net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <roque@di.fc.ul.pt>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 INDIRECT_CALLABLE_SCOPE
  86 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  87 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  88 INDIRECT_CALLABLE_SCOPE
  89 unsigned int            ip6_mtu(const struct dst_entry *dst);
  90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  91 static void             ip6_dst_destroy(struct dst_entry *);
  92 static void             ip6_dst_ifdown(struct dst_entry *,
  93                                        struct net_device *dev, int how);
  94 static int               ip6_dst_gc(struct dst_ops *ops);
  95
  96 static int              ip6_pkt_discard(struct sk_buff *skb);
  97 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  99 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
 100 static void             ip6_link_failure(struct sk_buff *skb);
 101 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 102                                            struct sk_buff *skb, u32 mtu,
 103                                            bool confirm_neigh);
 104 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 105                                         struct sk_buff *skb);
 106 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 107                            int strict);
 108 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 109 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 110                          struct fib6_info *rt, struct dst_entry *dst,
 111                          struct in6_addr *dest, struct in6_addr *src,
 112                          int iif, int type, u32 portid, u32 seq,
 113                          unsigned int flags);
 114 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 115                                            const struct in6_addr *daddr,
 116                                            const struct in6_addr *saddr);
 117
 118 #ifdef CONFIG_IPV6_ROUTE_INFO
 119 static struct fib6_info *rt6_add_route_info(struct net *net,
 120                                            const struct in6_addr *prefix, int prefixlen,
 121                                            const struct in6_addr *gwaddr,
 122                                            struct net_device *dev,
 123                                            unsigned int pref);
 124 static struct fib6_info *rt6_get_route_info(struct net *net,
 125                                            const struct in6_addr *prefix, int prefixlen,
 126                                            const struct in6_addr *gwaddr,
 127                                            struct net_device *dev);
 128 #endif
 129
 130 struct uncached_list {
 131         spinlock_t              lock;
 132         struct list_head        head;
 133 };
 134
 135 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 136
 137 void rt6_uncached_list_add(struct rt6_info *rt)
 138 {
 139         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 140
 141         rt->rt6i_uncached_list = ul;
 142
 143         spin_lock_bh(&ul->lock);
 144         list_add_tail(&rt->rt6i_uncached, &ul->head);
 145         spin_unlock_bh(&ul->lock);
 146 }
 147
 148 void rt6_uncached_list_del(struct rt6_info *rt)
 149 {
 150         if (!list_empty(&rt->rt6i_uncached)) {
 151                 struct uncached_list *ul = rt->rt6i_uncached_list;
 152                 struct net *net = dev_net(rt->dst.dev);
 153
 154                 spin_lock_bh(&ul->lock);
 155                 list_del(&rt->rt6i_uncached);
 156                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 157                 spin_unlock_bh(&ul->lock);
 158         }
 159 }
 160
 161 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 162 {
 163         struct net_device *loopback_dev = net->loopback_dev;
 164         int cpu;
 165
 166         if (dev == loopback_dev)
 167                 return;
 168
 169         for_each_possible_cpu(cpu) {
 170                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 171                 struct rt6_info *rt;
 172
 173                 spin_lock_bh(&ul->lock);
 174                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 175                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 176                         struct net_device *rt_dev = rt->dst.dev;
 177
 178                         if (rt_idev->dev == dev) {
 179                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 180                                 in6_dev_put(rt_idev);
 181                         }
 182
 183                         if (rt_dev == dev) {
 184                                 rt->dst.dev = blackhole_netdev;
 185                                 dev_hold(rt->dst.dev);
 186                                 dev_put(rt_dev);
 187                         }
 188                 }
 189                 spin_unlock_bh(&ul->lock);
 190         }
 191 }
 192
 193 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 194                                              struct sk_buff *skb,
 195                                              const void *daddr)
 196 {
 197         if (!ipv6_addr_any(p))
 198                 return (const void *) p;
 199         else if (skb)
 200                 return &ipv6_hdr(skb)->daddr;
 201         return daddr;
 202 }
 203
 204 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 205                                    struct net_device *dev,
 206                                    struct sk_buff *skb,
 207                                    const void *daddr)
 208 {
 209         struct neighbour *n;
 210
 211         daddr = choose_neigh_daddr(gw, skb, daddr);
 212         n = __ipv6_neigh_lookup(dev, daddr);
 213         if (n)
 214                 return n;
 215
 216         n = neigh_create(&nd_tbl, daddr, dev);
 217         return IS_ERR(n) ? NULL : n;
 218 }
 219
 220 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 221                                               struct sk_buff *skb,
 222                                               const void *daddr)
 223 {
 224         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 225
 226         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 227                                 dst->dev, skb, daddr);
 228 }
 229
 230 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 231 {
 232         struct net_device *dev = dst->dev;
 233         struct rt6_info *rt = (struct rt6_info *)dst;
 234
 235         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 236         if (!daddr)
 237                 return;
 238         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 239                 return;
 240         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 241                 return;
 242         __ipv6_confirm_neigh(dev, daddr);
 243 }
 244
 245 static struct dst_ops ip6_dst_ops_template = {
 246         .family                 =       AF_INET6,
 247         .gc                     =       ip6_dst_gc,
 248         .gc_thresh              =       1024,
 249         .check                  =       ip6_dst_check,
 250         .default_advmss         =       ip6_default_advmss,
 251         .mtu                    =       ip6_mtu,
 252         .cow_metrics            =       dst_cow_metrics_generic,
 253         .destroy                =       ip6_dst_destroy,
 254         .ifdown                 =       ip6_dst_ifdown,
 255         .negative_advice        =       ip6_negative_advice,
 256         .link_failure           =       ip6_link_failure,
 257         .update_pmtu            =       ip6_rt_update_pmtu,
 258         .redirect               =       rt6_do_redirect,
 259         .local_out              =       __ip6_local_out,
 260         .neigh_lookup           =       ip6_dst_neigh_lookup,
 261         .confirm_neigh          =       ip6_confirm_neigh,
 262 };
 263
 264 static struct dst_ops ip6_dst_blackhole_ops = {
 265         .family                 = AF_INET6,
 266         .default_advmss         = ip6_default_advmss,
 267         .neigh_lookup           = ip6_dst_neigh_lookup,
 268         .check                  = ip6_dst_check,
 269         .destroy                = ip6_dst_destroy,
 270         .cow_metrics            = dst_cow_metrics_generic,
 271         .update_pmtu            = dst_blackhole_update_pmtu,
 272         .redirect               = dst_blackhole_redirect,
 273         .mtu                    = dst_blackhole_mtu,
 274 };
 275
 276 static const u32 ip6_template_metrics[RTAX_MAX] = {
 277         [RTAX_HOPLIMIT - 1] = 0,
 278 };
 279
 280 static const struct fib6_info fib6_null_entry_template = {
 281         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 282         .fib6_protocol  = RTPROT_KERNEL,
 283         .fib6_metric    = ~(u32)0,
 284         .fib6_ref       = REFCOUNT_INIT(1),
 285         .fib6_type      = RTN_UNREACHABLE,
 286         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 287 };
 288
 289 static const struct rt6_info ip6_null_entry_template = {
 290         .dst = {
 291                 .__refcnt       = ATOMIC_INIT(1),
 292                 .__use          = 1,
 293                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 294                 .error          = -ENETUNREACH,
 295                 .input          = ip6_pkt_discard,
 296                 .output         = ip6_pkt_discard_out,
 297         },
 298         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 299 };
 300
 301 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 302
 303 static const struct rt6_info ip6_prohibit_entry_template = {
 304         .dst = {
 305                 .__refcnt       = ATOMIC_INIT(1),
 306                 .__use          = 1,
 307                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 308                 .error          = -EACCES,
 309                 .input          = ip6_pkt_prohibit,
 310                 .output         = ip6_pkt_prohibit_out,
 311         },
 312         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 313 };
 314
 315 static const struct rt6_info ip6_blk_hole_entry_template = {
 316         .dst = {
 317                 .__refcnt       = ATOMIC_INIT(1),
 318                 .__use          = 1,
 319                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 320                 .error          = -EINVAL,
 321                 .input          = dst_discard,
 322                 .output         = dst_discard_out,
 323         },
 324         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 325 };
 326
 327 #endif
 328
 329 static void rt6_info_init(struct rt6_info *rt)
 330 {
 331         struct dst_entry *dst = &rt->dst;
 332
 333         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 334         INIT_LIST_HEAD(&rt->rt6i_uncached);
 335 }
 336
 337 /* allocate dst with ip6_dst_ops */
 338 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 339                                int flags)
 340 {
 341         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 342                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 343
 344         if (rt) {
 345                 rt6_info_init(rt);
 346                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 347         }
 348
 349         return rt;
 350 }
 351 EXPORT_SYMBOL(ip6_dst_alloc);
 352
 353 static void ip6_dst_destroy(struct dst_entry *dst)
 354 {
 355         struct rt6_info *rt = (struct rt6_info *)dst;
 356         struct fib6_info *from;
 357         struct inet6_dev *idev;
 358
 359         ip_dst_metrics_put(dst);
 360         rt6_uncached_list_del(rt);
 361
 362         idev = rt->rt6i_idev;
 363         if (idev) {
 364                 rt->rt6i_idev = NULL;
 365                 in6_dev_put(idev);
 366         }
 367
 368         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 369         fib6_info_release(from);
 370 }
 371
 372 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 373                            int how)
 374 {
 375         struct rt6_info *rt = (struct rt6_info *)dst;
 376         struct inet6_dev *idev = rt->rt6i_idev;
 377         struct net_device *loopback_dev =
 378                 dev_net(dev)->loopback_dev;
 379
 380         if (idev && idev->dev != loopback_dev) {
 381                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 382                 if (loopback_idev) {
 383                         rt->rt6i_idev = loopback_idev;
 384                         in6_dev_put(idev);
 385                 }
 386         }
 387 }
 388
 389 static bool __rt6_check_expired(const struct rt6_info *rt)
 390 {
 391         if (rt->rt6i_flags & RTF_EXPIRES)
 392                 return time_after(jiffies, rt->dst.expires);
 393         else
 394                 return false;
 395 }
 396
 397 static bool rt6_check_expired(const struct rt6_info *rt)
 398 {
 399         struct fib6_info *from;
 400
 401         from = rcu_dereference(rt->from);
 402
 403         if (rt->rt6i_flags & RTF_EXPIRES) {
 404                 if (time_after(jiffies, rt->dst.expires))
 405                         return true;
 406         } else if (from) {
 407                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 408                         fib6_check_expired(from);
 409         }
 410         return false;
 411 }
 412
 413 void fib6_select_path(const struct net *net, struct fib6_result *res,
 414                       struct flowi6 *fl6, int oif, bool have_oif_match,
 415                       const struct sk_buff *skb, int strict)
 416 {
 417         struct fib6_info *sibling, *next_sibling;
 418         struct fib6_info *match = res->f6i;
 419
 420         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 421                 goto out;
 422
 423         if (match->nh && have_oif_match && res->nh)
 424                 return;
 425
 426         /* We might have already computed the hash for ICMPv6 errors. In such
 427          * case it will always be non-zero. Otherwise now is the time to do it.
 428          */
 429         if (!fl6->mp_hash &&
 430             (!match->nh || nexthop_is_multipath(match->nh)))
 431                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 432
 433         if (unlikely(match->nh)) {
 434                 nexthop_path_fib6_result(res, fl6->mp_hash);
 435                 return;
 436         }
 437
 438         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 439                 goto out;
 440
 441         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 442                                  fib6_siblings) {
 443                 const struct fib6_nh *nh = sibling->fib6_nh;
 444                 int nh_upper_bound;
 445
 446                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 447                 if (fl6->mp_hash > nh_upper_bound)
 448                         continue;
 449                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 450                         break;
 451                 match = sibling;
 452                 break;
 453         }
 454
 455 out:
 456         res->f6i = match;
 457         res->nh = match->fib6_nh;
 458 }
 459
 460 /*
 461  *      Route lookup. rcu_read_lock() should be held.
 462  */
 463
 464 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 465                                const struct in6_addr *saddr, int oif, int flags)
 466 {
 467         const struct net_device *dev;
 468
 469         if (nh->fib_nh_flags & RTNH_F_DEAD)
 470                 return false;
 471
 472         dev = nh->fib_nh_dev;
 473         if (oif) {
 474                 if (dev->ifindex == oif)
 475                         return true;
 476         } else {
 477                 if (ipv6_chk_addr(net, saddr, dev,
 478                                   flags & RT6_LOOKUP_F_IFACE))
 479                         return true;
 480         }
 481
 482         return false;
 483 }
 484
 485 struct fib6_nh_dm_arg {
 486         struct net              *net;
 487         const struct in6_addr   *saddr;
 488         int                     oif;
 489         int                     flags;
 490         struct fib6_nh          *nh;
 491 };
 492
 493 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 494 {
 495         struct fib6_nh_dm_arg *arg = _arg;
 496
 497         arg->nh = nh;
 498         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 499                                   arg->flags);
 500 }
 501
 502 /* returns fib6_nh from nexthop or NULL */
 503 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 504                                         struct fib6_result *res,
 505                                         const struct in6_addr *saddr,
 506                                         int oif, int flags)
 507 {
 508         struct fib6_nh_dm_arg arg = {
 509                 .net   = net,
 510                 .saddr = saddr,
 511                 .oif   = oif,
 512                 .flags = flags,
 513         };
 514
 515         if (nexthop_is_blackhole(nh))
 516                 return NULL;
 517
 518         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 519                 return arg.nh;
 520
 521         return NULL;
 522 }
 523
 524 static void rt6_device_match(struct net *net, struct fib6_result *res,
 525                              const struct in6_addr *saddr, int oif, int flags)
 526 {
 527         struct fib6_info *f6i = res->f6i;
 528         struct fib6_info *spf6i;
 529         struct fib6_nh *nh;
 530
 531         if (!oif && ipv6_addr_any(saddr)) {
 532                 if (unlikely(f6i->nh)) {
 533                         nh = nexthop_fib6_nh(f6i->nh);
 534                         if (nexthop_is_blackhole(f6i->nh))
 535                                 goto out_blackhole;
 536                 } else {
 537                         nh = f6i->fib6_nh;
 538                 }
 539                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 540                         goto out;
 541         }
 542
 543         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 544                 bool matched = false;
 545
 546                 if (unlikely(spf6i->nh)) {
 547                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 548                                               oif, flags);
 549                         if (nh)
 550                                 matched = true;
 551                 } else {
 552                         nh = spf6i->fib6_nh;
 553                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 554                                 matched = true;
 555                 }
 556                 if (matched) {
 557                         res->f6i = spf6i;
 558                         goto out;
 559                 }
 560         }
 561
 562         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 563                 res->f6i = net->ipv6.fib6_null_entry;
 564                 nh = res->f6i->fib6_nh;
 565                 goto out;
 566         }
 567
 568         if (unlikely(f6i->nh)) {
 569                 nh = nexthop_fib6_nh(f6i->nh);
 570                 if (nexthop_is_blackhole(f6i->nh))
 571                         goto out_blackhole;
 572         } else {
 573                 nh = f6i->fib6_nh;
 574         }
 575
 576         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 577                 res->f6i = net->ipv6.fib6_null_entry;
 578                 nh = res->f6i->fib6_nh;
 579         }
 580 out:
 581         res->nh = nh;
 582         res->fib6_type = res->f6i->fib6_type;
 583         res->fib6_flags = res->f6i->fib6_flags;
 584         return;
 585
 586 out_blackhole:
 587         res->fib6_flags |= RTF_REJECT;
 588         res->fib6_type = RTN_BLACKHOLE;
 589         res->nh = nh;
 590 }
 591
 592 #ifdef CONFIG_IPV6_ROUTER_PREF
 593 struct __rt6_probe_work {
 594         struct work_struct work;
 595         struct in6_addr target;
 596         struct net_device *dev;
 597 };
 598
 599 static void rt6_probe_deferred(struct work_struct *w)
 600 {
 601         struct in6_addr mcaddr;
 602         struct __rt6_probe_work *work =
 603                 container_of(w, struct __rt6_probe_work, work);
 604
 605         addrconf_addr_solict_mult(&work->target, &mcaddr);
 606         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 607         dev_put(work->dev);
 608         kfree(work);
 609 }
 610
 611 static void rt6_probe(struct fib6_nh *fib6_nh)
 612 {
 613         struct __rt6_probe_work *work = NULL;
 614         const struct in6_addr *nh_gw;
 615         unsigned long last_probe;
 616         struct neighbour *neigh;
 617         struct net_device *dev;
 618         struct inet6_dev *idev;
 619
 620         /*
 621          * Okay, this does not seem to be appropriate
 622          * for now, however, we need to check if it
 623          * is really so; aka Router Reachability Probing.
 624          *
 625          * Router Reachability Probe MUST be rate-limited
 626          * to no more than one per minute.
 627          */
 628         if (!fib6_nh->fib_nh_gw_family)
 629                 return;
 630
 631         nh_gw = &fib6_nh->fib_nh_gw6;
 632         dev = fib6_nh->fib_nh_dev;
 633         rcu_read_lock_bh();
 634         last_probe = READ_ONCE(fib6_nh->last_probe);
 635         idev = __in6_dev_get(dev);
 636         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 637         if (neigh) {
 638                 if (neigh->nud_state & NUD_VALID)
 639                         goto out;
 640
 641                 write_lock(&neigh->lock);
 642                 if (!(neigh->nud_state & NUD_VALID) &&
 643                     time_after(jiffies,
 644                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 645                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 646                         if (work)
 647                                 __neigh_set_probe_once(neigh);
 648                 }
 649                 write_unlock(&neigh->lock);
 650         } else if (time_after(jiffies, last_probe +
 651                                        idev->cnf.rtr_probe_interval)) {
 652                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 653         }
 654
 655         if (!work || cmpxchg(&fib6_nh->last_probe,
 656                              last_probe, jiffies) != last_probe) {
 657                 kfree(work);
 658         } else {
 659                 INIT_WORK(&work->work, rt6_probe_deferred);
 660                 work->target = *nh_gw;
 661                 dev_hold(dev);
 662                 work->dev = dev;
 663                 schedule_work(&work->work);
 664         }
 665
 666 out:
 667         rcu_read_unlock_bh();
 668 }
 669 #else
 670 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 671 {
 672 }
 673 #endif
 674
 675 /*
 676  * Default Router Selection (RFC 2461 6.3.6)
 677  */
 678 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 679 {
 680         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 681         struct neighbour *neigh;
 682
 683         rcu_read_lock_bh();
 684         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 685                                           &fib6_nh->fib_nh_gw6);
 686         if (neigh) {
 687                 read_lock(&neigh->lock);
 688                 if (neigh->nud_state & NUD_VALID)
 689                         ret = RT6_NUD_SUCCEED;
 690 #ifdef CONFIG_IPV6_ROUTER_PREF
 691                 else if (!(neigh->nud_state & NUD_FAILED))
 692                         ret = RT6_NUD_SUCCEED;
 693                 else
 694                         ret = RT6_NUD_FAIL_PROBE;
 695 #endif
 696                 read_unlock(&neigh->lock);
 697         } else {
 698                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 699                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 700         }
 701         rcu_read_unlock_bh();
 702
 703         return ret;
 704 }
 705
 706 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 707                            int strict)
 708 {
 709         int m = 0;
 710
 711         if (!oif || nh->fib_nh_dev->ifindex == oif)
 712                 m = 2;
 713
 714         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 715                 return RT6_NUD_FAIL_HARD;
 716 #ifdef CONFIG_IPV6_ROUTER_PREF
 717         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 718 #endif
 719         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 720             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 721                 int n = rt6_check_neigh(nh);
 722                 if (n < 0)
 723                         return n;
 724         }
 725         return m;
 726 }
 727
 728 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 729                        int oif, int strict, int *mpri, bool *do_rr)
 730 {
 731         bool match_do_rr = false;
 732         bool rc = false;
 733         int m;
 734
 735         if (nh->fib_nh_flags & RTNH_F_DEAD)
 736                 goto out;
 737
 738         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 739             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 740             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 741                 goto out;
 742
 743         m = rt6_score_route(nh, fib6_flags, oif, strict);
 744         if (m == RT6_NUD_FAIL_DO_RR) {
 745                 match_do_rr = true;
 746                 m = 0; /* lowest valid score */
 747         } else if (m == RT6_NUD_FAIL_HARD) {
 748                 goto out;
 749         }
 750
 751         if (strict & RT6_LOOKUP_F_REACHABLE)
 752                 rt6_probe(nh);
 753
 754         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 755         if (m > *mpri) {
 756                 *do_rr = match_do_rr;
 757                 *mpri = m;
 758                 rc = true;
 759         }
 760 out:
 761         return rc;
 762 }
 763
 764 struct fib6_nh_frl_arg {
 765         u32             flags;
 766         int             oif;
 767         int             strict;
 768         int             *mpri;
 769         bool            *do_rr;
 770         struct fib6_nh  *nh;
 771 };
 772
 773 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 774 {
 775         struct fib6_nh_frl_arg *arg = _arg;
 776
 777         arg->nh = nh;
 778         return find_match(nh, arg->flags, arg->oif, arg->strict,
 779                           arg->mpri, arg->do_rr);
 780 }
 781
 782 static void __find_rr_leaf(struct fib6_info *f6i_start,
 783                            struct fib6_info *nomatch, u32 metric,
 784                            struct fib6_result *res, struct fib6_info **cont,
 785                            int oif, int strict, bool *do_rr, int *mpri)
 786 {
 787         struct fib6_info *f6i;
 788
 789         for (f6i = f6i_start;
 790              f6i && f6i != nomatch;
 791              f6i = rcu_dereference(f6i->fib6_next)) {
 792                 bool matched = false;
 793                 struct fib6_nh *nh;
 794
 795                 if (cont && f6i->fib6_metric != metric) {
 796                         *cont = f6i;
 797                         return;
 798                 }
 799
 800                 if (fib6_check_expired(f6i))
 801                         continue;
 802
 803                 if (unlikely(f6i->nh)) {
 804                         struct fib6_nh_frl_arg arg = {
 805                                 .flags  = f6i->fib6_flags,
 806                                 .oif    = oif,
 807                                 .strict = strict,
 808                                 .mpri   = mpri,
 809                                 .do_rr  = do_rr
 810                         };
 811
 812                         if (nexthop_is_blackhole(f6i->nh)) {
 813                                 res->fib6_flags = RTF_REJECT;
 814                                 res->fib6_type = RTN_BLACKHOLE;
 815                                 res->f6i = f6i;
 816                                 res->nh = nexthop_fib6_nh(f6i->nh);
 817                                 return;
 818                         }
 819                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 820                                                      &arg)) {
 821                                 matched = true;
 822                                 nh = arg.nh;
 823                         }
 824                 } else {
 825                         nh = f6i->fib6_nh;
 826                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 827                                        mpri, do_rr))
 828                                 matched = true;
 829                 }
 830                 if (matched) {
 831                         res->f6i = f6i;
 832                         res->nh = nh;
 833                         res->fib6_flags = f6i->fib6_flags;
 834                         res->fib6_type = f6i->fib6_type;
 835                 }
 836         }
 837 }
 838
 839 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 840                          struct fib6_info *rr_head, int oif, int strict,
 841                          bool *do_rr, struct fib6_result *res)
 842 {
 843         u32 metric = rr_head->fib6_metric;
 844         struct fib6_info *cont = NULL;
 845         int mpri = -1;
 846
 847         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 848                        oif, strict, do_rr, &mpri);
 849
 850         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 851                        oif, strict, do_rr, &mpri);
 852
 853         if (res->f6i || !cont)
 854                 return;
 855
 856         __find_rr_leaf(cont, NULL, metric, res, NULL,
 857                        oif, strict, do_rr, &mpri);
 858 }
 859
 860 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 861                        struct fib6_result *res, int strict)
 862 {
 863         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 864         struct fib6_info *rt0;
 865         bool do_rr = false;
 866         int key_plen;
 867
 868         /* make sure this function or its helpers sets f6i */
 869         res->f6i = NULL;
 870
 871         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 872                 goto out;
 873
 874         rt0 = rcu_dereference(fn->rr_ptr);
 875         if (!rt0)
 876                 rt0 = leaf;
 877
 878         /* Double check to make sure fn is not an intermediate node
 879          * and fn->leaf does not points to its child's leaf
 880          * (This might happen if all routes under fn are deleted from
 881          * the tree and fib6_repair_tree() is called on the node.)
 882          */
 883         key_plen = rt0->fib6_dst.plen;
 884 #ifdef CONFIG_IPV6_SUBTREES
 885         if (rt0->fib6_src.plen)
 886                 key_plen = rt0->fib6_src.plen;
 887 #endif
 888         if (fn->fn_bit != key_plen)
 889                 goto out;
 890
 891         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 892         if (do_rr) {
 893                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 894
 895                 /* no entries matched; do round-robin */
 896                 if (!next || next->fib6_metric != rt0->fib6_metric)
 897                         next = leaf;
 898
 899                 if (next != rt0) {
 900                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 901                         /* make sure next is not being deleted from the tree */
 902                         if (next->fib6_node)
 903                                 rcu_assign_pointer(fn->rr_ptr, next);
 904                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 905                 }
 906         }
 907
 908 out:
 909         if (!res->f6i) {
 910                 res->f6i = net->ipv6.fib6_null_entry;
 911                 res->nh = res->f6i->fib6_nh;
 912                 res->fib6_flags = res->f6i->fib6_flags;
 913                 res->fib6_type = res->f6i->fib6_type;
 914         }
 915 }
 916
 917 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 918 {
 919         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 920                res->nh->fib_nh_gw_family;
 921 }
 922
 923 #ifdef CONFIG_IPV6_ROUTE_INFO
 924 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 925                   const struct in6_addr *gwaddr)
 926 {
 927         struct net *net = dev_net(dev);
 928         struct route_info *rinfo = (struct route_info *) opt;
 929         struct in6_addr prefix_buf, *prefix;
 930         unsigned int pref;
 931         unsigned long lifetime;
 932         struct fib6_info *rt;
 933
 934         if (len < sizeof(struct route_info)) {
 935                 return -EINVAL;
 936         }
 937
 938         /* Sanity check for prefix_len and length */
 939         if (rinfo->length > 3) {
 940                 return -EINVAL;
 941         } else if (rinfo->prefix_len > 128) {
 942                 return -EINVAL;
 943         } else if (rinfo->prefix_len > 64) {
 944                 if (rinfo->length < 2) {
 945                         return -EINVAL;
 946                 }
 947         } else if (rinfo->prefix_len > 0) {
 948                 if (rinfo->length < 1) {
 949                         return -EINVAL;
 950                 }
 951         }
 952
 953         pref = rinfo->route_pref;
 954         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 955                 return -EINVAL;
 956
 957         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 958
 959         if (rinfo->length == 3)
 960                 prefix = (struct in6_addr *)rinfo->prefix;
 961         else {
 962                 /* this function is safe */
 963                 ipv6_addr_prefix(&prefix_buf,
 964                                  (struct in6_addr *)rinfo->prefix,
 965                                  rinfo->prefix_len);
 966                 prefix = &prefix_buf;
 967         }
 968
 969         if (rinfo->prefix_len == 0)
 970                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 971         else
 972                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 973                                         gwaddr, dev);
 974
 975         if (rt && !lifetime) {
 976                 ip6_del_rt(net, rt, false);
 977                 rt = NULL;
 978         }
 979
 980         if (!rt && lifetime)
 981                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 982                                         dev, pref);
 983         else if (rt)
 984                 rt->fib6_flags = RTF_ROUTEINFO |
 985                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 986
 987         if (rt) {
 988                 if (!addrconf_finite_timeout(lifetime))
 989                         fib6_clean_expires(rt);
 990                 else
 991                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 992
 993                 fib6_info_release(rt);
 994         }
 995         return 0;
 996 }
 997 #endif
 998
 999 /*
1000  *      Misc support functions
1001  */
1002
1003 /* called with rcu_lock held */
1004 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1005 {
1006         struct net_device *dev = res->nh->fib_nh_dev;
1007
1008         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1009                 /* for copies of local routes, dst->dev needs to be the
1010                  * device if it is a master device, the master device if
1011                  * device is enslaved, and the loopback as the default
1012                  */
1013                 if (netif_is_l3_slave(dev) &&
1014                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1015                         dev = l3mdev_master_dev_rcu(dev);
1016                 else if (!netif_is_l3_master(dev))
1017                         dev = dev_net(dev)->loopback_dev;
1018                 /* last case is netif_is_l3_master(dev) is true in which
1019                  * case we want dev returned to be dev
1020                  */
1021         }
1022
1023         return dev;
1024 }
1025
1026 static const int fib6_prop[RTN_MAX + 1] = {
1027         [RTN_UNSPEC]    = 0,
1028         [RTN_UNICAST]   = 0,
1029         [RTN_LOCAL]     = 0,
1030         [RTN_BROADCAST] = 0,
1031         [RTN_ANYCAST]   = 0,
1032         [RTN_MULTICAST] = 0,
1033         [RTN_BLACKHOLE] = -EINVAL,
1034         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1035         [RTN_PROHIBIT]  = -EACCES,
1036         [RTN_THROW]     = -EAGAIN,
1037         [RTN_NAT]       = -EINVAL,
1038         [RTN_XRESOLVE]  = -EINVAL,
1039 };
1040
1041 static int ip6_rt_type_to_error(u8 fib6_type)
1042 {
1043         return fib6_prop[fib6_type];
1044 }
1045
1046 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1047 {
1048         unsigned short flags = 0;
1049
1050         if (rt->dst_nocount)
1051                 flags |= DST_NOCOUNT;
1052         if (rt->dst_nopolicy)
1053                 flags |= DST_NOPOLICY;
1054
1055         return flags;
1056 }
1057
1058 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1059 {
1060         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1061
1062         switch (fib6_type) {
1063         case RTN_BLACKHOLE:
1064                 rt->dst.output = dst_discard_out;
1065                 rt->dst.input = dst_discard;
1066                 break;
1067         case RTN_PROHIBIT:
1068                 rt->dst.output = ip6_pkt_prohibit_out;
1069                 rt->dst.input = ip6_pkt_prohibit;
1070                 break;
1071         case RTN_THROW:
1072         case RTN_UNREACHABLE:
1073         default:
1074                 rt->dst.output = ip6_pkt_discard_out;
1075                 rt->dst.input = ip6_pkt_discard;
1076                 break;
1077         }
1078 }
1079
1080 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1081 {
1082         struct fib6_info *f6i = res->f6i;
1083
1084         if (res->fib6_flags & RTF_REJECT) {
1085                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1086                 return;
1087         }
1088
1089         rt->dst.error = 0;
1090         rt->dst.output = ip6_output;
1091
1092         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1093                 rt->dst.input = ip6_input;
1094         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1095                 rt->dst.input = ip6_mc_input;
1096         } else {
1097                 rt->dst.input = ip6_forward;
1098         }
1099
1100         if (res->nh->fib_nh_lws) {
1101                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1102                 lwtunnel_set_redirect(&rt->dst);
1103         }
1104
1105         rt->dst.lastuse = jiffies;
1106 }
1107
1108 /* Caller must already hold reference to @from */
1109 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1110 {
1111         rt->rt6i_flags &= ~RTF_EXPIRES;
1112         rcu_assign_pointer(rt->from, from);
1113         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1114 }
1115
1116 /* Caller must already hold reference to f6i in result */
1117 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1118 {
1119         const struct fib6_nh *nh = res->nh;
1120         const struct net_device *dev = nh->fib_nh_dev;
1121         struct fib6_info *f6i = res->f6i;
1122
1123         ip6_rt_init_dst(rt, res);
1124
1125         rt->rt6i_dst = f6i->fib6_dst;
1126         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1127         rt->rt6i_flags = res->fib6_flags;
1128         if (nh->fib_nh_gw_family) {
1129                 rt->rt6i_gateway = nh->fib_nh_gw6;
1130                 rt->rt6i_flags |= RTF_GATEWAY;
1131         }
1132         rt6_set_from(rt, f6i);
1133 #ifdef CONFIG_IPV6_SUBTREES
1134         rt->rt6i_src = f6i->fib6_src;
1135 #endif
1136 }
1137
1138 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1139                                         struct in6_addr *saddr)
1140 {
1141         struct fib6_node *pn, *sn;
1142         while (1) {
1143                 if (fn->fn_flags & RTN_TL_ROOT)
1144                         return NULL;
1145                 pn = rcu_dereference(fn->parent);
1146                 sn = FIB6_SUBTREE(pn);
1147                 if (sn && sn != fn)
1148                         fn = fib6_node_lookup(sn, NULL, saddr);
1149                 else
1150                         fn = pn;
1151                 if (fn->fn_flags & RTN_RTINFO)
1152                         return fn;
1153         }
1154 }
1155
1156 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1157 {
1158         struct rt6_info *rt = *prt;
1159
1160         if (dst_hold_safe(&rt->dst))
1161                 return true;
1162         if (net) {
1163                 rt = net->ipv6.ip6_null_entry;
1164                 dst_hold(&rt->dst);
1165         } else {
1166                 rt = NULL;
1167         }
1168         *prt = rt;
1169         return false;
1170 }
1171
1172 /* called with rcu_lock held */
1173 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1174 {
1175         struct net_device *dev = res->nh->fib_nh_dev;
1176         struct fib6_info *f6i = res->f6i;
1177         unsigned short flags;
1178         struct rt6_info *nrt;
1179
1180         if (!fib6_info_hold_safe(f6i))
1181                 goto fallback;
1182
1183         flags = fib6_info_dst_flags(f6i);
1184         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1185         if (!nrt) {
1186                 fib6_info_release(f6i);
1187                 goto fallback;
1188         }
1189
1190         ip6_rt_copy_init(nrt, res);
1191         return nrt;
1192
1193 fallback:
1194         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1195         dst_hold(&nrt->dst);
1196         return nrt;
1197 }
1198
1199 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1200                                              struct fib6_table *table,
1201                                              struct flowi6 *fl6,
1202                                              const struct sk_buff *skb,
1203                                              int flags)
1204 {
1205         struct fib6_result res = {};
1206         struct fib6_node *fn;
1207         struct rt6_info *rt;
1208
1209         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1210                 flags &= ~RT6_LOOKUP_F_IFACE;
1211
1212         rcu_read_lock();
1213         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1214 restart:
1215         res.f6i = rcu_dereference(fn->leaf);
1216         if (!res.f6i)
1217                 res.f6i = net->ipv6.fib6_null_entry;
1218         else
1219                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1220                                  flags);
1221
1222         if (res.f6i == net->ipv6.fib6_null_entry) {
1223                 fn = fib6_backtrack(fn, &fl6->saddr);
1224                 if (fn)
1225                         goto restart;
1226
1227                 rt = net->ipv6.ip6_null_entry;
1228                 dst_hold(&rt->dst);
1229                 goto out;
1230         } else if (res.fib6_flags & RTF_REJECT) {
1231                 goto do_create;
1232         }
1233
1234         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1235                          fl6->flowi6_oif != 0, skb, flags);
1236
1237         /* Search through exception table */
1238         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1239         if (rt) {
1240                 if (ip6_hold_safe(net, &rt))
1241                         dst_use_noref(&rt->dst, jiffies);
1242         } else {
1243 do_create:
1244                 rt = ip6_create_rt_rcu(&res);
1245         }
1246
1247 out:
1248         trace_fib6_table_lookup(net, &res, table, fl6);
1249
1250         rcu_read_unlock();
1251
1252         return rt;
1253 }
1254
1255 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1256                                    const struct sk_buff *skb, int flags)
1257 {
1258         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1259 }
1260 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1261
1262 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1263                             const struct in6_addr *saddr, int oif,
1264                             const struct sk_buff *skb, int strict)
1265 {
1266         struct flowi6 fl6 = {
1267                 .flowi6_oif = oif,
1268                 .daddr = *daddr,
1269         };
1270         struct dst_entry *dst;
1271         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1272
1273         if (saddr) {
1274                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1275                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1276         }
1277
1278         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1279         if (dst->error == 0)
1280                 return (struct rt6_info *) dst;
1281
1282         dst_release(dst);
1283
1284         return NULL;
1285 }
1286 EXPORT_SYMBOL(rt6_lookup);
1287
1288 /* ip6_ins_rt is called with FREE table->tb6_lock.
1289  * It takes new route entry, the addition fails by any reason the
1290  * route is released.
1291  * Caller must hold dst before calling it.
1292  */
1293
1294 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1295                         struct netlink_ext_ack *extack)
1296 {
1297         int err;
1298         struct fib6_table *table;
1299
1300         table = rt->fib6_table;
1301         spin_lock_bh(&table->tb6_lock);
1302         err = fib6_add(&table->tb6_root, rt, info, extack);
1303         spin_unlock_bh(&table->tb6_lock);
1304
1305         return err;
1306 }
1307
1308 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1309 {
1310         struct nl_info info = { .nl_net = net, };
1311
1312         return __ip6_ins_rt(rt, &info, NULL);
1313 }
1314
1315 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1316                                            const struct in6_addr *daddr,
1317                                            const struct in6_addr *saddr)
1318 {
1319         struct fib6_info *f6i = res->f6i;
1320         struct net_device *dev;
1321         struct rt6_info *rt;
1322
1323         /*
1324          *      Clone the route.
1325          */
1326
1327         if (!fib6_info_hold_safe(f6i))
1328                 return NULL;
1329
1330         dev = ip6_rt_get_dev_rcu(res);
1331         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1332         if (!rt) {
1333                 fib6_info_release(f6i);
1334                 return NULL;
1335         }
1336
1337         ip6_rt_copy_init(rt, res);
1338         rt->rt6i_flags |= RTF_CACHE;
1339         rt->rt6i_dst.addr = *daddr;
1340         rt->rt6i_dst.plen = 128;
1341
1342         if (!rt6_is_gw_or_nonexthop(res)) {
1343                 if (f6i->fib6_dst.plen != 128 &&
1344                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1345                         rt->rt6i_flags |= RTF_ANYCAST;
1346 #ifdef CONFIG_IPV6_SUBTREES
1347                 if (rt->rt6i_src.plen && saddr) {
1348                         rt->rt6i_src.addr = *saddr;
1349                         rt->rt6i_src.plen = 128;
1350                 }
1351 #endif
1352         }
1353
1354         return rt;
1355 }
1356
1357 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1358 {
1359         struct fib6_info *f6i = res->f6i;
1360         unsigned short flags = fib6_info_dst_flags(f6i);
1361         struct net_device *dev;
1362         struct rt6_info *pcpu_rt;
1363
1364         if (!fib6_info_hold_safe(f6i))
1365                 return NULL;
1366
1367         rcu_read_lock();
1368         dev = ip6_rt_get_dev_rcu(res);
1369         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1370         rcu_read_unlock();
1371         if (!pcpu_rt) {
1372                 fib6_info_release(f6i);
1373                 return NULL;
1374         }
1375         ip6_rt_copy_init(pcpu_rt, res);
1376         pcpu_rt->rt6i_flags |= RTF_PCPU;
1377
1378         if (f6i->nh)
1379                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1380
1381         return pcpu_rt;
1382 }
1383
1384 static bool rt6_is_valid(const struct rt6_info *rt6)
1385 {
1386         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1387 }
1388
1389 /* It should be called with rcu_read_lock() acquired */
1390 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1391 {
1392         struct rt6_info *pcpu_rt;
1393
1394         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1395
1396         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1397                 struct rt6_info *prev, **p;
1398
1399                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1400                 prev = xchg(p, NULL);
1401                 if (prev) {
1402                         dst_dev_put(&prev->dst);
1403                         dst_release(&prev->dst);
1404                 }
1405
1406                 pcpu_rt = NULL;
1407         }
1408
1409         return pcpu_rt;
1410 }
1411
1412 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1413                                             const struct fib6_result *res)
1414 {
1415         struct rt6_info *pcpu_rt, *prev, **p;
1416
1417         pcpu_rt = ip6_rt_pcpu_alloc(res);
1418         if (!pcpu_rt)
1419                 return NULL;
1420
1421         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1422         prev = cmpxchg(p, NULL, pcpu_rt);
1423         BUG_ON(prev);
1424
1425         if (res->f6i->fib6_destroying) {
1426                 struct fib6_info *from;
1427
1428                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1429                 fib6_info_release(from);
1430         }
1431
1432         return pcpu_rt;
1433 }
1434
1435 /* exception hash table implementation
1436  */
1437 static DEFINE_SPINLOCK(rt6_exception_lock);
1438
1439 /* Remove rt6_ex from hash table and free the memory
1440  * Caller must hold rt6_exception_lock
1441  */
1442 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1443                                  struct rt6_exception *rt6_ex)
1444 {
1445         struct fib6_info *from;
1446         struct net *net;
1447
1448         if (!bucket || !rt6_ex)
1449                 return;
1450
1451         net = dev_net(rt6_ex->rt6i->dst.dev);
1452         net->ipv6.rt6_stats->fib_rt_cache--;
1453
1454         /* purge completely the exception to allow releasing the held resources:
1455          * some [sk] cache may keep the dst around for unlimited time
1456          */
1457         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1458         fib6_info_release(from);
1459         dst_dev_put(&rt6_ex->rt6i->dst);
1460
1461         hlist_del_rcu(&rt6_ex->hlist);
1462         dst_release(&rt6_ex->rt6i->dst);
1463         kfree_rcu(rt6_ex, rcu);
1464         WARN_ON_ONCE(!bucket->depth);
1465         bucket->depth--;
1466 }
1467
1468 /* Remove oldest rt6_ex in bucket and free the memory
1469  * Caller must hold rt6_exception_lock
1470  */
1471 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1472 {
1473         struct rt6_exception *rt6_ex, *oldest = NULL;
1474
1475         if (!bucket)
1476                 return;
1477
1478         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1479                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1480                         oldest = rt6_ex;
1481         }
1482         rt6_remove_exception(bucket, oldest);
1483 }
1484
1485 static u32 rt6_exception_hash(const struct in6_addr *dst,
1486                               const struct in6_addr *src)
1487 {
1488         static siphash_key_t rt6_exception_key __read_mostly;
1489         struct {
1490                 struct in6_addr dst;
1491                 struct in6_addr src;
1492         } __aligned(SIPHASH_ALIGNMENT) combined = {
1493                 .dst = *dst,
1494         };
1495         u64 val;
1496
1497         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1498
1499 #ifdef CONFIG_IPV6_SUBTREES
1500         if (src)
1501                 combined.src = *src;
1502 #endif
1503         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1504
1505         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1506 }
1507
1508 /* Helper function to find the cached rt in the hash table
1509  * and update bucket pointer to point to the bucket for this
1510  * (daddr, saddr) pair
1511  * Caller must hold rt6_exception_lock
1512  */
1513 static struct rt6_exception *
1514 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1515                               const struct in6_addr *daddr,
1516                               const struct in6_addr *saddr)
1517 {
1518         struct rt6_exception *rt6_ex;
1519         u32 hval;
1520
1521         if (!(*bucket) || !daddr)
1522                 return NULL;
1523
1524         hval = rt6_exception_hash(daddr, saddr);
1525         *bucket += hval;
1526
1527         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1528                 struct rt6_info *rt6 = rt6_ex->rt6i;
1529                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1530
1531 #ifdef CONFIG_IPV6_SUBTREES
1532                 if (matched && saddr)
1533                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1534 #endif
1535                 if (matched)
1536                         return rt6_ex;
1537         }
1538         return NULL;
1539 }
1540
1541 /* Helper function to find the cached rt in the hash table
1542  * and update bucket pointer to point to the bucket for this
1543  * (daddr, saddr) pair
1544  * Caller must hold rcu_read_lock()
1545  */
1546 static struct rt6_exception *
1547 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1548                          const struct in6_addr *daddr,
1549                          const struct in6_addr *saddr)
1550 {
1551         struct rt6_exception *rt6_ex;
1552         u32 hval;
1553
1554         WARN_ON_ONCE(!rcu_read_lock_held());
1555
1556         if (!(*bucket) || !daddr)
1557                 return NULL;
1558
1559         hval = rt6_exception_hash(daddr, saddr);
1560         *bucket += hval;
1561
1562         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1563                 struct rt6_info *rt6 = rt6_ex->rt6i;
1564                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1565
1566 #ifdef CONFIG_IPV6_SUBTREES
1567                 if (matched && saddr)
1568                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1569 #endif
1570                 if (matched)
1571                         return rt6_ex;
1572         }
1573         return NULL;
1574 }
1575
1576 static unsigned int fib6_mtu(const struct fib6_result *res)
1577 {
1578         const struct fib6_nh *nh = res->nh;
1579         unsigned int mtu;
1580
1581         if (res->f6i->fib6_pmtu) {
1582                 mtu = res->f6i->fib6_pmtu;
1583         } else {
1584                 struct net_device *dev = nh->fib_nh_dev;
1585                 struct inet6_dev *idev;
1586
1587                 rcu_read_lock();
1588                 idev = __in6_dev_get(dev);
1589                 mtu = idev->cnf.mtu6;
1590                 rcu_read_unlock();
1591         }
1592
1593         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1594
1595         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1596 }
1597
1598 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1599
1600 /* used when the flushed bit is not relevant, only access to the bucket
1601  * (ie., all bucket users except rt6_insert_exception);
1602  *
1603  * called under rcu lock; sometimes called with rt6_exception_lock held
1604  */
1605 static
1606 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1607                                                        spinlock_t *lock)
1608 {
1609         struct rt6_exception_bucket *bucket;
1610
1611         if (lock)
1612                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1613                                                    lockdep_is_held(lock));
1614         else
1615                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1616
1617         /* remove bucket flushed bit if set */
1618         if (bucket) {
1619                 unsigned long p = (unsigned long)bucket;
1620
1621                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1622                 bucket = (struct rt6_exception_bucket *)p;
1623         }
1624
1625         return bucket;
1626 }
1627
1628 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1629 {
1630         unsigned long p = (unsigned long)bucket;
1631
1632         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1633 }
1634
1635 /* called with rt6_exception_lock held */
1636 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1637                                               spinlock_t *lock)
1638 {
1639         struct rt6_exception_bucket *bucket;
1640         unsigned long p;
1641
1642         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1643                                            lockdep_is_held(lock));
1644
1645         p = (unsigned long)bucket;
1646         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1647         bucket = (struct rt6_exception_bucket *)p;
1648         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1649 }
1650
1651 static int rt6_insert_exception(struct rt6_info *nrt,
1652                                 const struct fib6_result *res)
1653 {
1654         struct net *net = dev_net(nrt->dst.dev);
1655         struct rt6_exception_bucket *bucket;
1656         struct fib6_info *f6i = res->f6i;
1657         struct in6_addr *src_key = NULL;
1658         struct rt6_exception *rt6_ex;
1659         struct fib6_nh *nh = res->nh;
1660         int err = 0;
1661
1662         spin_lock_bh(&rt6_exception_lock);
1663
1664         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1665                                           lockdep_is_held(&rt6_exception_lock));
1666         if (!bucket) {
1667                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1668                                  GFP_ATOMIC);
1669                 if (!bucket) {
1670                         err = -ENOMEM;
1671                         goto out;
1672                 }
1673                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1674         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1675                 err = -EINVAL;
1676                 goto out;
1677         }
1678
1679 #ifdef CONFIG_IPV6_SUBTREES
1680         /* fib6_src.plen != 0 indicates f6i is in subtree
1681          * and exception table is indexed by a hash of
1682          * both fib6_dst and fib6_src.
1683          * Otherwise, the exception table is indexed by
1684          * a hash of only fib6_dst.
1685          */
1686         if (f6i->fib6_src.plen)
1687                 src_key = &nrt->rt6i_src.addr;
1688 #endif
1689         /* rt6_mtu_change() might lower mtu on f6i.
1690          * Only insert this exception route if its mtu
1691          * is less than f6i's mtu value.
1692          */
1693         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1694                 err = -EINVAL;
1695                 goto out;
1696         }
1697
1698         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1699                                                src_key);
1700         if (rt6_ex)
1701                 rt6_remove_exception(bucket, rt6_ex);
1702
1703         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1704         if (!rt6_ex) {
1705                 err = -ENOMEM;
1706                 goto out;
1707         }
1708         rt6_ex->rt6i = nrt;
1709         rt6_ex->stamp = jiffies;
1710         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1711         bucket->depth++;
1712         net->ipv6.rt6_stats->fib_rt_cache++;
1713
1714         if (bucket->depth > FIB6_MAX_DEPTH)
1715                 rt6_exception_remove_oldest(bucket);
1716
1717 out:
1718         spin_unlock_bh(&rt6_exception_lock);
1719
1720         /* Update fn->fn_sernum to invalidate all cached dst */
1721         if (!err) {
1722                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1723                 fib6_update_sernum(net, f6i);
1724                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1725                 fib6_force_start_gc(net);
1726         }
1727
1728         return err;
1729 }
1730
1731 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1732 {
1733         struct rt6_exception_bucket *bucket;
1734         struct rt6_exception *rt6_ex;
1735         struct hlist_node *tmp;
1736         int i;
1737
1738         spin_lock_bh(&rt6_exception_lock);
1739
1740         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1741         if (!bucket)
1742                 goto out;
1743
1744         /* Prevent rt6_insert_exception() to recreate the bucket list */
1745         if (!from)
1746                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1747
1748         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1749                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1750                         if (!from ||
1751                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1752                                 rt6_remove_exception(bucket, rt6_ex);
1753                 }
1754                 WARN_ON_ONCE(!from && bucket->depth);
1755                 bucket++;
1756         }
1757 out:
1758         spin_unlock_bh(&rt6_exception_lock);
1759 }
1760
1761 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1762 {
1763         struct fib6_info *f6i = arg;
1764
1765         fib6_nh_flush_exceptions(nh, f6i);
1766
1767         return 0;
1768 }
1769
1770 void rt6_flush_exceptions(struct fib6_info *f6i)
1771 {
1772         if (f6i->nh)
1773                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1774                                          f6i);
1775         else
1776                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1777 }
1778
1779 /* Find cached rt in the hash table inside passed in rt
1780  * Caller has to hold rcu_read_lock()
1781  */
1782 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1783                                            const struct in6_addr *daddr,
1784                                            const struct in6_addr *saddr)
1785 {
1786         const struct in6_addr *src_key = NULL;
1787         struct rt6_exception_bucket *bucket;
1788         struct rt6_exception *rt6_ex;
1789         struct rt6_info *ret = NULL;
1790
1791 #ifdef CONFIG_IPV6_SUBTREES
1792         /* fib6i_src.plen != 0 indicates f6i is in subtree
1793          * and exception table is indexed by a hash of
1794          * both fib6_dst and fib6_src.
1795          * However, the src addr used to create the hash
1796          * might not be exactly the passed in saddr which
1797          * is a /128 addr from the flow.
1798          * So we need to use f6i->fib6_src to redo lookup
1799          * if the passed in saddr does not find anything.
1800          * (See the logic in ip6_rt_cache_alloc() on how
1801          * rt->rt6i_src is updated.)
1802          */
1803         if (res->f6i->fib6_src.plen)
1804                 src_key = saddr;
1805 find_ex:
1806 #endif
1807         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1808         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1809
1810         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1811                 ret = rt6_ex->rt6i;
1812
1813 #ifdef CONFIG_IPV6_SUBTREES
1814         /* Use fib6_src as src_key and redo lookup */
1815         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1816                 src_key = &res->f6i->fib6_src.addr;
1817                 goto find_ex;
1818         }
1819 #endif
1820
1821         return ret;
1822 }
1823
1824 /* Remove the passed in cached rt from the hash table that contains it */
1825 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1826                                     const struct rt6_info *rt)
1827 {
1828         const struct in6_addr *src_key = NULL;
1829         struct rt6_exception_bucket *bucket;
1830         struct rt6_exception *rt6_ex;
1831         int err;
1832
1833         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1834                 return -ENOENT;
1835
1836         spin_lock_bh(&rt6_exception_lock);
1837         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1838
1839 #ifdef CONFIG_IPV6_SUBTREES
1840         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1841          * and exception table is indexed by a hash of
1842          * both rt6i_dst and rt6i_src.
1843          * Otherwise, the exception table is indexed by
1844          * a hash of only rt6i_dst.
1845          */
1846         if (plen)
1847                 src_key = &rt->rt6i_src.addr;
1848 #endif
1849         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1850                                                &rt->rt6i_dst.addr,
1851                                                src_key);
1852         if (rt6_ex) {
1853                 rt6_remove_exception(bucket, rt6_ex);
1854                 err = 0;
1855         } else {
1856                 err = -ENOENT;
1857         }
1858
1859         spin_unlock_bh(&rt6_exception_lock);
1860         return err;
1861 }
1862
1863 struct fib6_nh_excptn_arg {
1864         struct rt6_info *rt;
1865         int             plen;
1866 };
1867
1868 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1869 {
1870         struct fib6_nh_excptn_arg *arg = _arg;
1871         int err;
1872
1873         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1874         if (err == 0)
1875                 return 1;
1876
1877         return 0;
1878 }
1879
1880 static int rt6_remove_exception_rt(struct rt6_info *rt)
1881 {
1882         struct fib6_info *from;
1883
1884         from = rcu_dereference(rt->from);
1885         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1886                 return -EINVAL;
1887
1888         if (from->nh) {
1889                 struct fib6_nh_excptn_arg arg = {
1890                         .rt = rt,
1891                         .plen = from->fib6_src.plen
1892                 };
1893                 int rc;
1894
1895                 /* rc = 1 means an entry was found */
1896                 rc = nexthop_for_each_fib6_nh(from->nh,
1897                                               rt6_nh_remove_exception_rt,
1898                                               &arg);
1899                 return rc ? 0 : -ENOENT;
1900         }
1901
1902         return fib6_nh_remove_exception(from->fib6_nh,
1903                                         from->fib6_src.plen, rt);
1904 }
1905
1906 /* Find rt6_ex which contains the passed in rt cache and
1907  * refresh its stamp
1908  */
1909 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1910                                      const struct rt6_info *rt)
1911 {
1912         const struct in6_addr *src_key = NULL;
1913         struct rt6_exception_bucket *bucket;
1914         struct rt6_exception *rt6_ex;
1915
1916         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1917 #ifdef CONFIG_IPV6_SUBTREES
1918         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1919          * and exception table is indexed by a hash of
1920          * both rt6i_dst and rt6i_src.
1921          * Otherwise, the exception table is indexed by
1922          * a hash of only rt6i_dst.
1923          */
1924         if (plen)
1925                 src_key = &rt->rt6i_src.addr;
1926 #endif
1927         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1928         if (rt6_ex)
1929                 rt6_ex->stamp = jiffies;
1930 }
1931
1932 struct fib6_nh_match_arg {
1933         const struct net_device *dev;
1934         const struct in6_addr   *gw;
1935         struct fib6_nh          *match;
1936 };
1937
1938 /* determine if fib6_nh has given device and gateway */
1939 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1940 {
1941         struct fib6_nh_match_arg *arg = _arg;
1942
1943         if (arg->dev != nh->fib_nh_dev ||
1944             (arg->gw && !nh->fib_nh_gw_family) ||
1945             (!arg->gw && nh->fib_nh_gw_family) ||
1946             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1947                 return 0;
1948
1949         arg->match = nh;
1950
1951         /* found a match, break the loop */
1952         return 1;
1953 }
1954
1955 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1956 {
1957         struct fib6_info *from;
1958         struct fib6_nh *fib6_nh;
1959
1960         rcu_read_lock();
1961
1962         from = rcu_dereference(rt->from);
1963         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1964                 goto unlock;
1965
1966         if (from->nh) {
1967                 struct fib6_nh_match_arg arg = {
1968                         .dev = rt->dst.dev,
1969                         .gw = &rt->rt6i_gateway,
1970                 };
1971
1972                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1973
1974                 if (!arg.match)
1975                         goto unlock;
1976                 fib6_nh = arg.match;
1977         } else {
1978                 fib6_nh = from->fib6_nh;
1979         }
1980         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1981 unlock:
1982         rcu_read_unlock();
1983 }
1984
1985 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1986                                          struct rt6_info *rt, int mtu)
1987 {
1988         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1989          * lowest MTU in the path: always allow updating the route PMTU to
1990          * reflect PMTU decreases.
1991          *
1992          * If the new MTU is higher, and the route PMTU is equal to the local
1993          * MTU, this means the old MTU is the lowest in the path, so allow
1994          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1995          * handle this.
1996          */
1997
1998         if (dst_mtu(&rt->dst) >= mtu)
1999                 return true;
2000
2001         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2002                 return true;
2003
2004         return false;
2005 }
2006
2007 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2008                                        const struct fib6_nh *nh, int mtu)
2009 {
2010         struct rt6_exception_bucket *bucket;
2011         struct rt6_exception *rt6_ex;
2012         int i;
2013
2014         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2015         if (!bucket)
2016                 return;
2017
2018         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2019                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2020                         struct rt6_info *entry = rt6_ex->rt6i;
2021
2022                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2023                          * route), the metrics of its rt->from have already
2024                          * been updated.
2025                          */
2026                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2027                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2028                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2029                 }
2030                 bucket++;
2031         }
2032 }
2033
2034 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2035
2036 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2037                                             const struct in6_addr *gateway)
2038 {
2039         struct rt6_exception_bucket *bucket;
2040         struct rt6_exception *rt6_ex;
2041         struct hlist_node *tmp;
2042         int i;
2043
2044         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2045                 return;
2046
2047         spin_lock_bh(&rt6_exception_lock);
2048         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2049         if (bucket) {
2050                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2051                         hlist_for_each_entry_safe(rt6_ex, tmp,
2052                                                   &bucket->chain, hlist) {
2053                                 struct rt6_info *entry = rt6_ex->rt6i;
2054
2055                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2056                                     RTF_CACHE_GATEWAY &&
2057                                     ipv6_addr_equal(gateway,
2058                                                     &entry->rt6i_gateway)) {
2059                                         rt6_remove_exception(bucket, rt6_ex);
2060                                 }
2061                         }
2062                         bucket++;
2063                 }
2064         }
2065
2066         spin_unlock_bh(&rt6_exception_lock);
2067 }
2068
2069 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2070                                       struct rt6_exception *rt6_ex,
2071                                       struct fib6_gc_args *gc_args,
2072                                       unsigned long now)
2073 {
2074         struct rt6_info *rt = rt6_ex->rt6i;
2075
2076         /* we are pruning and obsoleting aged-out and non gateway exceptions
2077          * even if others have still references to them, so that on next
2078          * dst_check() such references can be dropped.
2079          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2080          * expired, independently from their aging, as per RFC 8201 section 4
2081          */
2082         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2083                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2084                         RT6_TRACE("aging clone %p\n", rt);
2085                         rt6_remove_exception(bucket, rt6_ex);
2086                         return;
2087                 }
2088         } else if (time_after(jiffies, rt->dst.expires)) {
2089                 RT6_TRACE("purging expired route %p\n", rt);
2090                 rt6_remove_exception(bucket, rt6_ex);
2091                 return;
2092         }
2093
2094         if (rt->rt6i_flags & RTF_GATEWAY) {
2095                 struct neighbour *neigh;
2096
2097                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2098
2099                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2100                         RT6_TRACE("purging route %p via non-router but gateway\n",
2101                                   rt);
2102                         rt6_remove_exception(bucket, rt6_ex);
2103                         return;
2104                 }
2105         }
2106
2107         gc_args->more++;
2108 }
2109
2110 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2111                                    struct fib6_gc_args *gc_args,
2112                                    unsigned long now)
2113 {
2114         struct rt6_exception_bucket *bucket;
2115         struct rt6_exception *rt6_ex;
2116         struct hlist_node *tmp;
2117         int i;
2118
2119         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2120                 return;
2121
2122         rcu_read_lock_bh();
2123         spin_lock(&rt6_exception_lock);
2124         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2125         if (bucket) {
2126                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2127                         hlist_for_each_entry_safe(rt6_ex, tmp,
2128                                                   &bucket->chain, hlist) {
2129                                 rt6_age_examine_exception(bucket, rt6_ex,
2130                                                           gc_args, now);
2131                         }
2132                         bucket++;
2133                 }
2134         }
2135         spin_unlock(&rt6_exception_lock);
2136         rcu_read_unlock_bh();
2137 }
2138
2139 struct fib6_nh_age_excptn_arg {
2140         struct fib6_gc_args     *gc_args;
2141         unsigned long           now;
2142 };
2143
2144 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2145 {
2146         struct fib6_nh_age_excptn_arg *arg = _arg;
2147
2148         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2149         return 0;
2150 }
2151
2152 void rt6_age_exceptions(struct fib6_info *f6i,
2153                         struct fib6_gc_args *gc_args,
2154                         unsigned long now)
2155 {
2156         if (f6i->nh) {
2157                 struct fib6_nh_age_excptn_arg arg = {
2158                         .gc_args = gc_args,
2159                         .now = now
2160                 };
2161
2162                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2163                                          &arg);
2164         } else {
2165                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2166         }
2167 }
2168
2169 /* must be called with rcu lock held */
2170 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2171                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2172 {
2173         struct fib6_node *fn, *saved_fn;
2174
2175         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2176         saved_fn = fn;
2177
2178         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2179                 oif = 0;
2180
2181 redo_rt6_select:
2182         rt6_select(net, fn, oif, res, strict);
2183         if (res->f6i == net->ipv6.fib6_null_entry) {
2184                 fn = fib6_backtrack(fn, &fl6->saddr);
2185                 if (fn)
2186                         goto redo_rt6_select;
2187                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2188                         /* also consider unreachable route */
2189                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2190                         fn = saved_fn;
2191                         goto redo_rt6_select;
2192                 }
2193         }
2194
2195         trace_fib6_table_lookup(net, res, table, fl6);
2196
2197         return 0;
2198 }
2199
2200 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2201                                int oif, struct flowi6 *fl6,
2202                                const struct sk_buff *skb, int flags)
2203 {
2204         struct fib6_result res = {};
2205         struct rt6_info *rt = NULL;
2206         int strict = 0;
2207
2208         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2209                      !rcu_read_lock_held());
2210
2211         strict |= flags & RT6_LOOKUP_F_IFACE;
2212         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2213         if (net->ipv6.devconf_all->forwarding == 0)
2214                 strict |= RT6_LOOKUP_F_REACHABLE;
2215
2216         rcu_read_lock();
2217
2218         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2219         if (res.f6i == net->ipv6.fib6_null_entry)
2220                 goto out;
2221
2222         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2223
2224         /*Search through exception table */
2225         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2226         if (rt) {
2227                 goto out;
2228         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2229                             !res.nh->fib_nh_gw_family)) {
2230                 /* Create a RTF_CACHE clone which will not be
2231                  * owned by the fib6 tree.  It is for the special case where
2232                  * the daddr in the skb during the neighbor look-up is different
2233                  * from the fl6->daddr used to look-up route here.
2234                  */
2235                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2236
2237                 if (rt) {
2238                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2239                          * As rt6_uncached_list_add() does not consume refcnt,
2240                          * this refcnt is always returned to the caller even
2241                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2242                          */
2243                         rt6_uncached_list_add(rt);
2244                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2245                         rcu_read_unlock();
2246
2247                         return rt;
2248                 }
2249         } else {
2250                 /* Get a percpu copy */
2251                 local_bh_disable();
2252                 rt = rt6_get_pcpu_route(&res);
2253
2254                 if (!rt)
2255                         rt = rt6_make_pcpu_route(net, &res);
2256
2257                 local_bh_enable();
2258         }
2259 out:
2260         if (!rt)
2261                 rt = net->ipv6.ip6_null_entry;
2262         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2263                 ip6_hold_safe(net, &rt);
2264         rcu_read_unlock();
2265
2266         return rt;
2267 }
2268 EXPORT_SYMBOL_GPL(ip6_pol_route);
2269
2270 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2271                                             struct fib6_table *table,
2272                                             struct flowi6 *fl6,
2273                                             const struct sk_buff *skb,
2274                                             int flags)
2275 {
2276         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2277 }
2278
2279 struct dst_entry *ip6_route_input_lookup(struct net *net,
2280                                          struct net_device *dev,
2281                                          struct flowi6 *fl6,
2282                                          const struct sk_buff *skb,
2283                                          int flags)
2284 {
2285         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2286                 flags |= RT6_LOOKUP_F_IFACE;
2287
2288         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2289 }
2290 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2291
2292 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2293                                   struct flow_keys *keys,
2294                                   struct flow_keys *flkeys)
2295 {
2296         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2297         const struct ipv6hdr *key_iph = outer_iph;
2298         struct flow_keys *_flkeys = flkeys;
2299         const struct ipv6hdr *inner_iph;
2300         const struct icmp6hdr *icmph;
2301         struct ipv6hdr _inner_iph;
2302         struct icmp6hdr _icmph;
2303
2304         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2305                 goto out;
2306
2307         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2308                                    sizeof(_icmph), &_icmph);
2309         if (!icmph)
2310                 goto out;
2311
2312         if (!icmpv6_is_err(icmph->icmp6_type))
2313                 goto out;
2314
2315         inner_iph = skb_header_pointer(skb,
2316                                        skb_transport_offset(skb) + sizeof(*icmph),
2317                                        sizeof(_inner_iph), &_inner_iph);
2318         if (!inner_iph)
2319                 goto out;
2320
2321         key_iph = inner_iph;
2322         _flkeys = NULL;
2323 out:
2324         if (_flkeys) {
2325                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2326                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2327                 keys->tags.flow_label = _flkeys->tags.flow_label;
2328                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2329         } else {
2330                 keys->addrs.v6addrs.src = key_iph->saddr;
2331                 keys->addrs.v6addrs.dst = key_iph->daddr;
2332                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2333                 keys->basic.ip_proto = key_iph->nexthdr;
2334         }
2335 }
2336
2337 /* if skb is set it will be used and fl6 can be NULL */
2338 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2339                        const struct sk_buff *skb, struct flow_keys *flkeys)
2340 {
2341         struct flow_keys hash_keys;
2342         u32 mhash;
2343
2344         switch (ip6_multipath_hash_policy(net)) {
2345         case 0:
2346                 memset(&hash_keys, 0, sizeof(hash_keys));
2347                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2348                 if (skb) {
2349                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2350                 } else {
2351                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2352                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2353                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2354                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2355                 }
2356                 break;
2357         case 1:
2358                 if (skb) {
2359                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2360                         struct flow_keys keys;
2361
2362                         /* short-circuit if we already have L4 hash present */
2363                         if (skb->l4_hash)
2364                                 return skb_get_hash_raw(skb) >> 1;
2365
2366                         memset(&hash_keys, 0, sizeof(hash_keys));
2367
2368                         if (!flkeys) {
2369                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2370                                 flkeys = &keys;
2371                         }
2372                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2373                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2374                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2375                         hash_keys.ports.src = flkeys->ports.src;
2376                         hash_keys.ports.dst = flkeys->ports.dst;
2377                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2378                 } else {
2379                         memset(&hash_keys, 0, sizeof(hash_keys));
2380                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2381                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2382                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2383                         hash_keys.ports.src = fl6->fl6_sport;
2384                         hash_keys.ports.dst = fl6->fl6_dport;
2385                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2386                 }
2387                 break;
2388         case 2:
2389                 memset(&hash_keys, 0, sizeof(hash_keys));
2390                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2391                 if (skb) {
2392                         struct flow_keys keys;
2393
2394                         if (!flkeys) {
2395                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2396                                 flkeys = &keys;
2397                         }
2398
2399                         /* Inner can be v4 or v6 */
2400                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2401                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2402                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2403                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2404                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2405                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2406                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2407                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2408                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2409                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2410                         } else {
2411                                 /* Same as case 0 */
2412                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2413                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2414                         }
2415                 } else {
2416                         /* Same as case 0 */
2417                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2418                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2419                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2420                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2421                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2422                 }
2423                 break;
2424         }
2425         mhash = flow_hash_from_keys(&hash_keys);
2426
2427         return mhash >> 1;
2428 }
2429
2430 /* Called with rcu held */
2431 void ip6_route_input(struct sk_buff *skb)
2432 {
2433         const struct ipv6hdr *iph = ipv6_hdr(skb);
2434         struct net *net = dev_net(skb->dev);
2435         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2436         struct ip_tunnel_info *tun_info;
2437         struct flowi6 fl6 = {
2438                 .flowi6_iif = skb->dev->ifindex,
2439                 .daddr = iph->daddr,
2440                 .saddr = iph->saddr,
2441                 .flowlabel = ip6_flowinfo(iph),
2442                 .flowi6_mark = skb->mark,
2443                 .flowi6_proto = iph->nexthdr,
2444         };
2445         struct flow_keys *flkeys = NULL, _flkeys;
2446
2447         tun_info = skb_tunnel_info(skb);
2448         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2449                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2450
2451         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2452                 flkeys = &_flkeys;
2453
2454         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2455                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2456         skb_dst_drop(skb);
2457         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2458                                                       &fl6, skb, flags));
2459 }
2460
2461 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2462                                              struct fib6_table *table,
2463                                              struct flowi6 *fl6,
2464                                              const struct sk_buff *skb,
2465                                              int flags)
2466 {
2467         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2468 }
2469
2470 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2471                                                const struct sock *sk,
2472                                                struct flowi6 *fl6, int flags)
2473 {
2474         bool any_src;
2475
2476         if (ipv6_addr_type(&fl6->daddr) &
2477             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2478                 struct dst_entry *dst;
2479
2480                 /* This function does not take refcnt on the dst */
2481                 dst = l3mdev_link_scope_lookup(net, fl6);
2482                 if (dst)
2483                         return dst;
2484         }
2485
2486         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2487
2488         flags |= RT6_LOOKUP_F_DST_NOREF;
2489         any_src = ipv6_addr_any(&fl6->saddr);
2490         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2491             (fl6->flowi6_oif && any_src))
2492                 flags |= RT6_LOOKUP_F_IFACE;
2493
2494         if (!any_src)
2495                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2496         else if (sk)
2497                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2498
2499         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2500 }
2501 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2502
2503 struct dst_entry *ip6_route_output_flags(struct net *net,
2504                                          const struct sock *sk,
2505                                          struct flowi6 *fl6,
2506                                          int flags)
2507 {
2508         struct dst_entry *dst;
2509         struct rt6_info *rt6;
2510
2511         rcu_read_lock();
2512         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2513         rt6 = (struct rt6_info *)dst;
2514         /* For dst cached in uncached_list, refcnt is already taken. */
2515         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2516                 dst = &net->ipv6.ip6_null_entry->dst;
2517                 dst_hold(dst);
2518         }
2519         rcu_read_unlock();
2520
2521         return dst;
2522 }
2523 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2524
2525 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2526 {
2527         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2528         struct net_device *loopback_dev = net->loopback_dev;
2529         struct dst_entry *new = NULL;
2530
2531         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2532                        DST_OBSOLETE_DEAD, 0);
2533         if (rt) {
2534                 rt6_info_init(rt);
2535                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2536
2537                 new = &rt->dst;
2538                 new->__use = 1;
2539                 new->input = dst_discard;
2540                 new->output = dst_discard_out;
2541
2542                 dst_copy_metrics(new, &ort->dst);
2543
2544                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2545                 rt->rt6i_gateway = ort->rt6i_gateway;
2546                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2547
2548                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2549 #ifdef CONFIG_IPV6_SUBTREES
2550                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2551 #endif
2552         }
2553
2554         dst_release(dst_orig);
2555         return new ? new : ERR_PTR(-ENOMEM);
2556 }
2557
2558 /*
2559  *      Destination cache support functions
2560  */
2561
2562 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2563 {
2564         u32 rt_cookie = 0;
2565
2566         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2567                 return false;
2568
2569         if (fib6_check_expired(f6i))
2570                 return false;
2571
2572         return true;
2573 }
2574
2575 static struct dst_entry *rt6_check(struct rt6_info *rt,
2576                                    struct fib6_info *from,
2577                                    u32 cookie)
2578 {
2579         u32 rt_cookie = 0;
2580
2581         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2582             rt_cookie != cookie)
2583                 return NULL;
2584
2585         if (rt6_check_expired(rt))
2586                 return NULL;
2587
2588         return &rt->dst;
2589 }
2590
2591 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2592                                             struct fib6_info *from,
2593                                             u32 cookie)
2594 {
2595         if (!__rt6_check_expired(rt) &&
2596             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2597             fib6_check(from, cookie))
2598                 return &rt->dst;
2599         else
2600                 return NULL;
2601 }
2602
2603 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2604                                                         u32 cookie)
2605 {
2606         struct dst_entry *dst_ret;
2607         struct fib6_info *from;
2608         struct rt6_info *rt;
2609
2610         rt = container_of(dst, struct rt6_info, dst);
2611
2612         if (rt->sernum)
2613                 return rt6_is_valid(rt) ? dst : NULL;
2614
2615         rcu_read_lock();
2616
2617         /* All IPV6 dsts are created with ->obsolete set to the value
2618          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2619          * into this function always.
2620          */
2621
2622         from = rcu_dereference(rt->from);
2623
2624         if (from && (rt->rt6i_flags & RTF_PCPU ||
2625             unlikely(!list_empty(&rt->rt6i_uncached))))
2626                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2627         else
2628                 dst_ret = rt6_check(rt, from, cookie);
2629
2630         rcu_read_unlock();
2631
2632         return dst_ret;
2633 }
2634 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2635
2636 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2637 {
2638         struct rt6_info *rt = (struct rt6_info *) dst;
2639
2640         if (rt) {
2641                 if (rt->rt6i_flags & RTF_CACHE) {
2642                         rcu_read_lock();
2643                         if (rt6_check_expired(rt)) {
2644                                 rt6_remove_exception_rt(rt);
2645                                 dst = NULL;
2646                         }
2647                         rcu_read_unlock();
2648                 } else {
2649                         dst_release(dst);
2650                         dst = NULL;
2651                 }
2652         }
2653         return dst;
2654 }
2655
2656 static void ip6_link_failure(struct sk_buff *skb)
2657 {
2658         struct rt6_info *rt;
2659
2660         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2661
2662         rt = (struct rt6_info *) skb_dst(skb);
2663         if (rt) {
2664                 rcu_read_lock();
2665                 if (rt->rt6i_flags & RTF_CACHE) {
2666                         rt6_remove_exception_rt(rt);
2667                 } else {
2668                         struct fib6_info *from;
2669                         struct fib6_node *fn;
2670
2671                         from = rcu_dereference(rt->from);
2672                         if (from) {
2673                                 fn = rcu_dereference(from->fib6_node);
2674                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2675                                         fn->fn_sernum = -1;
2676                         }
2677                 }
2678                 rcu_read_unlock();
2679         }
2680 }
2681
2682 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2683 {
2684         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2685                 struct fib6_info *from;
2686
2687                 rcu_read_lock();
2688                 from = rcu_dereference(rt0->from);
2689                 if (from)
2690                         rt0->dst.expires = from->expires;
2691                 rcu_read_unlock();
2692         }
2693
2694         dst_set_expires(&rt0->dst, timeout);
2695         rt0->rt6i_flags |= RTF_EXPIRES;
2696 }
2697
2698 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2699 {
2700         struct net *net = dev_net(rt->dst.dev);
2701
2702         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2703         rt->rt6i_flags |= RTF_MODIFIED;
2704         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2705 }
2706
2707 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2708 {
2709         return !(rt->rt6i_flags & RTF_CACHE) &&
2710                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2711 }
2712
2713 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2714                                  const struct ipv6hdr *iph, u32 mtu,
2715                                  bool confirm_neigh)
2716 {
2717         const struct in6_addr *daddr, *saddr;
2718         struct rt6_info *rt6 = (struct rt6_info *)dst;
2719
2720         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2721          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2722          * [see also comment in rt6_mtu_change_route()]
2723          */
2724
2725         if (iph) {
2726                 daddr = &iph->daddr;
2727                 saddr = &iph->saddr;
2728         } else if (sk) {
2729                 daddr = &sk->sk_v6_daddr;
2730                 saddr = &inet6_sk(sk)->saddr;
2731         } else {
2732                 daddr = NULL;
2733                 saddr = NULL;
2734         }
2735
2736         if (confirm_neigh)
2737                 dst_confirm_neigh(dst, daddr);
2738
2739         if (mtu < IPV6_MIN_MTU)
2740                 return;
2741         if (mtu >= dst_mtu(dst))
2742                 return;
2743
2744         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2745                 rt6_do_update_pmtu(rt6, mtu);
2746                 /* update rt6_ex->stamp for cache */
2747                 if (rt6->rt6i_flags & RTF_CACHE)
2748                         rt6_update_exception_stamp_rt(rt6);
2749         } else if (daddr) {
2750                 struct fib6_result res = {};
2751                 struct rt6_info *nrt6;
2752
2753                 rcu_read_lock();
2754                 res.f6i = rcu_dereference(rt6->from);
2755                 if (!res.f6i)
2756                         goto out_unlock;
2757
2758                 res.fib6_flags = res.f6i->fib6_flags;
2759                 res.fib6_type = res.f6i->fib6_type;
2760
2761                 if (res.f6i->nh) {
2762                         struct fib6_nh_match_arg arg = {
2763                                 .dev = dst->dev,
2764                                 .gw = &rt6->rt6i_gateway,
2765                         };
2766
2767                         nexthop_for_each_fib6_nh(res.f6i->nh,
2768                                                  fib6_nh_find_match, &arg);
2769
2770                         /* fib6_info uses a nexthop that does not have fib6_nh
2771                          * using the dst->dev + gw. Should be impossible.
2772                          */
2773                         if (!arg.match)
2774                                 goto out_unlock;
2775
2776                         res.nh = arg.match;
2777                 } else {
2778                         res.nh = res.f6i->fib6_nh;
2779                 }
2780
2781                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2782                 if (nrt6) {
2783                         rt6_do_update_pmtu(nrt6, mtu);
2784                         if (rt6_insert_exception(nrt6, &res))
2785                                 dst_release_immediate(&nrt6->dst);
2786                 }
2787 out_unlock:
2788                 rcu_read_unlock();
2789         }
2790 }
2791
2792 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2793                                struct sk_buff *skb, u32 mtu,
2794                                bool confirm_neigh)
2795 {
2796         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2797                              confirm_neigh);
2798 }
2799
2800 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2801                      int oif, u32 mark, kuid_t uid)
2802 {
2803         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2804         struct dst_entry *dst;
2805         struct flowi6 fl6 = {
2806                 .flowi6_oif = oif,
2807                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2808                 .daddr = iph->daddr,
2809                 .saddr = iph->saddr,
2810                 .flowlabel = ip6_flowinfo(iph),
2811                 .flowi6_uid = uid,
2812         };
2813
2814         dst = ip6_route_output(net, NULL, &fl6);
2815         if (!dst->error)
2816                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2817         dst_release(dst);
2818 }
2819 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2820
2821 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2822 {
2823         int oif = sk->sk_bound_dev_if;
2824         struct dst_entry *dst;
2825
2826         if (!oif && skb->dev)
2827                 oif = l3mdev_master_ifindex(skb->dev);
2828
2829         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2830
2831         dst = __sk_dst_get(sk);
2832         if (!dst || !dst->obsolete ||
2833             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2834                 return;
2835
2836         bh_lock_sock(sk);
2837         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2838                 ip6_datagram_dst_update(sk, false);
2839         bh_unlock_sock(sk);
2840 }
2841 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2842
2843 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2844                            const struct flowi6 *fl6)
2845 {
2846 #ifdef CONFIG_IPV6_SUBTREES
2847         struct ipv6_pinfo *np = inet6_sk(sk);
2848 #endif
2849
2850         ip6_dst_store(sk, dst,
2851                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2852                       &sk->sk_v6_daddr : NULL,
2853 #ifdef CONFIG_IPV6_SUBTREES
2854                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2855                       &np->saddr :
2856 #endif
2857                       NULL);
2858 }
2859
2860 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2861                                   struct flowi6 *fl6,
2862                                   const struct in6_addr *gw,
2863                                   struct rt6_info **ret)
2864 {
2865         const struct fib6_nh *nh = res->nh;
2866
2867         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2868             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2869                 return false;
2870
2871         /* rt_cache's gateway might be different from its 'parent'
2872          * in the case of an ip redirect.
2873          * So we keep searching in the exception table if the gateway
2874          * is different.
2875          */
2876         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2877                 struct rt6_info *rt_cache;
2878
2879                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2880                 if (rt_cache &&
2881                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2882                         *ret = rt_cache;
2883                         return true;
2884                 }
2885                 return false;
2886         }
2887         return true;
2888 }
2889
2890 struct fib6_nh_rd_arg {
2891         struct fib6_result      *res;
2892         struct flowi6           *fl6;
2893         const struct in6_addr   *gw;
2894         struct rt6_info         **ret;
2895 };
2896
2897 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2898 {
2899         struct fib6_nh_rd_arg *arg = _arg;
2900
2901         arg->res->nh = nh;
2902         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2903 }
2904
2905 /* Handle redirects */
2906 struct ip6rd_flowi {
2907         struct flowi6 fl6;
2908         struct in6_addr gateway;
2909 };
2910
2911 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
2912                                              struct fib6_table *table,
2913                                              struct flowi6 *fl6,
2914                                              const struct sk_buff *skb,
2915                                              int flags)
2916 {
2917         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2918         struct rt6_info *ret = NULL;
2919         struct fib6_result res = {};
2920         struct fib6_nh_rd_arg arg = {
2921                 .res = &res,
2922                 .fl6 = fl6,
2923                 .gw  = &rdfl->gateway,
2924                 .ret = &ret
2925         };
2926         struct fib6_info *rt;
2927         struct fib6_node *fn;
2928
2929         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2930          * this case we must match on the real ingress device, so reset it
2931          */
2932         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2933                 fl6->flowi6_oif = skb->dev->ifindex;
2934
2935         /* Get the "current" route for this destination and
2936          * check if the redirect has come from appropriate router.
2937          *
2938          * RFC 4861 specifies that redirects should only be
2939          * accepted if they come from the nexthop to the target.
2940          * Due to the way the routes are chosen, this notion
2941          * is a bit fuzzy and one might need to check all possible
2942          * routes.
2943          */
2944
2945         rcu_read_lock();
2946         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2947 restart:
2948         for_each_fib6_node_rt_rcu(fn) {
2949                 res.f6i = rt;
2950                 if (fib6_check_expired(rt))
2951                         continue;
2952                 if (rt->fib6_flags & RTF_REJECT)
2953                         break;
2954                 if (unlikely(rt->nh)) {
2955                         if (nexthop_is_blackhole(rt->nh))
2956                                 continue;
2957                         /* on match, res->nh is filled in and potentially ret */
2958                         if (nexthop_for_each_fib6_nh(rt->nh,
2959                                                      fib6_nh_redirect_match,
2960                                                      &arg))
2961                                 goto out;
2962                 } else {
2963                         res.nh = rt->fib6_nh;
2964                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2965                                                   &ret))
2966                                 goto out;
2967                 }
2968         }
2969
2970         if (!rt)
2971                 rt = net->ipv6.fib6_null_entry;
2972         else if (rt->fib6_flags & RTF_REJECT) {
2973                 ret = net->ipv6.ip6_null_entry;
2974                 goto out;
2975         }
2976
2977         if (rt == net->ipv6.fib6_null_entry) {
2978                 fn = fib6_backtrack(fn, &fl6->saddr);
2979                 if (fn)
2980                         goto restart;
2981         }
2982
2983         res.f6i = rt;
2984         res.nh = rt->fib6_nh;
2985 out:
2986         if (ret) {
2987                 ip6_hold_safe(net, &ret);
2988         } else {
2989                 res.fib6_flags = res.f6i->fib6_flags;
2990                 res.fib6_type = res.f6i->fib6_type;
2991                 ret = ip6_create_rt_rcu(&res);
2992         }
2993
2994         rcu_read_unlock();
2995
2996         trace_fib6_table_lookup(net, &res, table, fl6);
2997         return ret;
2998 };
2999
3000 static struct dst_entry *ip6_route_redirect(struct net *net,
3001                                             const struct flowi6 *fl6,
3002                                             const struct sk_buff *skb,
3003                                             const struct in6_addr *gateway)
3004 {
3005         int flags = RT6_LOOKUP_F_HAS_SADDR;
3006         struct ip6rd_flowi rdfl;
3007
3008         rdfl.fl6 = *fl6;
3009         rdfl.gateway = *gateway;
3010
3011         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3012                                 flags, __ip6_route_redirect);
3013 }
3014
3015 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3016                   kuid_t uid)
3017 {
3018         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3019         struct dst_entry *dst;
3020         struct flowi6 fl6 = {
3021                 .flowi6_iif = LOOPBACK_IFINDEX,
3022                 .flowi6_oif = oif,
3023                 .flowi6_mark = mark,
3024                 .daddr = iph->daddr,
3025                 .saddr = iph->saddr,
3026                 .flowlabel = ip6_flowinfo(iph),
3027                 .flowi6_uid = uid,
3028         };
3029
3030         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3031         rt6_do_redirect(dst, NULL, skb);
3032         dst_release(dst);
3033 }
3034 EXPORT_SYMBOL_GPL(ip6_redirect);
3035
3036 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3037 {
3038         const struct ipv6hdr *iph = ipv6_hdr(skb);
3039         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3040         struct dst_entry *dst;
3041         struct flowi6 fl6 = {
3042                 .flowi6_iif = LOOPBACK_IFINDEX,
3043                 .flowi6_oif = oif,
3044                 .daddr = msg->dest,
3045                 .saddr = iph->daddr,
3046                 .flowi6_uid = sock_net_uid(net, NULL),
3047         };
3048
3049         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3050         rt6_do_redirect(dst, NULL, skb);
3051         dst_release(dst);
3052 }
3053
3054 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3055 {
3056         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3057                      sk->sk_uid);
3058 }
3059 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3060
3061 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3062 {
3063         struct net_device *dev = dst->dev;
3064         unsigned int mtu = dst_mtu(dst);
3065         struct net *net = dev_net(dev);
3066
3067         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3068
3069         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3070                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3071
3072         /*
3073          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3074          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3075          * IPV6_MAXPLEN is also valid and means: "any MSS,
3076          * rely only on pmtu discovery"
3077          */
3078         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3079                 mtu = IPV6_MAXPLEN;
3080         return mtu;
3081 }
3082
3083 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3084 {
3085         struct inet6_dev *idev;
3086         unsigned int mtu;
3087
3088         mtu = dst_metric_raw(dst, RTAX_MTU);
3089         if (mtu)
3090                 goto out;
3091
3092         mtu = IPV6_MIN_MTU;
3093
3094         rcu_read_lock();
3095         idev = __in6_dev_get(dst->dev);
3096         if (idev)
3097                 mtu = idev->cnf.mtu6;
3098         rcu_read_unlock();
3099
3100 out:
3101         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3102
3103         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3104 }
3105 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3106
3107 /* MTU selection:
3108  * 1. mtu on route is locked - use it
3109  * 2. mtu from nexthop exception
3110  * 3. mtu from egress device
3111  *
3112  * based on ip6_dst_mtu_forward and exception logic of
3113  * rt6_find_cached_rt; called with rcu_read_lock
3114  */
3115 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3116                       const struct in6_addr *daddr,
3117                       const struct in6_addr *saddr)
3118 {
3119         const struct fib6_nh *nh = res->nh;
3120         struct fib6_info *f6i = res->f6i;
3121         struct inet6_dev *idev;
3122         struct rt6_info *rt;
3123         u32 mtu = 0;
3124
3125         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3126                 mtu = f6i->fib6_pmtu;
3127                 if (mtu)
3128                         goto out;
3129         }
3130
3131         rt = rt6_find_cached_rt(res, daddr, saddr);
3132         if (unlikely(rt)) {
3133                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3134         } else {
3135                 struct net_device *dev = nh->fib_nh_dev;
3136
3137                 mtu = IPV6_MIN_MTU;
3138                 idev = __in6_dev_get(dev);
3139                 if (idev && idev->cnf.mtu6 > mtu)
3140                         mtu = idev->cnf.mtu6;
3141         }
3142
3143         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3144 out:
3145         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3146 }
3147
3148 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3149                                   struct flowi6 *fl6)
3150 {
3151         struct dst_entry *dst;
3152         struct rt6_info *rt;
3153         struct inet6_dev *idev = in6_dev_get(dev);
3154         struct net *net = dev_net(dev);
3155
3156         if (unlikely(!idev))
3157                 return ERR_PTR(-ENODEV);
3158
3159         rt = ip6_dst_alloc(net, dev, 0);
3160         if (unlikely(!rt)) {
3161                 in6_dev_put(idev);
3162                 dst = ERR_PTR(-ENOMEM);
3163                 goto out;
3164         }
3165
3166         rt->dst.input = ip6_input;
3167         rt->dst.output  = ip6_output;
3168         rt->rt6i_gateway  = fl6->daddr;
3169         rt->rt6i_dst.addr = fl6->daddr;
3170         rt->rt6i_dst.plen = 128;
3171         rt->rt6i_idev     = idev;
3172         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3173
3174         /* Add this dst into uncached_list so that rt6_disable_ip() can
3175          * do proper release of the net_device
3176          */
3177         rt6_uncached_list_add(rt);
3178         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3179
3180         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3181
3182 out:
3183         return dst;
3184 }
3185
3186 static int ip6_dst_gc(struct dst_ops *ops)
3187 {
3188         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3189         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3190         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3191         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3192         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3193         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3194         int entries;
3195
3196         entries = dst_entries_get_fast(ops);
3197         if (entries > rt_max_size)
3198                 entries = dst_entries_get_slow(ops);
3199
3200         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3201             entries <= rt_max_size)
3202                 goto out;
3203
3204         net->ipv6.ip6_rt_gc_expire++;
3205         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3206         entries = dst_entries_get_slow(ops);
3207         if (entries < ops->gc_thresh)
3208                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3209 out:
3210         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3211         return entries > rt_max_size;
3212 }
3213
3214 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3215                                const struct in6_addr *gw_addr, u32 tbid,
3216                                int flags, struct fib6_result *res)
3217 {
3218         struct flowi6 fl6 = {
3219                 .flowi6_oif = cfg->fc_ifindex,
3220                 .daddr = *gw_addr,
3221                 .saddr = cfg->fc_prefsrc,
3222         };
3223         struct fib6_table *table;
3224         int err;
3225
3226         table = fib6_get_table(net, tbid);
3227         if (!table)
3228                 return -EINVAL;
3229
3230         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3231                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3232
3233         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3234
3235         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3236         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3237                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3238                                  cfg->fc_ifindex != 0, NULL, flags);
3239
3240         return err;
3241 }
3242
3243 static int ip6_route_check_nh_onlink(struct net *net,
3244                                      struct fib6_config *cfg,
3245                                      const struct net_device *dev,
3246                                      struct netlink_ext_ack *extack)
3247 {
3248         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3249         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3250         struct fib6_result res = {};
3251         int err;
3252
3253         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3254         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3255             /* ignore match if it is the default route */
3256             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3257             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3258                 NL_SET_ERR_MSG(extack,
3259                                "Nexthop has invalid gateway or device mismatch");
3260                 err = -EINVAL;
3261         }
3262
3263         return err;
3264 }
3265
3266 static int ip6_route_check_nh(struct net *net,
3267                               struct fib6_config *cfg,
3268                               struct net_device **_dev,
3269                               struct inet6_dev **idev)
3270 {
3271         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3272         struct net_device *dev = _dev ? *_dev : NULL;
3273         int flags = RT6_LOOKUP_F_IFACE;
3274         struct fib6_result res = {};
3275         int err = -EHOSTUNREACH;
3276
3277         if (cfg->fc_table) {
3278                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3279                                           cfg->fc_table, flags, &res);
3280                 /* gw_addr can not require a gateway or resolve to a reject
3281                  * route. If a device is given, it must match the result.
3282                  */
3283                 if (err || res.fib6_flags & RTF_REJECT ||
3284                     res.nh->fib_nh_gw_family ||
3285                     (dev && dev != res.nh->fib_nh_dev))
3286                         err = -EHOSTUNREACH;
3287         }
3288
3289         if (err < 0) {
3290                 struct flowi6 fl6 = {
3291                         .flowi6_oif = cfg->fc_ifindex,
3292                         .daddr = *gw_addr,
3293                 };
3294
3295                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3296                 if (err || res.fib6_flags & RTF_REJECT ||
3297                     res.nh->fib_nh_gw_family)
3298                         err = -EHOSTUNREACH;
3299
3300                 if (err)
3301                         return err;
3302
3303                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3304                                  cfg->fc_ifindex != 0, NULL, flags);
3305         }
3306
3307         err = 0;
3308         if (dev) {
3309                 if (dev != res.nh->fib_nh_dev)
3310                         err = -EHOSTUNREACH;
3311         } else {
3312                 *_dev = dev = res.nh->fib_nh_dev;
3313                 dev_hold(dev);
3314                 *idev = in6_dev_get(dev);
3315         }
3316
3317         return err;
3318 }
3319
3320 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3321                            struct net_device **_dev, struct inet6_dev **idev,
3322                            struct netlink_ext_ack *extack)
3323 {
3324         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3325         int gwa_type = ipv6_addr_type(gw_addr);
3326         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3327         const struct net_device *dev = *_dev;
3328         bool need_addr_check = !dev;
3329         int err = -EINVAL;
3330
3331         /* if gw_addr is local we will fail to detect this in case
3332          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3333          * will return already-added prefix route via interface that
3334          * prefix route was assigned to, which might be non-loopback.
3335          */
3336         if (dev &&
3337             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3338                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3339                 goto out;
3340         }
3341
3342         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3343                 /* IPv6 strictly inhibits using not link-local
3344                  * addresses as nexthop address.
3345                  * Otherwise, router will not able to send redirects.
3346                  * It is very good, but in some (rare!) circumstances
3347                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3348                  * some exceptions. --ANK
3349                  * We allow IPv4-mapped nexthops to support RFC4798-type
3350                  * addressing
3351                  */
3352                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3353                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3354                         goto out;
3355                 }
3356
3357                 rcu_read_lock();
3358
3359                 if (cfg->fc_flags & RTNH_F_ONLINK)
3360                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3361                 else
3362                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3363
3364                 rcu_read_unlock();
3365
3366                 if (err)
3367                         goto out;
3368         }
3369
3370         /* reload in case device was changed */
3371         dev = *_dev;
3372
3373         err = -EINVAL;
3374         if (!dev) {
3375                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3376                 goto out;
3377         } else if (dev->flags & IFF_LOOPBACK) {
3378                 NL_SET_ERR_MSG(extack,
3379                                "Egress device can not be loopback device for this route");
3380                 goto out;
3381         }
3382
3383         /* if we did not check gw_addr above, do so now that the
3384          * egress device has been resolved.
3385          */
3386         if (need_addr_check &&
3387             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3388                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3389                 goto out;
3390         }
3391
3392         err = 0;
3393 out:
3394         return err;
3395 }
3396
3397 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3398 {
3399         if ((flags & RTF_REJECT) ||
3400             (dev && (dev->flags & IFF_LOOPBACK) &&
3401              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3402              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3403                 return true;
3404
3405         return false;
3406 }
3407
3408 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3409                  struct fib6_config *cfg, gfp_t gfp_flags,
3410                  struct netlink_ext_ack *extack)
3411 {
3412         struct net_device *dev = NULL;
3413         struct inet6_dev *idev = NULL;
3414         int addr_type;
3415         int err;
3416
3417         fib6_nh->fib_nh_family = AF_INET6;
3418 #ifdef CONFIG_IPV6_ROUTER_PREF
3419         fib6_nh->last_probe = jiffies;
3420 #endif
3421         if (cfg->fc_is_fdb) {
3422                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3423                 fib6_nh->fib_nh_gw_family = AF_INET6;
3424                 return 0;
3425         }
3426
3427         err = -ENODEV;
3428         if (cfg->fc_ifindex) {
3429                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3430                 if (!dev)
3431                         goto out;
3432                 idev = in6_dev_get(dev);
3433                 if (!idev)
3434                         goto out;
3435         }
3436
3437         if (cfg->fc_flags & RTNH_F_ONLINK) {
3438                 if (!dev) {
3439                         NL_SET_ERR_MSG(extack,
3440                                        "Nexthop device required for onlink");
3441                         goto out;
3442                 }
3443
3444                 if (!(dev->flags & IFF_UP)) {
3445                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3446                         err = -ENETDOWN;
3447                         goto out;
3448                 }
3449
3450                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3451         }
3452
3453         fib6_nh->fib_nh_weight = 1;
3454
3455         /* We cannot add true routes via loopback here,
3456          * they would result in kernel looping; promote them to reject routes
3457          */
3458         addr_type = ipv6_addr_type(&cfg->fc_dst);
3459         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3460                 /* hold loopback dev/idev if we haven't done so. */
3461                 if (dev != net->loopback_dev) {
3462                         if (dev) {
3463                                 dev_put(dev);
3464                                 in6_dev_put(idev);
3465                         }
3466                         dev = net->loopback_dev;
3467                         dev_hold(dev);
3468                         idev = in6_dev_get(dev);
3469                         if (!idev) {
3470                                 err = -ENODEV;
3471                                 goto out;
3472                         }
3473                 }
3474                 goto pcpu_alloc;
3475         }
3476
3477         if (cfg->fc_flags & RTF_GATEWAY) {
3478                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3479                 if (err)
3480                         goto out;
3481
3482                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3483                 fib6_nh->fib_nh_gw_family = AF_INET6;
3484         }
3485
3486         err = -ENODEV;
3487         if (!dev)
3488                 goto out;
3489
3490         if (idev->cnf.disable_ipv6) {
3491                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3492                 err = -EACCES;
3493                 goto out;
3494         }
3495
3496         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3497                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3498                 err = -ENETDOWN;
3499                 goto out;
3500         }
3501
3502         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3503             !netif_carrier_ok(dev))
3504                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3505
3506         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3507                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3508         if (err)
3509                 goto out;
3510
3511 pcpu_alloc:
3512         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3513         if (!fib6_nh->rt6i_pcpu) {
3514                 err = -ENOMEM;
3515                 goto out;
3516         }
3517
3518         fib6_nh->fib_nh_dev = dev;
3519         fib6_nh->fib_nh_oif = dev->ifindex;
3520         err = 0;
3521 out:
3522         if (idev)
3523                 in6_dev_put(idev);
3524
3525         if (err) {
3526                 lwtstate_put(fib6_nh->fib_nh_lws);
3527                 fib6_nh->fib_nh_lws = NULL;
3528                 if (dev)
3529                         dev_put(dev);
3530         }
3531
3532         return err;
3533 }
3534
3535 void fib6_nh_release(struct fib6_nh *fib6_nh)
3536 {
3537         struct rt6_exception_bucket *bucket;
3538
3539         rcu_read_lock();
3540
3541         fib6_nh_flush_exceptions(fib6_nh, NULL);
3542         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3543         if (bucket) {
3544                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3545                 kfree(bucket);
3546         }
3547
3548         rcu_read_unlock();
3549
3550         if (fib6_nh->rt6i_pcpu) {
3551                 int cpu;
3552
3553                 for_each_possible_cpu(cpu) {
3554                         struct rt6_info **ppcpu_rt;
3555                         struct rt6_info *pcpu_rt;
3556
3557                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3558                         pcpu_rt = *ppcpu_rt;
3559                         if (pcpu_rt) {
3560                                 dst_dev_put(&pcpu_rt->dst);
3561                                 dst_release(&pcpu_rt->dst);
3562                                 *ppcpu_rt = NULL;
3563                         }
3564                 }
3565
3566                 free_percpu(fib6_nh->rt6i_pcpu);
3567         }
3568
3569         fib_nh_common_release(&fib6_nh->nh_common);
3570 }
3571
3572 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3573                                               gfp_t gfp_flags,
3574                                               struct netlink_ext_ack *extack)
3575 {
3576         struct net *net = cfg->fc_nlinfo.nl_net;
3577         struct fib6_info *rt = NULL;
3578         struct nexthop *nh = NULL;
3579         struct fib6_table *table;
3580         struct fib6_nh *fib6_nh;
3581         int err = -EINVAL;
3582         int addr_type;
3583
3584         /* RTF_PCPU is an internal flag; can not be set by userspace */
3585         if (cfg->fc_flags & RTF_PCPU) {
3586                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3587                 goto out;
3588         }
3589
3590         /* RTF_CACHE is an internal flag; can not be set by userspace */
3591         if (cfg->fc_flags & RTF_CACHE) {
3592                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3593                 goto out;
3594         }
3595
3596         if (cfg->fc_type > RTN_MAX) {
3597                 NL_SET_ERR_MSG(extack, "Invalid route type");
3598                 goto out;
3599         }
3600
3601         if (cfg->fc_dst_len > 128) {
3602                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3603                 goto out;
3604         }
3605         if (cfg->fc_src_len > 128) {
3606                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3607                 goto out;
3608         }
3609 #ifndef CONFIG_IPV6_SUBTREES
3610         if (cfg->fc_src_len) {
3611                 NL_SET_ERR_MSG(extack,
3612                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3613                 goto out;
3614         }
3615 #endif
3616         if (cfg->fc_nh_id) {
3617                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3618                 if (!nh) {
3619                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3620                         goto out;
3621                 }
3622                 err = fib6_check_nexthop(nh, cfg, extack);
3623                 if (err)
3624                         goto out;
3625         }
3626
3627         err = -ENOBUFS;
3628         if (cfg->fc_nlinfo.nlh &&
3629             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3630                 table = fib6_get_table(net, cfg->fc_table);
3631                 if (!table) {
3632                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3633                         table = fib6_new_table(net, cfg->fc_table);
3634                 }
3635         } else {
3636                 table = fib6_new_table(net, cfg->fc_table);
3637         }
3638
3639         if (!table)
3640                 goto out;
3641
3642         err = -ENOMEM;
3643         rt = fib6_info_alloc(gfp_flags, !nh);
3644         if (!rt)
3645                 goto out;
3646
3647         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3648                                                extack);
3649         if (IS_ERR(rt->fib6_metrics)) {
3650                 err = PTR_ERR(rt->fib6_metrics);
3651                 /* Do not leave garbage there. */
3652                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3653                 goto out_free;
3654         }
3655
3656         if (cfg->fc_flags & RTF_ADDRCONF)
3657                 rt->dst_nocount = true;
3658
3659         if (cfg->fc_flags & RTF_EXPIRES)
3660                 fib6_set_expires(rt, jiffies +
3661                                 clock_t_to_jiffies(cfg->fc_expires));
3662         else
3663                 fib6_clean_expires(rt);
3664
3665         if (cfg->fc_protocol == RTPROT_UNSPEC)
3666                 cfg->fc_protocol = RTPROT_BOOT;
3667         rt->fib6_protocol = cfg->fc_protocol;
3668
3669         rt->fib6_table = table;
3670         rt->fib6_metric = cfg->fc_metric;
3671         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3672         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3673
3674         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3675         rt->fib6_dst.plen = cfg->fc_dst_len;
3676
3677 #ifdef CONFIG_IPV6_SUBTREES
3678         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3679         rt->fib6_src.plen = cfg->fc_src_len;
3680 #endif
3681         if (nh) {
3682                 if (rt->fib6_src.plen) {
3683                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3684                         goto out_free;
3685                 }
3686                 if (!nexthop_get(nh)) {
3687                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3688                         goto out_free;
3689                 }
3690                 rt->nh = nh;
3691                 fib6_nh = nexthop_fib6_nh(rt->nh);
3692         } else {
3693                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3694                 if (err)
3695                         goto out;
3696
3697                 fib6_nh = rt->fib6_nh;
3698
3699                 /* We cannot add true routes via loopback here, they would
3700                  * result in kernel looping; promote them to reject routes
3701                  */
3702                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3703                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3704                                    addr_type))
3705                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3706         }
3707
3708         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3709                 struct net_device *dev = fib6_nh->fib_nh_dev;
3710
3711                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3712                         NL_SET_ERR_MSG(extack, "Invalid source address");
3713                         err = -EINVAL;
3714                         goto out;
3715                 }
3716                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3717                 rt->fib6_prefsrc.plen = 128;
3718         } else
3719                 rt->fib6_prefsrc.plen = 0;
3720
3721         return rt;
3722 out:
3723         fib6_info_release(rt);
3724         return ERR_PTR(err);
3725 out_free:
3726         ip_fib_metrics_put(rt->fib6_metrics);
3727         kfree(rt);
3728         return ERR_PTR(err);
3729 }
3730
3731 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3732                   struct netlink_ext_ack *extack)
3733 {
3734         struct fib6_info *rt;
3735         int err;
3736
3737         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3738         if (IS_ERR(rt))
3739                 return PTR_ERR(rt);
3740
3741         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3742         fib6_info_release(rt);
3743
3744         return err;
3745 }
3746
3747 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3748 {
3749         struct net *net = info->nl_net;
3750         struct fib6_table *table;
3751         int err;
3752
3753         if (rt == net->ipv6.fib6_null_entry) {
3754                 err = -ENOENT;
3755                 goto out;
3756         }
3757
3758         table = rt->fib6_table;
3759         spin_lock_bh(&table->tb6_lock);
3760         err = fib6_del(rt, info);
3761         spin_unlock_bh(&table->tb6_lock);
3762
3763 out:
3764         fib6_info_release(rt);
3765         return err;
3766 }
3767
3768 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3769 {
3770         struct nl_info info = {
3771                 .nl_net = net,
3772                 .skip_notify = skip_notify
3773         };
3774
3775         return __ip6_del_rt(rt, &info);
3776 }
3777
3778 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3779 {
3780         struct nl_info *info = &cfg->fc_nlinfo;
3781         struct net *net = info->nl_net;
3782         struct sk_buff *skb = NULL;
3783         struct fib6_table *table;
3784         int err = -ENOENT;
3785
3786         if (rt == net->ipv6.fib6_null_entry)
3787                 goto out_put;
3788         table = rt->fib6_table;
3789         spin_lock_bh(&table->tb6_lock);
3790
3791         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3792                 struct fib6_info *sibling, *next_sibling;
3793                 struct fib6_node *fn;
3794
3795                 /* prefer to send a single notification with all hops */
3796                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3797                 if (skb) {
3798                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3799
3800                         if (rt6_fill_node(net, skb, rt, NULL,
3801                                           NULL, NULL, 0, RTM_DELROUTE,
3802                                           info->portid, seq, 0) < 0) {
3803                                 kfree_skb(skb);
3804                                 skb = NULL;
3805                         } else
3806                                 info->skip_notify = 1;
3807                 }
3808
3809                 /* 'rt' points to the first sibling route. If it is not the
3810                  * leaf, then we do not need to send a notification. Otherwise,
3811                  * we need to check if the last sibling has a next route or not
3812                  * and emit a replace or delete notification, respectively.
3813                  */
3814                 info->skip_notify_kernel = 1;
3815                 fn = rcu_dereference_protected(rt->fib6_node,
3816                                             lockdep_is_held(&table->tb6_lock));
3817                 if (rcu_access_pointer(fn->leaf) == rt) {
3818                         struct fib6_info *last_sibling, *replace_rt;
3819
3820                         last_sibling = list_last_entry(&rt->fib6_siblings,
3821                                                        struct fib6_info,
3822                                                        fib6_siblings);
3823                         replace_rt = rcu_dereference_protected(
3824                                             last_sibling->fib6_next,
3825                                             lockdep_is_held(&table->tb6_lock));
3826                         if (replace_rt)
3827                                 call_fib6_entry_notifiers_replace(net,
3828                                                                   replace_rt);
3829                         else
3830                                 call_fib6_multipath_entry_notifiers(net,
3831                                                        FIB_EVENT_ENTRY_DEL,
3832                                                        rt, rt->fib6_nsiblings,
3833                                                        NULL);
3834                 }
3835                 list_for_each_entry_safe(sibling, next_sibling,
3836                                          &rt->fib6_siblings,
3837                                          fib6_siblings) {
3838                         err = fib6_del(sibling, info);
3839                         if (err)
3840                                 goto out_unlock;
3841                 }
3842         }
3843
3844         err = fib6_del(rt, info);
3845 out_unlock:
3846         spin_unlock_bh(&table->tb6_lock);
3847 out_put:
3848         fib6_info_release(rt);
3849
3850         if (skb) {
3851                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3852                             info->nlh, gfp_any());
3853         }
3854         return err;
3855 }
3856
3857 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3858 {
3859         int rc = -ESRCH;
3860
3861         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3862                 goto out;
3863
3864         if (cfg->fc_flags & RTF_GATEWAY &&
3865             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3866                 goto out;
3867
3868         rc = rt6_remove_exception_rt(rt);
3869 out:
3870         return rc;
3871 }
3872
3873 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3874                              struct fib6_nh *nh)
3875 {
3876         struct fib6_result res = {
3877                 .f6i = rt,
3878                 .nh = nh,
3879         };
3880         struct rt6_info *rt_cache;
3881
3882         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3883         if (rt_cache)
3884                 return __ip6_del_cached_rt(rt_cache, cfg);
3885
3886         return 0;
3887 }
3888
3889 struct fib6_nh_del_cached_rt_arg {
3890         struct fib6_config *cfg;
3891         struct fib6_info *f6i;
3892 };
3893
3894 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3895 {
3896         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3897         int rc;
3898
3899         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3900         return rc != -ESRCH ? rc : 0;
3901 }
3902
3903 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3904 {
3905         struct fib6_nh_del_cached_rt_arg arg = {
3906                 .cfg = cfg,
3907                 .f6i = f6i
3908         };
3909
3910         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3911 }
3912
3913 static int ip6_route_del(struct fib6_config *cfg,
3914                          struct netlink_ext_ack *extack)
3915 {
3916         struct fib6_table *table;
3917         struct fib6_info *rt;
3918         struct fib6_node *fn;
3919         int err = -ESRCH;
3920
3921         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3922         if (!table) {
3923                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3924                 return err;
3925         }
3926
3927         rcu_read_lock();
3928
3929         fn = fib6_locate(&table->tb6_root,
3930                          &cfg->fc_dst, cfg->fc_dst_len,
3931                          &cfg->fc_src, cfg->fc_src_len,
3932                          !(cfg->fc_flags & RTF_CACHE));
3933
3934         if (fn) {
3935                 for_each_fib6_node_rt_rcu(fn) {
3936                         struct fib6_nh *nh;
3937
3938                         if (rt->nh && cfg->fc_nh_id &&
3939                             rt->nh->id != cfg->fc_nh_id)
3940                                 continue;
3941
3942                         if (cfg->fc_flags & RTF_CACHE) {
3943                                 int rc = 0;
3944
3945                                 if (rt->nh) {
3946                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3947                                 } else if (cfg->fc_nh_id) {
3948                                         continue;
3949                                 } else {
3950                                         nh = rt->fib6_nh;
3951                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3952                                 }
3953                                 if (rc != -ESRCH) {
3954                                         rcu_read_unlock();
3955                                         return rc;
3956                                 }
3957                                 continue;
3958                         }
3959
3960                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3961                                 continue;
3962                         if (cfg->fc_protocol &&
3963                             cfg->fc_protocol != rt->fib6_protocol)
3964                                 continue;
3965
3966                         if (rt->nh) {
3967                                 if (!fib6_info_hold_safe(rt))
3968                                         continue;
3969                                 rcu_read_unlock();
3970
3971                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3972                         }
3973                         if (cfg->fc_nh_id)
3974                                 continue;
3975
3976                         nh = rt->fib6_nh;
3977                         if (cfg->fc_ifindex &&
3978                             (!nh->fib_nh_dev ||
3979                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3980                                 continue;
3981                         if (cfg->fc_flags & RTF_GATEWAY &&
3982                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3983                                 continue;
3984                         if (!fib6_info_hold_safe(rt))
3985                                 continue;
3986                         rcu_read_unlock();
3987
3988                         /* if gateway was specified only delete the one hop */
3989                         if (cfg->fc_flags & RTF_GATEWAY)
3990                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3991
3992                         return __ip6_del_rt_siblings(rt, cfg);
3993                 }
3994         }
3995         rcu_read_unlock();
3996
3997         return err;
3998 }
3999
4000 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4001 {
4002         struct netevent_redirect netevent;
4003         struct rt6_info *rt, *nrt = NULL;
4004         struct fib6_result res = {};
4005         struct ndisc_options ndopts;
4006         struct inet6_dev *in6_dev;
4007         struct neighbour *neigh;
4008         struct rd_msg *msg;
4009         int optlen, on_link;
4010         u8 *lladdr;
4011
4012         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4013         optlen -= sizeof(*msg);
4014
4015         if (optlen < 0) {
4016                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4017                 return;
4018         }
4019
4020         msg = (struct rd_msg *)icmp6_hdr(skb);
4021
4022         if (ipv6_addr_is_multicast(&msg->dest)) {
4023                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4024                 return;
4025         }
4026
4027         on_link = 0;
4028         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4029                 on_link = 1;
4030         } else if (ipv6_addr_type(&msg->target) !=
4031                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4032                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4033                 return;
4034         }
4035
4036         in6_dev = __in6_dev_get(skb->dev);
4037         if (!in6_dev)
4038                 return;
4039         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4040                 return;
4041
4042         /* RFC2461 8.1:
4043          *      The IP source address of the Redirect MUST be the same as the current
4044          *      first-hop router for the specified ICMP Destination Address.
4045          */
4046
4047         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4048                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4049                 return;
4050         }
4051
4052         lladdr = NULL;
4053         if (ndopts.nd_opts_tgt_lladdr) {
4054                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4055                                              skb->dev);
4056                 if (!lladdr) {
4057                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4058                         return;
4059                 }
4060         }
4061
4062         rt = (struct rt6_info *) dst;
4063         if (rt->rt6i_flags & RTF_REJECT) {
4064                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4065                 return;
4066         }
4067
4068         /* Redirect received -> path was valid.
4069          * Look, redirects are sent only in response to data packets,
4070          * so that this nexthop apparently is reachable. --ANK
4071          */
4072         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4073
4074         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4075         if (!neigh)
4076                 return;
4077
4078         /*
4079          *      We have finally decided to accept it.
4080          */
4081
4082         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4083                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4084                      NEIGH_UPDATE_F_OVERRIDE|
4085                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4086                                      NEIGH_UPDATE_F_ISROUTER)),
4087                      NDISC_REDIRECT, &ndopts);
4088
4089         rcu_read_lock();
4090         res.f6i = rcu_dereference(rt->from);
4091         if (!res.f6i)
4092                 goto out;
4093
4094         if (res.f6i->nh) {
4095                 struct fib6_nh_match_arg arg = {
4096                         .dev = dst->dev,
4097                         .gw = &rt->rt6i_gateway,
4098                 };
4099
4100                 nexthop_for_each_fib6_nh(res.f6i->nh,
4101                                          fib6_nh_find_match, &arg);
4102
4103                 /* fib6_info uses a nexthop that does not have fib6_nh
4104                  * using the dst->dev. Should be impossible
4105                  */
4106                 if (!arg.match)
4107                         goto out;
4108                 res.nh = arg.match;
4109         } else {
4110                 res.nh = res.f6i->fib6_nh;
4111         }
4112
4113         res.fib6_flags = res.f6i->fib6_flags;
4114         res.fib6_type = res.f6i->fib6_type;
4115         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4116         if (!nrt)
4117                 goto out;
4118
4119         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4120         if (on_link)
4121                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4122
4123         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4124
4125         /* rt6_insert_exception() will take care of duplicated exceptions */
4126         if (rt6_insert_exception(nrt, &res)) {
4127                 dst_release_immediate(&nrt->dst);
4128                 goto out;
4129         }
4130
4131         netevent.old = &rt->dst;
4132         netevent.new = &nrt->dst;
4133         netevent.daddr = &msg->dest;
4134         netevent.neigh = neigh;
4135         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4136
4137 out:
4138         rcu_read_unlock();
4139         neigh_release(neigh);
4140 }
4141
4142 #ifdef CONFIG_IPV6_ROUTE_INFO
4143 static struct fib6_info *rt6_get_route_info(struct net *net,
4144                                            const struct in6_addr *prefix, int prefixlen,
4145                                            const struct in6_addr *gwaddr,
4146                                            struct net_device *dev)
4147 {
4148         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4149         int ifindex = dev->ifindex;
4150         struct fib6_node *fn;
4151         struct fib6_info *rt = NULL;
4152         struct fib6_table *table;
4153
4154         table = fib6_get_table(net, tb_id);
4155         if (!table)
4156                 return NULL;
4157
4158         rcu_read_lock();
4159         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4160         if (!fn)
4161                 goto out;
4162
4163         for_each_fib6_node_rt_rcu(fn) {
4164                 /* these routes do not use nexthops */
4165                 if (rt->nh)
4166                         continue;
4167                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4168                         continue;
4169                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4170                     !rt->fib6_nh->fib_nh_gw_family)
4171                         continue;
4172                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4173                         continue;
4174                 if (!fib6_info_hold_safe(rt))
4175                         continue;
4176                 break;
4177         }
4178 out:
4179         rcu_read_unlock();
4180         return rt;
4181 }
4182
4183 static struct fib6_info *rt6_add_route_info(struct net *net,
4184                                            const struct in6_addr *prefix, int prefixlen,
4185                                            const struct in6_addr *gwaddr,
4186                                            struct net_device *dev,
4187                                            unsigned int pref)
4188 {
4189         struct fib6_config cfg = {
4190                 .fc_metric      = IP6_RT_PRIO_USER,
4191                 .fc_ifindex     = dev->ifindex,
4192                 .fc_dst_len     = prefixlen,
4193                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4194                                   RTF_UP | RTF_PREF(pref),
4195                 .fc_protocol = RTPROT_RA,
4196                 .fc_type = RTN_UNICAST,
4197                 .fc_nlinfo.portid = 0,
4198                 .fc_nlinfo.nlh = NULL,
4199                 .fc_nlinfo.nl_net = net,
4200         };
4201
4202         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4203         cfg.fc_dst = *prefix;
4204         cfg.fc_gateway = *gwaddr;
4205
4206         /* We should treat it as a default route if prefix length is 0. */
4207         if (!prefixlen)
4208                 cfg.fc_flags |= RTF_DEFAULT;
4209
4210         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4211
4212         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4213 }
4214 #endif
4215
4216 struct fib6_info *rt6_get_dflt_router(struct net *net,
4217                                      const struct in6_addr *addr,
4218                                      struct net_device *dev)
4219 {
4220         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4221         struct fib6_info *rt;
4222         struct fib6_table *table;
4223
4224         table = fib6_get_table(net, tb_id);
4225         if (!table)
4226                 return NULL;
4227
4228         rcu_read_lock();
4229         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4230                 struct fib6_nh *nh;
4231
4232                 /* RA routes do not use nexthops */
4233                 if (rt->nh)
4234                         continue;
4235
4236                 nh = rt->fib6_nh;
4237                 if (dev == nh->fib_nh_dev &&
4238                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4239                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4240                         break;
4241         }
4242         if (rt && !fib6_info_hold_safe(rt))
4243                 rt = NULL;
4244         rcu_read_unlock();
4245         return rt;
4246 }
4247
4248 struct fib6_info *rt6_add_dflt_router(struct net *net,
4249                                      const struct in6_addr *gwaddr,
4250                                      struct net_device *dev,
4251                                      unsigned int pref,
4252                                      u32 defrtr_usr_metric)
4253 {
4254         struct fib6_config cfg = {
4255                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4256                 .fc_metric      = defrtr_usr_metric,
4257                 .fc_ifindex     = dev->ifindex,
4258                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4259                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4260                 .fc_protocol = RTPROT_RA,
4261                 .fc_type = RTN_UNICAST,
4262                 .fc_nlinfo.portid = 0,
4263                 .fc_nlinfo.nlh = NULL,
4264                 .fc_nlinfo.nl_net = net,
4265         };
4266
4267         cfg.fc_gateway = *gwaddr;
4268
4269         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4270                 struct fib6_table *table;
4271
4272                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4273                 if (table)
4274                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4275         }
4276
4277         return rt6_get_dflt_router(net, gwaddr, dev);
4278 }
4279
4280 static void __rt6_purge_dflt_routers(struct net *net,
4281                                      struct fib6_table *table)
4282 {
4283         struct fib6_info *rt;
4284
4285 restart:
4286         rcu_read_lock();
4287         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4288                 struct net_device *dev = fib6_info_nh_dev(rt);
4289                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4290
4291                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4292                     (!idev || idev->cnf.accept_ra != 2) &&
4293                     fib6_info_hold_safe(rt)) {
4294                         rcu_read_unlock();
4295                         ip6_del_rt(net, rt, false);
4296                         goto restart;
4297                 }
4298         }
4299         rcu_read_unlock();
4300
4301         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4302 }
4303
4304 void rt6_purge_dflt_routers(struct net *net)
4305 {
4306         struct fib6_table *table;
4307         struct hlist_head *head;
4308         unsigned int h;
4309
4310         rcu_read_lock();
4311
4312         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4313                 head = &net->ipv6.fib_table_hash[h];
4314                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4315                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4316                                 __rt6_purge_dflt_routers(net, table);
4317                 }
4318         }
4319
4320         rcu_read_unlock();
4321 }
4322
4323 static void rtmsg_to_fib6_config(struct net *net,
4324                                  struct in6_rtmsg *rtmsg,
4325                                  struct fib6_config *cfg)
4326 {
4327         *cfg = (struct fib6_config){
4328                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4329                          : RT6_TABLE_MAIN,
4330                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4331                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4332                 .fc_expires = rtmsg->rtmsg_info,
4333                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4334                 .fc_src_len = rtmsg->rtmsg_src_len,
4335                 .fc_flags = rtmsg->rtmsg_flags,
4336                 .fc_type = rtmsg->rtmsg_type,
4337
4338                 .fc_nlinfo.nl_net = net,
4339
4340                 .fc_dst = rtmsg->rtmsg_dst,
4341                 .fc_src = rtmsg->rtmsg_src,
4342                 .fc_gateway = rtmsg->rtmsg_gateway,
4343         };
4344 }
4345
4346 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4347 {
4348         struct fib6_config cfg;
4349         int err;
4350
4351         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4352                 return -EINVAL;
4353         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4354                 return -EPERM;
4355
4356         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4357
4358         rtnl_lock();
4359         switch (cmd) {
4360         case SIOCADDRT:
4361                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4362                 break;
4363         case SIOCDELRT:
4364                 err = ip6_route_del(&cfg, NULL);
4365                 break;
4366         }
4367         rtnl_unlock();
4368         return err;
4369 }
4370
4371 /*
4372  *      Drop the packet on the floor
4373  */
4374
4375 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4376 {
4377         struct dst_entry *dst = skb_dst(skb);
4378         struct net *net = dev_net(dst->dev);
4379         struct inet6_dev *idev;
4380         int type;
4381
4382         if (netif_is_l3_master(skb->dev) &&
4383             dst->dev == net->loopback_dev)
4384                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4385         else
4386                 idev = ip6_dst_idev(dst);
4387
4388         switch (ipstats_mib_noroutes) {
4389         case IPSTATS_MIB_INNOROUTES:
4390                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4391                 if (type == IPV6_ADDR_ANY) {
4392                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4393                         break;
4394                 }
4395                 fallthrough;
4396         case IPSTATS_MIB_OUTNOROUTES:
4397                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4398                 break;
4399         }
4400
4401         /* Start over by dropping the dst for l3mdev case */
4402         if (netif_is_l3_master(skb->dev))
4403                 skb_dst_drop(skb);
4404
4405         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4406         kfree_skb(skb);
4407         return 0;
4408 }
4409
4410 static int ip6_pkt_discard(struct sk_buff *skb)
4411 {
4412         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4413 }
4414
4415 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4416 {
4417         skb->dev = skb_dst(skb)->dev;
4418         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4419 }
4420
4421 static int ip6_pkt_prohibit(struct sk_buff *skb)
4422 {
4423         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4424 }
4425
4426 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4427 {
4428         skb->dev = skb_dst(skb)->dev;
4429         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4430 }
4431
4432 /*
4433  *      Allocate a dst for local (unicast / anycast) address.
4434  */
4435
4436 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4437                                      struct inet6_dev *idev,
4438                                      const struct in6_addr *addr,
4439                                      bool anycast, gfp_t gfp_flags)
4440 {
4441         struct fib6_config cfg = {
4442                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4443                 .fc_ifindex = idev->dev->ifindex,
4444                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4445                 .fc_dst = *addr,
4446                 .fc_dst_len = 128,
4447                 .fc_protocol = RTPROT_KERNEL,
4448                 .fc_nlinfo.nl_net = net,
4449                 .fc_ignore_dev_down = true,
4450         };
4451         struct fib6_info *f6i;
4452
4453         if (anycast) {
4454                 cfg.fc_type = RTN_ANYCAST;
4455                 cfg.fc_flags |= RTF_ANYCAST;
4456         } else {
4457                 cfg.fc_type = RTN_LOCAL;
4458                 cfg.fc_flags |= RTF_LOCAL;
4459         }
4460
4461         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4462         if (!IS_ERR(f6i))
4463                 f6i->dst_nocount = true;
4464         return f6i;
4465 }
4466
4467 /* remove deleted ip from prefsrc entries */
4468 struct arg_dev_net_ip {
4469         struct net_device *dev;
4470         struct net *net;
4471         struct in6_addr *addr;
4472 };
4473
4474 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4475 {
4476         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4477         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4478         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4479
4480         if (!rt->nh &&
4481             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4482             rt != net->ipv6.fib6_null_entry &&
4483             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4484                 spin_lock_bh(&rt6_exception_lock);
4485                 /* remove prefsrc entry */
4486                 rt->fib6_prefsrc.plen = 0;
4487                 spin_unlock_bh(&rt6_exception_lock);
4488         }
4489         return 0;
4490 }
4491
4492 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4493 {
4494         struct net *net = dev_net(ifp->idev->dev);
4495         struct arg_dev_net_ip adni = {
4496                 .dev = ifp->idev->dev,
4497                 .net = net,
4498                 .addr = &ifp->addr,
4499         };
4500         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4501 }
4502
4503 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4504
4505 /* Remove routers and update dst entries when gateway turn into host. */
4506 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4507 {
4508         struct in6_addr *gateway = (struct in6_addr *)arg;
4509         struct fib6_nh *nh;
4510
4511         /* RA routes do not use nexthops */
4512         if (rt->nh)
4513                 return 0;
4514
4515         nh = rt->fib6_nh;
4516         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4517             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4518                 return -1;
4519
4520         /* Further clean up cached routes in exception table.
4521          * This is needed because cached route may have a different
4522          * gateway than its 'parent' in the case of an ip redirect.
4523          */
4524         fib6_nh_exceptions_clean_tohost(nh, gateway);
4525
4526         return 0;
4527 }
4528
4529 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4530 {
4531         fib6_clean_all(net, fib6_clean_tohost, gateway);
4532 }
4533
4534 struct arg_netdev_event {
4535         const struct net_device *dev;
4536         union {
4537                 unsigned char nh_flags;
4538                 unsigned long event;
4539         };
4540 };
4541
4542 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4543 {
4544         struct fib6_info *iter;
4545         struct fib6_node *fn;
4546
4547         fn = rcu_dereference_protected(rt->fib6_node,
4548                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4549         iter = rcu_dereference_protected(fn->leaf,
4550                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4551         while (iter) {
4552                 if (iter->fib6_metric == rt->fib6_metric &&
4553                     rt6_qualify_for_ecmp(iter))
4554                         return iter;
4555                 iter = rcu_dereference_protected(iter->fib6_next,
4556                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4557         }
4558
4559         return NULL;
4560 }
4561
4562 /* only called for fib entries with builtin fib6_nh */
4563 static bool rt6_is_dead(const struct fib6_info *rt)
4564 {
4565         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4566             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4567              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4568                 return true;
4569
4570         return false;
4571 }
4572
4573 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4574 {
4575         struct fib6_info *iter;
4576         int total = 0;
4577
4578         if (!rt6_is_dead(rt))
4579                 total += rt->fib6_nh->fib_nh_weight;
4580
4581         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4582                 if (!rt6_is_dead(iter))
4583                         total += iter->fib6_nh->fib_nh_weight;
4584         }
4585
4586         return total;
4587 }
4588
4589 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4590 {
4591         int upper_bound = -1;
4592
4593         if (!rt6_is_dead(rt)) {
4594                 *weight += rt->fib6_nh->fib_nh_weight;
4595                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4596                                                     total) - 1;
4597         }
4598         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4599 }
4600
4601 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4602 {
4603         struct fib6_info *iter;
4604         int weight = 0;
4605
4606         rt6_upper_bound_set(rt, &weight, total);
4607
4608         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4609                 rt6_upper_bound_set(iter, &weight, total);
4610 }
4611
4612 void rt6_multipath_rebalance(struct fib6_info *rt)
4613 {
4614         struct fib6_info *first;
4615         int total;
4616
4617         /* In case the entire multipath route was marked for flushing,
4618          * then there is no need to rebalance upon the removal of every
4619          * sibling route.
4620          */
4621         if (!rt->fib6_nsiblings || rt->should_flush)
4622                 return;
4623
4624         /* During lookup routes are evaluated in order, so we need to
4625          * make sure upper bounds are assigned from the first sibling
4626          * onwards.
4627          */
4628         first = rt6_multipath_first_sibling(rt);
4629         if (WARN_ON_ONCE(!first))
4630                 return;
4631
4632         total = rt6_multipath_total_weight(first);
4633         rt6_multipath_upper_bound_set(first, total);
4634 }
4635
4636 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4637 {
4638         const struct arg_netdev_event *arg = p_arg;
4639         struct net *net = dev_net(arg->dev);
4640
4641         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4642             rt->fib6_nh->fib_nh_dev == arg->dev) {
4643                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4644                 fib6_update_sernum_upto_root(net, rt);
4645                 rt6_multipath_rebalance(rt);
4646         }
4647
4648         return 0;
4649 }
4650
4651 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4652 {
4653         struct arg_netdev_event arg = {
4654                 .dev = dev,
4655                 {
4656                         .nh_flags = nh_flags,
4657                 },
4658         };
4659
4660         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4661                 arg.nh_flags |= RTNH_F_LINKDOWN;
4662
4663         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4664 }
4665
4666 /* only called for fib entries with inline fib6_nh */
4667 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4668                                    const struct net_device *dev)
4669 {
4670         struct fib6_info *iter;
4671
4672         if (rt->fib6_nh->fib_nh_dev == dev)
4673                 return true;
4674         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4675                 if (iter->fib6_nh->fib_nh_dev == dev)
4676                         return true;
4677
4678         return false;
4679 }
4680
4681 static void rt6_multipath_flush(struct fib6_info *rt)
4682 {
4683         struct fib6_info *iter;
4684
4685         rt->should_flush = 1;
4686         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4687                 iter->should_flush = 1;
4688 }
4689
4690 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4691                                              const struct net_device *down_dev)
4692 {
4693         struct fib6_info *iter;
4694         unsigned int dead = 0;
4695
4696         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4697             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4698                 dead++;
4699         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4700                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4701                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4702                         dead++;
4703
4704         return dead;
4705 }
4706
4707 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4708                                        const struct net_device *dev,
4709                                        unsigned char nh_flags)
4710 {
4711         struct fib6_info *iter;
4712
4713         if (rt->fib6_nh->fib_nh_dev == dev)
4714                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4715         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4716                 if (iter->fib6_nh->fib_nh_dev == dev)
4717                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4718 }
4719
4720 /* called with write lock held for table with rt */
4721 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4722 {
4723         const struct arg_netdev_event *arg = p_arg;
4724         const struct net_device *dev = arg->dev;
4725         struct net *net = dev_net(dev);
4726
4727         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4728                 return 0;
4729
4730         switch (arg->event) {
4731         case NETDEV_UNREGISTER:
4732                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4733         case NETDEV_DOWN:
4734                 if (rt->should_flush)
4735                         return -1;
4736                 if (!rt->fib6_nsiblings)
4737                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4738                 if (rt6_multipath_uses_dev(rt, dev)) {
4739                         unsigned int count;
4740
4741                         count = rt6_multipath_dead_count(rt, dev);
4742                         if (rt->fib6_nsiblings + 1 == count) {
4743                                 rt6_multipath_flush(rt);
4744                                 return -1;
4745                         }
4746                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4747                                                    RTNH_F_LINKDOWN);
4748                         fib6_update_sernum(net, rt);
4749                         rt6_multipath_rebalance(rt);
4750                 }
4751                 return -2;
4752         case NETDEV_CHANGE:
4753                 if (rt->fib6_nh->fib_nh_dev != dev ||
4754                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4755                         break;
4756                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4757                 rt6_multipath_rebalance(rt);
4758                 break;
4759         }
4760
4761         return 0;
4762 }
4763
4764 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4765 {
4766         struct arg_netdev_event arg = {
4767                 .dev = dev,
4768                 {
4769                         .event = event,
4770                 },
4771         };
4772         struct net *net = dev_net(dev);
4773
4774         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4775                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4776         else
4777                 fib6_clean_all(net, fib6_ifdown, &arg);
4778 }
4779
4780 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4781 {
4782         rt6_sync_down_dev(dev, event);
4783         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4784         neigh_ifdown(&nd_tbl, dev);
4785 }
4786
4787 struct rt6_mtu_change_arg {
4788         struct net_device *dev;
4789         unsigned int mtu;
4790         struct fib6_info *f6i;
4791 };
4792
4793 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4794 {
4795         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4796         struct fib6_info *f6i = arg->f6i;
4797
4798         /* For administrative MTU increase, there is no way to discover
4799          * IPv6 PMTU increase, so PMTU increase should be updated here.
4800          * Since RFC 1981 doesn't include administrative MTU increase
4801          * update PMTU increase is a MUST. (i.e. jumbo frame)
4802          */
4803         if (nh->fib_nh_dev == arg->dev) {
4804                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4805                 u32 mtu = f6i->fib6_pmtu;
4806
4807                 if (mtu >= arg->mtu ||
4808                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4809                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4810
4811                 spin_lock_bh(&rt6_exception_lock);
4812                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4813                 spin_unlock_bh(&rt6_exception_lock);
4814         }
4815
4816         return 0;
4817 }
4818
4819 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4820 {
4821         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4822         struct inet6_dev *idev;
4823
4824         /* In IPv6 pmtu discovery is not optional,
4825            so that RTAX_MTU lock cannot disable it.
4826            We still use this lock to block changes
4827            caused by addrconf/ndisc.
4828         */
4829
4830         idev = __in6_dev_get(arg->dev);
4831         if (!idev)
4832                 return 0;
4833
4834         if (fib6_metric_locked(f6i, RTAX_MTU))
4835                 return 0;
4836
4837         arg->f6i = f6i;
4838         if (f6i->nh) {
4839                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4840                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4841                                                 arg);
4842         }
4843
4844         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4845 }
4846
4847 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4848 {
4849         struct rt6_mtu_change_arg arg = {
4850                 .dev = dev,
4851                 .mtu = mtu,
4852         };
4853
4854         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4855 }
4856
4857 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4858         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4859         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4860         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4861         [RTA_OIF]               = { .type = NLA_U32 },
4862         [RTA_IIF]               = { .type = NLA_U32 },
4863         [RTA_PRIORITY]          = { .type = NLA_U32 },
4864         [RTA_METRICS]           = { .type = NLA_NESTED },
4865         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4866         [RTA_PREF]              = { .type = NLA_U8 },
4867         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4868         [RTA_ENCAP]             = { .type = NLA_NESTED },
4869         [RTA_EXPIRES]           = { .type = NLA_U32 },
4870         [RTA_UID]               = { .type = NLA_U32 },
4871         [RTA_MARK]              = { .type = NLA_U32 },
4872         [RTA_TABLE]             = { .type = NLA_U32 },
4873         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4874         [RTA_SPORT]             = { .type = NLA_U16 },
4875         [RTA_DPORT]             = { .type = NLA_U16 },
4876         [RTA_NH_ID]             = { .type = NLA_U32 },
4877 };
4878
4879 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4880                               struct fib6_config *cfg,
4881                               struct netlink_ext_ack *extack)
4882 {
4883         struct rtmsg *rtm;
4884         struct nlattr *tb[RTA_MAX+1];
4885         unsigned int pref;
4886         int err;
4887
4888         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4889                                      rtm_ipv6_policy, extack);
4890         if (err < 0)
4891                 goto errout;
4892
4893         err = -EINVAL;
4894         rtm = nlmsg_data(nlh);
4895
4896         *cfg = (struct fib6_config){
4897                 .fc_table = rtm->rtm_table,
4898                 .fc_dst_len = rtm->rtm_dst_len,
4899                 .fc_src_len = rtm->rtm_src_len,
4900                 .fc_flags = RTF_UP,
4901                 .fc_protocol = rtm->rtm_protocol,
4902                 .fc_type = rtm->rtm_type,
4903
4904                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4905                 .fc_nlinfo.nlh = nlh,
4906                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4907         };
4908
4909         if (rtm->rtm_type == RTN_UNREACHABLE ||
4910             rtm->rtm_type == RTN_BLACKHOLE ||
4911             rtm->rtm_type == RTN_PROHIBIT ||
4912             rtm->rtm_type == RTN_THROW)
4913                 cfg->fc_flags |= RTF_REJECT;
4914
4915         if (rtm->rtm_type == RTN_LOCAL)
4916                 cfg->fc_flags |= RTF_LOCAL;
4917
4918         if (rtm->rtm_flags & RTM_F_CLONED)
4919                 cfg->fc_flags |= RTF_CACHE;
4920
4921         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4922
4923         if (tb[RTA_NH_ID]) {
4924                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4925                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4926                         NL_SET_ERR_MSG(extack,
4927                                        "Nexthop specification and nexthop id are mutually exclusive");
4928                         goto errout;
4929                 }
4930                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4931         }
4932
4933         if (tb[RTA_GATEWAY]) {
4934                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4935                 cfg->fc_flags |= RTF_GATEWAY;
4936         }
4937         if (tb[RTA_VIA]) {
4938                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4939                 goto errout;
4940         }
4941
4942         if (tb[RTA_DST]) {
4943                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4944
4945                 if (nla_len(tb[RTA_DST]) < plen)
4946                         goto errout;
4947
4948                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4949         }
4950
4951         if (tb[RTA_SRC]) {
4952                 int plen = (rtm->rtm_src_len + 7) >> 3;
4953
4954                 if (nla_len(tb[RTA_SRC]) < plen)
4955                         goto errout;
4956
4957                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4958         }
4959
4960         if (tb[RTA_PREFSRC])
4961                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4962
4963         if (tb[RTA_OIF])
4964                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4965
4966         if (tb[RTA_PRIORITY])
4967                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4968
4969         if (tb[RTA_METRICS]) {
4970                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4971                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4972         }
4973
4974         if (tb[RTA_TABLE])
4975                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4976
4977         if (tb[RTA_MULTIPATH]) {
4978                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4979                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4980
4981                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4982                                                      cfg->fc_mp_len, extack);
4983                 if (err < 0)
4984                         goto errout;
4985         }
4986
4987         if (tb[RTA_PREF]) {
4988                 pref = nla_get_u8(tb[RTA_PREF]);
4989                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4990                     pref != ICMPV6_ROUTER_PREF_HIGH)
4991                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4992                 cfg->fc_flags |= RTF_PREF(pref);
4993         }
4994
4995         if (tb[RTA_ENCAP])
4996                 cfg->fc_encap = tb[RTA_ENCAP];
4997
4998         if (tb[RTA_ENCAP_TYPE]) {
4999                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5000
5001                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5002                 if (err < 0)
5003                         goto errout;
5004         }
5005
5006         if (tb[RTA_EXPIRES]) {
5007                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5008
5009                 if (addrconf_finite_timeout(timeout)) {
5010                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5011                         cfg->fc_flags |= RTF_EXPIRES;
5012                 }
5013         }
5014
5015         err = 0;
5016 errout:
5017         return err;
5018 }
5019
5020 struct rt6_nh {
5021         struct fib6_info *fib6_info;
5022         struct fib6_config r_cfg;
5023         struct list_head next;
5024 };
5025
5026 static int ip6_route_info_append(struct net *net,
5027                                  struct list_head *rt6_nh_list,
5028                                  struct fib6_info *rt,
5029                                  struct fib6_config *r_cfg)
5030 {
5031         struct rt6_nh *nh;
5032         int err = -EEXIST;
5033
5034         list_for_each_entry(nh, rt6_nh_list, next) {
5035                 /* check if fib6_info already exists */
5036                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5037                         return err;
5038         }
5039
5040         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5041         if (!nh)
5042                 return -ENOMEM;
5043         nh->fib6_info = rt;
5044         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5045         list_add_tail(&nh->next, rt6_nh_list);
5046
5047         return 0;
5048 }
5049
5050 static void ip6_route_mpath_notify(struct fib6_info *rt,
5051                                    struct fib6_info *rt_last,
5052                                    struct nl_info *info,
5053                                    __u16 nlflags)
5054 {
5055         /* if this is an APPEND route, then rt points to the first route
5056          * inserted and rt_last points to last route inserted. Userspace
5057          * wants a consistent dump of the route which starts at the first
5058          * nexthop. Since sibling routes are always added at the end of
5059          * the list, find the first sibling of the last route appended
5060          */
5061         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5062                 rt = list_first_entry(&rt_last->fib6_siblings,
5063                                       struct fib6_info,
5064                                       fib6_siblings);
5065         }
5066
5067         if (rt)
5068                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5069 }
5070
5071 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5072 {
5073         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5074         bool should_notify = false;
5075         struct fib6_info *leaf;
5076         struct fib6_node *fn;
5077
5078         rcu_read_lock();
5079         fn = rcu_dereference(rt->fib6_node);
5080         if (!fn)
5081                 goto out;
5082
5083         leaf = rcu_dereference(fn->leaf);
5084         if (!leaf)
5085                 goto out;
5086
5087         if (rt == leaf ||
5088             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5089              rt6_qualify_for_ecmp(leaf)))
5090                 should_notify = true;
5091 out:
5092         rcu_read_unlock();
5093
5094         return should_notify;
5095 }
5096
5097 static int ip6_route_multipath_add(struct fib6_config *cfg,
5098                                    struct netlink_ext_ack *extack)
5099 {
5100         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5101         struct nl_info *info = &cfg->fc_nlinfo;
5102         struct fib6_config r_cfg;
5103         struct rtnexthop *rtnh;
5104         struct fib6_info *rt;
5105         struct rt6_nh *err_nh;
5106         struct rt6_nh *nh, *nh_safe;
5107         __u16 nlflags;
5108         int remaining;
5109         int attrlen;
5110         int err = 1;
5111         int nhn = 0;
5112         int replace = (cfg->fc_nlinfo.nlh &&
5113                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5114         LIST_HEAD(rt6_nh_list);
5115
5116         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5117         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5118                 nlflags |= NLM_F_APPEND;
5119
5120         remaining = cfg->fc_mp_len;
5121         rtnh = (struct rtnexthop *)cfg->fc_mp;
5122
5123         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5124          * fib6_info structs per nexthop
5125          */
5126         while (rtnh_ok(rtnh, remaining)) {
5127                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5128                 if (rtnh->rtnh_ifindex)
5129                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5130
5131                 attrlen = rtnh_attrlen(rtnh);
5132                 if (attrlen > 0) {
5133                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5134
5135                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5136                         if (nla) {
5137                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5138                                 r_cfg.fc_flags |= RTF_GATEWAY;
5139                         }
5140                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5141                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5142                         if (nla)
5143                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5144                 }
5145
5146                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5147                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5148                 if (IS_ERR(rt)) {
5149                         err = PTR_ERR(rt);
5150                         rt = NULL;
5151                         goto cleanup;
5152                 }
5153                 if (!rt6_qualify_for_ecmp(rt)) {
5154                         err = -EINVAL;
5155                         NL_SET_ERR_MSG(extack,
5156                                        "Device only routes can not be added for IPv6 using the multipath API.");
5157                         fib6_info_release(rt);
5158                         goto cleanup;
5159                 }
5160
5161                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5162
5163                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5164                                             rt, &r_cfg);
5165                 if (err) {
5166                         fib6_info_release(rt);
5167                         goto cleanup;
5168                 }
5169
5170                 rtnh = rtnh_next(rtnh, &remaining);
5171         }
5172
5173         if (list_empty(&rt6_nh_list)) {
5174                 NL_SET_ERR_MSG(extack,
5175                                "Invalid nexthop configuration - no valid nexthops");
5176                 return -EINVAL;
5177         }
5178
5179         /* for add and replace send one notification with all nexthops.
5180          * Skip the notification in fib6_add_rt2node and send one with
5181          * the full route when done
5182          */
5183         info->skip_notify = 1;
5184
5185         /* For add and replace, send one notification with all nexthops. For
5186          * append, send one notification with all appended nexthops.
5187          */
5188         info->skip_notify_kernel = 1;
5189
5190         err_nh = NULL;
5191         list_for_each_entry(nh, &rt6_nh_list, next) {
5192                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5193                 fib6_info_release(nh->fib6_info);
5194
5195                 if (!err) {
5196                         /* save reference to last route successfully inserted */
5197                         rt_last = nh->fib6_info;
5198
5199                         /* save reference to first route for notification */
5200                         if (!rt_notif)
5201                                 rt_notif = nh->fib6_info;
5202                 }
5203
5204                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5205                 nh->fib6_info = NULL;
5206                 if (err) {
5207                         if (replace && nhn)
5208                                 NL_SET_ERR_MSG_MOD(extack,
5209                                                    "multipath route replace failed (check consistency of installed routes)");
5210                         err_nh = nh;
5211                         goto add_errout;
5212                 }
5213
5214                 /* Because each route is added like a single route we remove
5215                  * these flags after the first nexthop: if there is a collision,
5216                  * we have already failed to add the first nexthop:
5217                  * fib6_add_rt2node() has rejected it; when replacing, old
5218                  * nexthops have been replaced by first new, the rest should
5219                  * be added to it.
5220                  */
5221                 if (cfg->fc_nlinfo.nlh) {
5222                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5223                                                              NLM_F_REPLACE);
5224                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5225                 }
5226                 nhn++;
5227         }
5228
5229         /* An in-kernel notification should only be sent in case the new
5230          * multipath route is added as the first route in the node, or if
5231          * it was appended to it. We pass 'rt_notif' since it is the first
5232          * sibling and might allow us to skip some checks in the replace case.
5233          */
5234         if (ip6_route_mpath_should_notify(rt_notif)) {
5235                 enum fib_event_type fib_event;
5236
5237                 if (rt_notif->fib6_nsiblings != nhn - 1)
5238                         fib_event = FIB_EVENT_ENTRY_APPEND;
5239                 else
5240                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5241
5242                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5243                                                           fib_event, rt_notif,
5244                                                           nhn - 1, extack);
5245                 if (err) {
5246                         /* Delete all the siblings that were just added */
5247                         err_nh = NULL;
5248                         goto add_errout;
5249                 }
5250         }
5251
5252         /* success ... tell user about new route */
5253         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5254         goto cleanup;
5255
5256 add_errout:
5257         /* send notification for routes that were added so that
5258          * the delete notifications sent by ip6_route_del are
5259          * coherent
5260          */
5261         if (rt_notif)
5262                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5263
5264         /* Delete routes that were already added */
5265         list_for_each_entry(nh, &rt6_nh_list, next) {
5266                 if (err_nh == nh)
5267                         break;
5268                 ip6_route_del(&nh->r_cfg, extack);
5269         }
5270
5271 cleanup:
5272         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5273                 if (nh->fib6_info)
5274                         fib6_info_release(nh->fib6_info);
5275                 list_del(&nh->next);
5276                 kfree(nh);
5277         }
5278
5279         return err;
5280 }
5281
5282 static int ip6_route_multipath_del(struct fib6_config *cfg,
5283                                    struct netlink_ext_ack *extack)
5284 {
5285         struct fib6_config r_cfg;
5286         struct rtnexthop *rtnh;
5287         int last_err = 0;
5288         int remaining;
5289         int attrlen;
5290         int err;
5291
5292         remaining = cfg->fc_mp_len;
5293         rtnh = (struct rtnexthop *)cfg->fc_mp;
5294
5295         /* Parse a Multipath Entry */
5296         while (rtnh_ok(rtnh, remaining)) {
5297                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5298                 if (rtnh->rtnh_ifindex)
5299                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5300
5301                 attrlen = rtnh_attrlen(rtnh);
5302                 if (attrlen > 0) {
5303                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5304
5305                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5306                         if (nla) {
5307                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5308                                 r_cfg.fc_flags |= RTF_GATEWAY;
5309                         }
5310                 }
5311                 err = ip6_route_del(&r_cfg, extack);
5312                 if (err)
5313                         last_err = err;
5314
5315                 rtnh = rtnh_next(rtnh, &remaining);
5316         }
5317
5318         return last_err;
5319 }
5320
5321 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5322                               struct netlink_ext_ack *extack)
5323 {
5324         struct fib6_config cfg;
5325         int err;
5326
5327         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5328         if (err < 0)
5329                 return err;
5330
5331         if (cfg.fc_nh_id &&
5332             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5333                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5334                 return -EINVAL;
5335         }
5336
5337         if (cfg.fc_mp)
5338                 return ip6_route_multipath_del(&cfg, extack);
5339         else {
5340                 cfg.fc_delete_all_nh = 1;
5341                 return ip6_route_del(&cfg, extack);
5342         }
5343 }
5344
5345 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5346                               struct netlink_ext_ack *extack)
5347 {
5348         struct fib6_config cfg;
5349         int err;
5350
5351         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5352         if (err < 0)
5353                 return err;
5354
5355         if (cfg.fc_metric == 0)
5356                 cfg.fc_metric = IP6_RT_PRIO_USER;
5357
5358         if (cfg.fc_mp)
5359                 return ip6_route_multipath_add(&cfg, extack);
5360         else
5361                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5362 }
5363
5364 /* add the overhead of this fib6_nh to nexthop_len */
5365 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5366 {
5367         int *nexthop_len = arg;
5368
5369         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5370                      + NLA_ALIGN(sizeof(struct rtnexthop))
5371                      + nla_total_size(16); /* RTA_GATEWAY */
5372
5373         if (nh->fib_nh_lws) {
5374                 /* RTA_ENCAP_TYPE */
5375                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5376                 /* RTA_ENCAP */
5377                 *nexthop_len += nla_total_size(2);
5378         }
5379
5380         return 0;
5381 }
5382
5383 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5384 {
5385         int nexthop_len;
5386
5387         if (f6i->nh) {
5388                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5389                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5390                                          &nexthop_len);
5391         } else {
5392                 struct fib6_nh *nh = f6i->fib6_nh;
5393
5394                 nexthop_len = 0;
5395                 if (f6i->fib6_nsiblings) {
5396                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5397                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5398                                     + nla_total_size(16) /* RTA_GATEWAY */
5399                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5400
5401                         nexthop_len *= f6i->fib6_nsiblings;
5402                 }
5403                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5404         }
5405
5406         return NLMSG_ALIGN(sizeof(struct rtmsg))
5407                + nla_total_size(16) /* RTA_SRC */
5408                + nla_total_size(16) /* RTA_DST */
5409                + nla_total_size(16) /* RTA_GATEWAY */
5410                + nla_total_size(16) /* RTA_PREFSRC */
5411                + nla_total_size(4) /* RTA_TABLE */
5412                + nla_total_size(4) /* RTA_IIF */
5413                + nla_total_size(4) /* RTA_OIF */
5414                + nla_total_size(4) /* RTA_PRIORITY */
5415                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5416                + nla_total_size(sizeof(struct rta_cacheinfo))
5417                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5418                + nla_total_size(1) /* RTA_PREF */
5419                + nexthop_len;
5420 }
5421
5422 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5423                                  unsigned char *flags)
5424 {
5425         if (nexthop_is_multipath(nh)) {
5426                 struct nlattr *mp;
5427
5428                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5429                 if (!mp)
5430                         goto nla_put_failure;
5431
5432                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5433                         goto nla_put_failure;
5434
5435                 nla_nest_end(skb, mp);
5436         } else {
5437                 struct fib6_nh *fib6_nh;
5438
5439                 fib6_nh = nexthop_fib6_nh(nh);
5440                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5441                                      flags, false) < 0)
5442                         goto nla_put_failure;
5443         }
5444
5445         return 0;
5446
5447 nla_put_failure:
5448         return -EMSGSIZE;
5449 }
5450
5451 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5452                          struct fib6_info *rt, struct dst_entry *dst,
5453                          struct in6_addr *dest, struct in6_addr *src,
5454                          int iif, int type, u32 portid, u32 seq,
5455                          unsigned int flags)
5456 {
5457         struct rt6_info *rt6 = (struct rt6_info *)dst;
5458         struct rt6key *rt6_dst, *rt6_src;
5459         u32 *pmetrics, table, rt6_flags;
5460         unsigned char nh_flags = 0;
5461         struct nlmsghdr *nlh;
5462         struct rtmsg *rtm;
5463         long expires = 0;
5464
5465         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5466         if (!nlh)
5467                 return -EMSGSIZE;
5468
5469         if (rt6) {
5470                 rt6_dst = &rt6->rt6i_dst;
5471                 rt6_src = &rt6->rt6i_src;
5472                 rt6_flags = rt6->rt6i_flags;
5473         } else {
5474                 rt6_dst = &rt->fib6_dst;
5475                 rt6_src = &rt->fib6_src;
5476                 rt6_flags = rt->fib6_flags;
5477         }
5478
5479         rtm = nlmsg_data(nlh);
5480         rtm->rtm_family = AF_INET6;
5481         rtm->rtm_dst_len = rt6_dst->plen;
5482         rtm->rtm_src_len = rt6_src->plen;
5483         rtm->rtm_tos = 0;
5484         if (rt->fib6_table)
5485                 table = rt->fib6_table->tb6_id;
5486         else
5487                 table = RT6_TABLE_UNSPEC;
5488         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5489         if (nla_put_u32(skb, RTA_TABLE, table))
5490                 goto nla_put_failure;
5491
5492         rtm->rtm_type = rt->fib6_type;
5493         rtm->rtm_flags = 0;
5494         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5495         rtm->rtm_protocol = rt->fib6_protocol;
5496
5497         if (rt6_flags & RTF_CACHE)
5498                 rtm->rtm_flags |= RTM_F_CLONED;
5499
5500         if (dest) {
5501                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5502                         goto nla_put_failure;
5503                 rtm->rtm_dst_len = 128;
5504         } else if (rtm->rtm_dst_len)
5505                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5506                         goto nla_put_failure;
5507 #ifdef CONFIG_IPV6_SUBTREES
5508         if (src) {
5509                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5510                         goto nla_put_failure;
5511                 rtm->rtm_src_len = 128;
5512         } else if (rtm->rtm_src_len &&
5513                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5514                 goto nla_put_failure;
5515 #endif
5516         if (iif) {
5517 #ifdef CONFIG_IPV6_MROUTE
5518                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5519                         int err = ip6mr_get_route(net, skb, rtm, portid);
5520
5521                         if (err == 0)
5522                                 return 0;
5523                         if (err < 0)
5524                                 goto nla_put_failure;
5525                 } else
5526 #endif
5527                         if (nla_put_u32(skb, RTA_IIF, iif))
5528                                 goto nla_put_failure;
5529         } else if (dest) {
5530                 struct in6_addr saddr_buf;
5531                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5532                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5533                         goto nla_put_failure;
5534         }
5535
5536         if (rt->fib6_prefsrc.plen) {
5537                 struct in6_addr saddr_buf;
5538                 saddr_buf = rt->fib6_prefsrc.addr;
5539                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5540                         goto nla_put_failure;
5541         }
5542
5543         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5544         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5545                 goto nla_put_failure;
5546
5547         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5548                 goto nla_put_failure;
5549
5550         /* For multipath routes, walk the siblings list and add
5551          * each as a nexthop within RTA_MULTIPATH.
5552          */
5553         if (rt6) {
5554                 if (rt6_flags & RTF_GATEWAY &&
5555                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5556                         goto nla_put_failure;
5557
5558                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5559                         goto nla_put_failure;
5560
5561                 if (dst->lwtstate &&
5562                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5563                         goto nla_put_failure;
5564         } else if (rt->fib6_nsiblings) {
5565                 struct fib6_info *sibling, *next_sibling;
5566                 struct nlattr *mp;
5567
5568                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5569                 if (!mp)
5570                         goto nla_put_failure;
5571
5572                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5573                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5574                         goto nla_put_failure;
5575
5576                 list_for_each_entry_safe(sibling, next_sibling,
5577                                          &rt->fib6_siblings, fib6_siblings) {
5578                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5579                                             sibling->fib6_nh->fib_nh_weight,
5580                                             AF_INET6) < 0)
5581                                 goto nla_put_failure;
5582                 }
5583
5584                 nla_nest_end(skb, mp);
5585         } else if (rt->nh) {
5586                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5587                         goto nla_put_failure;
5588
5589                 if (nexthop_is_blackhole(rt->nh))
5590                         rtm->rtm_type = RTN_BLACKHOLE;
5591
5592                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5593                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5594                         goto nla_put_failure;
5595
5596                 rtm->rtm_flags |= nh_flags;
5597         } else {
5598                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5599                                      &nh_flags, false) < 0)
5600                         goto nla_put_failure;
5601
5602                 rtm->rtm_flags |= nh_flags;
5603         }
5604
5605         if (rt6_flags & RTF_EXPIRES) {
5606                 expires = dst ? dst->expires : rt->expires;
5607                 expires -= jiffies;
5608         }
5609
5610         if (!dst) {
5611                 if (rt->offload)
5612                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5613                 if (rt->trap)
5614                         rtm->rtm_flags |= RTM_F_TRAP;
5615                 if (rt->offload_failed)
5616                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5617         }
5618
5619         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5620                 goto nla_put_failure;
5621
5622         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5623                 goto nla_put_failure;
5624
5625
5626         nlmsg_end(skb, nlh);
5627         return 0;
5628
5629 nla_put_failure:
5630         nlmsg_cancel(skb, nlh);
5631         return -EMSGSIZE;
5632 }
5633
5634 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5635 {
5636         const struct net_device *dev = arg;
5637
5638         if (nh->fib_nh_dev == dev)
5639                 return 1;
5640
5641         return 0;
5642 }
5643
5644 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5645                                const struct net_device *dev)
5646 {
5647         if (f6i->nh) {
5648                 struct net_device *_dev = (struct net_device *)dev;
5649
5650                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5651                                                   fib6_info_nh_uses_dev,
5652                                                   _dev);
5653         }
5654
5655         if (f6i->fib6_nh->fib_nh_dev == dev)
5656                 return true;
5657
5658         if (f6i->fib6_nsiblings) {
5659                 struct fib6_info *sibling, *next_sibling;
5660
5661                 list_for_each_entry_safe(sibling, next_sibling,
5662                                          &f6i->fib6_siblings, fib6_siblings) {
5663                         if (sibling->fib6_nh->fib_nh_dev == dev)
5664                                 return true;
5665                 }
5666         }
5667
5668         return false;
5669 }
5670
5671 struct fib6_nh_exception_dump_walker {
5672         struct rt6_rtnl_dump_arg *dump;
5673         struct fib6_info *rt;
5674         unsigned int flags;
5675         unsigned int skip;
5676         unsigned int count;
5677 };
5678
5679 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5680 {
5681         struct fib6_nh_exception_dump_walker *w = arg;
5682         struct rt6_rtnl_dump_arg *dump = w->dump;
5683         struct rt6_exception_bucket *bucket;
5684         struct rt6_exception *rt6_ex;
5685         int i, err;
5686
5687         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5688         if (!bucket)
5689                 return 0;
5690
5691         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5692                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5693                         if (w->skip) {
5694                                 w->skip--;
5695                                 continue;
5696                         }
5697
5698                         /* Expiration of entries doesn't bump sernum, insertion
5699                          * does. Removal is triggered by insertion, so we can
5700                          * rely on the fact that if entries change between two
5701                          * partial dumps, this node is scanned again completely,
5702                          * see rt6_insert_exception() and fib6_dump_table().
5703                          *
5704                          * Count expired entries we go through as handled
5705                          * entries that we'll skip next time, in case of partial
5706                          * node dump. Otherwise, if entries expire meanwhile,
5707                          * we'll skip the wrong amount.
5708                          */
5709                         if (rt6_check_expired(rt6_ex->rt6i)) {
5710                                 w->count++;
5711                                 continue;
5712                         }
5713
5714                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5715                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5716                                             RTM_NEWROUTE,
5717                                             NETLINK_CB(dump->cb->skb).portid,
5718                                             dump->cb->nlh->nlmsg_seq, w->flags);
5719                         if (err)
5720                                 return err;
5721
5722                         w->count++;
5723                 }
5724                 bucket++;
5725         }
5726
5727         return 0;
5728 }
5729
5730 /* Return -1 if done with node, number of handled routes on partial dump */
5731 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5732 {
5733         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5734         struct fib_dump_filter *filter = &arg->filter;
5735         unsigned int flags = NLM_F_MULTI;
5736         struct net *net = arg->net;
5737         int count = 0;
5738
5739         if (rt == net->ipv6.fib6_null_entry)
5740                 return -1;
5741
5742         if ((filter->flags & RTM_F_PREFIX) &&
5743             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5744                 /* success since this is not a prefix route */
5745                 return -1;
5746         }
5747         if (filter->filter_set &&
5748             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5749              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5750              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5751                 return -1;
5752         }
5753
5754         if (filter->filter_set ||
5755             !filter->dump_routes || !filter->dump_exceptions) {
5756                 flags |= NLM_F_DUMP_FILTERED;
5757         }
5758
5759         if (filter->dump_routes) {
5760                 if (skip) {
5761                         skip--;
5762                 } else {
5763                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5764                                           0, RTM_NEWROUTE,
5765                                           NETLINK_CB(arg->cb->skb).portid,
5766                                           arg->cb->nlh->nlmsg_seq, flags)) {
5767                                 return 0;
5768                         }
5769                         count++;
5770                 }
5771         }
5772
5773         if (filter->dump_exceptions) {
5774                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5775                                                            .rt = rt,
5776                                                            .flags = flags,
5777                                                            .skip = skip,
5778                                                            .count = 0 };
5779                 int err;
5780
5781                 rcu_read_lock();
5782                 if (rt->nh) {
5783                         err = nexthop_for_each_fib6_nh(rt->nh,
5784                                                        rt6_nh_dump_exceptions,
5785                                                        &w);
5786                 } else {
5787                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5788                 }
5789                 rcu_read_unlock();
5790
5791                 if (err)
5792                         return count += w.count;
5793         }
5794
5795         return -1;
5796 }
5797
5798 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5799                                         const struct nlmsghdr *nlh,
5800                                         struct nlattr **tb,
5801                                         struct netlink_ext_ack *extack)
5802 {
5803         struct rtmsg *rtm;
5804         int i, err;
5805
5806         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5807                 NL_SET_ERR_MSG_MOD(extack,
5808                                    "Invalid header for get route request");
5809                 return -EINVAL;
5810         }
5811
5812         if (!netlink_strict_get_check(skb))
5813                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5814                                               rtm_ipv6_policy, extack);
5815
5816         rtm = nlmsg_data(nlh);
5817         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5818             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5819             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5820             rtm->rtm_type) {
5821                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5822                 return -EINVAL;
5823         }
5824         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5825                 NL_SET_ERR_MSG_MOD(extack,
5826                                    "Invalid flags for get route request");
5827                 return -EINVAL;
5828         }
5829
5830         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5831                                             rtm_ipv6_policy, extack);
5832         if (err)
5833                 return err;
5834
5835         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5836             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5837                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5838                 return -EINVAL;
5839         }
5840
5841         for (i = 0; i <= RTA_MAX; i++) {
5842                 if (!tb[i])
5843                         continue;
5844
5845                 switch (i) {
5846                 case RTA_SRC:
5847                 case RTA_DST:
5848                 case RTA_IIF:
5849                 case RTA_OIF:
5850                 case RTA_MARK:
5851                 case RTA_UID:
5852                 case RTA_SPORT:
5853                 case RTA_DPORT:
5854                 case RTA_IP_PROTO:
5855                         break;
5856                 default:
5857                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5858                         return -EINVAL;
5859                 }
5860         }
5861
5862         return 0;
5863 }
5864
5865 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5866                               struct netlink_ext_ack *extack)
5867 {
5868         struct net *net = sock_net(in_skb->sk);
5869         struct nlattr *tb[RTA_MAX+1];
5870         int err, iif = 0, oif = 0;
5871         struct fib6_info *from;
5872         struct dst_entry *dst;
5873         struct rt6_info *rt;
5874         struct sk_buff *skb;
5875         struct rtmsg *rtm;
5876         struct flowi6 fl6 = {};
5877         bool fibmatch;
5878
5879         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5880         if (err < 0)
5881                 goto errout;
5882
5883         err = -EINVAL;
5884         rtm = nlmsg_data(nlh);
5885         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5886         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5887
5888         if (tb[RTA_SRC]) {
5889                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5890                         goto errout;
5891
5892                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5893         }
5894
5895         if (tb[RTA_DST]) {
5896                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5897                         goto errout;
5898
5899                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5900         }
5901
5902         if (tb[RTA_IIF])
5903                 iif = nla_get_u32(tb[RTA_IIF]);
5904
5905         if (tb[RTA_OIF])
5906                 oif = nla_get_u32(tb[RTA_OIF]);
5907
5908         if (tb[RTA_MARK])
5909                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5910
5911         if (tb[RTA_UID])
5912                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5913                                            nla_get_u32(tb[RTA_UID]));
5914         else
5915                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5916
5917         if (tb[RTA_SPORT])
5918                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5919
5920         if (tb[RTA_DPORT])
5921                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5922
5923         if (tb[RTA_IP_PROTO]) {
5924                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5925                                                   &fl6.flowi6_proto, AF_INET6,
5926                                                   extack);
5927                 if (err)
5928                         goto errout;
5929         }
5930
5931         if (iif) {
5932                 struct net_device *dev;
5933                 int flags = 0;
5934
5935                 rcu_read_lock();
5936
5937                 dev = dev_get_by_index_rcu(net, iif);
5938                 if (!dev) {
5939                         rcu_read_unlock();
5940                         err = -ENODEV;
5941                         goto errout;
5942                 }
5943
5944                 fl6.flowi6_iif = iif;
5945
5946                 if (!ipv6_addr_any(&fl6.saddr))
5947                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5948
5949                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5950
5951                 rcu_read_unlock();
5952         } else {
5953                 fl6.flowi6_oif = oif;
5954
5955                 dst = ip6_route_output(net, NULL, &fl6);
5956         }
5957
5958
5959         rt = container_of(dst, struct rt6_info, dst);
5960         if (rt->dst.error) {
5961                 err = rt->dst.error;
5962                 ip6_rt_put(rt);
5963                 goto errout;
5964         }
5965
5966         if (rt == net->ipv6.ip6_null_entry) {
5967                 err = rt->dst.error;
5968                 ip6_rt_put(rt);
5969                 goto errout;
5970         }
5971
5972         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5973         if (!skb) {
5974                 ip6_rt_put(rt);
5975                 err = -ENOBUFS;
5976                 goto errout;
5977         }
5978
5979         skb_dst_set(skb, &rt->dst);
5980
5981         rcu_read_lock();
5982         from = rcu_dereference(rt->from);
5983         if (from) {
5984                 if (fibmatch)
5985                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5986                                             iif, RTM_NEWROUTE,
5987                                             NETLINK_CB(in_skb).portid,
5988                                             nlh->nlmsg_seq, 0);
5989                 else
5990                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5991                                             &fl6.saddr, iif, RTM_NEWROUTE,
5992                                             NETLINK_CB(in_skb).portid,
5993                                             nlh->nlmsg_seq, 0);
5994         } else {
5995                 err = -ENETUNREACH;
5996         }
5997         rcu_read_unlock();
5998
5999         if (err < 0) {
6000                 kfree_skb(skb);
6001                 goto errout;
6002         }
6003
6004         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6005 errout:
6006         return err;
6007 }
6008
6009 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6010                      unsigned int nlm_flags)
6011 {
6012         struct sk_buff *skb;
6013         struct net *net = info->nl_net;
6014         u32 seq;
6015         int err;
6016
6017         err = -ENOBUFS;
6018         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6019
6020         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6021         if (!skb)
6022                 goto errout;
6023
6024         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6025                             event, info->portid, seq, nlm_flags);
6026         if (err < 0) {
6027                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6028                 WARN_ON(err == -EMSGSIZE);
6029                 kfree_skb(skb);
6030                 goto errout;
6031         }
6032         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6033                     info->nlh, gfp_any());
6034         return;
6035 errout:
6036         if (err < 0)
6037                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6038 }
6039
6040 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6041                     struct nl_info *info)
6042 {
6043         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6044         struct sk_buff *skb;
6045         int err = -ENOBUFS;
6046
6047         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6048         if (!skb)
6049                 goto errout;
6050
6051         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6052                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6053         if (err < 0) {
6054                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6055                 WARN_ON(err == -EMSGSIZE);
6056                 kfree_skb(skb);
6057                 goto errout;
6058         }
6059         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6060                     info->nlh, gfp_any());
6061         return;
6062 errout:
6063         if (err < 0)
6064                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6065 }
6066
6067 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6068                             bool offload, bool trap, bool offload_failed)
6069 {
6070         struct sk_buff *skb;
6071         int err;
6072
6073         if (f6i->offload == offload && f6i->trap == trap &&
6074             f6i->offload_failed == offload_failed)
6075                 return;
6076
6077         f6i->offload = offload;
6078         f6i->trap = trap;
6079
6080         /* 2 means send notifications only if offload_failed was changed. */
6081         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6082             f6i->offload_failed == offload_failed)
6083                 return;
6084
6085         f6i->offload_failed = offload_failed;
6086
6087         if (!rcu_access_pointer(f6i->fib6_node))
6088                 /* The route was removed from the tree, do not send
6089                  * notification.
6090                  */
6091                 return;
6092
6093         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6094                 return;
6095
6096         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6097         if (!skb) {
6098                 err = -ENOBUFS;
6099                 goto errout;
6100         }
6101
6102         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6103                             0, 0);
6104         if (err < 0) {
6105                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6106                 WARN_ON(err == -EMSGSIZE);
6107                 kfree_skb(skb);
6108                 goto errout;
6109         }
6110
6111         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6112         return;
6113
6114 errout:
6115         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6116 }
6117 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6118
6119 static int ip6_route_dev_notify(struct notifier_block *this,
6120                                 unsigned long event, void *ptr)
6121 {
6122         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6123         struct net *net = dev_net(dev);
6124
6125         if (!(dev->flags & IFF_LOOPBACK))
6126                 return NOTIFY_OK;
6127
6128         if (event == NETDEV_REGISTER) {
6129                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6130                 net->ipv6.ip6_null_entry->dst.dev = dev;
6131                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6132 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6133                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6134                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6135                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6136                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6137 #endif
6138          } else if (event == NETDEV_UNREGISTER &&
6139                     dev->reg_state != NETREG_UNREGISTERED) {
6140                 /* NETDEV_UNREGISTER could be fired for multiple times by
6141                  * netdev_wait_allrefs(). Make sure we only call this once.
6142                  */
6143                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6144 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6145                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6146                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6147 #endif
6148         }
6149
6150         return NOTIFY_OK;
6151 }
6152
6153 /*
6154  *      /proc
6155  */
6156
6157 #ifdef CONFIG_PROC_FS
6158 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6159 {
6160         struct net *net = (struct net *)seq->private;
6161         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6162                    net->ipv6.rt6_stats->fib_nodes,
6163                    net->ipv6.rt6_stats->fib_route_nodes,
6164                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6165                    net->ipv6.rt6_stats->fib_rt_entries,
6166                    net->ipv6.rt6_stats->fib_rt_cache,
6167                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6168                    net->ipv6.rt6_stats->fib_discarded_routes);
6169
6170         return 0;
6171 }
6172 #endif  /* CONFIG_PROC_FS */
6173
6174 #ifdef CONFIG_SYSCTL
6175
6176 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6177                               void *buffer, size_t *lenp, loff_t *ppos)
6178 {
6179         struct net *net;
6180         int delay;
6181         int ret;
6182         if (!write)
6183                 return -EINVAL;
6184
6185         net = (struct net *)ctl->extra1;
6186         delay = net->ipv6.sysctl.flush_delay;
6187         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6188         if (ret)
6189                 return ret;
6190
6191         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6192         return 0;
6193 }
6194
6195 static struct ctl_table ipv6_route_table_template[] = {
6196         {
6197                 .procname       =       "flush",
6198                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6199                 .maxlen         =       sizeof(int),
6200                 .mode           =       0200,
6201                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6202         },
6203         {
6204                 .procname       =       "gc_thresh",
6205                 .data           =       &ip6_dst_ops_template.gc_thresh,
6206                 .maxlen         =       sizeof(int),
6207                 .mode           =       0644,
6208                 .proc_handler   =       proc_dointvec,
6209         },
6210         {
6211                 .procname       =       "max_size",
6212                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6213                 .maxlen         =       sizeof(int),
6214                 .mode           =       0644,
6215                 .proc_handler   =       proc_dointvec,
6216         },
6217         {
6218                 .procname       =       "gc_min_interval",
6219                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6220                 .maxlen         =       sizeof(int),
6221                 .mode           =       0644,
6222                 .proc_handler   =       proc_dointvec_jiffies,
6223         },
6224         {
6225                 .procname       =       "gc_timeout",
6226                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6227                 .maxlen         =       sizeof(int),
6228                 .mode           =       0644,
6229                 .proc_handler   =       proc_dointvec_jiffies,
6230         },
6231         {
6232                 .procname       =       "gc_interval",
6233                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6234                 .maxlen         =       sizeof(int),
6235                 .mode           =       0644,
6236                 .proc_handler   =       proc_dointvec_jiffies,
6237         },
6238         {
6239                 .procname       =       "gc_elasticity",
6240                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6241                 .maxlen         =       sizeof(int),
6242                 .mode           =       0644,
6243                 .proc_handler   =       proc_dointvec,
6244         },
6245         {
6246                 .procname       =       "mtu_expires",
6247                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6248                 .maxlen         =       sizeof(int),
6249                 .mode           =       0644,
6250                 .proc_handler   =       proc_dointvec_jiffies,
6251         },
6252         {
6253                 .procname       =       "min_adv_mss",
6254                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6255                 .maxlen         =       sizeof(int),
6256                 .mode           =       0644,
6257                 .proc_handler   =       proc_dointvec,
6258         },
6259         {
6260                 .procname       =       "gc_min_interval_ms",
6261                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6262                 .maxlen         =       sizeof(int),
6263                 .mode           =       0644,
6264                 .proc_handler   =       proc_dointvec_ms_jiffies,
6265         },
6266         {
6267                 .procname       =       "skip_notify_on_dev_down",
6268                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6269                 .maxlen         =       sizeof(int),
6270                 .mode           =       0644,
6271                 .proc_handler   =       proc_dointvec_minmax,
6272                 .extra1         =       SYSCTL_ZERO,
6273                 .extra2         =       SYSCTL_ONE,
6274         },
6275         { }
6276 };
6277
6278 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6279 {
6280         struct ctl_table *table;
6281
6282         table = kmemdup(ipv6_route_table_template,
6283                         sizeof(ipv6_route_table_template),
6284                         GFP_KERNEL);
6285
6286         if (table) {
6287                 table[0].data = &net->ipv6.sysctl.flush_delay;
6288                 table[0].extra1 = net;
6289                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6290                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6291                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6292                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6293                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6294                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6295                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6296                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6297                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6298                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6299
6300                 /* Don't export sysctls to unprivileged users */
6301                 if (net->user_ns != &init_user_ns)
6302                         table[0].procname = NULL;
6303         }
6304
6305         return table;
6306 }
6307 #endif
6308
6309 static int __net_init ip6_route_net_init(struct net *net)
6310 {
6311         int ret = -ENOMEM;
6312
6313         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6314                sizeof(net->ipv6.ip6_dst_ops));
6315
6316         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6317                 goto out_ip6_dst_ops;
6318
6319         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6320         if (!net->ipv6.fib6_null_entry)
6321                 goto out_ip6_dst_entries;
6322         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6323                sizeof(*net->ipv6.fib6_null_entry));
6324
6325         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6326                                            sizeof(*net->ipv6.ip6_null_entry),
6327                                            GFP_KERNEL);
6328         if (!net->ipv6.ip6_null_entry)
6329                 goto out_fib6_null_entry;
6330         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6331         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6332                          ip6_template_metrics, true);
6333         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6334
6335 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6336         net->ipv6.fib6_has_custom_rules = false;
6337         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6338                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6339                                                GFP_KERNEL);
6340         if (!net->ipv6.ip6_prohibit_entry)
6341                 goto out_ip6_null_entry;
6342         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6343         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6344                          ip6_template_metrics, true);
6345         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6346
6347         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6348                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6349                                                GFP_KERNEL);
6350         if (!net->ipv6.ip6_blk_hole_entry)
6351                 goto out_ip6_prohibit_entry;
6352         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6353         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6354                          ip6_template_metrics, true);
6355         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6356 #ifdef CONFIG_IPV6_SUBTREES
6357         net->ipv6.fib6_routes_require_src = 0;
6358 #endif
6359 #endif
6360
6361         net->ipv6.sysctl.flush_delay = 0;
6362         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6363         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6364         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6365         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6366         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6367         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6368         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6369         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6370
6371         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6372
6373         ret = 0;
6374 out:
6375         return ret;
6376
6377 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6378 out_ip6_prohibit_entry:
6379         kfree(net->ipv6.ip6_prohibit_entry);
6380 out_ip6_null_entry:
6381         kfree(net->ipv6.ip6_null_entry);
6382 #endif
6383 out_fib6_null_entry:
6384         kfree(net->ipv6.fib6_null_entry);
6385 out_ip6_dst_entries:
6386         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6387 out_ip6_dst_ops:
6388         goto out;
6389 }
6390
6391 static void __net_exit ip6_route_net_exit(struct net *net)
6392 {
6393         kfree(net->ipv6.fib6_null_entry);
6394         kfree(net->ipv6.ip6_null_entry);
6395 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6396         kfree(net->ipv6.ip6_prohibit_entry);
6397         kfree(net->ipv6.ip6_blk_hole_entry);
6398 #endif
6399         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6400 }
6401
6402 static int __net_init ip6_route_net_init_late(struct net *net)
6403 {
6404 #ifdef CONFIG_PROC_FS
6405         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6406                         sizeof(struct ipv6_route_iter));
6407         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6408                         rt6_stats_seq_show, NULL);
6409 #endif
6410         return 0;
6411 }
6412
6413 static void __net_exit ip6_route_net_exit_late(struct net *net)
6414 {
6415 #ifdef CONFIG_PROC_FS
6416         remove_proc_entry("ipv6_route", net->proc_net);
6417         remove_proc_entry("rt6_stats", net->proc_net);
6418 #endif
6419 }
6420
6421 static struct pernet_operations ip6_route_net_ops = {
6422         .init = ip6_route_net_init,
6423         .exit = ip6_route_net_exit,
6424 };
6425
6426 static int __net_init ipv6_inetpeer_init(struct net *net)
6427 {
6428         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6429
6430         if (!bp)
6431                 return -ENOMEM;
6432         inet_peer_base_init(bp);
6433         net->ipv6.peers = bp;
6434         return 0;
6435 }
6436
6437 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6438 {
6439         struct inet_peer_base *bp = net->ipv6.peers;
6440
6441         net->ipv6.peers = NULL;
6442         inetpeer_invalidate_tree(bp);
6443         kfree(bp);
6444 }
6445
6446 static struct pernet_operations ipv6_inetpeer_ops = {
6447         .init   =       ipv6_inetpeer_init,
6448         .exit   =       ipv6_inetpeer_exit,
6449 };
6450
6451 static struct pernet_operations ip6_route_net_late_ops = {
6452         .init = ip6_route_net_init_late,
6453         .exit = ip6_route_net_exit_late,
6454 };
6455
6456 static struct notifier_block ip6_route_dev_notifier = {
6457         .notifier_call = ip6_route_dev_notify,
6458         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6459 };
6460
6461 void __init ip6_route_init_special_entries(void)
6462 {
6463         /* Registering of the loopback is done before this portion of code,
6464          * the loopback reference in rt6_info will not be taken, do it
6465          * manually for init_net */
6466         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6467         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6468         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6469   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6470         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6471         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6472         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6473         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6474   #endif
6475 }
6476
6477 #if IS_BUILTIN(CONFIG_IPV6)
6478 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6479 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6480
6481 BTF_ID_LIST(btf_fib6_info_id)
6482 BTF_ID(struct, fib6_info)
6483
6484 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6485         .seq_ops                = &ipv6_route_seq_ops,
6486         .init_seq_private       = bpf_iter_init_seq_net,
6487         .fini_seq_private       = bpf_iter_fini_seq_net,
6488         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6489 };
6490
6491 static struct bpf_iter_reg ipv6_route_reg_info = {
6492         .target                 = "ipv6_route",
6493         .ctx_arg_info_size      = 1,
6494         .ctx_arg_info           = {
6495                 { offsetof(struct bpf_iter__ipv6_route, rt),
6496                   PTR_TO_BTF_ID_OR_NULL },
6497         },
6498         .seq_info               = &ipv6_route_seq_info,
6499 };
6500
6501 static int __init bpf_iter_register(void)
6502 {
6503         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6504         return bpf_iter_reg_target(&ipv6_route_reg_info);
6505 }
6506
6507 static void bpf_iter_unregister(void)
6508 {
6509         bpf_iter_unreg_target(&ipv6_route_reg_info);
6510 }
6511 #endif
6512 #endif
6513
6514 int __init ip6_route_init(void)
6515 {
6516         int ret;
6517         int cpu;
6518
6519         ret = -ENOMEM;
6520         ip6_dst_ops_template.kmem_cachep =
6521                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6522                                   SLAB_HWCACHE_ALIGN, NULL);
6523         if (!ip6_dst_ops_template.kmem_cachep)
6524                 goto out;
6525
6526         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6527         if (ret)
6528                 goto out_kmem_cache;
6529
6530         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6531         if (ret)
6532                 goto out_dst_entries;
6533
6534         ret = register_pernet_subsys(&ip6_route_net_ops);
6535         if (ret)
6536                 goto out_register_inetpeer;
6537
6538         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6539
6540         ret = fib6_init();
6541         if (ret)
6542                 goto out_register_subsys;
6543
6544         ret = xfrm6_init();
6545         if (ret)
6546                 goto out_fib6_init;
6547
6548         ret = fib6_rules_init();
6549         if (ret)
6550                 goto xfrm6_init;
6551
6552         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6553         if (ret)
6554                 goto fib6_rules_init;
6555
6556         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6557                                    inet6_rtm_newroute, NULL, 0);
6558         if (ret < 0)
6559                 goto out_register_late_subsys;
6560
6561         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6562                                    inet6_rtm_delroute, NULL, 0);
6563         if (ret < 0)
6564                 goto out_register_late_subsys;
6565
6566         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6567                                    inet6_rtm_getroute, NULL,
6568                                    RTNL_FLAG_DOIT_UNLOCKED);
6569         if (ret < 0)
6570                 goto out_register_late_subsys;
6571
6572         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6573         if (ret)
6574                 goto out_register_late_subsys;
6575
6576 #if IS_BUILTIN(CONFIG_IPV6)
6577 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6578         ret = bpf_iter_register();
6579         if (ret)
6580                 goto out_register_late_subsys;
6581 #endif
6582 #endif
6583
6584         for_each_possible_cpu(cpu) {
6585                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6586
6587                 INIT_LIST_HEAD(&ul->head);
6588                 spin_lock_init(&ul->lock);
6589         }
6590
6591 out:
6592         return ret;
6593
6594 out_register_late_subsys:
6595         rtnl_unregister_all(PF_INET6);
6596         unregister_pernet_subsys(&ip6_route_net_late_ops);
6597 fib6_rules_init:
6598         fib6_rules_cleanup();
6599 xfrm6_init:
6600         xfrm6_fini();
6601 out_fib6_init:
6602         fib6_gc_cleanup();
6603 out_register_subsys:
6604         unregister_pernet_subsys(&ip6_route_net_ops);
6605 out_register_inetpeer:
6606         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6607 out_dst_entries:
6608         dst_entries_destroy(&ip6_dst_blackhole_ops);
6609 out_kmem_cache:
6610         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6611         goto out;
6612 }
6613
6614 void ip6_route_cleanup(void)
6615 {
6616 #if IS_BUILTIN(CONFIG_IPV6)
6617 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6618         bpf_iter_unregister();
6619 #endif
6620 #endif
6621         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6622         unregister_pernet_subsys(&ip6_route_net_late_ops);
6623         fib6_rules_cleanup();
6624         xfrm6_fini();
6625         fib6_gc_cleanup();
6626         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6627         unregister_pernet_subsys(&ip6_route_net_ops);
6628         dst_entries_destroy(&ip6_dst_blackhole_ops);
6629         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6630 }