saddr)
3113 {
3114         const struct fib6_nh *nh = res->nh;
3115         struct fib6_info *f6i = res->f6i;
3116         struct inet6_dev *idev;
3117         struct rt6_info *rt;
3118         u32 mtu = 0;
3119
3120         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3121                 mtu = f6i->fib6_pmtu;
3122                 if (mtu)
3123                         goto out;
3124         }
3125
3126         rt = rt6_find_cached_rt(res, daddr, saddr);
3127         if (unlikely(rt)) {
3128                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3129         } else {
3130                 struct net_device *dev = nh->fib_nh_dev;
3131
3132                 mtu = IPV6_MIN_MTU;
3133                 idev = __in6_dev_get(dev);
3134                 if (idev && idev->cnf.mtu6 > mtu)
3135                         mtu = idev->cnf.mtu6;
3136         }
3137
3138         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3139 out:
3140         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3141 }
3142
3143 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3144                                   struct flowi6 *fl6)
3145 {
3146         struct dst_entry *dst;
3147         struct rt6_info *rt;
3148         struct inet6_dev *idev = in6_dev_get(dev);
3149         struct net *net = dev_net(dev);
3150
3151         if (unlikely(!idev))
3152                 return ERR_PTR(-ENODEV);
3153
3154         rt = ip6_dst_alloc(net, dev, 0);
3155         if (unlikely(!rt)) {
3156                 in6_dev_put(idev);
3157                 dst = ERR_PTR(-ENOMEM);
3158                 goto out;
3159         }
3160
3161         rt->dst.input = ip6_input;
3162         rt->dst.output  = ip6_output;
3163         rt->rt6i_gateway  = fl6->daddr;
3164         rt->rt6i_dst.addr = fl6->daddr;
3165         rt->rt6i_dst.plen = 128;
3166         rt->rt6i_idev     = idev;
3167         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3168
3169         /* Add this dst into uncached_list so that rt6_disable_ip() can
3170          * do proper release of the net_device
3171          */
3172         rt6_uncached_list_add(rt);
3173         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3174
3175         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3176
3177 out:
3178         return dst;
3179 }
3180
3181 static int ip6_dst_gc(struct dst_ops *ops)
3182 {
3183         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3184         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3185         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3186         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3187         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3188         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3189         int entries;
3190
3191         entries = dst_entries_get_fast(ops);
3192         if (entries > rt_max_size)
3193                 entries = dst_entries_get_slow(ops);
3194
3195         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3196             entries <= rt_max_size)
3197                 goto out;
3198
3199         net->ipv6.ip6_rt_gc_expire++;
3200         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3201         entries = dst_entries_get_slow(ops);
3202         if (entries < ops->gc_thresh)
3203                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3204 out:
3205         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3206         return entries > rt_max_size;
3207 }
3208
3209 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3210                                const struct in6_addr *gw_addr, u32 tbid,
3211                                int flags, struct fib6_result *res)
3212 {
3213         struct flowi6 fl6 = {
3214                 .flowi6_oif = cfg->fc_ifindex,
3215                 .daddr = *gw_addr,
3216                 .saddr = cfg->fc_prefsrc,
3217         };
3218         struct fib6_table *table;
3219         int err;
3220
3221         table = fib6_get_table(net, tbid);
3222         if (!table)
3223                 return -EINVAL;
3224
3225         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3226                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3227
3228         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3229
3230         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3231         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3232                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3233                                  cfg->fc_ifindex != 0, NULL, flags);
3234
3235         return err;
3236 }
3237
3238 static int ip6_route_check_nh_onlink(struct net *net,
3239                                      struct fib6_config *cfg,
3240                                      const struct net_device *dev,
3241                                      struct netlink_ext_ack *extack)
3242 {
3243         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3244         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3245         struct fib6_result res = {};
3246         int err;
3247
3248         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3249         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3250             /* ignore match if it is the default route */
3251             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3252             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3253                 NL_SET_ERR_MSG(extack,
3254                                "Nexthop has invalid gateway or device mismatch");
3255                 err = -EINVAL;
3256         }
3257
3258         return err;
3259 }
3260
3261 static int ip6_route_check_nh(struct net *net,
3262                               struct fib6_config *cfg,
3263                               struct net_device **_dev,
3264                               struct inet6_dev **idev)
3265 {
3266         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3267         struct net_device *dev = _dev ? *_dev : NULL;
3268         int flags = RT6_LOOKUP_F_IFACE;
3269         struct fib6_result res = {};
3270         int err = -EHOSTUNREACH;
3271
3272         if (cfg->fc_table) {
3273                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3274                                           cfg->fc_table, flags, &res);
3275                 /* gw_addr can not require a gateway or resolve to a reject
3276                  * route. If a device is given, it must match the result.
3277                  */
3278                 if (err || res.fib6_flags & RTF_REJECT ||
3279                     res.nh->fib_nh_gw_family ||
3280                     (dev && dev != res.nh->fib_nh_dev))
3281                         err = -EHOSTUNREACH;
3282         }
3283
3284         if (err < 0) {
3285                 struct flowi6 fl6 = {
3286                         .flowi6_oif = cfg->fc_ifindex,
3287                         .daddr = *gw_addr,
3288                 };
3289
3290                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3291                 if (err || res.fib6_flags & RTF_REJECT ||
3292                     res.nh->fib_nh_gw_family)
3293                         err = -EHOSTUNREACH;
3294
3295                 if (err)
3296                         return err;
3297
3298                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3299                                  cfg->fc_ifindex != 0, NULL, flags);
3300         }
3301
3302         err = 0;
3303         if (dev) {
3304                 if (dev != res.nh->fib_nh_dev)
3305                         err = -EHOSTUNREACH;
3306         } else {
3307                 *_dev = dev = res.nh->fib_nh_dev;
3308                 dev_hold(dev);
3309                 *idev = in6_dev_get(dev);
3310         }
3311
3312         return err;
3313 }
3314
3315 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3316                            struct net_device **_dev, struct inet6_dev **idev,
3317                            struct netlink_ext_ack *extack)
3318 {
3319         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3320         int gwa_type = ipv6_addr_type(gw_addr);
3321         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3322         const struct net_device *dev = *_dev;
3323         bool need_addr_check = !dev;
3324         int err = -EINVAL;
3325
3326         /* if gw_addr is local we will fail to detect this in case
3327          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3328          * will return already-added prefix route via interface that
3329          * prefix route was assigned to, which might be non-loopback.
3330          */
3331         if (dev &&
3332             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3333                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3334                 goto out;
3335         }
3336
3337         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3338                 /* IPv6 strictly inhibits using not link-local
3339                  * addresses as nexthop address.
3340                  * Otherwise, router will not able to send redirects.
3341                  * It is very good, but in some (rare!) circumstances
3342                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3343                  * some exceptions. --ANK
3344                  * We allow IPv4-mapped nexthops to support RFC4798-type
3345                  * addressing
3346                  */
3347                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3348                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3349                         goto out;
3350                 }
3351
3352                 rcu_read_lock();
3353
3354                 if (cfg->fc_flags & RTNH_F_ONLINK)
3355                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3356                 else
3357                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3358
3359                 rcu_read_unlock();
3360
3361                 if (err)
3362                         goto out;
3363         }
3364
3365         /* reload in case device was changed */
3366         dev = *_dev;
3367
3368         err = -EINVAL;
3369         if (!dev) {
3370                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3371                 goto out;
3372         } else if (dev->flags & IFF_LOOPBACK) {
3373                 NL_SET_ERR_MSG(extack,
3374                                "Egress device can not be loopback device for this route");
3375                 goto out;
3376         }
3377
3378         /* if we did not check gw_addr above, do so now that the
3379          * egress device has been resolved.
3380          */
3381         if (need_addr_check &&
3382             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3383                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3384                 goto out;
3385         }
3386
3387         err = 0;
3388 out:
3389         return err;
3390 }
3391
3392 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3393 {
3394         if ((flags & RTF_REJECT) ||
3395             (dev && (dev->flags & IFF_LOOPBACK) &&
3396              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3397              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3398                 return true;
3399
3400         return false;
3401 }
3402
3403 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3404                  struct fib6_config *cfg, gfp_t gfp_flags,
3405                  struct netlink_ext_ack *extack)
3406 {
3407         struct net_device *dev = NULL;
3408         struct inet6_dev *idev = NULL;
3409         int addr_type;
3410         int err;
3411
3412         fib6_nh->fib_nh_family = AF_INET6;
3413 #ifdef CONFIG_IPV6_ROUTER_PREF
3414         fib6_nh->last_probe = jiffies;
3415 #endif
3416         if (cfg->fc_is_fdb) {
3417                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3418                 fib6_nh->fib_nh_gw_family = AF_INET6;
3419                 return 0;
3420         }
3421
3422         err = -ENODEV;
3423         if (cfg->fc_ifindex) {
3424                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3425                 if (!dev)
3426                         goto out;
3427                 idev = in6_dev_get(dev);
3428                 if (!idev)
3429                         goto out;
3430         }
3431
3432         if (cfg->fc_flags & RTNH_F_ONLINK) {
3433                 if (!dev) {
3434                         NL_SET_ERR_MSG(extack,
3435                                        "Nexthop device required for onlink");
3436                         goto out;
3437                 }
3438
3439                 if (!(dev->flags & IFF_UP)) {
3440                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3441                         err = -ENETDOWN;
3442                         goto out;
3443                 }
3444
3445                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3446         }
3447
3448         fib6_nh->fib_nh_weight = 1;
3449
3450         /* We cannot add true routes via loopback here,
3451          * they would result in kernel looping; promote them to reject routes
3452          */
3453         addr_type = ipv6_addr_type(&cfg->fc_dst);
3454         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3455                 /* hold loopback dev/idev if we haven't done so. */
3456                 if (dev != net->loopback_dev) {
3457                         if (dev) {
3458                                 dev_put(dev);
3459                                 in6_dev_put(idev);
3460                         }
3461                         dev = net->loopback_dev;
3462                         dev_hold(dev);
3463                         idev = in6_dev_get(dev);
3464                         if (!idev) {
3465                                 err = -ENODEV;
3466                                 goto out;
3467                         }
3468                 }
3469                 goto pcpu_alloc;
3470         }
3471
3472         if (cfg->fc_flags & RTF_GATEWAY) {
3473                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3474                 if (err)
3475                         goto out;
3476
3477                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3478                 fib6_nh->fib_nh_gw_family = AF_INET6;
3479         }
3480
3481         err = -ENODEV;
3482         if (!dev)
3483                 goto out;
3484
3485         if (idev->cnf.disable_ipv6) {
3486                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3487                 err = -EACCES;
3488                 goto out;
3489         }
3490
3491         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3492                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3493                 err = -ENETDOWN;
3494                 goto out;
3495         }
3496
3497         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3498             !netif_carrier_ok(dev))
3499                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3500
3501         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3502                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3503         if (err)
3504                 goto out;
3505
3506 pcpu_alloc:
3507         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3508         if (!fib6_nh->rt6i_pcpu) {
3509                 err = -ENOMEM;
3510                 goto out;
3511         }
3512
3513         fib6_nh->fib_nh_dev = dev;
3514         fib6_nh->fib_nh_oif = dev->ifindex;
3515         err = 0;
3516 out:
3517         if (idev)
3518                 in6_dev_put(idev);
3519
3520         if (err) {
3521                 lwtstate_put(fib6_nh->fib_nh_lws);
3522                 fib6_nh->fib_nh_lws = NULL;
3523                 if (dev)
3524                         dev_put(dev);
3525         }
3526
3527         return err;
3528 }
3529
3530 void fib6_nh_release(struct fib6_nh *fib6_nh)
3531 {
3532         struct rt6_exception_bucket *bucket;
3533
3534         rcu_read_lock();
3535
3536         fib6_nh_flush_exceptions(fib6_nh, NULL);
3537         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3538         if (bucket) {
3539                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3540                 kfree(bucket);
3541         }
3542
3543         rcu_read_unlock();
3544
3545         if (fib6_nh->rt6i_pcpu) {
3546                 int cpu;
3547
3548                 for_each_possible_cpu(cpu) {
3549                         struct rt6_info **ppcpu_rt;
3550                         struct rt6_info *pcpu_rt;
3551
3552                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3553                         pcpu_rt = *ppcpu_rt;
3554                         if (pcpu_rt) {
3555                                 dst_dev_put(&pcpu_rt->dst);
3556                                 dst_release(&pcpu_rt->dst);
3557                                 *ppcpu_rt = NULL;
3558                         }
3559                 }
3560
3561                 free_percpu(fib6_nh->rt6i_pcpu);
3562         }
3563
3564         fib_nh_common_release(&fib6_nh->nh_common);
3565 }
3566
3567 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3568                                               gfp_t gfp_flags,
3569                                               struct netlink_ext_ack *extack)
3570 {
3571         struct net *net = cfg->fc_nlinfo.nl_net;
3572         struct fib6_info *rt = NULL;
3573         struct nexthop *nh = NULL;
3574         struct fib6_table *table;
3575         struct fib6_nh *fib6_nh;
3576         int err = -EINVAL;
3577         int addr_type;
3578
3579         /* RTF_PCPU is an internal flag; can not be set by userspace */
3580         if (cfg->fc_flags & RTF_PCPU) {
3581                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3582                 goto out;
3583         }
3584
3585         /* RTF_CACHE is an internal flag; can not be set by userspace */
3586         if (cfg->fc_flags & RTF_CACHE) {
3587                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3588                 goto out;
3589         }
3590
3591         if (cfg->fc_type > RTN_MAX) {
3592                 NL_SET_ERR_MSG(extack, "Invalid route type");
3593                 goto out;
3594         }
3595
3596         if (cfg->fc_dst_len > 128) {
3597                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3598                 goto out;
3599         }
3600         if (cfg->fc_src_len > 128) {
3601                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3602                 goto out;
3603         }
3604 #ifndef CONFIG_IPV6_SUBTREES
3605         if (cfg->fc_src_len) {
3606                 NL_SET_ERR_MSG(extack,
3607                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3608                 goto out;
3609         }
3610 #endif
3611         if (cfg->fc_nh_id) {
3612                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3613                 if (!nh) {
3614                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3615                         goto out;
3616                 }
3617                 err = fib6_check_nexthop(nh, cfg, extack);
3618                 if (err)
3619                         goto out;
3620         }
3621
3622         err = -ENOBUFS;
3623         if (cfg->fc_nlinfo.nlh &&
3624             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3625                 table = fib6_get_table(net, cfg->fc_table);
3626                 if (!table) {
3627                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3628                         table = fib6_new_table(net, cfg->fc_table);
3629                 }
3630         } else {
3631                 table = fib6_new_table(net, cfg->fc_table);
3632         }
3633
3634         if (!table)
3635                 goto out;
3636
3637         err = -ENOMEM;
3638         rt = fib6_info_alloc(gfp_flags, !nh);
3639         if (!rt)
3640                 goto out;
3641
3642         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3643                                                extack);
3644         if (IS_ERR(rt->fib6_metrics)) {
3645                 err = PTR_ERR(rt->fib6_metrics);
3646                 /* Do not leave garbage there. */
3647                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3648                 goto out;
3649         }
3650
3651         if (cfg->fc_flags & RTF_ADDRCONF)
3652                 rt->dst_nocount = true;
3653
3654         if (cfg->fc_flags & RTF_EXPIRES)
3655                 fib6_set_expires(rt, jiffies +
3656                                 clock_t_to_jiffies(cfg->fc_expires));
3657         else
3658                 fib6_clean_expires(rt);
3659
3660         if (cfg->fc_protocol == RTPROT_UNSPEC)
3661                 cfg->fc_protocol = RTPROT_BOOT;
3662         rt->fib6_protocol = cfg->fc_protocol;
3663
3664         rt->fib6_table = table;
3665         rt->fib6_metric = cfg->fc_metric;
3666         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3667         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3668
3669         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3670         rt->fib6_dst.plen = cfg->fc_dst_len;
3671
3672 #ifdef CONFIG_IPV6_SUBTREES
3673         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3674         rt->fib6_src.plen = cfg->fc_src_len;
3675 #endif
3676         if (nh) {
3677                 if (rt->fib6_src.plen) {
3678                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3679                         goto out;
3680                 }
3681                 if (!nexthop_get(nh)) {
3682                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3683                         goto out;
3684                 }
3685                 rt->nh = nh;
3686                 fib6_nh = nexthop_fib6_nh(rt->nh);
3687         } else {
3688                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3689                 if (err)
3690                         goto out;
3691
3692                 fib6_nh = rt->fib6_nh;
3693
3694                 /* We cannot add true routes via loopback here, they would
3695                  * result in kernel looping; promote them to reject routes
3696                  */
3697                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3698                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3699                                    addr_type))
3700                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3701         }
3702
3703         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3704                 struct net_device *dev = fib6_nh->fib_nh_dev;
3705
3706                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3707                         NL_SET_ERR_MSG(extack, "Invalid source address");
3708                         err = -EINVAL;
3709                         goto out;
3710                 }
3711                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3712                 rt->fib6_prefsrc.plen = 128;
3713         } else
3714                 rt->fib6_prefsrc.plen = 0;
3715
3716         return rt;
3717 out:
3718         fib6_info_release(rt);
3719         return ERR_PTR(err);
3720 }
3721
3722 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3723                   struct netlink_ext_ack *extack)
3724 {
3725         struct fib6_info *rt;
3726         int err;
3727
3728         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3729         if (IS_ERR(rt))
3730                 return PTR_ERR(rt);
3731
3732         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3733         fib6_info_release(rt);
3734
3735         return err;
3736 }
3737
3738 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3739 {
3740         struct net *net = info->nl_net;
3741         struct fib6_table *table;
3742         int err;
3743
3744         if (rt == net->ipv6.fib6_null_entry) {
3745                 err = -ENOENT;
3746                 goto out;
3747         }
3748
3749         table = rt->fib6_table;
3750         spin_lock_bh(&table->tb6_lock);
3751         err = fib6_del(rt, info);
3752         spin_unlock_bh(&table->tb6_lock);
3753
3754 out:
3755         fib6_info_release(rt);
3756         return err;
3757 }
3758
3759 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3760 {
3761         struct nl_info info = {
3762                 .nl_net = net,
3763                 .skip_notify = skip_notify
3764         };
3765
3766         return __ip6_del_rt(rt, &info);
3767 }
3768
3769 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3770 {
3771         struct nl_info *info = &cfg->fc_nlinfo;
3772         struct net *net = info->nl_net;
3773         struct sk_buff *skb = NULL;
3774         struct fib6_table *table;
3775         int err = -ENOENT;
3776
3777         if (rt == net->ipv6.fib6_null_entry)
3778                 goto out_put;
3779         table = rt->fib6_table;
3780         spin_lock_bh(&table->tb6_lock);
3781
3782         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3783                 struct fib6_info *sibling, *next_sibling;
3784                 struct fib6_node *fn;
3785
3786                 /* prefer to send a single notification with all hops */
3787                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3788                 if (skb) {
3789                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3790
3791                         if (rt6_fill_node(net, skb, rt, NULL,
3792                                           NULL, NULL, 0, RTM_DELROUTE,
3793                                           info->portid, seq, 0) < 0) {
3794                                 kfree_skb(skb);
3795                                 skb = NULL;
3796                         } else
3797                                 info->skip_notify = 1;
3798                 }
3799
3800                 /* 'rt' points to the first sibling route. If it is not the
3801                  * leaf, then we do not need to send a notification. Otherwise,
3802                  * we need to check if the last sibling has a next route or not
3803                  * and emit a replace or delete notification, respectively.
3804                  */
3805                 info->skip_notify_kernel = 1;
3806                 fn = rcu_dereference_protected(rt->fib6_node,
3807                                             lockdep_is_held(&table->tb6_lock));
3808                 if (rcu_access_pointer(fn->leaf) == rt) {
3809                         struct fib6_info *last_sibling, *replace_rt;
3810
3811                         last_sibling = list_last_entry(&rt->fib6_siblings,
3812                                                        struct fib6_info,
3813                                                        fib6_siblings);
3814                         replace_rt = rcu_dereference_protected(
3815                                             last_sibling->fib6_next,
3816                                             lockdep_is_held(&table->tb6_lock));
3817                         if (replace_rt)
3818                                 call_fib6_entry_notifiers_replace(net,
3819                                                                   replace_rt);
3820                         else
3821                                 call_fib6_multipath_entry_notifiers(net,
3822                                                        FIB_EVENT_ENTRY_DEL,
3823                                                        rt, rt->fib6_nsiblings,
3824                                                        NULL);
3825                 }
3826                 list_for_each_entry_safe(sibling, next_sibling,
3827                                          &rt->fib6_siblings,
3828                                          fib6_siblings) {
3829                         err = fib6_del(sibling, info);
3830                         if (err)
3831                                 goto out_unlock;
3832                 }
3833         }
3834
3835         err = fib6_del(rt, info);
3836 out_unlock:
3837         spin_unlock_bh(&table->tb6_lock);
3838 out_put:
3839         fib6_info_release(rt);
3840
3841         if (skb) {
3842                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3843                             info->nlh, gfp_any());
3844         }
3845         return err;
3846 }
3847
3848 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3849 {
3850         int rc = -ESRCH;
3851
3852         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3853                 goto out;
3854
3855         if (cfg->fc_flags & RTF_GATEWAY &&
3856             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3857                 goto out;
3858
3859         rc = rt6_remove_exception_rt(rt);
3860 out:
3861         return rc;
3862 }
3863
3864 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3865                              struct fib6_nh *nh)
3866 {
3867         struct fib6_result res = {
3868                 .f6i = rt,
3869                 .nh = nh,
3870         };
3871         struct rt6_info *rt_cache;
3872
3873         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3874         if (rt_cache)
3875                 return __ip6_del_cached_rt(rt_cache, cfg);
3876
3877         return 0;
3878 }
3879
3880 struct fib6_nh_del_cached_rt_arg {
3881         struct fib6_config *cfg;
3882         struct fib6_info *f6i;
3883 };
3884
3885 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3886 {
3887         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3888         int rc;
3889
3890         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3891         return rc != -ESRCH ? rc : 0;
3892 }
3893
3894 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3895 {
3896         struct fib6_nh_del_cached_rt_arg arg = {
3897                 .cfg = cfg,
3898                 .f6i = f6i
3899         };
3900
3901         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3902 }
3903
3904 static int ip6_route_del(struct fib6_config *cfg,
3905                          struct netlink_ext_ack *extack)
3906 {
3907         struct fib6_table *table;
3908         struct fib6_info *rt;
3909         struct fib6_node *fn;
3910         int err = -ESRCH;
3911
3912         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3913         if (!table) {
3914                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3915                 return err;
3916         }
3917
3918         rcu_read_lock();
3919
3920         fn = fib6_locate(&table->tb6_root,
3921                          &cfg->fc_dst, cfg->fc_dst_len,
3922                          &cfg->fc_src, cfg->fc_src_len,
3923                          !(cfg->fc_flags & RTF_CACHE));
3924
3925         if (fn) {
3926                 for_each_fib6_node_rt_rcu(fn) {
3927                         struct fib6_nh *nh;
3928
3929                         if (rt->nh && cfg->fc_nh_id &&
3930                             rt->nh->id != cfg->fc_nh_id)
3931                                 continue;
3932
3933                         if (cfg->fc_flags & RTF_CACHE) {
3934                                 int rc = 0;
3935
3936                                 if (rt->nh) {
3937                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3938                                 } else if (cfg->fc_nh_id) {
3939                                         continue;
3940                                 } else {
3941                                         nh = rt->fib6_nh;
3942                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3943                                 }
3944                                 if (rc != -ESRCH) {
3945                                         rcu_read_unlock();
3946                                         return rc;
3947                                 }
3948                                 continue;
3949                         }
3950
3951                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3952                                 continue;
3953                         if (cfg->fc_protocol &&
3954                             cfg->fc_protocol != rt->fib6_protocol)
3955                                 continue;
3956
3957                         if (rt->nh) {
3958                                 if (!fib6_info_hold_safe(rt))
3959                                         continue;
3960                                 rcu_read_unlock();
3961
3962                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3963                         }
3964                         if (cfg->fc_nh_id)
3965                                 continue;
3966
3967                         nh = rt->fib6_nh;
3968                         if (cfg->fc_ifindex &&
3969                             (!nh->fib_nh_dev ||
3970                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3971                                 continue;
3972                         if (cfg->fc_flags & RTF_GATEWAY &&
3973                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3974                                 continue;
3975                         if (!fib6_info_hold_safe(rt))
3976                                 continue;
3977                         rcu_read_unlock();
3978
3979                         /* if gateway was specified only delete the one hop */
3980                         if (cfg->fc_flags & RTF_GATEWAY)
3981                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3982
3983                         return __ip6_del_rt_siblings(rt, cfg);
3984                 }
3985         }
3986         rcu_read_unlock();
3987
3988         return err;
3989 }
3990
3991 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3992 {
3993         struct netevent_redirect netevent;
3994         struct rt6_info *rt, *nrt = NULL;
3995         struct fib6_result res = {};
3996         struct ndisc_options ndopts;
3997         struct inet6_dev *in6_dev;
3998         struct neighbour *neigh;
3999         struct rd_msg *msg;
4000         int optlen, on_link;
4001         u8 *lladdr;
4002
4003         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4004         optlen -= sizeof(*msg);
4005
4006         if (optlen < 0) {
4007                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4008                 return;
4009         }
4010
4011         msg = (struct rd_msg *)icmp6_hdr(skb);
4012
4013         if (ipv6_addr_is_multicast(&msg->dest)) {
4014                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4015                 return;
4016         }
4017
4018         on_link = 0;
4019         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4020                 on_link = 1;
4021         } else if (ipv6_addr_type(&msg->target) !=
4022                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4023                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4024                 return;
4025         }
4026
4027         in6_dev = __in6_dev_get(skb->dev);
4028         if (!in6_dev)
4029                 return;
4030         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4031                 return;
4032
4033         /* RFC2461 8.1:
4034          *      The IP source address of the Redirect MUST be the same as the current
4035          *      first-hop router for the specified ICMP Destination Address.
4036          */
4037
4038         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4039                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4040                 return;
4041         }
4042
4043         lladdr = NULL;
4044         if (ndopts.nd_opts_tgt_lladdr) {
4045                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4046                                              skb->dev);
4047                 if (!lladdr) {
4048                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4049                         return;
4050                 }
4051         }
4052
4053         rt = (struct rt6_info *) dst;
4054         if (rt->rt6i_flags & RTF_REJECT) {
4055                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4056                 return;
4057         }
4058
4059         /* Redirect received -> path was valid.
4060          * Look, redirects are sent only in response to data packets,
4061          * so that this nexthop apparently is reachable. --ANK
4062          */
4063         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4064
4065         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4066         if (!neigh)
4067                 return;
4068
4069         /*
4070          *      We have finally decided to accept it.
4071          */
4072
4073         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4074                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4075                      NEIGH_UPDATE_F_OVERRIDE|
4076                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4077                                      NEIGH_UPDATE_F_ISROUTER)),
4078                      NDISC_REDIRECT, &ndopts);
4079
4080         rcu_read_lock();
4081         res.f6i = rcu_dereference(rt->from);
4082         if (!res.f6i)
4083                 goto out;
4084
4085         if (res.f6i->nh) {
4086                 struct fib6_nh_match_arg arg = {
4087                         .dev = dst->dev,
4088                         .gw = &rt->rt6i_gateway,
4089                 };
4090
4091                 nexthop_for_each_fib6_nh(res.f6i->nh,
4092                                          fib6_nh_find_match, &arg);
4093
4094                 /* fib6_info uses a nexthop that does not have fib6_nh
4095                  * using the dst->dev. Should be impossible
4096                  */
4097                 if (!arg.match)
4098                         goto out;
4099                 res.nh = arg.match;
4100         } else {
4101                 res.nh = res.f6i->fib6_nh;
4102         }
4103
4104         res.fib6_flags = res.f6i->fib6_flags;
4105         res.fib6_type = res.f6i->fib6_type;
4106         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4107         if (!nrt)
4108                 goto out;
4109
4110         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4111         if (on_link)
4112                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4113
4114         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4115
4116         /* rt6_insert_exception() will take care of duplicated exceptions */
4117         if (rt6_insert_exception(nrt, &res)) {
4118                 dst_release_immediate(&nrt->dst);
4119                 goto out;
4120         }
4121
4122         netevent.old = &rt->dst;
4123         netevent.new = &nrt->dst;
4124         netevent.daddr = &msg->dest;
4125         netevent.neigh = neigh;
4126         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4127
4128 out:
4129         rcu_read_unlock();
4130         neigh_release(neigh);
4131 }
4132
4133 #ifdef CONFIG_IPV6_ROUTE_INFO
4134 static struct fib6_info *rt6_get_route_info(struct net *net,
4135                                            const struct in6_addr *prefix, int prefixlen,
4136                                            const struct in6_addr *gwaddr,
4137                                            struct net_device *dev)
4138 {
4139         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4140         int ifindex = dev->ifindex;
4141         struct fib6_node *fn;
4142         struct fib6_info *rt = NULL;
4143         struct fib6_table *table;
4144
4145         table = fib6_get_table(net, tb_id);
4146         if (!table)
4147                 return NULL;
4148
4149         rcu_read_lock();
4150         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4151         if (!fn)
4152                 goto out;
4153
4154         for_each_fib6_node_rt_rcu(fn) {
4155                 /* these routes do not use nexthops */
4156                 if (rt->nh)
4157                         continue;
4158                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4159                         continue;
4160                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4161                     !rt->fib6_nh->fib_nh_gw_family)
4162                         continue;
4163                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4164                         continue;
4165                 if (!fib6_info_hold_safe(rt))
4166                         continue;
4167                 break;
4168         }
4169 out:
4170         rcu_read_unlock();
4171         return rt;
4172 }
4173
4174 static struct fib6_info *rt6_add_route_info(struct net *net,
4175                                            const struct in6_addr *prefix, int prefixlen,
4176                                            const struct in6_addr *gwaddr,
4177                                            struct net_device *dev,
4178                                            unsigned int pref)
4179 {
4180         struct fib6_config cfg = {
4181                 .fc_metric      = IP6_RT_PRIO_USER,
4182                 .fc_ifindex     = dev->ifindex,
4183                 .fc_dst_len     = prefixlen,
4184                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4185                                   RTF_UP | RTF_PREF(pref),
4186                 .fc_protocol = RTPROT_RA,
4187                 .fc_type = RTN_UNICAST,
4188                 .fc_nlinfo.portid = 0,
4189                 .fc_nlinfo.nlh = NULL,
4190                 .fc_nlinfo.nl_net = net,
4191         };
4192
4193         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4194         cfg.fc_dst = *prefix;
4195         cfg.fc_gateway = *gwaddr;
4196
4197         /* We should treat it as a default route if prefix length is 0. */
4198         if (!prefixlen)
4199                 cfg.fc_flags |= RTF_DEFAULT;
4200
4201         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4202
4203         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4204 }
4205 #endif
4206
4207 struct fib6_info *rt6_get_dflt_router(struct net *net,
4208                                      const struct in6_addr *addr,
4209                                      struct net_device *dev)
4210 {
4211         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4212         struct fib6_info *rt;
4213         struct fib6_table *table;
4214
4215         table = fib6_get_table(net, tb_id);
4216         if (!table)
4217                 return NULL;
4218
4219         rcu_read_lock();
4220         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4221                 struct fib6_nh *nh;
4222
4223                 /* RA routes do not use nexthops */
4224                 if (rt->nh)
4225                         continue;
4226
4227                 nh = rt->fib6_nh;
4228                 if (dev == nh->fib_nh_dev &&
4229                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4230                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4231                         break;
4232         }
4233         if (rt && !fib6_info_hold_safe(rt))
4234                 rt = NULL;
4235         rcu_read_unlock();
4236         return rt;
4237 }
4238
4239 struct fib6_info *rt6_add_dflt_router(struct net *net,
4240                                      const struct in6_addr *gwaddr,
4241                                      struct net_device *dev,
4242                                      unsigned int pref,
4243                                      u32 defrtr_usr_metric)
4244 {
4245         struct fib6_config cfg = {
4246                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4247                 .fc_metric      = defrtr_usr_metric,
4248                 .fc_ifindex     = dev->ifindex,
4249                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4250                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4251                 .fc_protocol = RTPROT_RA,
4252                 .fc_type = RTN_UNICAST,
4253                 .fc_nlinfo.portid = 0,
4254                 .fc_nlinfo.nlh = NULL,
4255                 .fc_nlinfo.nl_net = net,
4256         };
4257
4258         cfg.fc_gateway = *gwaddr;
4259
4260         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4261                 struct fib6_table *table;
4262
4263                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4264                 if (table)
4265                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4266         }
4267
4268         return rt6_get_dflt_router(net, gwaddr, dev);
4269 }
4270
4271 static void __rt6_purge_dflt_routers(struct net *net,
4272                                      struct fib6_table *table)
4273 {
4274         struct fib6_info *rt;
4275
4276 restart:
4277         rcu_read_lock();
4278         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4279                 struct net_device *dev = fib6_info_nh_dev(rt);
4280                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4281
4282                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4283                     (!idev || idev->cnf.accept_ra != 2) &&
4284                     fib6_info_hold_safe(rt)) {
4285                         rcu_read_unlock();
4286                         ip6_del_rt(net, rt, false);
4287                         goto restart;
4288                 }
4289         }
4290         rcu_read_unlock();
4291
4292         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4293 }
4294
4295 void rt6_purge_dflt_routers(struct net *net)
4296 {
4297         struct fib6_table *table;
4298         struct hlist_head *head;
4299         unsigned int h;
4300
4301         rcu_read_lock();
4302
4303         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4304                 head = &net->ipv6.fib_table_hash[h];
4305                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4306                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4307                                 __rt6_purge_dflt_routers(net, table);
4308                 }
4309         }
4310
4311         rcu_read_unlock();
4312 }
4313
4314 static void rtmsg_to_fib6_config(struct net *net,
4315                                  struct in6_rtmsg *rtmsg,
4316                                  struct fib6_config *cfg)
4317 {
4318         *cfg = (struct fib6_config){
4319                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4320                          : RT6_TABLE_MAIN,
4321                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4322                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4323                 .fc_expires = rtmsg->rtmsg_info,
4324                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4325                 .fc_src_len = rtmsg->rtmsg_src_len,
4326                 .fc_flags = rtmsg->rtmsg_flags,
4327                 .fc_type = rtmsg->rtmsg_type,
4328
4329                 .fc_nlinfo.nl_net = net,
4330
4331                 .fc_dst = rtmsg->rtmsg_dst,
4332                 .fc_src = rtmsg->rtmsg_src,
4333                 .fc_gateway = rtmsg->rtmsg_gateway,
4334         };
4335 }
4336
4337 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4338 {
4339         struct fib6_config cfg;
4340         int err;
4341
4342         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4343                 return -EINVAL;
4344         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4345                 return -EPERM;
4346
4347         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4348
4349         rtnl_lock();
4350         switch (cmd) {
4351         case SIOCADDRT:
4352                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4353                 break;
4354         case SIOCDELRT:
4355                 err = ip6_route_del(&cfg, NULL);
4356                 break;
4357         }
4358         rtnl_unlock();
4359         return err;
4360 }
4361
4362 /*
4363  *      Drop the packet on the floor
4364  */
4365
4366 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4367 {
4368         struct dst_entry *dst = skb_dst(skb);
4369         struct net *net = dev_net(dst->dev);
4370         struct inet6_dev *idev;
4371         int type;
4372
4373         if (netif_is_l3_master(skb->dev) &&
4374             dst->dev == net->loopback_dev)
4375                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4376         else
4377                 idev = ip6_dst_idev(dst);
4378
4379         switch (ipstats_mib_noroutes) {
4380         case IPSTATS_MIB_INNOROUTES:
4381                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4382                 if (type == IPV6_ADDR_ANY) {
4383                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4384                         break;
4385                 }
4386                 fallthrough;
4387         case IPSTATS_MIB_OUTNOROUTES:
4388                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4389                 break;
4390         }
4391
4392         /* Start over by dropping the dst for l3mdev case */
4393         if (netif_is_l3_master(skb->dev))
4394                 skb_dst_drop(skb);
4395
4396         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4397         kfree_skb(skb);
4398         return 0;
4399 }
4400
4401 static int ip6_pkt_discard(struct sk_buff *skb)
4402 {
4403         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4404 }
4405
4406 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4407 {
4408         skb->dev = skb_dst(skb)->dev;
4409         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4410 }
4411
4412 static int ip6_pkt_prohibit(struct sk_buff *skb)
4413 {
4414         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4415 }
4416
4417 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4418 {
4419         skb->dev = skb_dst(skb)->dev;
4420         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4421 }
4422
4423 /*
4424  *      Allocate a dst for local (unicast / anycast) address.
4425  */
4426
4427 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4428                                      struct inet6_dev *idev,
4429                                      const struct in6_addr *addr,
4430                                      bool anycast, gfp_t gfp_flags)
4431 {
4432         struct fib6_config cfg = {
4433                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4434                 .fc_ifindex = idev->dev->ifindex,
4435                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4436                 .fc_dst = *addr,
4437                 .fc_dst_len = 128,
4438                 .fc_protocol = RTPROT_KERNEL,
4439                 .fc_nlinfo.nl_net = net,
4440                 .fc_ignore_dev_down = true,
4441         };
4442         struct fib6_info *f6i;
4443
4444         if (anycast) {
4445                 cfg.fc_type = RTN_ANYCAST;
4446                 cfg.fc_flags |= RTF_ANYCAST;
4447         } else {
4448                 cfg.fc_type = RTN_LOCAL;
4449                 cfg.fc_flags |= RTF_LOCAL;
4450         }
4451
4452         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4453         if (!IS_ERR(f6i))
4454                 f6i->dst_nocount = true;
4455         return f6i;
4456 }
4457
4458 /* remove deleted ip from prefsrc entries */
4459 struct arg_dev_net_ip {
4460         struct net_device *dev;
4461         struct net *net;
4462         struct in6_addr *addr;
4463 };
4464
4465 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4466 {
4467         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4468         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4469         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4470
4471         if (!rt->nh &&
4472             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4473             rt != net->ipv6.fib6_null_entry &&
4474             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4475                 spin_lock_bh(&rt6_exception_lock);
4476                 /* remove prefsrc entry */
4477                 rt->fib6_prefsrc.plen = 0;
4478                 spin_unlock_bh(&rt6_exception_lock);
4479         }
4480         return 0;
4481 }
4482
4483 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4484 {
4485         struct net *net = dev_net(ifp->idev->dev);
4486         struct arg_dev_net_ip adni = {
4487                 .dev = ifp->idev->dev,
4488                 .net = net,
4489                 .addr = &ifp->addr,
4490         };
4491         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4492 }
4493
4494 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4495
4496 /* Remove routers and update dst entries when gateway turn into host. */
4497 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4498 {
4499         struct in6_addr *gateway = (struct in6_addr *)arg;
4500         struct fib6_nh *nh;
4501
4502         /* RA routes do not use nexthops */
4503         if (rt->nh)
4504                 return 0;
4505
4506         nh = rt->fib6_nh;
4507         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4508             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4509                 return -1;
4510
4511         /* Further clean up cached routes in exception table.
4512          * This is needed because cached route may have a different
4513          * gateway than its 'parent' in the case of an ip redirect.
4514          */
4515         fib6_nh_exceptions_clean_tohost(nh, gateway);
4516
4517         return 0;
4518 }
4519
4520 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4521 {
4522         fib6_clean_all(net, fib6_clean_tohost, gateway);
4523 }
4524
4525 struct arg_netdev_event {
4526         const struct net_device *dev;
4527         union {
4528                 unsigned char nh_flags;
4529                 unsigned long event;
4530         };
4531 };
4532
4533 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4534 {
4535         struct fib6_info *iter;
4536         struct fib6_node *fn;
4537
4538         fn = rcu_dereference_protected(rt->fib6_node,
4539                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4540         iter = rcu_dereference_protected(fn->leaf,
4541                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4542         while (iter) {
4543                 if (iter->fib6_metric == rt->fib6_metric &&
4544                     rt6_qualify_for_ecmp(iter))
4545                         return iter;
4546                 iter = rcu_dereference_protected(iter->fib6_next,
4547                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4548         }
4549
4550         return NULL;
4551 }
4552
4553 /* only called for fib entries with builtin fib6_nh */
4554 static bool rt6_is_dead(const struct fib6_info *rt)
4555 {
4556         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4557             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4558              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4559                 return true;
4560
4561         return false;
4562 }
4563
4564 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4565 {
4566         struct fib6_info *iter;
4567         int total = 0;
4568
4569         if (!rt6_is_dead(rt))
4570                 total += rt->fib6_nh->fib_nh_weight;
4571
4572         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4573                 if (!rt6_is_dead(iter))
4574                         total += iter->fib6_nh->fib_nh_weight;
4575         }
4576
4577         return total;
4578 }
4579
4580 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4581 {
4582         int upper_bound = -1;
4583
4584         if (!rt6_is_dead(rt)) {
4585                 *weight += rt->fib6_nh->fib_nh_weight;
4586                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4587                                                     total) - 1;
4588         }
4589         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4590 }
4591
4592 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4593 {
4594         struct fib6_info *iter;
4595         int weight = 0;
4596
4597         rt6_upper_bound_set(rt, &weight, total);
4598
4599         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4600                 rt6_upper_bound_set(iter, &weight, total);
4601 }
4602
4603 void rt6_multipath_rebalance(struct fib6_info *rt)
4604 {
4605         struct fib6_info *first;
4606         int total;
4607
4608         /* In case the entire multipath route was marked for flushing,
4609          * then there is no need to rebalance upon the removal of every
4610          * sibling route.
4611          */
4612         if (!rt->fib6_nsiblings || rt->should_flush)
4613                 return;
4614
4615         /* During lookup routes are evaluated in order, so we need to
4616          * make sure upper bounds are assigned from the first sibling
4617          * onwards.
4618          */
4619         first = rt6_multipath_first_sibling(rt);
4620         if (WARN_ON_ONCE(!first))
4621                 return;
4622
4623         total = rt6_multipath_total_weight(first);
4624         rt6_multipath_upper_bound_set(first, total);
4625 }
4626
4627 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4628 {
4629         const struct arg_netdev_event *arg = p_arg;
4630         struct net *net = dev_net(arg->dev);
4631
4632         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4633             rt->fib6_nh->fib_nh_dev == arg->dev) {
4634                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4635                 fib6_update_sernum_upto_root(net, rt);
4636                 rt6_multipath_rebalance(rt);
4637         }
4638
4639         return 0;
4640 }
4641
4642 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4643 {
4644         struct arg_netdev_event arg = {
4645                 .dev = dev,
4646                 {
4647                         .nh_flags = nh_flags,
4648                 },
4649         };
4650
4651         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4652                 arg.nh_flags |= RTNH_F_LINKDOWN;
4653
4654         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4655 }
4656
4657 /* only called for fib entries with inline fib6_nh */
4658 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4659                                    const struct net_device *dev)
4660 {
4661         struct fib6_info *iter;
4662
4663         if (rt->fib6_nh->fib_nh_dev == dev)
4664                 return true;
4665         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4666                 if (iter->fib6_nh->fib_nh_dev == dev)
4667                         return true;
4668
4669         return false;
4670 }
4671
4672 static void rt6_multipath_flush(struct fib6_info *rt)
4673 {
4674         struct fib6_info *iter;
4675
4676         rt->should_flush = 1;
4677         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4678                 iter->should_flush = 1;
4679 }
4680
4681 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4682                                              const struct net_device *down_dev)
4683 {
4684         struct fib6_info *iter;
4685         unsigned int dead = 0;
4686
4687         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4688             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4689                 dead++;
4690         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4691                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4692                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4693                         dead++;
4694
4695         return dead;
4696 }
4697
4698 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4699                                        const struct net_device *dev,
4700                                        unsigned char nh_flags)
4701 {
4702         struct fib6_info *iter;
4703
4704         if (rt->fib6_nh->fib_nh_dev == dev)
4705                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4706         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4707                 if (iter->fib6_nh->fib_nh_dev == dev)
4708                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4709 }
4710
4711 /* called with write lock held for table with rt */
4712 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4713 {
4714         const struct arg_netdev_event *arg = p_arg;
4715         const struct net_device *dev = arg->dev;
4716         struct net *net = dev_net(dev);
4717
4718         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4719                 return 0;
4720
4721         switch (arg->event) {
4722         case NETDEV_UNREGISTER:
4723                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4724         case NETDEV_DOWN:
4725                 if (rt->should_flush)
4726                         return -1;
4727                 if (!rt->fib6_nsiblings)
4728                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4729                 if (rt6_multipath_uses_dev(rt, dev)) {
4730                         unsigned int count;
4731
4732                         count = rt6_multipath_dead_count(rt, dev);
4733                         if (rt->fib6_nsiblings + 1 == count) {
4734                                 rt6_multipath_flush(rt);
4735                                 return -1;
4736                         }
4737                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4738                                                    RTNH_F_LINKDOWN);
4739                         fib6_update_sernum(net, rt);
4740                         rt6_multipath_rebalance(rt);
4741                 }
4742                 return -2;
4743         case NETDEV_CHANGE:
4744                 if (rt->fib6_nh->fib_nh_dev != dev ||
4745                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4746                         break;
4747                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4748                 rt6_multipath_rebalance(rt);
4749                 break;
4750         }
4751
4752         return 0;
4753 }
4754
4755 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4756 {
4757         struct arg_netdev_event arg = {
4758                 .dev = dev,
4759                 {
4760                         .event = event,
4761                 },
4762         };
4763         struct net *net = dev_net(dev);
4764
4765         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4766                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4767         else
4768                 fib6_clean_all(net, fib6_ifdown, &arg);
4769 }
4770
4771 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4772 {
4773         rt6_sync_down_dev(dev, event);
4774         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4775         neigh_ifdown(&nd_tbl, dev);
4776 }
4777
4778 struct rt6_mtu_change_arg {
4779         struct net_device *dev;
4780         unsigned int mtu;
4781         struct fib6_info *f6i;
4782 };
4783
4784 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4785 {
4786         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4787         struct fib6_info *f6i = arg->f6i;
4788
4789         /* For administrative MTU increase, there is no way to discover
4790          * IPv6 PMTU increase, so PMTU increase should be updated here.
4791          * Since RFC 1981 doesn't include administrative MTU increase
4792          * update PMTU increase is a MUST. (i.e. jumbo frame)
4793          */
4794         if (nh->fib_nh_dev == arg->dev) {
4795                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4796                 u32 mtu = f6i->fib6_pmtu;
4797
4798                 if (mtu >= arg->mtu ||
4799                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4800                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4801
4802                 spin_lock_bh(&rt6_exception_lock);
4803                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4804                 spin_unlock_bh(&rt6_exception_lock);
4805         }
4806
4807         return 0;
4808 }
4809
4810 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4811 {
4812         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4813         struct inet6_dev *idev;
4814
4815         /* In IPv6 pmtu discovery is not optional,
4816            so that RTAX_MTU lock cannot disable it.
4817            We still use this lock to block changes
4818            caused by addrconf/ndisc.
4819         */
4820
4821         idev = __in6_dev_get(arg->dev);
4822         if (!idev)
4823                 return 0;
4824
4825         if (fib6_metric_locked(f6i, RTAX_MTU))
4826                 return 0;
4827
4828         arg->f6i = f6i;
4829         if (f6i->nh) {
4830                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4831                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4832                                                 arg);
4833         }
4834
4835         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4836 }
4837
4838 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4839 {
4840         struct rt6_mtu_change_arg arg = {
4841                 .dev = dev,
4842                 .mtu = mtu,
4843         };
4844
4845         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4846 }
4847
4848 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4849         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4850         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4851         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4852         [RTA_OIF]               = { .type = NLA_U32 },
4853         [RTA_IIF]               = { .type = NLA_U32 },
4854         [RTA_PRIORITY]          = { .type = NLA_U32 },
4855         [RTA_METRICS]           = { .type = NLA_NESTED },
4856         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4857         [RTA_PREF]              = { .type = NLA_U8 },
4858         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4859         [RTA_ENCAP]             = { .type = NLA_NESTED },
4860         [RTA_EXPIRES]           = { .type = NLA_U32 },
4861         [RTA_UID]               = { .type = NLA_U32 },
4862         [RTA_MARK]              = { .type = NLA_U32 },
4863         [RTA_TABLE]             = { .type = NLA_U32 },
4864         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4865         [RTA_SPORT]             = { .type = NLA_U16 },
4866         [RTA_DPORT]             = { .type = NLA_U16 },
4867         [RTA_NH_ID]             = { .type = NLA_U32 },
4868 };
4869
4870 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4871                               struct fib6_config *cfg,
4872                               struct netlink_ext_ack *extack)
4873 {
4874         struct rtmsg *rtm;
4875         struct nlattr *tb[RTA_MAX+1];
4876         unsigned int pref;
4877         int err;
4878
4879         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4880                                      rtm_ipv6_policy, extack);
4881         if (err < 0)
4882                 goto errout;
4883
4884         err = -EINVAL;
4885         rtm = nlmsg_data(nlh);
4886
4887         *cfg = (struct fib6_config){
4888                 .fc_table = rtm->rtm_table,
4889                 .fc_dst_len = rtm->rtm_dst_len,
4890                 .fc_src_len = rtm->rtm_src_len,
4891                 .fc_flags = RTF_UP,
4892                 .fc_protocol = rtm->rtm_protocol,
4893                 .fc_type = rtm->rtm_type,
4894
4895                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4896                 .fc_nlinfo.nlh = nlh,
4897                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4898         };
4899
4900         if (rtm->rtm_type == RTN_UNREACHABLE ||
4901             rtm->rtm_type == RTN_BLACKHOLE ||
4902             rtm->rtm_type == RTN_PROHIBIT ||
4903             rtm->rtm_type == RTN_THROW)
4904                 cfg->fc_flags |= RTF_REJECT;
4905
4906         if (rtm->rtm_type == RTN_LOCAL)
4907                 cfg->fc_flags |= RTF_LOCAL;
4908
4909         if (rtm->rtm_flags & RTM_F_CLONED)
4910                 cfg->fc_flags |= RTF_CACHE;
4911
4912         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4913
4914         if (tb[RTA_NH_ID]) {
4915                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4916                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4917                         NL_SET_ERR_MSG(extack,
4918                                        "Nexthop specification and nexthop id are mutually exclusive");
4919                         goto errout;
4920                 }
4921                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4922         }
4923
4924         if (tb[RTA_GATEWAY]) {
4925                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4926                 cfg->fc_flags |= RTF_GATEWAY;
4927         }
4928         if (tb[RTA_VIA]) {
4929                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4930                 goto errout;
4931         }
4932
4933         if (tb[RTA_DST]) {
4934                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4935
4936                 if (nla_len(tb[RTA_DST]) < plen)
4937                         goto errout;
4938
4939                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4940         }
4941
4942         if (tb[RTA_SRC]) {
4943                 int plen = (rtm->rtm_src_len + 7) >> 3;
4944
4945                 if (nla_len(tb[RTA_SRC]) < plen)
4946                         goto errout;
4947
4948                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4949         }
4950
4951         if (tb[RTA_PREFSRC])
4952                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4953
4954         if (tb[RTA_OIF])
4955                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4956
4957         if (tb[RTA_PRIORITY])
4958                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4959
4960         if (tb[RTA_METRICS]) {
4961                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4962                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4963         }
4964
4965         if (tb[RTA_TABLE])
4966                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4967
4968         if (tb[RTA_MULTIPATH]) {
4969                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4970                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4971
4972                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4973                                                      cfg->fc_mp_len, extack);
4974                 if (err < 0)
4975                         goto errout;
4976         }
4977
4978         if (tb[RTA_PREF]) {
4979                 pref = nla_get_u8(tb[RTA_PREF]);
4980                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4981                     pref != ICMPV6_ROUTER_PREF_HIGH)
4982                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4983                 cfg->fc_flags |= RTF_PREF(pref);
4984         }
4985
4986         if (tb[RTA_ENCAP])
4987                 cfg->fc_encap = tb[RTA_ENCAP];
4988
4989         if (tb[RTA_ENCAP_TYPE]) {
4990                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4991
4992                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4993                 if (err < 0)
4994                         goto errout;
4995         }
4996
4997         if (tb[RTA_EXPIRES]) {
4998                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4999
5000                 if (addrconf_finite_timeout(timeout)) {
5001                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5002                         cfg->fc_flags |= RTF_EXPIRES;
5003                 }
5004         }
5005
5006         err = 0;
5007 errout:
5008         return err;
5009 }
5010
5011 struct rt6_nh {
5012         struct fib6_info *fib6_info;
5013         struct fib6_config r_cfg;
5014         struct list_head next;
5015 };
5016
5017 static int ip6_route_info_append(struct net *net,
5018                                  struct list_head *rt6_nh_list,
5019                                  struct fib6_info *rt,
5020                                  struct fib6_config *r_cfg)
5021 {
5022         struct rt6_nh *nh;
5023         int err = -EEXIST;
5024
5025         list_for_each_entry(nh, rt6_nh_list, next) {
5026                 /* check if fib6_info already exists */
5027                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5028                         return err;
5029         }
5030
5031         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5032         if (!nh)
5033                 return -ENOMEM;
5034         nh->fib6_info = rt;
5035         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5036         list_add_tail(&nh->next, rt6_nh_list);
5037
5038         return 0;
5039 }
5040
5041 static void ip6_route_mpath_notify(struct fib6_info *rt,
5042                                    struct fib6_info *rt_last,
5043                                    struct nl_info *info,
5044                                    __u16 nlflags)
5045 {
5046         /* if this is an APPEND route, then rt points to the first route
5047          * inserted and rt_last points to last route inserted. Userspace
5048          * wants a consistent dump of the route which starts at the first
5049          * nexthop. Since sibling routes are always added at the end of
5050          * the list, find the first sibling of the last route appended
5051          */
5052         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5053                 rt = list_first_entry(&rt_last->fib6_siblings,
5054                                       struct fib6_info,
5055                                       fib6_siblings);
5056         }
5057
5058         if (rt)
5059                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5060 }
5061
5062 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5063 {
5064         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5065         bool should_notify = false;
5066         struct fib6_info *leaf;
5067         struct fib6_node *fn;
5068
5069         rcu_read_lock();
5070         fn = rcu_dereference(rt->fib6_node);
5071         if (!fn)
5072                 goto out;
5073
5074         leaf = rcu_dereference(fn->leaf);
5075         if (!leaf)
5076                 goto out;
5077
5078         if (rt == leaf ||
5079             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5080              rt6_qualify_for_ecmp(leaf)))
5081                 should_notify = true;
5082 out:
5083         rcu_read_unlock();
5084
5085         return should_notify;
5086 }
5087
5088 static int ip6_route_multipath_add(struct fib6_config *cfg,
5089                                    struct netlink_ext_ack *extack)
5090 {
5091         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5092         struct nl_info *info = &cfg->fc_nlinfo;
5093         struct fib6_config r_cfg;
5094         struct rtnexthop *rtnh;
5095         struct fib6_info *rt;
5096         struct rt6_nh *err_nh;
5097         struct rt6_nh *nh, *nh_safe;
5098         __u16 nlflags;
5099         int remaining;
5100         int attrlen;
5101         int err = 1;
5102         int nhn = 0;
5103         int replace = (cfg->fc_nlinfo.nlh &&
5104                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5105         LIST_HEAD(rt6_nh_list);
5106
5107         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5108         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5109                 nlflags |= NLM_F_APPEND;
5110
5111         remaining = cfg->fc_mp_len;
5112         rtnh = (struct rtnexthop *)cfg->fc_mp;
5113
5114         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5115          * fib6_info structs per nexthop
5116          */
5117         while (rtnh_ok(rtnh, remaining)) {
5118                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5119                 if (rtnh->rtnh_ifindex)
5120                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5121
5122                 attrlen = rtnh_attrlen(rtnh);
5123                 if (attrlen > 0) {
5124                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5125
5126                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5127                         if (nla) {
5128                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5129                                 r_cfg.fc_flags |= RTF_GATEWAY;
5130                         }
5131                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5132                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5133                         if (nla)
5134                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5135                 }
5136
5137                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5138                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5139                 if (IS_ERR(rt)) {
5140                         err = PTR_ERR(rt);
5141                         rt = NULL;
5142                         goto cleanup;
5143                 }
5144                 if (!rt6_qualify_for_ecmp(rt)) {
5145                         err = -EINVAL;
5146                         NL_SET_ERR_MSG(extack,
5147                                        "Device only routes can not be added for IPv6 using the multipath API.");
5148                         fib6_info_release(rt);
5149                         goto cleanup;
5150                 }
5151
5152                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5153
5154                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5155                                             rt, &r_cfg);
5156                 if (err) {
5157                         fib6_info_release(rt);
5158                         goto cleanup;
5159                 }
5160
5161                 rtnh = rtnh_next(rtnh, &remaining);
5162         }
5163
5164         if (list_empty(&rt6_nh_list)) {
5165                 NL_SET_ERR_MSG(extack,
5166                                "Invalid nexthop configuration - no valid nexthops");
5167                 return -EINVAL;
5168         }
5169
5170         /* for add and replace send one notification with all nexthops.
5171          * Skip the notification in fib6_add_rt2node and send one with
5172          * the full route when done
5173          */
5174         info->skip_notify = 1;
5175
5176         /* For add and replace, send one notification with all nexthops. For
5177          * append, send one notification with all appended nexthops.
5178          */
5179         info->skip_notify_kernel = 1;
5180
5181         err_nh = NULL;
5182         list_for_each_entry(nh, &rt6_nh_list, next) {
5183                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5184                 fib6_info_release(nh->fib6_info);
5185
5186                 if (!err) {
5187                         /* save reference to last route successfully inserted */
5188                         rt_last = nh->fib6_info;
5189
5190                         /* save reference to first route for notification */
5191                         if (!rt_notif)
5192                                 rt_notif = nh->fib6_info;
5193                 }
5194
5195                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5196                 nh->fib6_info = NULL;
5197                 if (err) {
5198                         if (replace && nhn)
5199                                 NL_SET_ERR_MSG_MOD(extack,
5200                                                    "multipath route replace failed (check consistency of installed routes)");
5201                         err_nh = nh;
5202                         goto add_errout;
5203                 }
5204
5205                 /* Because each route is added like a single route we remove
5206                  * these flags after the first nexthop: if there is a collision,
5207                  * we have already failed to add the first nexthop:
5208                  * fib6_add_rt2node() has rejected it; when replacing, old
5209                  * nexthops have been replaced by first new, the rest should
5210                  * be added to it.
5211                  */
5212                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5213                                                      NLM_F_REPLACE);
5214                 cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5215                 nhn++;
5216         }
5217
5218         /* An in-kernel notification should only be sent in case the new
5219          * multipath route is added as the first route in the node, or if
5220          * it was appended to it. We pass 'rt_notif' since it is the first
5221          * sibling and might allow us to skip some checks in the replace case.
5222          */
5223         if (ip6_route_mpath_should_notify(rt_notif)) {
5224                 enum fib_event_type fib_event;
5225
5226                 if (rt_notif->fib6_nsiblings != nhn - 1)
5227                         fib_event = FIB_EVENT_ENTRY_APPEND;
5228                 else
5229                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5230
5231                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5232                                                           fib_event, rt_notif,
5233                                                           nhn - 1, extack);
5234                 if (err) {
5235                         /* Delete all the siblings that were just added */
5236                         err_nh = NULL;
5237                         goto add_errout;
5238                 }
5239         }
5240
5241         /* success ... tell user about new route */
5242         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5243         goto cleanup;
5244
5245 add_errout:
5246         /* send notification for routes that were added so that
5247          * the delete notifications sent by ip6_route_del are
5248          * coherent
5249          */
5250         if (rt_notif)
5251                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5252
5253         /* Delete routes that were already added */
5254         list_for_each_entry(nh, &rt6_nh_list, next) {
5255                 if (err_nh == nh)
5256                         break;
5257                 ip6_route_del(&nh->r_cfg, extack);
5258         }
5259
5260 cleanup:
5261         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5262                 if (nh->fib6_info)
5263                         fib6_info_release(nh->fib6_info);
5264                 list_del(&nh->next);
5265                 kfree(nh);
5266         }
5267
5268         return err;
5269 }
5270
5271 static int ip6_route_multipath_del(struct fib6_config *cfg,
5272                                    struct netlink_ext_ack *extack)
5273 {
5274         struct fib6_config r_cfg;
5275         struct rtnexthop *rtnh;
5276         int last_err = 0;
5277         int remaining;
5278         int attrlen;
5279         int err;
5280
5281         remaining = cfg->fc_mp_len;
5282         rtnh = (struct rtnexthop *)cfg->fc_mp;
5283
5284         /* Parse a Multipath Entry */
5285         while (rtnh_ok(rtnh, remaining)) {
5286                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5287                 if (rtnh->rtnh_ifindex)
5288                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5289
5290                 attrlen = rtnh_attrlen(rtnh);
5291                 if (attrlen > 0) {
5292                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5293
5294                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5295                         if (nla) {
5296                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5297                                 r_cfg.fc_flags |= RTF_GATEWAY;
5298                         }
5299                 }
5300                 err = ip6_route_del(&r_cfg, extack);
5301                 if (err)
5302                         last_err = err;
5303
5304                 rtnh = rtnh_next(rtnh, &remaining);
5305         }
5306
5307         return last_err;
5308 }
5309
5310 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5311                               struct netlink_ext_ack *extack)
5312 {
5313         struct fib6_config cfg;
5314         int err;
5315
5316         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5317         if (err < 0)
5318                 return err;
5319
5320         if (cfg.fc_nh_id &&
5321             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5322                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5323                 return -EINVAL;
5324         }
5325
5326         if (cfg.fc_mp)
5327                 return ip6_route_multipath_del(&cfg, extack);
5328         else {
5329                 cfg.fc_delete_all_nh = 1;
5330                 return ip6_route_del(&cfg, extack);
5331         }
5332 }
5333
5334 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5335                               struct netlink_ext_ack *extack)
5336 {
5337         struct fib6_config cfg;
5338         int err;
5339
5340         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5341         if (err < 0)
5342                 return err;
5343
5344         if (cfg.fc_metric == 0)
5345                 cfg.fc_metric = IP6_RT_PRIO_USER;
5346
5347         if (cfg.fc_mp)
5348                 return ip6_route_multipath_add(&cfg, extack);
5349         else
5350                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5351 }
5352
5353 /* add the overhead of this fib6_nh to nexthop_len */
5354 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5355 {
5356         int *nexthop_len = arg;
5357
5358         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5359                      + NLA_ALIGN(sizeof(struct rtnexthop))
5360                      + nla_total_size(16); /* RTA_GATEWAY */
5361
5362         if (nh->fib_nh_lws) {
5363                 /* RTA_ENCAP_TYPE */
5364                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5365                 /* RTA_ENCAP */
5366                 *nexthop_len += nla_total_size(2);
5367         }
5368
5369         return 0;
5370 }
5371
5372 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5373 {
5374         int nexthop_len;
5375
5376         if (f6i->nh) {
5377                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5378                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5379                                          &nexthop_len);
5380         } else {
5381                 struct fib6_nh *nh = f6i->fib6_nh;
5382
5383                 nexthop_len = 0;
5384                 if (f6i->fib6_nsiblings) {
5385                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5386                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5387                                     + nla_total_size(16) /* RTA_GATEWAY */
5388                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5389
5390                         nexthop_len *= f6i->fib6_nsiblings;
5391                 }
5392                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5393         }
5394
5395         return NLMSG_ALIGN(sizeof(struct rtmsg))
5396                + nla_total_size(16) /* RTA_SRC */
5397                + nla_total_size(16) /* RTA_DST */
5398                + nla_total_size(16) /* RTA_GATEWAY */
5399                + nla_total_size(16) /* RTA_PREFSRC */
5400                + nla_total_size(4) /* RTA_TABLE */
5401                + nla_total_size(4) /* RTA_IIF */
5402                + nla_total_size(4) /* RTA_OIF */
5403                + nla_total_size(4) /* RTA_PRIORITY */
5404                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5405                + nla_total_size(sizeof(struct rta_cacheinfo))
5406                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5407                + nla_total_size(1) /* RTA_PREF */
5408                + nexthop_len;
5409 }
5410
5411 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5412                                  unsigned char *flags)
5413 {
5414         if (nexthop_is_multipath(nh)) {
5415                 struct nlattr *mp;
5416
5417                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5418                 if (!mp)
5419                         goto nla_put_failure;
5420
5421                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5422                         goto nla_put_failure;
5423
5424                 nla_nest_end(skb, mp);
5425         } else {
5426                 struct fib6_nh *fib6_nh;
5427
5428                 fib6_nh = nexthop_fib6_nh(nh);
5429                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5430                                      flags, false) < 0)
5431                         goto nla_put_failure;
5432         }
5433
5434         return 0;
5435
5436 nla_put_failure:
5437         return -EMSGSIZE;
5438 }
5439
5440 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5441                          struct fib6_info *rt, struct dst_entry *dst,
5442                          struct in6_addr *dest, struct in6_addr *src,
5443                          int iif, int type, u32 portid, u32 seq,
5444                          unsigned int flags)
5445 {
5446         struct rt6_info *rt6 = (struct rt6_info *)dst;
5447         struct rt6key *rt6_dst, *rt6_src;
5448         u32 *pmetrics, table, rt6_flags;
5449         unsigned char nh_flags = 0;
5450         struct nlmsghdr *nlh;
5451         struct rtmsg *rtm;
5452         long expires = 0;
5453
5454         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5455         if (!nlh)
5456                 return -EMSGSIZE;
5457
5458         if (rt6) {
5459                 rt6_dst = &rt6->rt6i_dst;
5460                 rt6_src = &rt6->rt6i_src;
5461                 rt6_flags = rt6->rt6i_flags;
5462         } else {
5463                 rt6_dst = &rt->fib6_dst;
5464                 rt6_src = &rt->fib6_src;
5465                 rt6_flags = rt->fib6_flags;
5466         }
5467
5468         rtm = nlmsg_data(nlh);
5469         rtm->rtm_family = AF_INET6;
5470         rtm->rtm_dst_len = rt6_dst->plen;
5471         rtm->rtm_src_len = rt6_src->plen;
5472         rtm->rtm_tos = 0;
5473         if (rt->fib6_table)
5474                 table = rt->fib6_table->tb6_id;
5475         else
5476                 table = RT6_TABLE_UNSPEC;
5477         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5478         if (nla_put_u32(skb, RTA_TABLE, table))
5479                 goto nla_put_failure;
5480
5481         rtm->rtm_type = rt->fib6_type;
5482         rtm->rtm_flags = 0;
5483         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5484         rtm->rtm_protocol = rt->fib6_protocol;
5485
5486         if (rt6_flags & RTF_CACHE)
5487                 rtm->rtm_flags |= RTM_F_CLONED;
5488
5489         if (dest) {
5490                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5491                         goto nla_put_failure;
5492                 rtm->rtm_dst_len = 128;
5493         } else if (rtm->rtm_dst_len)
5494                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5495                         goto nla_put_failure;
5496 #ifdef CONFIG_IPV6_SUBTREES
5497         if (src) {
5498                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5499                         goto nla_put_failure;
5500                 rtm->rtm_src_len = 128;
5501         } else if (rtm->rtm_src_len &&
5502                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5503                 goto nla_put_failure;
5504 #endif
5505         if (iif) {
5506 #ifdef CONFIG_IPV6_MROUTE
5507                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5508                         int err = ip6mr_get_route(net, skb, rtm, portid);
5509
5510                         if (err == 0)
5511                                 return 0;
5512                         if (err < 0)
5513                                 goto nla_put_failure;
5514                 } else
5515 #endif
5516                         if (nla_put_u32(skb, RTA_IIF, iif))
5517                                 goto nla_put_failure;
5518         } else if (dest) {
5519                 struct in6_addr saddr_buf;
5520                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5521                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5522                         goto nla_put_failure;
5523         }
5524
5525         if (rt->fib6_prefsrc.plen) {
5526                 struct in6_addr saddr_buf;
5527                 saddr_buf = rt->fib6_prefsrc.addr;
5528                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5529                         goto nla_put_failure;
5530         }
5531
5532         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5533         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5534                 goto nla_put_failure;
5535
5536         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5537                 goto nla_put_failure;
5538
5539         /* For multipath routes, walk the siblings list and add
5540          * each as a nexthop within RTA_MULTIPATH.
5541          */
5542         if (rt6) {
5543                 if (rt6_flags & RTF_GATEWAY &&
5544                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5545                         goto nla_put_failure;
5546
5547                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5548                         goto nla_put_failure;
5549
5550                 if (dst->lwtstate &&
5551                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5552                         goto nla_put_failure;
5553         } else if (rt->fib6_nsiblings) {
5554                 struct fib6_info *sibling, *next_sibling;
5555                 struct nlattr *mp;
5556
5557                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5558                 if (!mp)
5559                         goto nla_put_failure;
5560
5561                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5562                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5563                         goto nla_put_failure;
5564
5565                 list_for_each_entry_safe(sibling, next_sibling,
5566                                          &rt->fib6_siblings, fib6_siblings) {
5567                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5568                                             sibling->fib6_nh->fib_nh_weight,
5569                                             AF_INET6) < 0)
5570                                 goto nla_put_failure;
5571                 }
5572
5573                 nla_nest_end(skb, mp);
5574         } else if (rt->nh) {
5575                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5576                         goto nla_put_failure;
5577
5578                 if (nexthop_is_blackhole(rt->nh))
5579                         rtm->rtm_type = RTN_BLACKHOLE;
5580
5581                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5582                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5583                         goto nla_put_failure;
5584
5585                 rtm->rtm_flags |= nh_flags;
5586         } else {
5587                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5588                                      &nh_flags, false) < 0)
5589                         goto nla_put_failure;
5590
5591                 rtm->rtm_flags |= nh_flags;
5592         }
5593
5594         if (rt6_flags & RTF_EXPIRES) {
5595                 expires = dst ? dst->expires : rt->expires;
5596                 expires -= jiffies;
5597         }
5598
5599         if (!dst) {
5600                 if (rt->offload)
5601                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5602                 if (rt->trap)
5603                         rtm->rtm_flags |= RTM_F_TRAP;
5604                 if (rt->offload_failed)
5605                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5606         }
5607
5608         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5609                 goto nla_put_failure;
5610
5611         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5612                 goto nla_put_failure;
5613
5614
5615         nlmsg_end(skb, nlh);
5616         return 0;
5617
5618 nla_put_failure:
5619         nlmsg_cancel(skb, nlh);
5620         return -EMSGSIZE;
5621 }
5622
5623 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5624 {
5625         const struct net_device *dev = arg;
5626
5627         if (nh->fib_nh_dev == dev)
5628                 return 1;
5629
5630         return 0;
5631 }
5632
5633 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5634                                const struct net_device *dev)
5635 {
5636         if (f6i->nh) {
5637                 struct net_device *_dev = (struct net_device *)dev;
5638
5639                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5640                                                   fib6_info_nh_uses_dev,
5641                                                   _dev);
5642         }
5643
5644         if (f6i->fib6_nh->fib_nh_dev == dev)
5645                 return true;
5646
5647         if (f6i->fib6_nsiblings) {
5648                 struct fib6_info *sibling, *next_sibling;
5649
5650                 list_for_each_entry_safe(sibling, next_sibling,
5651                                          &f6i->fib6_siblings, fib6_siblings) {
5652                         if (sibling->fib6_nh->fib_nh_dev == dev)
5653                                 return true;
5654                 }
5655         }
5656
5657         return false;
5658 }
5659
5660 struct fib6_nh_exception_dump_walker {
5661         struct rt6_rtnl_dump_arg *dump;
5662         struct fib6_info *rt;
5663         unsigned int flags;
5664         unsigned int skip;
5665         unsigned int count;
5666 };
5667
5668 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5669 {
5670         struct fib6_nh_exception_dump_walker *w = arg;
5671         struct rt6_rtnl_dump_arg *dump = w->dump;
5672         struct rt6_exception_bucket *bucket;
5673         struct rt6_exception *rt6_ex;
5674         int i, err;
5675
5676         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5677         if (!bucket)
5678                 return 0;
5679
5680         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5681                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5682                         if (w->skip) {
5683                                 w->skip--;
5684                                 continue;
5685                         }
5686
5687                         /* Expiration of entries doesn't bump sernum, insertion
5688                          * does. Removal is triggered by insertion, so we can
5689                          * rely on the fact that if entries change between two
5690                          * partial dumps, this node is scanned again completely,
5691                          * see rt6_insert_exception() and fib6_dump_table().
5692                          *
5693                          * Count expired entries we go through as handled
5694                          * entries that we'll skip next time, in case of partial
5695                          * node dump. Otherwise, if entries expire meanwhile,
5696                          * we'll skip the wrong amount.
5697                          */
5698                         if (rt6_check_expired(rt6_ex->rt6i)) {
5699                                 w->count++;
5700                                 continue;
5701                         }
5702
5703                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5704                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5705                                             RTM_NEWROUTE,
5706                                             NETLINK_CB(dump->cb->skb).portid,
5707                                             dump->cb->nlh->nlmsg_seq, w->flags);
5708                         if (err)
5709                                 return err;
5710
5711                         w->count++;
5712                 }
5713                 bucket++;
5714         }
5715
5716         return 0;
5717 }
5718
5719 /* Return -1 if done with node, number of handled routes on partial dump */
5720 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5721 {
5722         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5723         struct fib_dump_filter *filter = &arg->filter;
5724         unsigned int flags = NLM_F_MULTI;
5725         struct net *net = arg->net;
5726         int count = 0;
5727
5728         if (rt == net->ipv6.fib6_null_entry)
5729                 return -1;
5730
5731         if ((filter->flags & RTM_F_PREFIX) &&
5732             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5733                 /* success since this is not a prefix route */
5734                 return -1;
5735         }
5736         if (filter->filter_set &&
5737             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5738              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5739              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5740                 return -1;
5741         }
5742
5743         if (filter->filter_set ||
5744             !filter->dump_routes || !filter->dump_exceptions) {
5745                 flags |= NLM_F_DUMP_FILTERED;
5746         }
5747
5748         if (filter->dump_routes) {
5749                 if (skip) {
5750                         skip--;
5751                 } else {
5752                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5753                                           0, RTM_NEWROUTE,
5754                                           NETLINK_CB(arg->cb->skb).portid,
5755                                           arg->cb->nlh->nlmsg_seq, flags)) {
5756                                 return 0;
5757                         }
5758                         count++;
5759                 }
5760         }
5761
5762         if (filter->dump_exceptions) {
5763                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5764                                                            .rt = rt,
5765                                                            .flags = flags,
5766                                                            .skip = skip,
5767                                                            .count = 0 };
5768                 int err;
5769
5770                 rcu_read_lock();
5771                 if (rt->nh) {
5772                         err = nexthop_for_each_fib6_nh(rt->nh,
5773                                                        rt6_nh_dump_exceptions,
5774                                                        &w);
5775                 } else {
5776                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5777                 }
5778                 rcu_read_unlock();
5779
5780                 if (err)
5781                         return count += w.count;
5782         }
5783
5784         return -1;
5785 }
5786
5787 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5788                                         const struct nlmsghdr *nlh,
5789                                         struct nlattr **tb,
5790                                         struct netlink_ext_ack *extack)
5791 {
5792         struct rtmsg *rtm;
5793         int i, err;
5794
5795         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5796                 NL_SET_ERR_MSG_MOD(extack,
5797                                    "Invalid header for get route request");
5798                 return -EINVAL;
5799         }
5800
5801         if (!netlink_strict_get_check(skb))
5802                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5803                                               rtm_ipv6_policy, extack);
5804
5805         rtm = nlmsg_data(nlh);
5806         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5807             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5808             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5809             rtm->rtm_type) {
5810                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5811                 return -EINVAL;
5812         }
5813         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5814                 NL_SET_ERR_MSG_MOD(extack,
5815                                    "Invalid flags for get route request");
5816                 return -EINVAL;
5817         }
5818
5819         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5820                                             rtm_ipv6_policy, extack);
5821         if (err)
5822                 return err;
5823
5824         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5825             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5826                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5827                 return -EINVAL;
5828         }
5829
5830         for (i = 0; i <= RTA_MAX; i++) {
5831                 if (!tb[i])
5832                         continue;
5833
5834                 switch (i) {
5835                 case RTA_SRC:
5836                 case RTA_DST:
5837                 case RTA_IIF:
5838                 case RTA_OIF:
5839                 case RTA_MARK:
5840                 case RTA_UID:
5841                 case RTA_SPORT:
5842                 case RTA_DPORT:
5843                 case RTA_IP_PROTO:
5844                         break;
5845                 default:
5846                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5847                         return -EINVAL;
5848                 }
5849         }
5850
5851         return 0;
5852 }
5853
5854 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5855                               struct netlink_ext_ack *extack)
5856 {
5857         struct net *net = sock_net(in_skb->sk);
5858         struct nlattr *tb[RTA_MAX+1];
5859         int err, iif = 0, oif = 0;
5860         struct fib6_info *from;
5861         struct dst_entry *dst;
5862         struct rt6_info *rt;
5863         struct sk_buff *skb;
5864         struct rtmsg *rtm;
5865         struct flowi6 fl6 = {};
5866         bool fibmatch;
5867
5868         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5869         if (err < 0)
5870                 goto errout;
5871
5872         err = -EINVAL;
5873         rtm = nlmsg_data(nlh);
5874         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5875         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5876
5877         if (tb[RTA_SRC]) {
5878                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5879                         goto errout;
5880
5881                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5882         }
5883
5884         if (tb[RTA_DST]) {
5885                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5886                         goto errout;
5887
5888                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5889         }
5890
5891         if (tb[RTA_IIF])
5892                 iif = nla_get_u32(tb[RTA_IIF]);
5893
5894         if (tb[RTA_OIF])
5895                 oif = nla_get_u32(tb[RTA_OIF]);
5896
5897         if (tb[RTA_MARK])
5898                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5899
5900         if (tb[RTA_UID])
5901                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5902                                            nla_get_u32(tb[RTA_UID]));
5903         else
5904                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5905
5906         if (tb[RTA_SPORT])
5907                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5908
5909         if (tb[RTA_DPORT])
5910                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5911
5912         if (tb[RTA_IP_PROTO]) {
5913                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5914                                                   &fl6.flowi6_proto, AF_INET6,
5915                                                   extack);
5916                 if (err)
5917                         goto errout;
5918         }
5919
5920         if (iif) {
5921                 struct net_device *dev;
5922                 int flags = 0;
5923
5924                 rcu_read_lock();
5925
5926                 dev = dev_get_by_index_rcu(net, iif);
5927                 if (!dev) {
5928                         rcu_read_unlock();
5929                         err = -ENODEV;
5930                         goto errout;
5931                 }
5932
5933                 fl6.flowi6_iif = iif;
5934
5935                 if (!ipv6_addr_any(&fl6.saddr))
5936                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5937
5938                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5939
5940                 rcu_read_unlock();
5941         } else {
5942                 fl6.flowi6_oif = oif;
5943
5944                 dst = ip6_route_output(net, NULL, &fl6);
5945         }
5946
5947
5948         rt = container_of(dst, struct rt6_info, dst);
5949         if (rt->dst.error) {
5950                 err = rt->dst.error;
5951                 ip6_rt_put(rt);
5952                 goto errout;
5953         }
5954
5955         if (rt == net->ipv6.ip6_null_entry) {
5956                 err = rt->dst.error;
5957                 ip6_rt_put(rt);
5958                 goto errout;
5959         }
5960
5961         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5962         if (!skb) {
5963                 ip6_rt_put(rt);
5964                 err = -ENOBUFS;
5965                 goto errout;
5966         }
5967
5968         skb_dst_set(skb, &rt->dst);
5969
5970         rcu_read_lock();
5971         from = rcu_dereference(rt->from);
5972         if (from) {
5973                 if (fibmatch)
5974                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5975                                             iif, RTM_NEWROUTE,
5976                                             NETLINK_CB(in_skb).portid,
5977                                             nlh->nlmsg_seq, 0);
5978                 else
5979                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5980                                             &fl6.saddr, iif, RTM_NEWROUTE,
5981                                             NETLINK_CB(in_skb).portid,
5982                                             nlh->nlmsg_seq, 0);
5983         } else {
5984                 err = -ENETUNREACH;
5985         }
5986         rcu_read_unlock();
5987
5988         if (err < 0) {
5989                 kfree_skb(skb);
5990                 goto errout;
5991         }
5992
5993         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5994 errout:
5995         return err;
5996 }
5997
5998 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
5999                      unsigned int nlm_flags)
6000 {
6001         struct sk_buff *skb;
6002         struct net *net = info->nl_net;
6003         u32 seq;
6004         int err;
6005
6006         err = -ENOBUFS;
6007         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6008
6009         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6010         if (!skb)
6011                 goto errout;
6012
6013         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6014                             event, info->portid, seq, nlm_flags);
6015         if (err < 0) {
6016                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6017                 WARN_ON(err == -EMSGSIZE);
6018                 kfree_skb(skb);
6019                 goto errout;
6020         }
6021         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6022                     info->nlh, gfp_any());
6023         return;
6024 errout:
6025         if (err < 0)
6026                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6027 }
6028
6029 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6030                     struct nl_info *info)
6031 {
6032         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6033         struct sk_buff *skb;
6034         int err = -ENOBUFS;
6035
6036         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6037         if (!skb)
6038                 goto errout;
6039
6040         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6041                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6042         if (err < 0) {
6043                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6044                 WARN_ON(err == -EMSGSIZE);
6045                 kfree_skb(skb);
6046                 goto errout;
6047         }
6048         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6049                     info->nlh, gfp_any());
6050         return;
6051 errout:
6052         if (err < 0)
6053                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6054 }
6055
6056 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6057                             bool offload, bool trap, bool offload_failed)
6058 {
6059         struct sk_buff *skb;
6060         int err;
6061
6062         if (f6i->offload == offload && f6i->trap == trap &&
6063             f6i->offload_failed == offload_failed)
6064                 return;
6065
6066         f6i->offload = offload;
6067         f6i->trap = trap;
6068
6069         /* 2 means send notifications only if offload_failed was changed. */
6070         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6071             f6i->offload_failed == offload_failed)
6072                 return;
6073
6074         f6i->offload_failed = offload_failed;
6075
6076         if (!rcu_access_pointer(f6i->fib6_node))
6077                 /* The route was removed from the tree, do not send
6078                  * notfication.
6079                  */
6080                 return;
6081
6082         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6083                 return;
6084
6085         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6086         if (!skb) {
6087                 err = -ENOBUFS;
6088                 goto errout;
6089         }
6090
6091         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6092                             0, 0);
6093         if (err < 0) {
6094                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6095                 WARN_ON(err == -EMSGSIZE);
6096                 kfree_skb(skb);
6097                 goto errout;
6098         }
6099
6100         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6101         return;
6102
6103 errout:
6104         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6105 }
6106 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6107
6108 static int ip6_route_dev_notify(struct notifier_block *this,
6109                                 unsigned long event, void *ptr)
6110 {
6111         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6112         struct net *net = dev_net(dev);
6113
6114         if (!(dev->flags & IFF_LOOPBACK))
6115                 return NOTIFY_OK;
6116
6117         if (event == NETDEV_REGISTER) {
6118                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6119                 net->ipv6.ip6_null_entry->dst.dev = dev;
6120                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6121 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6122                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6123                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6124                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6125                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6126 #endif
6127          } else if (event == NETDEV_UNREGISTER &&
6128                     dev->reg_state != NETREG_UNREGISTERED) {
6129                 /* NETDEV_UNREGISTER could be fired for multiple times by
6130                  * netdev_wait_allrefs(). Make sure we only call this once.
6131                  */
6132                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6133 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6134                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6135                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6136 #endif
6137         }
6138
6139         return NOTIFY_OK;
6140 }
6141
6142 /*
6143  *      /proc
6144  */
6145
6146 #ifdef CONFIG_PROC_FS
6147 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6148 {
6149         struct net *net = (struct net *)seq->private;
6150         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6151                    net->ipv6.rt6_stats->fib_nodes,
6152                    net->ipv6.rt6_stats->fib_route_nodes,
6153                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6154                    net->ipv6.rt6_stats->fib_rt_entries,
6155                    net->ipv6.rt6_stats->fib_rt_cache,
6156                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6157                    net->ipv6.rt6_stats->fib_discarded_routes);
6158
6159         return 0;
6160 }
6161 #endif  /* CONFIG_PROC_FS */
6162
6163 #ifdef CONFIG_SYSCTL
6164
6165 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6166                               void *buffer, size_t *lenp, loff_t *ppos)
6167 {
6168         struct net *net;
6169         int delay;
6170         int ret;
6171         if (!write)
6172                 return -EINVAL;
6173
6174         net = (struct net *)ctl->extra1;
6175         delay = net->ipv6.sysctl.flush_delay;
6176         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6177         if (ret)
6178                 return ret;
6179
6180         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6181         return 0;
6182 }
6183
6184 static struct ctl_table ipv6_route_table_template[] = {
6185         {
6186                 .procname       =       "flush",
6187                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6188                 .maxlen         =       sizeof(int),
6189                 .mode           =       0200,
6190                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6191         },
6192         {
6193                 .procname       =       "gc_thresh",
6194                 .data           =       &ip6_dst_ops_template.gc_thresh,
6195                 .maxlen         =       sizeof(int),
6196                 .mode           =       0644,
6197                 .proc_handler   =       proc_dointvec,
6198         },
6199         {
6200                 .procname       =       "max_size",
6201                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6202                 .maxlen         =       sizeof(int),
6203                 .mode           =       0644,
6204                 .proc_handler   =       proc_dointvec,
6205         },
6206         {
6207                 .procname       =       "gc_min_interval",
6208                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6209                 .maxlen         =       sizeof(int),
6210                 .mode           =       0644,
6211                 .proc_handler   =       proc_dointvec_jiffies,
6212         },
6213         {
6214                 .procname       =       "gc_timeout",
6215                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6216                 .maxlen         =       sizeof(int),
6217                 .mode           =       0644,
6218                 .proc_handler   =       proc_dointvec_jiffies,
6219         },
6220         {
6221                 .procname       =       "gc_interval",
6222                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6223                 .maxlen         =       sizeof(int),
6224                 .mode           =       0644,
6225                 .proc_handler   =       proc_dointvec_jiffies,
6226         },
6227         {
6228                 .procname       =       "gc_elasticity",
6229                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6230                 .maxlen         =       sizeof(int),
6231                 .mode           =       0644,
6232                 .proc_handler   =       proc_dointvec,
6233         },
6234         {
6235                 .procname       =       "mtu_expires",
6236                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6237                 .maxlen         =       sizeof(int),
6238                 .mode           =       0644,
6239                 .proc_handler   =       proc_dointvec_jiffies,
6240         },
6241         {
6242                 .procname       =       "min_adv_mss",
6243                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6244                 .maxlen         =       sizeof(int),
6245                 .mode           =       0644,
6246                 .proc_handler   =       proc_dointvec,
6247         },
6248         {
6249                 .procname       =       "gc_min_interval_ms",
6250                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6251                 .maxlen         =       sizeof(int),
6252                 .mode           =       0644,
6253                 .proc_handler   =       proc_dointvec_ms_jiffies,
6254         },
6255         {
6256                 .procname       =       "skip_notify_on_dev_down",
6257                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6258                 .maxlen         =       sizeof(int),
6259                 .mode           =       0644,
6260                 .proc_handler   =       proc_dointvec_minmax,
6261                 .extra1         =       SYSCTL_ZERO,
6262                 .extra2         =       SYSCTL_ONE,
6263         },
6264         { }
6265 };
6266
6267 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6268 {
6269         struct ctl_table *table;
6270
6271         table = kmemdup(ipv6_route_table_template,
6272                         sizeof(ipv6_route_table_template),
6273                         GFP_KERNEL);
6274
6275         if (table) {
6276                 table[0].data = &net->ipv6.sysctl.flush_delay;
6277                 table[0].extra1 = net;
6278                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6279                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6280                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6281                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6282                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6283                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6284                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6285                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6286                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6287                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6288
6289                 /* Don't export sysctls to unprivileged users */
6290                 if (net->user_ns != &init_user_ns)
6291                         table[0].procname = NULL;
6292         }
6293
6294         return table;
6295 }
6296 #endif
6297
6298 static int __net_init ip6_route_net_init(struct net *net)
6299 {
6300         int ret = -ENOMEM;
6301
6302         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6303                sizeof(net->ipv6.ip6_dst_ops));
6304
6305         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6306                 goto out_ip6_dst_ops;
6307
6308         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6309         if (!net->ipv6.fib6_null_entry)
6310                 goto out_ip6_dst_entries;
6311         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6312                sizeof(*net->ipv6.fib6_null_entry));
6313
6314         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6315                                            sizeof(*net->ipv6.ip6_null_entry),
6316                                            GFP_KERNEL);
6317         if (!net->ipv6.ip6_null_entry)
6318                 goto out_fib6_null_entry;
6319         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6320         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6321                          ip6_template_metrics, true);
6322         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6323
6324 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6325         net->ipv6.fib6_has_custom_rules = false;
6326         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6327                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6328                                                GFP_KERNEL);
6329         if (!net->ipv6.ip6_prohibit_entry)
6330                 goto out_ip6_null_entry;
6331         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6332         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6333                          ip6_template_metrics, true);
6334         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6335
6336         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6337                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6338                                                GFP_KERNEL);
6339         if (!net->ipv6.ip6_blk_hole_entry)
6340                 goto out_ip6_prohibit_entry;
6341         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6342         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6343                          ip6_template_metrics, true);
6344         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6345 #ifdef CONFIG_IPV6_SUBTREES
6346         net->ipv6.fib6_routes_require_src = 0;
6347 #endif
6348 #endif
6349
6350         net->ipv6.sysctl.flush_delay = 0;
6351         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6352         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6353         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6354         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6355         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6356         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6357         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6358         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6359
6360         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6361
6362         ret = 0;
6363 out:
6364         return ret;
6365
6366 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6367 out_ip6_prohibit_entry:
6368         kfree(net->ipv6.ip6_prohibit_entry);
6369 out_ip6_null_entry:
6370         kfree(net->ipv6.ip6_null_entry);
6371 #endif
6372 out_fib6_null_entry:
6373         kfree(net->ipv6.fib6_null_entry);
6374 out_ip6_dst_entries:
6375         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6376 out_ip6_dst_ops:
6377         goto out;
6378 }
6379
6380 static void __net_exit ip6_route_net_exit(struct net *net)
6381 {
6382         kfree(net->ipv6.fib6_null_entry);
6383         kfree(net->ipv6.ip6_null_entry);
6384 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6385         kfree(net->ipv6.ip6_prohibit_entry);
6386         kfree(net->ipv6.ip6_blk_hole_entry);
6387 #endif
6388         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6389 }
6390
6391 static int __net_init ip6_route_net_init_late(struct net *net)
6392 {
6393 #ifdef CONFIG_PROC_FS
6394         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6395                         sizeof(struct ipv6_route_iter));
6396         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6397                         rt6_stats_seq_show, NULL);
6398 #endif
6399         return 0;
6400 }
6401
6402 static void __net_exit ip6_route_net_exit_late(struct net *net)
6403 {
6404 #ifdef CONFIG_PROC_FS
6405         remove_proc_entry("ipv6_route", net->proc_net);
6406         remove_proc_entry("rt6_stats", net->proc_net);
6407 #endif
6408 }
6409
6410 static struct pernet_operations ip6_route_net_ops = {
6411         .init = ip6_route_net_init,
6412         .exit = ip6_route_net_exit,
6413 };
6414
6415 static int __net_init ipv6_inetpeer_init(struct net *net)
6416 {
6417         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6418
6419         if (!bp)
6420                 return -ENOMEM;
6421         inet_peer_base_init(bp);
6422         net->ipv6.peers = bp;
6423         return 0;
6424 }
6425
6426 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6427 {
6428         struct inet_peer_base *bp = net->ipv6.peers;
6429
6430         net->ipv6.peers = NULL;
6431         inetpeer_invalidate_tree(bp);
6432         kfree(bp);
6433 }
6434
6435 static struct pernet_operations ipv6_inetpeer_ops = {
6436         .init   =       ipv6_inetpeer_init,
6437         .exit   =       ipv6_inetpeer_exit,
6438 };
6439
6440 static struct pernet_operations ip6_route_net_late_ops = {
6441         .init = ip6_route_net_init_late,
6442         .exit = ip6_route_net_exit_late,
6443 };
6444
6445 static struct notifier_block ip6_route_dev_notifier = {
6446         .notifier_call = ip6_route_dev_notify,
6447         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6448 };
6449
6450 void __init ip6_route_init_special_entries(void)
6451 {
6452         /* Registering of the loopback is done before this portion of code,
6453          * the loopback reference in rt6_info will not be taken, do it
6454          * manually for init_net */
6455         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6456         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6457         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6458   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6459         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6460         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6461         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6462         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6463   #endif
6464 }
6465
6466 #if IS_BUILTIN(CONFIG_IPV6)
6467 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6468 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6469
6470 BTF_ID_LIST(btf_fib6_info_id)
6471 BTF_ID(struct, fib6_info)
6472
6473 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6474         .seq_ops                = &ipv6_route_seq_ops,
6475         .init_seq_private       = bpf_iter_init_seq_net,
6476         .fini_seq_private       = bpf_iter_fini_seq_net,
6477         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6478 };
6479
6480 static struct bpf_iter_reg ipv6_route_reg_info = {
6481         .target                 = "ipv6_route",
6482         .ctx_arg_info_size      = 1,
6483         .ctx_arg_info           = {
6484                 { offsetof(struct bpf_iter__ipv6_route, rt),
6485                   PTR_TO_BTF_ID_OR_NULL },
6486         },
6487         .seq_info               = &ipv6_route_seq_info,
6488 };
6489
6490 static int __init bpf_iter_register(void)
6491 {
6492         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6493         return bpf_iter_reg_target(&ipv6_route_reg_info);
6494 }
6495
6496 static void bpf_iter_unregister(void)
6497 {
6498         bpf_iter_unreg_target(&ipv6_route_reg_info);
6499 }
6500 #endif
6501 #endif
6502
6503 int __init ip6_route_init(void)
6504 {
6505         int ret;
6506         int cpu;
6507
6508         ret = -ENOMEM;
6509         ip6_dst_ops_template.kmem_cachep =
6510                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6511                                   SLAB_HWCACHE_ALIGN, NULL);
6512         if (!ip6_dst_ops_template.kmem_cachep)
6513                 goto out;
6514
6515         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6516         if (ret)
6517                 goto out_kmem_cache;
6518
6519         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6520         if (ret)
6521                 goto out_dst_entries;
6522
6523         ret = register_pernet_subsys(&ip6_route_net_ops);
6524         if (ret)
6525                 goto out_register_inetpeer;
6526
6527         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6528
6529         ret = fib6_init();
6530         if (ret)
6531                 goto out_register_subsys;
6532
6533         ret = xfrm6_init();
6534         if (ret)
6535                 goto out_fib6_init;
6536
6537         ret = fib6_rules_init();
6538         if (ret)
6539                 goto xfrm6_init;
6540
6541         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6542         if (ret)
6543                 goto fib6_rules_init;
6544
6545         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6546                                    inet6_rtm_newroute, NULL, 0);
6547         if (ret < 0)
6548                 goto out_register_late_subsys;
6549
6550         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6551                                    inet6_rtm_delroute, NULL, 0);
6552         if (ret < 0)
6553                 goto out_register_late_subsys;
6554
6555         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6556                                    inet6_rtm_getroute, NULL,
6557                                    RTNL_FLAG_DOIT_UNLOCKED);
6558         if (ret < 0)
6559                 goto out_register_late_subsys;
6560
6561         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6562         if (ret)
6563                 goto out_register_late_subsys;
6564
6565 #if IS_BUILTIN(CONFIG_IPV6)
6566 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6567         ret = bpf_iter_register();
6568         if (ret)
6569                 goto out_register_late_subsys;
6570 #endif
6571 #endif
6572
6573         for_each_possible_cpu(cpu) {
6574                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6575
6576                 INIT_LIST_HEAD(&ul->head);
6577                 spin_lock_init(&ul->lock);
6578         }
6579
6580 out:
6581         return ret;
6582
6583 out_register_late_subsys:
6584         rtnl_unregister_all(PF_INET6);
6585         unregister_pernet_subsys(&ip6_route_net_late_ops);
6586 fib6_rules_init:
6587         fib6_rules_cleanup();
6588 xfrm6_init:
6589         xfrm6_fini();
6590 out_fib6_init:
6591         fib6_gc_cleanup();
6592 out_register_subsys:
6593         unregister_pernet_subsys(&ip6_route_net_ops);
6594 out_register_inetpeer:
6595         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6596 out_dst_entries:
6597         dst_entries_destroy(&ip6_dst_blackhole_ops);
6598 out_kmem_cache:
6599         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6600         goto out;
6601 }
6602
6603 void ip6_route_cleanup(void)
6604 {
6605 #if IS_BUILTIN(CONFIG_IPV6)
6606 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6607         bpf_iter_unregister();
6608 #endif
6609 #endif
6610         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6611         unregister_pernet_subsys(&ip6_route_net_late_ops);
6612         fib6_rules_cleanup();
6613         xfrm6_fini();
6614         fib6_gc_cleanup();
6615         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6616         unregister_pernet_subsys(&ip6_route_net_ops);
6617         dst_entries_destroy(&ip6_dst_blackhole_ops);
6618         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6619 }