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 max_depth;
1661         int err = 0;
1662
1663         spin_lock_bh(&rt6_exception_lock);
1664
1665         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1666                                           lockdep_is_held(&rt6_exception_lock));
1667         if (!bucket) {
1668                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1669                                  GFP_ATOMIC);
1670                 if (!bucket) {
1671                         err = -ENOMEM;
1672                         goto out;
1673                 }
1674                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1675         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1676                 err = -EINVAL;
1677                 goto out;
1678         }
1679
1680 #ifdef CONFIG_IPV6_SUBTREES
1681         /* fib6_src.plen != 0 indicates f6i is in subtree
1682          * and exception table is indexed by a hash of
1683          * both fib6_dst and fib6_src.
1684          * Otherwise, the exception table is indexed by
1685          * a hash of only fib6_dst.
1686          */
1687         if (f6i->fib6_src.plen)
1688                 src_key = &nrt->rt6i_src.addr;
1689 #endif
1690         /* rt6_mtu_change() might lower mtu on f6i.
1691          * Only insert this exception route if its mtu
1692          * is less than f6i's mtu value.
1693          */
1694         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1695                 err = -EINVAL;
1696                 goto out;
1697         }
1698
1699         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1700                                                src_key);
1701         if (rt6_ex)
1702                 rt6_remove_exception(bucket, rt6_ex);
1703
1704         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1705         if (!rt6_ex) {
1706                 err = -ENOMEM;
1707                 goto out;
1708         }
1709         rt6_ex->rt6i = nrt;
1710         rt6_ex->stamp = jiffies;
1711         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1712         bucket->depth++;
1713         net->ipv6.rt6_stats->fib_rt_cache++;
1714
1715         /* Randomize max depth to avoid some side channels attacks. */
1716         max_depth = FIB6_MAX_DEPTH + prandom_u32_max(FIB6_MAX_DEPTH);
1717         while (bucket->depth > max_depth)
1718                 rt6_exception_remove_oldest(bucket);
1719
1720 out:
1721         spin_unlock_bh(&rt6_exception_lock);
1722
1723         /* Update fn->fn_sernum to invalidate all cached dst */
1724         if (!err) {
1725                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1726                 fib6_update_sernum(net, f6i);
1727                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1728                 fib6_force_start_gc(net);
1729         }
1730
1731         return err;
1732 }
1733
1734 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1735 {
1736         struct rt6_exception_bucket *bucket;
1737         struct rt6_exception *rt6_ex;
1738         struct hlist_node *tmp;
1739         int i;
1740
1741         spin_lock_bh(&rt6_exception_lock);
1742
1743         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1744         if (!bucket)
1745                 goto out;
1746
1747         /* Prevent rt6_insert_exception() to recreate the bucket list */
1748         if (!from)
1749                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1750
1751         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1752                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1753                         if (!from ||
1754                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1755                                 rt6_remove_exception(bucket, rt6_ex);
1756                 }
1757                 WARN_ON_ONCE(!from && bucket->depth);
1758                 bucket++;
1759         }
1760 out:
1761         spin_unlock_bh(&rt6_exception_lock);
1762 }
1763
1764 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1765 {
1766         struct fib6_info *f6i = arg;
1767
1768         fib6_nh_flush_exceptions(nh, f6i);
1769
1770         return 0;
1771 }
1772
1773 void rt6_flush_exceptions(struct fib6_info *f6i)
1774 {
1775         if (f6i->nh)
1776                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1777                                          f6i);
1778         else
1779                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1780 }
1781
1782 /* Find cached rt in the hash table inside passed in rt
1783  * Caller has to hold rcu_read_lock()
1784  */
1785 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1786                                            const struct in6_addr *daddr,
1787                                            const struct in6_addr *saddr)
1788 {
1789         const struct in6_addr *src_key = NULL;
1790         struct rt6_exception_bucket *bucket;
1791         struct rt6_exception *rt6_ex;
1792         struct rt6_info *ret = NULL;
1793
1794 #ifdef CONFIG_IPV6_SUBTREES
1795         /* fib6i_src.plen != 0 indicates f6i is in subtree
1796          * and exception table is indexed by a hash of
1797          * both fib6_dst and fib6_src.
1798          * However, the src addr used to create the hash
1799          * might not be exactly the passed in saddr which
1800          * is a /128 addr from the flow.
1801          * So we need to use f6i->fib6_src to redo lookup
1802          * if the passed in saddr does not find anything.
1803          * (See the logic in ip6_rt_cache_alloc() on how
1804          * rt->rt6i_src is updated.)
1805          */
1806         if (res->f6i->fib6_src.plen)
1807                 src_key = saddr;
1808 find_ex:
1809 #endif
1810         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1811         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1812
1813         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1814                 ret = rt6_ex->rt6i;
1815
1816 #ifdef CONFIG_IPV6_SUBTREES
1817         /* Use fib6_src as src_key and redo lookup */
1818         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1819                 src_key = &res->f6i->fib6_src.addr;
1820                 goto find_ex;
1821         }
1822 #endif
1823
1824         return ret;
1825 }
1826
1827 /* Remove the passed in cached rt from the hash table that contains it */
1828 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1829                                     const struct rt6_info *rt)
1830 {
1831         const struct in6_addr *src_key = NULL;
1832         struct rt6_exception_bucket *bucket;
1833         struct rt6_exception *rt6_ex;
1834         int err;
1835
1836         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1837                 return -ENOENT;
1838
1839         spin_lock_bh(&rt6_exception_lock);
1840         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1841
1842 #ifdef CONFIG_IPV6_SUBTREES
1843         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1844          * and exception table is indexed by a hash of
1845          * both rt6i_dst and rt6i_src.
1846          * Otherwise, the exception table is indexed by
1847          * a hash of only rt6i_dst.
1848          */
1849         if (plen)
1850                 src_key = &rt->rt6i_src.addr;
1851 #endif
1852         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1853                                                &rt->rt6i_dst.addr,
1854                                                src_key);
1855         if (rt6_ex) {
1856                 rt6_remove_exception(bucket, rt6_ex);
1857                 err = 0;
1858         } else {
1859                 err = -ENOENT;
1860         }
1861
1862         spin_unlock_bh(&rt6_exception_lock);
1863         return err;
1864 }
1865
1866 struct fib6_nh_excptn_arg {
1867         struct rt6_info *rt;
1868         int             plen;
1869 };
1870
1871 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1872 {
1873         struct fib6_nh_excptn_arg *arg = _arg;
1874         int err;
1875
1876         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1877         if (err == 0)
1878                 return 1;
1879
1880         return 0;
1881 }
1882
1883 static int rt6_remove_exception_rt(struct rt6_info *rt)
1884 {
1885         struct fib6_info *from;
1886
1887         from = rcu_dereference(rt->from);
1888         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1889                 return -EINVAL;
1890
1891         if (from->nh) {
1892                 struct fib6_nh_excptn_arg arg = {
1893                         .rt = rt,
1894                         .plen = from->fib6_src.plen
1895                 };
1896                 int rc;
1897
1898                 /* rc = 1 means an entry was found */
1899                 rc = nexthop_for_each_fib6_nh(from->nh,
1900                                               rt6_nh_remove_exception_rt,
1901                                               &arg);
1902                 return rc ? 0 : -ENOENT;
1903         }
1904
1905         return fib6_nh_remove_exception(from->fib6_nh,
1906                                         from->fib6_src.plen, rt);
1907 }
1908
1909 /* Find rt6_ex which contains the passed in rt cache and
1910  * refresh its stamp
1911  */
1912 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1913                                      const struct rt6_info *rt)
1914 {
1915         const struct in6_addr *src_key = NULL;
1916         struct rt6_exception_bucket *bucket;
1917         struct rt6_exception *rt6_ex;
1918
1919         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1920 #ifdef CONFIG_IPV6_SUBTREES
1921         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1922          * and exception table is indexed by a hash of
1923          * both rt6i_dst and rt6i_src.
1924          * Otherwise, the exception table is indexed by
1925          * a hash of only rt6i_dst.
1926          */
1927         if (plen)
1928                 src_key = &rt->rt6i_src.addr;
1929 #endif
1930         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1931         if (rt6_ex)
1932                 rt6_ex->stamp = jiffies;
1933 }
1934
1935 struct fib6_nh_match_arg {
1936         const struct net_device *dev;
1937         const struct in6_addr   *gw;
1938         struct fib6_nh          *match;
1939 };
1940
1941 /* determine if fib6_nh has given device and gateway */
1942 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1943 {
1944         struct fib6_nh_match_arg *arg = _arg;
1945
1946         if (arg->dev != nh->fib_nh_dev ||
1947             (arg->gw && !nh->fib_nh_gw_family) ||
1948             (!arg->gw && nh->fib_nh_gw_family) ||
1949             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1950                 return 0;
1951
1952         arg->match = nh;
1953
1954         /* found a match, break the loop */
1955         return 1;
1956 }
1957
1958 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1959 {
1960         struct fib6_info *from;
1961         struct fib6_nh *fib6_nh;
1962
1963         rcu_read_lock();
1964
1965         from = rcu_dereference(rt->from);
1966         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1967                 goto unlock;
1968
1969         if (from->nh) {
1970                 struct fib6_nh_match_arg arg = {
1971                         .dev = rt->dst.dev,
1972                         .gw = &rt->rt6i_gateway,
1973                 };
1974
1975                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1976
1977                 if (!arg.match)
1978                         goto unlock;
1979                 fib6_nh = arg.match;
1980         } else {
1981                 fib6_nh = from->fib6_nh;
1982         }
1983         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1984 unlock:
1985         rcu_read_unlock();
1986 }
1987
1988 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1989                                          struct rt6_info *rt, int mtu)
1990 {
1991         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1992          * lowest MTU in the path: always allow updating the route PMTU to
1993          * reflect PMTU decreases.
1994          *
1995          * If the new MTU is higher, and the route PMTU is equal to the local
1996          * MTU, this means the old MTU is the lowest in the path, so allow
1997          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1998          * handle this.
1999          */
2000
2001         if (dst_mtu(&rt->dst) >= mtu)
2002                 return true;
2003
2004         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2005                 return true;
2006
2007         return false;
2008 }
2009
2010 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2011                                        const struct fib6_nh *nh, int mtu)
2012 {
2013         struct rt6_exception_bucket *bucket;
2014         struct rt6_exception *rt6_ex;
2015         int i;
2016
2017         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2018         if (!bucket)
2019                 return;
2020
2021         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2022                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2023                         struct rt6_info *entry = rt6_ex->rt6i;
2024
2025                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2026                          * route), the metrics of its rt->from have already
2027                          * been updated.
2028                          */
2029                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2030                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2031                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2032                 }
2033                 bucket++;
2034         }
2035 }
2036
2037 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2038
2039 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2040                                             const struct in6_addr *gateway)
2041 {
2042         struct rt6_exception_bucket *bucket;
2043         struct rt6_exception *rt6_ex;
2044         struct hlist_node *tmp;
2045         int i;
2046
2047         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2048                 return;
2049
2050         spin_lock_bh(&rt6_exception_lock);
2051         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2052         if (bucket) {
2053                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2054                         hlist_for_each_entry_safe(rt6_ex, tmp,
2055                                                   &bucket->chain, hlist) {
2056                                 struct rt6_info *entry = rt6_ex->rt6i;
2057
2058                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2059                                     RTF_CACHE_GATEWAY &&
2060                                     ipv6_addr_equal(gateway,
2061                                                     &entry->rt6i_gateway)) {
2062                                         rt6_remove_exception(bucket, rt6_ex);
2063                                 }
2064                         }
2065                         bucket++;
2066                 }
2067         }
2068
2069         spin_unlock_bh(&rt6_exception_lock);
2070 }
2071
2072 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2073                                       struct rt6_exception *rt6_ex,
2074                                       struct fib6_gc_args *gc_args,
2075                                       unsigned long now)
2076 {
2077         struct rt6_info *rt = rt6_ex->rt6i;
2078
2079         /* we are pruning and obsoleting aged-out and non gateway exceptions
2080          * even if others have still references to them, so that on next
2081          * dst_check() such references can be dropped.
2082          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2083          * expired, independently from their aging, as per RFC 8201 section 4
2084          */
2085         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2086                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2087                         RT6_TRACE("aging clone %p\n", rt);
2088                         rt6_remove_exception(bucket, rt6_ex);
2089                         return;
2090                 }
2091         } else if (time_after(jiffies, rt->dst.expires)) {
2092                 RT6_TRACE("purging expired route %p\n", rt);
2093                 rt6_remove_exception(bucket, rt6_ex);
2094                 return;
2095         }
2096
2097         if (rt->rt6i_flags & RTF_GATEWAY) {
2098                 struct neighbour *neigh;
2099
2100                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2101
2102                 if (!(neigh && (neigh->flags & NTF_ROUTER))) {
2103                         RT6_TRACE("purging route %p via non-router but gateway\n",
2104                                   rt);
2105                         rt6_remove_exception(bucket, rt6_ex);
2106                         return;
2107                 }
2108         }
2109
2110         gc_args->more++;
2111 }
2112
2113 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2114                                    struct fib6_gc_args *gc_args,
2115                                    unsigned long now)
2116 {
2117         struct rt6_exception_bucket *bucket;
2118         struct rt6_exception *rt6_ex;
2119         struct hlist_node *tmp;
2120         int i;
2121
2122         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2123                 return;
2124
2125         rcu_read_lock_bh();
2126         spin_lock(&rt6_exception_lock);
2127         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2128         if (bucket) {
2129                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2130                         hlist_for_each_entry_safe(rt6_ex, tmp,
2131                                                   &bucket->chain, hlist) {
2132                                 rt6_age_examine_exception(bucket, rt6_ex,
2133                                                           gc_args, now);
2134                         }
2135                         bucket++;
2136                 }
2137         }
2138         spin_unlock(&rt6_exception_lock);
2139         rcu_read_unlock_bh();
2140 }
2141
2142 struct fib6_nh_age_excptn_arg {
2143         struct fib6_gc_args     *gc_args;
2144         unsigned long           now;
2145 };
2146
2147 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2148 {
2149         struct fib6_nh_age_excptn_arg *arg = _arg;
2150
2151         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2152         return 0;
2153 }
2154
2155 void rt6_age_exceptions(struct fib6_info *f6i,
2156                         struct fib6_gc_args *gc_args,
2157                         unsigned long now)
2158 {
2159         if (f6i->nh) {
2160                 struct fib6_nh_age_excptn_arg arg = {
2161                         .gc_args = gc_args,
2162                         .now = now
2163                 };
2164
2165                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2166                                          &arg);
2167         } else {
2168                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2169         }
2170 }
2171
2172 /* must be called with rcu lock held */
2173 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2174                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2175 {
2176         struct fib6_node *fn, *saved_fn;
2177
2178         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2179         saved_fn = fn;
2180
2181         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2182                 oif = 0;
2183
2184 redo_rt6_select:
2185         rt6_select(net, fn, oif, res, strict);
2186         if (res->f6i == net->ipv6.fib6_null_entry) {
2187                 fn = fib6_backtrack(fn, &fl6->saddr);
2188                 if (fn)
2189                         goto redo_rt6_select;
2190                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2191                         /* also consider unreachable route */
2192                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2193                         fn = saved_fn;
2194                         goto redo_rt6_select;
2195                 }
2196         }
2197
2198         trace_fib6_table_lookup(net, res, table, fl6);
2199
2200         return 0;
2201 }
2202
2203 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2204                                int oif, struct flowi6 *fl6,
2205                                const struct sk_buff *skb, int flags)
2206 {
2207         struct fib6_result res = {};
2208         struct rt6_info *rt = NULL;
2209         int strict = 0;
2210
2211         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2212                      !rcu_read_lock_held());
2213
2214         strict |= flags & RT6_LOOKUP_F_IFACE;
2215         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2216         if (net->ipv6.devconf_all->forwarding == 0)
2217                 strict |= RT6_LOOKUP_F_REACHABLE;
2218
2219         rcu_read_lock();
2220
2221         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2222         if (res.f6i == net->ipv6.fib6_null_entry)
2223                 goto out;
2224
2225         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2226
2227         /*Search through exception table */
2228         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2229         if (rt) {
2230                 goto out;
2231         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2232                             !res.nh->fib_nh_gw_family)) {
2233                 /* Create a RTF_CACHE clone which will not be
2234                  * owned by the fib6 tree.  It is for the special case where
2235                  * the daddr in the skb during the neighbor look-up is different
2236                  * from the fl6->daddr used to look-up route here.
2237                  */
2238                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2239
2240                 if (rt) {
2241                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2242                          * As rt6_uncached_list_add() does not consume refcnt,
2243                          * this refcnt is always returned to the caller even
2244                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2245                          */
2246                         rt6_uncached_list_add(rt);
2247                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2248                         rcu_read_unlock();
2249
2250                         return rt;
2251                 }
2252         } else {
2253                 /* Get a percpu copy */
2254                 local_bh_disable();
2255                 rt = rt6_get_pcpu_route(&res);
2256
2257                 if (!rt)
2258                         rt = rt6_make_pcpu_route(net, &res);
2259
2260                 local_bh_enable();
2261         }
2262 out:
2263         if (!rt)
2264                 rt = net->ipv6.ip6_null_entry;
2265         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2266                 ip6_hold_safe(net, &rt);
2267         rcu_read_unlock();
2268
2269         return rt;
2270 }
2271 EXPORT_SYMBOL_GPL(ip6_pol_route);
2272
2273 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2274                                             struct fib6_table *table,
2275                                             struct flowi6 *fl6,
2276                                             const struct sk_buff *skb,
2277                                             int flags)
2278 {
2279         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2280 }
2281
2282 struct dst_entry *ip6_route_input_lookup(struct net *net,
2283                                          struct net_device *dev,
2284                                          struct flowi6 *fl6,
2285                                          const struct sk_buff *skb,
2286                                          int flags)
2287 {
2288         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2289                 flags |= RT6_LOOKUP_F_IFACE;
2290
2291         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2292 }
2293 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2294
2295 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2296                                   struct flow_keys *keys,
2297                                   struct flow_keys *flkeys)
2298 {
2299         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2300         const struct ipv6hdr *key_iph = outer_iph;
2301         struct flow_keys *_flkeys = flkeys;
2302         const struct ipv6hdr *inner_iph;
2303         const struct icmp6hdr *icmph;
2304         struct ipv6hdr _inner_iph;
2305         struct icmp6hdr _icmph;
2306
2307         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2308                 goto out;
2309
2310         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2311                                    sizeof(_icmph), &_icmph);
2312         if (!icmph)
2313                 goto out;
2314
2315         if (!icmpv6_is_err(icmph->icmp6_type))
2316                 goto out;
2317
2318         inner_iph = skb_header_pointer(skb,
2319                                        skb_transport_offset(skb) + sizeof(*icmph),
2320                                        sizeof(_inner_iph), &_inner_iph);
2321         if (!inner_iph)
2322                 goto out;
2323
2324         key_iph = inner_iph;
2325         _flkeys = NULL;
2326 out:
2327         if (_flkeys) {
2328                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2329                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2330                 keys->tags.flow_label = _flkeys->tags.flow_label;
2331                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2332         } else {
2333                 keys->addrs.v6addrs.src = key_iph->saddr;
2334                 keys->addrs.v6addrs.dst = key_iph->daddr;
2335                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2336                 keys->basic.ip_proto = key_iph->nexthdr;
2337         }
2338 }
2339
2340 static u32 rt6_multipath_custom_hash_outer(const struct net *net,
2341                                            const struct sk_buff *skb,
2342                                            bool *p_has_inner)
2343 {
2344         u32 hash_fields = ip6_multipath_hash_fields(net);
2345         struct flow_keys keys, hash_keys;
2346
2347         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2348                 return 0;
2349
2350         memset(&hash_keys, 0, sizeof(hash_keys));
2351         skb_flow_dissect_flow_keys(skb, &keys, FLOW_DISSECTOR_F_STOP_AT_ENCAP);
2352
2353         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2354         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2355                 hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2356         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2357                 hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2358         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2359                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2360         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2361                 hash_keys.tags.flow_label = keys.tags.flow_label;
2362         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2363                 hash_keys.ports.src = keys.ports.src;
2364         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2365                 hash_keys.ports.dst = keys.ports.dst;
2366
2367         *p_has_inner = !!(keys.control.flags & FLOW_DIS_ENCAPSULATION);
2368         return flow_hash_from_keys(&hash_keys);
2369 }
2370
2371 static u32 rt6_multipath_custom_hash_inner(const struct net *net,
2372                                            const struct sk_buff *skb,
2373                                            bool has_inner)
2374 {
2375         u32 hash_fields = ip6_multipath_hash_fields(net);
2376         struct flow_keys keys, hash_keys;
2377
2378         /* We assume the packet carries an encapsulation, but if none was
2379          * encountered during dissection of the outer flow, then there is no
2380          * point in calling the flow dissector again.
2381          */
2382         if (!has_inner)
2383                 return 0;
2384
2385         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_MASK))
2386                 return 0;
2387
2388         memset(&hash_keys, 0, sizeof(hash_keys));
2389         skb_flow_dissect_flow_keys(skb, &keys, 0);
2390
2391         if (!(keys.control.flags & FLOW_DIS_ENCAPSULATION))
2392                 return 0;
2393
2394         if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2395                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2396                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2397                         hash_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src;
2398                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2399                         hash_keys.addrs.v4addrs.dst = keys.addrs.v4addrs.dst;
2400         } else if (keys.control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2401                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2402                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_IP)
2403                         hash_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src;
2404                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_IP)
2405                         hash_keys.addrs.v6addrs.dst = keys.addrs.v6addrs.dst;
2406                 if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_FLOWLABEL)
2407                         hash_keys.tags.flow_label = keys.tags.flow_label;
2408         }
2409
2410         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_IP_PROTO)
2411                 hash_keys.basic.ip_proto = keys.basic.ip_proto;
2412         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_SRC_PORT)
2413                 hash_keys.ports.src = keys.ports.src;
2414         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_INNER_DST_PORT)
2415                 hash_keys.ports.dst = keys.ports.dst;
2416
2417         return flow_hash_from_keys(&hash_keys);
2418 }
2419
2420 static u32 rt6_multipath_custom_hash_skb(const struct net *net,
2421                                          const struct sk_buff *skb)
2422 {
2423         u32 mhash, mhash_inner;
2424         bool has_inner = true;
2425
2426         mhash = rt6_multipath_custom_hash_outer(net, skb, &has_inner);
2427         mhash_inner = rt6_multipath_custom_hash_inner(net, skb, has_inner);
2428
2429         return jhash_2words(mhash, mhash_inner, 0);
2430 }
2431
2432 static u32 rt6_multipath_custom_hash_fl6(const struct net *net,
2433                                          const struct flowi6 *fl6)
2434 {
2435         u32 hash_fields = ip6_multipath_hash_fields(net);
2436         struct flow_keys hash_keys;
2437
2438         if (!(hash_fields & FIB_MULTIPATH_HASH_FIELD_OUTER_MASK))
2439                 return 0;
2440
2441         memset(&hash_keys, 0, sizeof(hash_keys));
2442         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2443         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_IP)
2444                 hash_keys.addrs.v6addrs.src = fl6->saddr;
2445         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_IP)
2446                 hash_keys.addrs.v6addrs.dst = fl6->daddr;
2447         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_IP_PROTO)
2448                 hash_keys.basic.ip_proto = fl6->flowi6_proto;
2449         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_FLOWLABEL)
2450                 hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2451         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_SRC_PORT)
2452                 hash_keys.ports.src = fl6->fl6_sport;
2453         if (hash_fields & FIB_MULTIPATH_HASH_FIELD_DST_PORT)
2454                 hash_keys.ports.dst = fl6->fl6_dport;
2455
2456         return flow_hash_from_keys(&hash_keys);
2457 }
2458
2459 /* if skb is set it will be used and fl6 can be NULL */
2460 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2461                        const struct sk_buff *skb, struct flow_keys *flkeys)
2462 {
2463         struct flow_keys hash_keys;
2464         u32 mhash = 0;
2465
2466         switch (ip6_multipath_hash_policy(net)) {
2467         case 0:
2468                 memset(&hash_keys, 0, sizeof(hash_keys));
2469                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2470                 if (skb) {
2471                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2472                 } else {
2473                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2474                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2475                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2476                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2477                 }
2478                 mhash = flow_hash_from_keys(&hash_keys);
2479                 break;
2480         case 1:
2481                 if (skb) {
2482                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2483                         struct flow_keys keys;
2484
2485                         /* short-circuit if we already have L4 hash present */
2486                         if (skb->l4_hash)
2487                                 return skb_get_hash_raw(skb) >> 1;
2488
2489                         memset(&hash_keys, 0, sizeof(hash_keys));
2490
2491                         if (!flkeys) {
2492                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2493                                 flkeys = &keys;
2494                         }
2495                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2496                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2497                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2498                         hash_keys.ports.src = flkeys->ports.src;
2499                         hash_keys.ports.dst = flkeys->ports.dst;
2500                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2501                 } else {
2502                         memset(&hash_keys, 0, sizeof(hash_keys));
2503                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2504                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2505                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2506                         hash_keys.ports.src = fl6->fl6_sport;
2507                         hash_keys.ports.dst = fl6->fl6_dport;
2508                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2509                 }
2510                 mhash = flow_hash_from_keys(&hash_keys);
2511                 break;
2512         case 2:
2513                 memset(&hash_keys, 0, sizeof(hash_keys));
2514                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2515                 if (skb) {
2516                         struct flow_keys keys;
2517
2518                         if (!flkeys) {
2519                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2520                                 flkeys = &keys;
2521                         }
2522
2523                         /* Inner can be v4 or v6 */
2524                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2525                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2526                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2527                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2528                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2529                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2530                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2531                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2532                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2533                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2534                         } else {
2535                                 /* Same as case 0 */
2536                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2537                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2538                         }
2539                 } else {
2540                         /* Same as case 0 */
2541                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2542                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2543                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2544                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2545                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2546                 }
2547                 mhash = flow_hash_from_keys(&hash_keys);
2548                 break;
2549         case 3:
2550                 if (skb)
2551                         mhash = rt6_multipath_custom_hash_skb(net, skb);
2552                 else
2553                         mhash = rt6_multipath_custom_hash_fl6(net, fl6);
2554                 break;
2555         }
2556
2557         return mhash >> 1;
2558 }
2559
2560 /* Called with rcu held */
2561 void ip6_route_input(struct sk_buff *skb)
2562 {
2563         const struct ipv6hdr *iph = ipv6_hdr(skb);
2564         struct net *net = dev_net(skb->dev);
2565         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2566         struct ip_tunnel_info *tun_info;
2567         struct flowi6 fl6 = {
2568                 .flowi6_iif = skb->dev->ifindex,
2569                 .daddr = iph->daddr,
2570                 .saddr = iph->saddr,
2571                 .flowlabel = ip6_flowinfo(iph),
2572                 .flowi6_mark = skb->mark,
2573                 .flowi6_proto = iph->nexthdr,
2574         };
2575         struct flow_keys *flkeys = NULL, _flkeys;
2576
2577         tun_info = skb_tunnel_info(skb);
2578         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2579                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2580
2581         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2582                 flkeys = &_flkeys;
2583
2584         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2585                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2586         skb_dst_drop(skb);
2587         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2588                                                       &fl6, skb, flags));
2589 }
2590
2591 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2592                                              struct fib6_table *table,
2593                                              struct flowi6 *fl6,
2594                                              const struct sk_buff *skb,
2595                                              int flags)
2596 {
2597         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2598 }
2599
2600 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2601                                                const struct sock *sk,
2602                                                struct flowi6 *fl6, int flags)
2603 {
2604         bool any_src;
2605
2606         if (ipv6_addr_type(&fl6->daddr) &
2607             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2608                 struct dst_entry *dst;
2609
2610                 /* This function does not take refcnt on the dst */
2611                 dst = l3mdev_link_scope_lookup(net, fl6);
2612                 if (dst)
2613                         return dst;
2614         }
2615
2616         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2617
2618         flags |= RT6_LOOKUP_F_DST_NOREF;
2619         any_src = ipv6_addr_any(&fl6->saddr);
2620         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2621             (fl6->flowi6_oif && any_src))
2622                 flags |= RT6_LOOKUP_F_IFACE;
2623
2624         if (!any_src)
2625                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2626         else if (sk)
2627                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2628
2629         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2630 }
2631 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2632
2633 struct dst_entry *ip6_route_output_flags(struct net *net,
2634                                          const struct sock *sk,
2635                                          struct flowi6 *fl6,
2636                                          int flags)
2637 {
2638         struct dst_entry *dst;
2639         struct rt6_info *rt6;
2640
2641         rcu_read_lock();
2642         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2643         rt6 = (struct rt6_info *)dst;
2644         /* For dst cached in uncached_list, refcnt is already taken. */
2645         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2646                 dst = &net->ipv6.ip6_null_entry->dst;
2647                 dst_hold(dst);
2648         }
2649         rcu_read_unlock();
2650
2651         return dst;
2652 }
2653 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2654
2655 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2656 {
2657         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2658         struct net_device *loopback_dev = net->loopback_dev;
2659         struct dst_entry *new = NULL;
2660
2661         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2662                        DST_OBSOLETE_DEAD, 0);
2663         if (rt) {
2664                 rt6_info_init(rt);
2665                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2666
2667                 new = &rt->dst;
2668                 new->__use = 1;
2669                 new->input = dst_discard;
2670                 new->output = dst_discard_out;
2671
2672                 dst_copy_metrics(new, &ort->dst);
2673
2674                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2675                 rt->rt6i_gateway = ort->rt6i_gateway;
2676                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2677
2678                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2679 #ifdef CONFIG_IPV6_SUBTREES
2680                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2681 #endif
2682         }
2683
2684         dst_release(dst_orig);
2685         return new ? new : ERR_PTR(-ENOMEM);
2686 }
2687
2688 /*
2689  *      Destination cache support functions
2690  */
2691
2692 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2693 {
2694         u32 rt_cookie = 0;
2695
2696         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2697                 return false;
2698
2699         if (fib6_check_expired(f6i))
2700                 return false;
2701
2702         return true;
2703 }
2704
2705 static struct dst_entry *rt6_check(struct rt6_info *rt,
2706                                    struct fib6_info *from,
2707                                    u32 cookie)
2708 {
2709         u32 rt_cookie = 0;
2710
2711         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2712             rt_cookie != cookie)
2713                 return NULL;
2714
2715         if (rt6_check_expired(rt))
2716                 return NULL;
2717
2718         return &rt->dst;
2719 }
2720
2721 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2722                                             struct fib6_info *from,
2723                                             u32 cookie)
2724 {
2725         if (!__rt6_check_expired(rt) &&
2726             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2727             fib6_check(from, cookie))
2728                 return &rt->dst;
2729         else
2730                 return NULL;
2731 }
2732
2733 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2734                                                         u32 cookie)
2735 {
2736         struct dst_entry *dst_ret;
2737         struct fib6_info *from;
2738         struct rt6_info *rt;
2739
2740         rt = container_of(dst, struct rt6_info, dst);
2741
2742         if (rt->sernum)
2743                 return rt6_is_valid(rt) ? dst : NULL;
2744
2745         rcu_read_lock();
2746
2747         /* All IPV6 dsts are created with ->obsolete set to the value
2748          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2749          * into this function always.
2750          */
2751
2752         from = rcu_dereference(rt->from);
2753
2754         if (from && (rt->rt6i_flags & RTF_PCPU ||
2755             unlikely(!list_empty(&rt->rt6i_uncached))))
2756                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2757         else
2758                 dst_ret = rt6_check(rt, from, cookie);
2759
2760         rcu_read_unlock();
2761
2762         return dst_ret;
2763 }
2764 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2765
2766 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2767 {
2768         struct rt6_info *rt = (struct rt6_info *) dst;
2769
2770         if (rt) {
2771                 if (rt->rt6i_flags & RTF_CACHE) {
2772                         rcu_read_lock();
2773                         if (rt6_check_expired(rt)) {
2774                                 rt6_remove_exception_rt(rt);
2775                                 dst = NULL;
2776                         }
2777                         rcu_read_unlock();
2778                 } else {
2779                         dst_release(dst);
2780                         dst = NULL;
2781                 }
2782         }
2783         return dst;
2784 }
2785
2786 static void ip6_link_failure(struct sk_buff *skb)
2787 {
2788         struct rt6_info *rt;
2789
2790         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2791
2792         rt = (struct rt6_info *) skb_dst(skb);
2793         if (rt) {
2794                 rcu_read_lock();
2795                 if (rt->rt6i_flags & RTF_CACHE) {
2796                         rt6_remove_exception_rt(rt);
2797                 } else {
2798                         struct fib6_info *from;
2799                         struct fib6_node *fn;
2800
2801                         from = rcu_dereference(rt->from);
2802                         if (from) {
2803                                 fn = rcu_dereference(from->fib6_node);
2804                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2805                                         fn->fn_sernum = -1;
2806                         }
2807                 }
2808                 rcu_read_unlock();
2809         }
2810 }
2811
2812 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2813 {
2814         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2815                 struct fib6_info *from;
2816
2817                 rcu_read_lock();
2818                 from = rcu_dereference(rt0->from);
2819                 if (from)
2820                         rt0->dst.expires = from->expires;
2821                 rcu_read_unlock();
2822         }
2823
2824         dst_set_expires(&rt0->dst, timeout);
2825         rt0->rt6i_flags |= RTF_EXPIRES;
2826 }
2827
2828 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2829 {
2830         struct net *net = dev_net(rt->dst.dev);
2831
2832         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2833         rt->rt6i_flags |= RTF_MODIFIED;
2834         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2835 }
2836
2837 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2838 {
2839         return !(rt->rt6i_flags & RTF_CACHE) &&
2840                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2841 }
2842
2843 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2844                                  const struct ipv6hdr *iph, u32 mtu,
2845                                  bool confirm_neigh)
2846 {
2847         const struct in6_addr *daddr, *saddr;
2848         struct rt6_info *rt6 = (struct rt6_info *)dst;
2849
2850         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2851          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2852          * [see also comment in rt6_mtu_change_route()]
2853          */
2854
2855         if (iph) {
2856                 daddr = &iph->daddr;
2857                 saddr = &iph->saddr;
2858         } else if (sk) {
2859                 daddr = &sk->sk_v6_daddr;
2860                 saddr = &inet6_sk(sk)->saddr;
2861         } else {
2862                 daddr = NULL;
2863                 saddr = NULL;
2864         }
2865
2866         if (confirm_neigh)
2867                 dst_confirm_neigh(dst, daddr);
2868
2869         if (mtu < IPV6_MIN_MTU)
2870                 return;
2871         if (mtu >= dst_mtu(dst))
2872                 return;
2873
2874         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2875                 rt6_do_update_pmtu(rt6, mtu);
2876                 /* update rt6_ex->stamp for cache */
2877                 if (rt6->rt6i_flags & RTF_CACHE)
2878                         rt6_update_exception_stamp_rt(rt6);
2879         } else if (daddr) {
2880                 struct fib6_result res = {};
2881                 struct rt6_info *nrt6;
2882
2883                 rcu_read_lock();
2884                 res.f6i = rcu_dereference(rt6->from);
2885                 if (!res.f6i)
2886                         goto out_unlock;
2887
2888                 res.fib6_flags = res.f6i->fib6_flags;
2889                 res.fib6_type = res.f6i->fib6_type;
2890
2891                 if (res.f6i->nh) {
2892                         struct fib6_nh_match_arg arg = {
2893                                 .dev = dst->dev,
2894                                 .gw = &rt6->rt6i_gateway,
2895                         };
2896
2897                         nexthop_for_each_fib6_nh(res.f6i->nh,
2898                                                  fib6_nh_find_match, &arg);
2899
2900                         /* fib6_info uses a nexthop that does not have fib6_nh
2901                          * using the dst->dev + gw. Should be impossible.
2902                          */
2903                         if (!arg.match)
2904                                 goto out_unlock;
2905
2906                         res.nh = arg.match;
2907                 } else {
2908                         res.nh = res.f6i->fib6_nh;
2909                 }
2910
2911                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2912                 if (nrt6) {
2913                         rt6_do_update_pmtu(nrt6, mtu);
2914                         if (rt6_insert_exception(nrt6, &res))
2915                                 dst_release_immediate(&nrt6->dst);
2916                 }
2917 out_unlock:
2918                 rcu_read_unlock();
2919         }
2920 }
2921
2922 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2923                                struct sk_buff *skb, u32 mtu,
2924                                bool confirm_neigh)
2925 {
2926         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2927                              confirm_neigh);
2928 }
2929
2930 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2931                      int oif, u32 mark, kuid_t uid)
2932 {
2933         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2934         struct dst_entry *dst;
2935         struct flowi6 fl6 = {
2936                 .flowi6_oif = oif,
2937                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2938                 .daddr = iph->daddr,
2939                 .saddr = iph->saddr,
2940                 .flowlabel = ip6_flowinfo(iph),
2941                 .flowi6_uid = uid,
2942         };
2943
2944         dst = ip6_route_output(net, NULL, &fl6);
2945         if (!dst->error)
2946                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2947         dst_release(dst);
2948 }
2949 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2950
2951 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2952 {
2953         int oif = sk->sk_bound_dev_if;
2954         struct dst_entry *dst;
2955
2956         if (!oif && skb->dev)
2957                 oif = l3mdev_master_ifindex(skb->dev);
2958
2959         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2960
2961         dst = __sk_dst_get(sk);
2962         if (!dst || !dst->obsolete ||
2963             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2964                 return;
2965
2966         bh_lock_sock(sk);
2967         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2968                 ip6_datagram_dst_update(sk, false);
2969         bh_unlock_sock(sk);
2970 }
2971 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2972
2973 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2974                            const struct flowi6 *fl6)
2975 {
2976 #ifdef CONFIG_IPV6_SUBTREES
2977         struct ipv6_pinfo *np = inet6_sk(sk);
2978 #endif
2979
2980         ip6_dst_store(sk, dst,
2981                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2982                       &sk->sk_v6_daddr : NULL,
2983 #ifdef CONFIG_IPV6_SUBTREES
2984                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2985                       &np->saddr :
2986 #endif
2987                       NULL);
2988 }
2989
2990 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2991                                   struct flowi6 *fl6,
2992                                   const struct in6_addr *gw,
2993                                   struct rt6_info **ret)
2994 {
2995         const struct fib6_nh *nh = res->nh;
2996
2997         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2998             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2999                 return false;
3000
3001         /* rt_cache's gateway might be different from its 'parent'
3002          * in the case of an ip redirect.
3003          * So we keep searching in the exception table if the gateway
3004          * is different.
3005          */
3006         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
3007                 struct rt6_info *rt_cache;
3008
3009                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
3010                 if (rt_cache &&
3011                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
3012                         *ret = rt_cache;
3013                         return true;
3014                 }
3015                 return false;
3016         }
3017         return true;
3018 }
3019
3020 struct fib6_nh_rd_arg {
3021         struct fib6_result      *res;
3022         struct flowi6           *fl6;
3023         const struct in6_addr   *gw;
3024         struct rt6_info         **ret;
3025 };
3026
3027 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
3028 {
3029         struct fib6_nh_rd_arg *arg = _arg;
3030
3031         arg->res->nh = nh;
3032         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
3033 }
3034
3035 /* Handle redirects */
3036 struct ip6rd_flowi {
3037         struct flowi6 fl6;
3038         struct in6_addr gateway;
3039 };
3040
3041 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
3042                                              struct fib6_table *table,
3043                                              struct flowi6 *fl6,
3044                                              const struct sk_buff *skb,
3045                                              int flags)
3046 {
3047         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
3048         struct rt6_info *ret = NULL;
3049         struct fib6_result res = {};
3050         struct fib6_nh_rd_arg arg = {
3051                 .res = &res,
3052                 .fl6 = fl6,
3053                 .gw  = &rdfl->gateway,
3054                 .ret = &ret
3055         };
3056         struct fib6_info *rt;
3057         struct fib6_node *fn;
3058
3059         /* l3mdev_update_flow overrides oif if the device is enslaved; in
3060          * this case we must match on the real ingress device, so reset it
3061          */
3062         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
3063                 fl6->flowi6_oif = skb->dev->ifindex;
3064
3065         /* Get the "current" route for this destination and
3066          * check if the redirect has come from appropriate router.
3067          *
3068          * RFC 4861 specifies that redirects should only be
3069          * accepted if they come from the nexthop to the target.
3070          * Due to the way the routes are chosen, this notion
3071          * is a bit fuzzy and one might need to check all possible
3072          * routes.
3073          */
3074
3075         rcu_read_lock();
3076         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
3077 restart:
3078         for_each_fib6_node_rt_rcu(fn) {
3079                 res.f6i = rt;
3080                 if (fib6_check_expired(rt))
3081                         continue;
3082                 if (rt->fib6_flags & RTF_REJECT)
3083                         break;
3084                 if (unlikely(rt->nh)) {
3085                         if (nexthop_is_blackhole(rt->nh))
3086                                 continue;
3087                         /* on match, res->nh is filled in and potentially ret */
3088                         if (nexthop_for_each_fib6_nh(rt->nh,
3089                                                      fib6_nh_redirect_match,
3090                                                      &arg))
3091                                 goto out;
3092                 } else {
3093                         res.nh = rt->fib6_nh;
3094                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
3095                                                   &ret))
3096                                 goto out;
3097                 }
3098         }
3099
3100         if (!rt)
3101                 rt = net->ipv6.fib6_null_entry;
3102         else if (rt->fib6_flags & RTF_REJECT) {
3103                 ret = net->ipv6.ip6_null_entry;
3104                 goto out;
3105         }
3106
3107         if (rt == net->ipv6.fib6_null_entry) {
3108                 fn = fib6_backtrack(fn, &fl6->saddr);
3109                 if (fn)
3110                         goto restart;
3111         }
3112
3113         res.f6i = rt;
3114         res.nh = rt->fib6_nh;
3115 out:
3116         if (ret) {
3117                 ip6_hold_safe(net, &ret);
3118         } else {
3119                 res.fib6_flags = res.f6i->fib6_flags;
3120                 res.fib6_type = res.f6i->fib6_type;
3121                 ret = ip6_create_rt_rcu(&res);
3122         }
3123
3124         rcu_read_unlock();
3125
3126         trace_fib6_table_lookup(net, &res, table, fl6);
3127         return ret;
3128 };
3129
3130 static struct dst_entry *ip6_route_redirect(struct net *net,
3131                                             const struct flowi6 *fl6,
3132                                             const struct sk_buff *skb,
3133                                             const struct in6_addr *gateway)
3134 {
3135         int flags = RT6_LOOKUP_F_HAS_SADDR;
3136         struct ip6rd_flowi rdfl;
3137
3138         rdfl.fl6 = *fl6;
3139         rdfl.gateway = *gateway;
3140
3141         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3142                                 flags, __ip6_route_redirect);
3143 }
3144
3145 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3146                   kuid_t uid)
3147 {
3148         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3149         struct dst_entry *dst;
3150         struct flowi6 fl6 = {
3151                 .flowi6_iif = LOOPBACK_IFINDEX,
3152                 .flowi6_oif = oif,
3153                 .flowi6_mark = mark,
3154                 .daddr = iph->daddr,
3155                 .saddr = iph->saddr,
3156                 .flowlabel = ip6_flowinfo(iph),
3157                 .flowi6_uid = uid,
3158         };
3159
3160         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3161         rt6_do_redirect(dst, NULL, skb);
3162         dst_release(dst);
3163 }
3164 EXPORT_SYMBOL_GPL(ip6_redirect);
3165
3166 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3167 {
3168         const struct ipv6hdr *iph = ipv6_hdr(skb);
3169         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3170         struct dst_entry *dst;
3171         struct flowi6 fl6 = {
3172                 .flowi6_iif = LOOPBACK_IFINDEX,
3173                 .flowi6_oif = oif,
3174                 .daddr = msg->dest,
3175                 .saddr = iph->daddr,
3176                 .flowi6_uid = sock_net_uid(net, NULL),
3177         };
3178
3179         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3180         rt6_do_redirect(dst, NULL, skb);
3181         dst_release(dst);
3182 }
3183
3184 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3185 {
3186         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3187                      sk->sk_uid);
3188 }
3189 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3190
3191 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3192 {
3193         struct net_device *dev = dst->dev;
3194         unsigned int mtu = dst_mtu(dst);
3195         struct net *net = dev_net(dev);
3196
3197         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3198
3199         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3200                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3201
3202         /*
3203          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3204          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3205          * IPV6_MAXPLEN is also valid and means: "any MSS,
3206          * rely only on pmtu discovery"
3207          */
3208         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3209                 mtu = IPV6_MAXPLEN;
3210         return mtu;
3211 }
3212
3213 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3214 {
3215         return ip6_dst_mtu_maybe_forward(dst, false);
3216 }
3217 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3218
3219 /* MTU selection:
3220  * 1. mtu on route is locked - use it
3221  * 2. mtu from nexthop exception
3222  * 3. mtu from egress device
3223  *
3224  * based on ip6_dst_mtu_forward and exception logic of
3225  * rt6_find_cached_rt; called with rcu_read_lock
3226  */
3227 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3228                       const struct in6_addr *daddr,
3229                       const struct in6_addr *saddr)
3230 {
3231         const struct fib6_nh *nh = res->nh;
3232         struct fib6_info *f6i = res->f6i;
3233         struct inet6_dev *idev;
3234         struct rt6_info *rt;
3235         u32 mtu = 0;
3236
3237         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3238                 mtu = f6i->fib6_pmtu;
3239                 if (mtu)
3240                         goto out;
3241         }
3242
3243         rt = rt6_find_cached_rt(res, daddr, saddr);
3244         if (unlikely(rt)) {
3245                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3246         } else {
3247                 struct net_device *dev = nh->fib_nh_dev;
3248
3249                 mtu = IPV6_MIN_MTU;
3250                 idev = __in6_dev_get(dev);
3251                 if (idev && idev->cnf.mtu6 > mtu)
3252                         mtu = idev->cnf.mtu6;
3253         }
3254
3255         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3256 out:
3257         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3258 }
3259
3260 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3261                                   struct flowi6 *fl6)
3262 {
3263         struct dst_entry *dst;
3264         struct rt6_info *rt;
3265         struct inet6_dev *idev = in6_dev_get(dev);
3266         struct net *net = dev_net(dev);
3267
3268         if (unlikely(!idev))
3269                 return ERR_PTR(-ENODEV);
3270
3271         rt = ip6_dst_alloc(net, dev, 0);
3272         if (unlikely(!rt)) {
3273                 in6_dev_put(idev);
3274                 dst = ERR_PTR(-ENOMEM);
3275                 goto out;
3276         }
3277
3278         rt->dst.input = ip6_input;
3279         rt->dst.output  = ip6_output;
3280         rt->rt6i_gateway  = fl6->daddr;
3281         rt->rt6i_dst.addr = fl6->daddr;
3282         rt->rt6i_dst.plen = 128;
3283         rt->rt6i_idev     = idev;
3284         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3285
3286         /* Add this dst into uncached_list so that rt6_disable_ip() can
3287          * do proper release of the net_device
3288          */
3289         rt6_uncached_list_add(rt);
3290         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3291
3292         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3293
3294 out:
3295         return dst;
3296 }
3297
3298 static int ip6_dst_gc(struct dst_ops *ops)
3299 {
3300         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3301         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3302         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3303         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3304         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3305         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3306         int entries;
3307
3308         entries = dst_entries_get_fast(ops);
3309         if (entries > rt_max_size)
3310                 entries = dst_entries_get_slow(ops);
3311
3312         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3313             entries <= rt_max_size)
3314                 goto out;
3315
3316         net->ipv6.ip6_rt_gc_expire++;
3317         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3318         entries = dst_entries_get_slow(ops);
3319         if (entries < ops->gc_thresh)
3320                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3321 out:
3322         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3323         return entries > rt_max_size;
3324 }
3325
3326 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3327                                const struct in6_addr *gw_addr, u32 tbid,
3328                                int flags, struct fib6_result *res)
3329 {
3330         struct flowi6 fl6 = {
3331                 .flowi6_oif = cfg->fc_ifindex,
3332                 .daddr = *gw_addr,
3333                 .saddr = cfg->fc_prefsrc,
3334         };
3335         struct fib6_table *table;
3336         int err;
3337
3338         table = fib6_get_table(net, tbid);
3339         if (!table)
3340                 return -EINVAL;
3341
3342         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3343                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3344
3345         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3346
3347         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3348         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3349                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3350                                  cfg->fc_ifindex != 0, NULL, flags);
3351
3352         return err;
3353 }
3354
3355 static int ip6_route_check_nh_onlink(struct net *net,
3356                                      struct fib6_config *cfg,
3357                                      const struct net_device *dev,
3358                                      struct netlink_ext_ack *extack)
3359 {
3360         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3361         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3362         struct fib6_result res = {};
3363         int err;
3364
3365         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3366         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3367             /* ignore match if it is the default route */
3368             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3369             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3370                 NL_SET_ERR_MSG(extack,
3371                                "Nexthop has invalid gateway or device mismatch");
3372                 err = -EINVAL;
3373         }
3374
3375         return err;
3376 }
3377
3378 static int ip6_route_check_nh(struct net *net,
3379                               struct fib6_config *cfg,
3380                               struct net_device **_dev,
3381                               struct inet6_dev **idev)
3382 {
3383         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3384         struct net_device *dev = _dev ? *_dev : NULL;
3385         int flags = RT6_LOOKUP_F_IFACE;
3386         struct fib6_result res = {};
3387         int err = -EHOSTUNREACH;
3388
3389         if (cfg->fc_table) {
3390                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3391                                           cfg->fc_table, flags, &res);
3392                 /* gw_addr can not require a gateway or resolve to a reject
3393                  * route. If a device is given, it must match the result.
3394                  */
3395                 if (err || res.fib6_flags & RTF_REJECT ||
3396                     res.nh->fib_nh_gw_family ||
3397                     (dev && dev != res.nh->fib_nh_dev))
3398                         err = -EHOSTUNREACH;
3399         }
3400
3401         if (err < 0) {
3402                 struct flowi6 fl6 = {
3403                         .flowi6_oif = cfg->fc_ifindex,
3404                         .daddr = *gw_addr,
3405                 };
3406
3407                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3408                 if (err || res.fib6_flags & RTF_REJECT ||
3409                     res.nh->fib_nh_gw_family)
3410                         err = -EHOSTUNREACH;
3411
3412                 if (err)
3413                         return err;
3414
3415                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3416                                  cfg->fc_ifindex != 0, NULL, flags);
3417         }
3418
3419         err = 0;
3420         if (dev) {
3421                 if (dev != res.nh->fib_nh_dev)
3422                         err = -EHOSTUNREACH;
3423         } else {
3424                 *_dev = dev = res.nh->fib_nh_dev;
3425                 dev_hold(dev);
3426                 *idev = in6_dev_get(dev);
3427         }
3428
3429         return err;
3430 }
3431
3432 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3433                            struct net_device **_dev, struct inet6_dev **idev,
3434                            struct netlink_ext_ack *extack)
3435 {
3436         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3437         int gwa_type = ipv6_addr_type(gw_addr);
3438         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3439         const struct net_device *dev = *_dev;
3440         bool need_addr_check = !dev;
3441         int err = -EINVAL;
3442
3443         /* if gw_addr is local we will fail to detect this in case
3444          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3445          * will return already-added prefix route via interface that
3446          * prefix route was assigned to, which might be non-loopback.
3447          */
3448         if (dev &&
3449             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3450                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3451                 goto out;
3452         }
3453
3454         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3455                 /* IPv6 strictly inhibits using not link-local
3456                  * addresses as nexthop address.
3457                  * Otherwise, router will not able to send redirects.
3458                  * It is very good, but in some (rare!) circumstances
3459                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3460                  * some exceptions. --ANK
3461                  * We allow IPv4-mapped nexthops to support RFC4798-type
3462                  * addressing
3463                  */
3464                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3465                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3466                         goto out;
3467                 }
3468
3469                 rcu_read_lock();
3470
3471                 if (cfg->fc_flags & RTNH_F_ONLINK)
3472                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3473                 else
3474                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3475
3476                 rcu_read_unlock();
3477
3478                 if (err)
3479                         goto out;
3480         }
3481
3482         /* reload in case device was changed */
3483         dev = *_dev;
3484
3485         err = -EINVAL;
3486         if (!dev) {
3487                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3488                 goto out;
3489         } else if (dev->flags & IFF_LOOPBACK) {
3490                 NL_SET_ERR_MSG(extack,
3491                                "Egress device can not be loopback device for this route");
3492                 goto out;
3493         }
3494
3495         /* if we did not check gw_addr above, do so now that the
3496          * egress device has been resolved.
3497          */
3498         if (need_addr_check &&
3499             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3500                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3501                 goto out;
3502         }
3503
3504         err = 0;
3505 out:
3506         return err;
3507 }
3508
3509 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3510 {
3511         if ((flags & RTF_REJECT) ||
3512             (dev && (dev->flags & IFF_LOOPBACK) &&
3513              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3514              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3515                 return true;
3516
3517         return false;
3518 }
3519
3520 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3521                  struct fib6_config *cfg, gfp_t gfp_flags,
3522                  struct netlink_ext_ack *extack)
3523 {
3524         struct net_device *dev = NULL;
3525         struct inet6_dev *idev = NULL;
3526         int addr_type;
3527         int err;
3528
3529         fib6_nh->fib_nh_family = AF_INET6;
3530 #ifdef CONFIG_IPV6_ROUTER_PREF
3531         fib6_nh->last_probe = jiffies;
3532 #endif
3533         if (cfg->fc_is_fdb) {
3534                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3535                 fib6_nh->fib_nh_gw_family = AF_INET6;
3536                 return 0;
3537         }
3538
3539         err = -ENODEV;
3540         if (cfg->fc_ifindex) {
3541                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3542                 if (!dev)
3543                         goto out;
3544                 idev = in6_dev_get(dev);
3545                 if (!idev)
3546                         goto out;
3547         }
3548
3549         if (cfg->fc_flags & RTNH_F_ONLINK) {
3550                 if (!dev) {
3551                         NL_SET_ERR_MSG(extack,
3552                                        "Nexthop device required for onlink");
3553                         goto out;
3554                 }
3555
3556                 if (!(dev->flags & IFF_UP)) {
3557                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3558                         err = -ENETDOWN;
3559                         goto out;
3560                 }
3561
3562                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3563         }
3564
3565         fib6_nh->fib_nh_weight = 1;
3566
3567         /* We cannot add true routes via loopback here,
3568          * they would result in kernel looping; promote them to reject routes
3569          */
3570         addr_type = ipv6_addr_type(&cfg->fc_dst);
3571         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3572                 /* hold loopback dev/idev if we haven't done so. */
3573                 if (dev != net->loopback_dev) {
3574                         if (dev) {
3575                                 dev_put(dev);
3576                                 in6_dev_put(idev);
3577                         }
3578                         dev = net->loopback_dev;
3579                         dev_hold(dev);
3580                         idev = in6_dev_get(dev);
3581                         if (!idev) {
3582                                 err = -ENODEV;
3583                                 goto out;
3584                         }
3585                 }
3586                 goto pcpu_alloc;
3587         }
3588
3589         if (cfg->fc_flags & RTF_GATEWAY) {
3590                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3591                 if (err)
3592                         goto out;
3593
3594                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3595                 fib6_nh->fib_nh_gw_family = AF_INET6;
3596         }
3597
3598         err = -ENODEV;
3599         if (!dev)
3600                 goto out;
3601
3602         if (idev->cnf.disable_ipv6) {
3603                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3604                 err = -EACCES;
3605                 goto out;
3606         }
3607
3608         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3609                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3610                 err = -ENETDOWN;
3611                 goto out;
3612         }
3613
3614         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3615             !netif_carrier_ok(dev))
3616                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3617
3618         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3619                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3620         if (err)
3621                 goto out;
3622
3623 pcpu_alloc:
3624         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3625         if (!fib6_nh->rt6i_pcpu) {
3626                 err = -ENOMEM;
3627                 goto out;
3628         }
3629
3630         fib6_nh->fib_nh_dev = dev;
3631         fib6_nh->fib_nh_oif = dev->ifindex;
3632         err = 0;
3633 out:
3634         if (idev)
3635                 in6_dev_put(idev);
3636
3637         if (err) {
3638                 lwtstate_put(fib6_nh->fib_nh_lws);
3639                 fib6_nh->fib_nh_lws = NULL;
3640                 dev_put(dev);
3641         }
3642
3643         return err;
3644 }
3645
3646 void fib6_nh_release(struct fib6_nh *fib6_nh)
3647 {
3648         struct rt6_exception_bucket *bucket;
3649
3650         rcu_read_lock();
3651
3652         fib6_nh_flush_exceptions(fib6_nh, NULL);
3653         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3654         if (bucket) {
3655                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3656                 kfree(bucket);
3657         }
3658
3659         rcu_read_unlock();
3660
3661         if (fib6_nh->rt6i_pcpu) {
3662                 int cpu;
3663
3664                 for_each_possible_cpu(cpu) {
3665                         struct rt6_info **ppcpu_rt;
3666                         struct rt6_info *pcpu_rt;
3667
3668                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3669                         pcpu_rt = *ppcpu_rt;
3670                         if (pcpu_rt) {
3671                                 dst_dev_put(&pcpu_rt->dst);
3672                                 dst_release(&pcpu_rt->dst);
3673                                 *ppcpu_rt = NULL;
3674                         }
3675                 }
3676
3677                 free_percpu(fib6_nh->rt6i_pcpu);
3678         }
3679
3680         fib_nh_common_release(&fib6_nh->nh_common);
3681 }
3682
3683 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3684                                               gfp_t gfp_flags,
3685                                               struct netlink_ext_ack *extack)
3686 {
3687         struct net *net = cfg->fc_nlinfo.nl_net;
3688         struct fib6_info *rt = NULL;
3689         struct nexthop *nh = NULL;
3690         struct fib6_table *table;
3691         struct fib6_nh *fib6_nh;
3692         int err = -EINVAL;
3693         int addr_type;
3694
3695         /* RTF_PCPU is an internal flag; can not be set by userspace */
3696         if (cfg->fc_flags & RTF_PCPU) {
3697                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3698                 goto out;
3699         }
3700
3701         /* RTF_CACHE is an internal flag; can not be set by userspace */
3702         if (cfg->fc_flags & RTF_CACHE) {
3703                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3704                 goto out;
3705         }
3706
3707         if (cfg->fc_type > RTN_MAX) {
3708                 NL_SET_ERR_MSG(extack, "Invalid route type");
3709                 goto out;
3710         }
3711
3712         if (cfg->fc_dst_len > 128) {
3713                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3714                 goto out;
3715         }
3716         if (cfg->fc_src_len > 128) {
3717                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3718                 goto out;
3719         }
3720 #ifndef CONFIG_IPV6_SUBTREES
3721         if (cfg->fc_src_len) {
3722                 NL_SET_ERR_MSG(extack,
3723                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3724                 goto out;
3725         }
3726 #endif
3727         if (cfg->fc_nh_id) {
3728                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3729                 if (!nh) {
3730                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3731                         goto out;
3732                 }
3733                 err = fib6_check_nexthop(nh, cfg, extack);
3734                 if (err)
3735                         goto out;
3736         }
3737
3738         err = -ENOBUFS;
3739         if (cfg->fc_nlinfo.nlh &&
3740             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3741                 table = fib6_get_table(net, cfg->fc_table);
3742                 if (!table) {
3743                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3744                         table = fib6_new_table(net, cfg->fc_table);
3745                 }
3746         } else {
3747                 table = fib6_new_table(net, cfg->fc_table);
3748         }
3749
3750         if (!table)
3751                 goto out;
3752
3753         err = -ENOMEM;
3754         rt = fib6_info_alloc(gfp_flags, !nh);
3755         if (!rt)
3756                 goto out;
3757
3758         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3759                                                extack);
3760         if (IS_ERR(rt->fib6_metrics)) {
3761                 err = PTR_ERR(rt->fib6_metrics);
3762                 /* Do not leave garbage there. */
3763                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3764                 goto out_free;
3765         }
3766
3767         if (cfg->fc_flags & RTF_ADDRCONF)
3768                 rt->dst_nocount = true;
3769
3770         if (cfg->fc_flags & RTF_EXPIRES)
3771                 fib6_set_expires(rt, jiffies +
3772                                 clock_t_to_jiffies(cfg->fc_expires));
3773         else
3774                 fib6_clean_expires(rt);
3775
3776         if (cfg->fc_protocol == RTPROT_UNSPEC)
3777                 cfg->fc_protocol = RTPROT_BOOT;
3778         rt->fib6_protocol = cfg->fc_protocol;
3779
3780         rt->fib6_table = table;
3781         rt->fib6_metric = cfg->fc_metric;
3782         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3783         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3784
3785         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3786         rt->fib6_dst.plen = cfg->fc_dst_len;
3787
3788 #ifdef CONFIG_IPV6_SUBTREES
3789         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3790         rt->fib6_src.plen = cfg->fc_src_len;
3791 #endif
3792         if (nh) {
3793                 if (rt->fib6_src.plen) {
3794                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3795                         goto out_free;
3796                 }
3797                 if (!nexthop_get(nh)) {
3798                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3799                         goto out_free;
3800                 }
3801                 rt->nh = nh;
3802                 fib6_nh = nexthop_fib6_nh(rt->nh);
3803         } else {
3804                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3805                 if (err)
3806                         goto out;
3807
3808                 fib6_nh = rt->fib6_nh;
3809
3810                 /* We cannot add true routes via loopback here, they would
3811                  * result in kernel looping; promote them to reject routes
3812                  */
3813                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3814                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3815                                    addr_type))
3816                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3817         }
3818
3819         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3820                 struct net_device *dev = fib6_nh->fib_nh_dev;
3821
3822                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3823                         NL_SET_ERR_MSG(extack, "Invalid source address");
3824                         err = -EINVAL;
3825                         goto out;
3826                 }
3827                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3828                 rt->fib6_prefsrc.plen = 128;
3829         } else
3830                 rt->fib6_prefsrc.plen = 0;
3831
3832         return rt;
3833 out:
3834         fib6_info_release(rt);
3835         return ERR_PTR(err);
3836 out_free:
3837         ip_fib_metrics_put(rt->fib6_metrics);
3838         kfree(rt);
3839         return ERR_PTR(err);
3840 }
3841
3842 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3843                   struct netlink_ext_ack *extack)
3844 {
3845         struct fib6_info *rt;
3846         int err;
3847
3848         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3849         if (IS_ERR(rt))
3850                 return PTR_ERR(rt);
3851
3852         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3853         fib6_info_release(rt);
3854
3855         return err;
3856 }
3857
3858 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3859 {
3860         struct net *net = info->nl_net;
3861         struct fib6_table *table;
3862         int err;
3863
3864         if (rt == net->ipv6.fib6_null_entry) {
3865                 err = -ENOENT;
3866                 goto out;
3867         }
3868
3869         table = rt->fib6_table;
3870         spin_lock_bh(&table->tb6_lock);
3871         err = fib6_del(rt, info);
3872         spin_unlock_bh(&table->tb6_lock);
3873
3874 out:
3875         fib6_info_release(rt);
3876         return err;
3877 }
3878
3879 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3880 {
3881         struct nl_info info = {
3882                 .nl_net = net,
3883                 .skip_notify = skip_notify
3884         };
3885
3886         return __ip6_del_rt(rt, &info);
3887 }
3888
3889 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3890 {
3891         struct nl_info *info = &cfg->fc_nlinfo;
3892         struct net *net = info->nl_net;
3893         struct sk_buff *skb = NULL;
3894         struct fib6_table *table;
3895         int err = -ENOENT;
3896
3897         if (rt == net->ipv6.fib6_null_entry)
3898                 goto out_put;
3899         table = rt->fib6_table;
3900         spin_lock_bh(&table->tb6_lock);
3901
3902         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3903                 struct fib6_info *sibling, *next_sibling;
3904                 struct fib6_node *fn;
3905
3906                 /* prefer to send a single notification with all hops */
3907                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3908                 if (skb) {
3909                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3910
3911                         if (rt6_fill_node(net, skb, rt, NULL,
3912                                           NULL, NULL, 0, RTM_DELROUTE,
3913                                           info->portid, seq, 0) < 0) {
3914                                 kfree_skb(skb);
3915                                 skb = NULL;
3916                         } else
3917                                 info->skip_notify = 1;
3918                 }
3919
3920                 /* 'rt' points to the first sibling route. If it is not the
3921                  * leaf, then we do not need to send a notification. Otherwise,
3922                  * we need to check if the last sibling has a next route or not
3923                  * and emit a replace or delete notification, respectively.
3924                  */
3925                 info->skip_notify_kernel = 1;
3926                 fn = rcu_dereference_protected(rt->fib6_node,
3927                                             lockdep_is_held(&table->tb6_lock));
3928                 if (rcu_access_pointer(fn->leaf) == rt) {
3929                         struct fib6_info *last_sibling, *replace_rt;
3930
3931                         last_sibling = list_last_entry(&rt->fib6_siblings,
3932                                                        struct fib6_info,
3933                                                        fib6_siblings);
3934                         replace_rt = rcu_dereference_protected(
3935                                             last_sibling->fib6_next,
3936                                             lockdep_is_held(&table->tb6_lock));
3937                         if (replace_rt)
3938                                 call_fib6_entry_notifiers_replace(net,
3939                                                                   replace_rt);
3940                         else
3941                                 call_fib6_multipath_entry_notifiers(net,
3942                                                        FIB_EVENT_ENTRY_DEL,
3943                                                        rt, rt->fib6_nsiblings,
3944                                                        NULL);
3945                 }
3946                 list_for_each_entry_safe(sibling, next_sibling,
3947                                          &rt->fib6_siblings,
3948                                          fib6_siblings) {
3949                         err = fib6_del(sibling, info);
3950                         if (err)
3951                                 goto out_unlock;
3952                 }
3953         }
3954
3955         err = fib6_del(rt, info);
3956 out_unlock:
3957         spin_unlock_bh(&table->tb6_lock);
3958 out_put:
3959         fib6_info_release(rt);
3960
3961         if (skb) {
3962                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3963                             info->nlh, gfp_any());
3964         }
3965         return err;
3966 }
3967
3968 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3969 {
3970         int rc = -ESRCH;
3971
3972         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3973                 goto out;
3974
3975         if (cfg->fc_flags & RTF_GATEWAY &&
3976             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3977                 goto out;
3978
3979         rc = rt6_remove_exception_rt(rt);
3980 out:
3981         return rc;
3982 }
3983
3984 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3985                              struct fib6_nh *nh)
3986 {
3987         struct fib6_result res = {
3988                 .f6i = rt,
3989                 .nh = nh,
3990         };
3991         struct rt6_info *rt_cache;
3992
3993         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3994         if (rt_cache)
3995                 return __ip6_del_cached_rt(rt_cache, cfg);
3996
3997         return 0;
3998 }
3999
4000 struct fib6_nh_del_cached_rt_arg {
4001         struct fib6_config *cfg;
4002         struct fib6_info *f6i;
4003 };
4004
4005 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
4006 {
4007         struct fib6_nh_del_cached_rt_arg *arg = _arg;
4008         int rc;
4009
4010         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
4011         return rc != -ESRCH ? rc : 0;
4012 }
4013
4014 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
4015 {
4016         struct fib6_nh_del_cached_rt_arg arg = {
4017                 .cfg = cfg,
4018                 .f6i = f6i
4019         };
4020
4021         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
4022 }
4023
4024 static int ip6_route_del(struct fib6_config *cfg,
4025                          struct netlink_ext_ack *extack)
4026 {
4027         struct fib6_table *table;
4028         struct fib6_info *rt;
4029         struct fib6_node *fn;
4030         int err = -ESRCH;
4031
4032         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
4033         if (!table) {
4034                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
4035                 return err;
4036         }
4037
4038         rcu_read_lock();
4039
4040         fn = fib6_locate(&table->tb6_root,
4041                          &cfg->fc_dst, cfg->fc_dst_len,
4042                          &cfg->fc_src, cfg->fc_src_len,
4043                          !(cfg->fc_flags & RTF_CACHE));
4044
4045         if (fn) {
4046                 for_each_fib6_node_rt_rcu(fn) {
4047                         struct fib6_nh *nh;
4048
4049                         if (rt->nh && cfg->fc_nh_id &&
4050                             rt->nh->id != cfg->fc_nh_id)
4051                                 continue;
4052
4053                         if (cfg->fc_flags & RTF_CACHE) {
4054                                 int rc = 0;
4055
4056                                 if (rt->nh) {
4057                                         rc = ip6_del_cached_rt_nh(cfg, rt);
4058                                 } else if (cfg->fc_nh_id) {
4059                                         continue;
4060                                 } else {
4061                                         nh = rt->fib6_nh;
4062                                         rc = ip6_del_cached_rt(cfg, rt, nh);
4063                                 }
4064                                 if (rc != -ESRCH) {
4065                                         rcu_read_unlock();
4066                                         return rc;
4067                                 }
4068                                 continue;
4069                         }
4070
4071                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
4072                                 continue;
4073                         if (cfg->fc_protocol &&
4074                             cfg->fc_protocol != rt->fib6_protocol)
4075                                 continue;
4076
4077                         if (rt->nh) {
4078                                 if (!fib6_info_hold_safe(rt))
4079                                         continue;
4080                                 rcu_read_unlock();
4081
4082                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4083                         }
4084                         if (cfg->fc_nh_id)
4085                                 continue;
4086
4087                         nh = rt->fib6_nh;
4088                         if (cfg->fc_ifindex &&
4089                             (!nh->fib_nh_dev ||
4090                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4091                                 continue;
4092                         if (cfg->fc_flags & RTF_GATEWAY &&
4093                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4094                                 continue;
4095                         if (!fib6_info_hold_safe(rt))
4096                                 continue;
4097                         rcu_read_unlock();
4098
4099                         /* if gateway was specified only delete the one hop */
4100                         if (cfg->fc_flags & RTF_GATEWAY)
4101                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4102
4103                         return __ip6_del_rt_siblings(rt, cfg);
4104                 }
4105         }
4106         rcu_read_unlock();
4107
4108         return err;
4109 }
4110
4111 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4112 {
4113         struct netevent_redirect netevent;
4114         struct rt6_info *rt, *nrt = NULL;
4115         struct fib6_result res = {};
4116         struct ndisc_options ndopts;
4117         struct inet6_dev *in6_dev;
4118         struct neighbour *neigh;
4119         struct rd_msg *msg;
4120         int optlen, on_link;
4121         u8 *lladdr;
4122
4123         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4124         optlen -= sizeof(*msg);
4125
4126         if (optlen < 0) {
4127                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4128                 return;
4129         }
4130
4131         msg = (struct rd_msg *)icmp6_hdr(skb);
4132
4133         if (ipv6_addr_is_multicast(&msg->dest)) {
4134                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4135                 return;
4136         }
4137
4138         on_link = 0;
4139         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4140                 on_link = 1;
4141         } else if (ipv6_addr_type(&msg->target) !=
4142                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4143                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4144                 return;
4145         }
4146
4147         in6_dev = __in6_dev_get(skb->dev);
4148         if (!in6_dev)
4149                 return;
4150         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4151                 return;
4152
4153         /* RFC2461 8.1:
4154          *      The IP source address of the Redirect MUST be the same as the current
4155          *      first-hop router for the specified ICMP Destination Address.
4156          */
4157
4158         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4159                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4160                 return;
4161         }
4162
4163         lladdr = NULL;
4164         if (ndopts.nd_opts_tgt_lladdr) {
4165                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4166                                              skb->dev);
4167                 if (!lladdr) {
4168                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4169                         return;
4170                 }
4171         }
4172
4173         rt = (struct rt6_info *) dst;
4174         if (rt->rt6i_flags & RTF_REJECT) {
4175                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4176                 return;
4177         }
4178
4179         /* Redirect received -> path was valid.
4180          * Look, redirects are sent only in response to data packets,
4181          * so that this nexthop apparently is reachable. --ANK
4182          */
4183         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4184
4185         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4186         if (!neigh)
4187                 return;
4188
4189         /*
4190          *      We have finally decided to accept it.
4191          */
4192
4193         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4194                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4195                      NEIGH_UPDATE_F_OVERRIDE|
4196                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4197                                      NEIGH_UPDATE_F_ISROUTER)),
4198                      NDISC_REDIRECT, &ndopts);
4199
4200         rcu_read_lock();
4201         res.f6i = rcu_dereference(rt->from);
4202         if (!res.f6i)
4203                 goto out;
4204
4205         if (res.f6i->nh) {
4206                 struct fib6_nh_match_arg arg = {
4207                         .dev = dst->dev,
4208                         .gw = &rt->rt6i_gateway,
4209                 };
4210
4211                 nexthop_for_each_fib6_nh(res.f6i->nh,
4212                                          fib6_nh_find_match, &arg);
4213
4214                 /* fib6_info uses a nexthop that does not have fib6_nh
4215                  * using the dst->dev. Should be impossible
4216                  */
4217                 if (!arg.match)
4218                         goto out;
4219                 res.nh = arg.match;
4220         } else {
4221                 res.nh = res.f6i->fib6_nh;
4222         }
4223
4224         res.fib6_flags = res.f6i->fib6_flags;
4225         res.fib6_type = res.f6i->fib6_type;
4226         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4227         if (!nrt)
4228                 goto out;
4229
4230         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4231         if (on_link)
4232                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4233
4234         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4235
4236         /* rt6_insert_exception() will take care of duplicated exceptions */
4237         if (rt6_insert_exception(nrt, &res)) {
4238                 dst_release_immediate(&nrt->dst);
4239                 goto out;
4240         }
4241
4242         netevent.old = &rt->dst;
4243         netevent.new = &nrt->dst;
4244         netevent.daddr = &msg->dest;
4245         netevent.neigh = neigh;
4246         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4247
4248 out:
4249         rcu_read_unlock();
4250         neigh_release(neigh);
4251 }
4252
4253 #ifdef CONFIG_IPV6_ROUTE_INFO
4254 static struct fib6_info *rt6_get_route_info(struct net *net,
4255                                            const struct in6_addr *prefix, int prefixlen,
4256                                            const struct in6_addr *gwaddr,
4257                                            struct net_device *dev)
4258 {
4259         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4260         int ifindex = dev->ifindex;
4261         struct fib6_node *fn;
4262         struct fib6_info *rt = NULL;
4263         struct fib6_table *table;
4264
4265         table = fib6_get_table(net, tb_id);
4266         if (!table)
4267                 return NULL;
4268
4269         rcu_read_lock();
4270         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4271         if (!fn)
4272                 goto out;
4273
4274         for_each_fib6_node_rt_rcu(fn) {
4275                 /* these routes do not use nexthops */
4276                 if (rt->nh)
4277                         continue;
4278                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4279                         continue;
4280                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4281                     !rt->fib6_nh->fib_nh_gw_family)
4282                         continue;
4283                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4284                         continue;
4285                 if (!fib6_info_hold_safe(rt))
4286                         continue;
4287                 break;
4288         }
4289 out:
4290         rcu_read_unlock();
4291         return rt;
4292 }
4293
4294 static struct fib6_info *rt6_add_route_info(struct net *net,
4295                                            const struct in6_addr *prefix, int prefixlen,
4296                                            const struct in6_addr *gwaddr,
4297                                            struct net_device *dev,
4298                                            unsigned int pref)
4299 {
4300         struct fib6_config cfg = {
4301                 .fc_metric      = IP6_RT_PRIO_USER,
4302                 .fc_ifindex     = dev->ifindex,
4303                 .fc_dst_len     = prefixlen,
4304                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4305                                   RTF_UP | RTF_PREF(pref),
4306                 .fc_protocol = RTPROT_RA,
4307                 .fc_type = RTN_UNICAST,
4308                 .fc_nlinfo.portid = 0,
4309                 .fc_nlinfo.nlh = NULL,
4310                 .fc_nlinfo.nl_net = net,
4311         };
4312
4313         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4314         cfg.fc_dst = *prefix;
4315         cfg.fc_gateway = *gwaddr;
4316
4317         /* We should treat it as a default route if prefix length is 0. */
4318         if (!prefixlen)
4319                 cfg.fc_flags |= RTF_DEFAULT;
4320
4321         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4322
4323         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4324 }
4325 #endif
4326
4327 struct fib6_info *rt6_get_dflt_router(struct net *net,
4328                                      const struct in6_addr *addr,
4329                                      struct net_device *dev)
4330 {
4331         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4332         struct fib6_info *rt;
4333         struct fib6_table *table;
4334
4335         table = fib6_get_table(net, tb_id);
4336         if (!table)
4337                 return NULL;
4338
4339         rcu_read_lock();
4340         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4341                 struct fib6_nh *nh;
4342
4343                 /* RA routes do not use nexthops */
4344                 if (rt->nh)
4345                         continue;
4346
4347                 nh = rt->fib6_nh;
4348                 if (dev == nh->fib_nh_dev &&
4349                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4350                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4351                         break;
4352         }
4353         if (rt && !fib6_info_hold_safe(rt))
4354                 rt = NULL;
4355         rcu_read_unlock();
4356         return rt;
4357 }
4358
4359 struct fib6_info *rt6_add_dflt_router(struct net *net,
4360                                      const struct in6_addr *gwaddr,
4361                                      struct net_device *dev,
4362                                      unsigned int pref,
4363                                      u32 defrtr_usr_metric)
4364 {
4365         struct fib6_config cfg = {
4366                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4367                 .fc_metric      = defrtr_usr_metric,
4368                 .fc_ifindex     = dev->ifindex,
4369                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4370                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4371                 .fc_protocol = RTPROT_RA,
4372                 .fc_type = RTN_UNICAST,
4373                 .fc_nlinfo.portid = 0,
4374                 .fc_nlinfo.nlh = NULL,
4375                 .fc_nlinfo.nl_net = net,
4376         };
4377
4378         cfg.fc_gateway = *gwaddr;
4379
4380         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4381                 struct fib6_table *table;
4382
4383                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4384                 if (table)
4385                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4386         }
4387
4388         return rt6_get_dflt_router(net, gwaddr, dev);
4389 }
4390
4391 static void __rt6_purge_dflt_routers(struct net *net,
4392                                      struct fib6_table *table)
4393 {
4394         struct fib6_info *rt;
4395
4396 restart:
4397         rcu_read_lock();
4398         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4399                 struct net_device *dev = fib6_info_nh_dev(rt);
4400                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4401
4402                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4403                     (!idev || idev->cnf.accept_ra != 2) &&
4404                     fib6_info_hold_safe(rt)) {
4405                         rcu_read_unlock();
4406                         ip6_del_rt(net, rt, false);
4407                         goto restart;
4408                 }
4409         }
4410         rcu_read_unlock();
4411
4412         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4413 }
4414
4415 void rt6_purge_dflt_routers(struct net *net)
4416 {
4417         struct fib6_table *table;
4418         struct hlist_head *head;
4419         unsigned int h;
4420
4421         rcu_read_lock();
4422
4423         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4424                 head = &net->ipv6.fib_table_hash[h];
4425                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4426                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4427                                 __rt6_purge_dflt_routers(net, table);
4428                 }
4429         }
4430
4431         rcu_read_unlock();
4432 }
4433
4434 static void rtmsg_to_fib6_config(struct net *net,
4435                                  struct in6_rtmsg *rtmsg,
4436                                  struct fib6_config *cfg)
4437 {
4438         *cfg = (struct fib6_config){
4439                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4440                          : RT6_TABLE_MAIN,
4441                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4442                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4443                 .fc_expires = rtmsg->rtmsg_info,
4444                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4445                 .fc_src_len = rtmsg->rtmsg_src_len,
4446                 .fc_flags = rtmsg->rtmsg_flags,
4447                 .fc_type = rtmsg->rtmsg_type,
4448
4449                 .fc_nlinfo.nl_net = net,
4450
4451                 .fc_dst = rtmsg->rtmsg_dst,
4452                 .fc_src = rtmsg->rtmsg_src,
4453                 .fc_gateway = rtmsg->rtmsg_gateway,
4454         };
4455 }
4456
4457 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4458 {
4459         struct fib6_config cfg;
4460         int err;
4461
4462         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4463                 return -EINVAL;
4464         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4465                 return -EPERM;
4466
4467         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4468
4469         rtnl_lock();
4470         switch (cmd) {
4471         case SIOCADDRT:
4472                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4473                 break;
4474         case SIOCDELRT:
4475                 err = ip6_route_del(&cfg, NULL);
4476                 break;
4477         }
4478         rtnl_unlock();
4479         return err;
4480 }
4481
4482 /*
4483  *      Drop the packet on the floor
4484  */
4485
4486 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4487 {
4488         struct dst_entry *dst = skb_dst(skb);
4489         struct net *net = dev_net(dst->dev);
4490         struct inet6_dev *idev;
4491         int type;
4492
4493         if (netif_is_l3_master(skb->dev) &&
4494             dst->dev == net->loopback_dev)
4495                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4496         else
4497                 idev = ip6_dst_idev(dst);
4498
4499         switch (ipstats_mib_noroutes) {
4500         case IPSTATS_MIB_INNOROUTES:
4501                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4502                 if (type == IPV6_ADDR_ANY) {
4503                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4504                         break;
4505                 }
4506                 fallthrough;
4507         case IPSTATS_MIB_OUTNOROUTES:
4508                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4509                 break;
4510         }
4511
4512         /* Start over by dropping the dst for l3mdev case */
4513         if (netif_is_l3_master(skb->dev))
4514                 skb_dst_drop(skb);
4515
4516         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4517         kfree_skb(skb);
4518         return 0;
4519 }
4520
4521 static int ip6_pkt_discard(struct sk_buff *skb)
4522 {
4523         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4524 }
4525
4526 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4527 {
4528         skb->dev = skb_dst(skb)->dev;
4529         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4530 }
4531
4532 static int ip6_pkt_prohibit(struct sk_buff *skb)
4533 {
4534         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4535 }
4536
4537 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4538 {
4539         skb->dev = skb_dst(skb)->dev;
4540         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4541 }
4542
4543 /*
4544  *      Allocate a dst for local (unicast / anycast) address.
4545  */
4546
4547 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4548                                      struct inet6_dev *idev,
4549                                      const struct in6_addr *addr,
4550                                      bool anycast, gfp_t gfp_flags)
4551 {
4552         struct fib6_config cfg = {
4553                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4554                 .fc_ifindex = idev->dev->ifindex,
4555                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4556                 .fc_dst = *addr,
4557                 .fc_dst_len = 128,
4558                 .fc_protocol = RTPROT_KERNEL,
4559                 .fc_nlinfo.nl_net = net,
4560                 .fc_ignore_dev_down = true,
4561         };
4562         struct fib6_info *f6i;
4563
4564         if (anycast) {
4565                 cfg.fc_type = RTN_ANYCAST;
4566                 cfg.fc_flags |= RTF_ANYCAST;
4567         } else {
4568                 cfg.fc_type = RTN_LOCAL;
4569                 cfg.fc_flags |= RTF_LOCAL;
4570         }
4571
4572         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4573         if (!IS_ERR(f6i))
4574                 f6i->dst_nocount = true;
4575         return f6i;
4576 }
4577
4578 /* remove deleted ip from prefsrc entries */
4579 struct arg_dev_net_ip {
4580         struct net_device *dev;
4581         struct net *net;
4582         struct in6_addr *addr;
4583 };
4584
4585 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4586 {
4587         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4588         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4589         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4590
4591         if (!rt->nh &&
4592             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4593             rt != net->ipv6.fib6_null_entry &&
4594             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4595                 spin_lock_bh(&rt6_exception_lock);
4596                 /* remove prefsrc entry */
4597                 rt->fib6_prefsrc.plen = 0;
4598                 spin_unlock_bh(&rt6_exception_lock);
4599         }
4600         return 0;
4601 }
4602
4603 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4604 {
4605         struct net *net = dev_net(ifp->idev->dev);
4606         struct arg_dev_net_ip adni = {
4607                 .dev = ifp->idev->dev,
4608                 .net = net,
4609                 .addr = &ifp->addr,
4610         };
4611         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4612 }
4613
4614 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4615
4616 /* Remove routers and update dst entries when gateway turn into host. */
4617 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4618 {
4619         struct in6_addr *gateway = (struct in6_addr *)arg;
4620         struct fib6_nh *nh;
4621
4622         /* RA routes do not use nexthops */
4623         if (rt->nh)
4624                 return 0;
4625
4626         nh = rt->fib6_nh;
4627         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4628             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4629                 return -1;
4630
4631         /* Further clean up cached routes in exception table.
4632          * This is needed because cached route may have a different
4633          * gateway than its 'parent' in the case of an ip redirect.
4634          */
4635         fib6_nh_exceptions_clean_tohost(nh, gateway);
4636
4637         return 0;
4638 }
4639
4640 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4641 {
4642         fib6_clean_all(net, fib6_clean_tohost, gateway);
4643 }
4644
4645 struct arg_netdev_event {
4646         const struct net_device *dev;
4647         union {
4648                 unsigned char nh_flags;
4649                 unsigned long event;
4650         };
4651 };
4652
4653 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4654 {
4655         struct fib6_info *iter;
4656         struct fib6_node *fn;
4657
4658         fn = rcu_dereference_protected(rt->fib6_node,
4659                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4660         iter = rcu_dereference_protected(fn->leaf,
4661                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4662         while (iter) {
4663                 if (iter->fib6_metric == rt->fib6_metric &&
4664                     rt6_qualify_for_ecmp(iter))
4665                         return iter;
4666                 iter = rcu_dereference_protected(iter->fib6_next,
4667                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4668         }
4669
4670         return NULL;
4671 }
4672
4673 /* only called for fib entries with builtin fib6_nh */
4674 static bool rt6_is_dead(const struct fib6_info *rt)
4675 {
4676         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4677             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4678              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4679                 return true;
4680
4681         return false;
4682 }
4683
4684 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4685 {
4686         struct fib6_info *iter;
4687         int total = 0;
4688
4689         if (!rt6_is_dead(rt))
4690                 total += rt->fib6_nh->fib_nh_weight;
4691
4692         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4693                 if (!rt6_is_dead(iter))
4694                         total += iter->fib6_nh->fib_nh_weight;
4695         }
4696
4697         return total;
4698 }
4699
4700 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4701 {
4702         int upper_bound = -1;
4703
4704         if (!rt6_is_dead(rt)) {
4705                 *weight += rt->fib6_nh->fib_nh_weight;
4706                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4707                                                     total) - 1;
4708         }
4709         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4710 }
4711
4712 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4713 {
4714         struct fib6_info *iter;
4715         int weight = 0;
4716
4717         rt6_upper_bound_set(rt, &weight, total);
4718
4719         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4720                 rt6_upper_bound_set(iter, &weight, total);
4721 }
4722
4723 void rt6_multipath_rebalance(struct fib6_info *rt)
4724 {
4725         struct fib6_info *first;
4726         int total;
4727
4728         /* In case the entire multipath route was marked for flushing,
4729          * then there is no need to rebalance upon the removal of every
4730          * sibling route.
4731          */
4732         if (!rt->fib6_nsiblings || rt->should_flush)
4733                 return;
4734
4735         /* During lookup routes are evaluated in order, so we need to
4736          * make sure upper bounds are assigned from the first sibling
4737          * onwards.
4738          */
4739         first = rt6_multipath_first_sibling(rt);
4740         if (WARN_ON_ONCE(!first))
4741                 return;
4742
4743         total = rt6_multipath_total_weight(first);
4744         rt6_multipath_upper_bound_set(first, total);
4745 }
4746
4747 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4748 {
4749         const struct arg_netdev_event *arg = p_arg;
4750         struct net *net = dev_net(arg->dev);
4751
4752         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4753             rt->fib6_nh->fib_nh_dev == arg->dev) {
4754                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4755                 fib6_update_sernum_upto_root(net, rt);
4756                 rt6_multipath_rebalance(rt);
4757         }
4758
4759         return 0;
4760 }
4761
4762 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4763 {
4764         struct arg_netdev_event arg = {
4765                 .dev = dev,
4766                 {
4767                         .nh_flags = nh_flags,
4768                 },
4769         };
4770
4771         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4772                 arg.nh_flags |= RTNH_F_LINKDOWN;
4773
4774         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4775 }
4776
4777 /* only called for fib entries with inline fib6_nh */
4778 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4779                                    const struct net_device *dev)
4780 {
4781         struct fib6_info *iter;
4782
4783         if (rt->fib6_nh->fib_nh_dev == dev)
4784                 return true;
4785         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4786                 if (iter->fib6_nh->fib_nh_dev == dev)
4787                         return true;
4788
4789         return false;
4790 }
4791
4792 static void rt6_multipath_flush(struct fib6_info *rt)
4793 {
4794         struct fib6_info *iter;
4795
4796         rt->should_flush = 1;
4797         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4798                 iter->should_flush = 1;
4799 }
4800
4801 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4802                                              const struct net_device *down_dev)
4803 {
4804         struct fib6_info *iter;
4805         unsigned int dead = 0;
4806
4807         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4808             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4809                 dead++;
4810         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4811                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4812                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4813                         dead++;
4814
4815         return dead;
4816 }
4817
4818 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4819                                        const struct net_device *dev,
4820                                        unsigned char nh_flags)
4821 {
4822         struct fib6_info *iter;
4823
4824         if (rt->fib6_nh->fib_nh_dev == dev)
4825                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4826         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4827                 if (iter->fib6_nh->fib_nh_dev == dev)
4828                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4829 }
4830
4831 /* called with write lock held for table with rt */
4832 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4833 {
4834         const struct arg_netdev_event *arg = p_arg;
4835         const struct net_device *dev = arg->dev;
4836         struct net *net = dev_net(dev);
4837
4838         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4839                 return 0;
4840
4841         switch (arg->event) {
4842         case NETDEV_UNREGISTER:
4843                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4844         case NETDEV_DOWN:
4845                 if (rt->should_flush)
4846                         return -1;
4847                 if (!rt->fib6_nsiblings)
4848                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4849                 if (rt6_multipath_uses_dev(rt, dev)) {
4850                         unsigned int count;
4851
4852                         count = rt6_multipath_dead_count(rt, dev);
4853                         if (rt->fib6_nsiblings + 1 == count) {
4854                                 rt6_multipath_flush(rt);
4855                                 return -1;
4856                         }
4857                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4858                                                    RTNH_F_LINKDOWN);
4859                         fib6_update_sernum(net, rt);
4860                         rt6_multipath_rebalance(rt);
4861                 }
4862                 return -2;
4863         case NETDEV_CHANGE:
4864                 if (rt->fib6_nh->fib_nh_dev != dev ||
4865                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4866                         break;
4867                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4868                 rt6_multipath_rebalance(rt);
4869                 break;
4870         }
4871
4872         return 0;
4873 }
4874
4875 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4876 {
4877         struct arg_netdev_event arg = {
4878                 .dev = dev,
4879                 {
4880                         .event = event,
4881                 },
4882         };
4883         struct net *net = dev_net(dev);
4884
4885         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4886                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4887         else
4888                 fib6_clean_all(net, fib6_ifdown, &arg);
4889 }
4890
4891 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4892 {
4893         rt6_sync_down_dev(dev, event);
4894         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4895         neigh_ifdown(&nd_tbl, dev);
4896 }
4897
4898 struct rt6_mtu_change_arg {
4899         struct net_device *dev;
4900         unsigned int mtu;
4901         struct fib6_info *f6i;
4902 };
4903
4904 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4905 {
4906         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4907         struct fib6_info *f6i = arg->f6i;
4908
4909         /* For administrative MTU increase, there is no way to discover
4910          * IPv6 PMTU increase, so PMTU increase should be updated here.
4911          * Since RFC 1981 doesn't include administrative MTU increase
4912          * update PMTU increase is a MUST. (i.e. jumbo frame)
4913          */
4914         if (nh->fib_nh_dev == arg->dev) {
4915                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4916                 u32 mtu = f6i->fib6_pmtu;
4917
4918                 if (mtu >= arg->mtu ||
4919                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4920                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4921
4922                 spin_lock_bh(&rt6_exception_lock);
4923                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4924                 spin_unlock_bh(&rt6_exception_lock);
4925         }
4926
4927         return 0;
4928 }
4929
4930 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4931 {
4932         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4933         struct inet6_dev *idev;
4934
4935         /* In IPv6 pmtu discovery is not optional,
4936            so that RTAX_MTU lock cannot disable it.
4937            We still use this lock to block changes
4938            caused by addrconf/ndisc.
4939         */
4940
4941         idev = __in6_dev_get(arg->dev);
4942         if (!idev)
4943                 return 0;
4944
4945         if (fib6_metric_locked(f6i, RTAX_MTU))
4946                 return 0;
4947
4948         arg->f6i = f6i;
4949         if (f6i->nh) {
4950                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4951                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4952                                                 arg);
4953         }
4954
4955         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4956 }
4957
4958 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4959 {
4960         struct rt6_mtu_change_arg arg = {
4961                 .dev = dev,
4962                 .mtu = mtu,
4963         };
4964
4965         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4966 }
4967
4968 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4969         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4970         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4971         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4972         [RTA_OIF]               = { .type = NLA_U32 },
4973         [RTA_IIF]               = { .type = NLA_U32 },
4974         [RTA_PRIORITY]          = { .type = NLA_U32 },
4975         [RTA_METRICS]           = { .type = NLA_NESTED },
4976         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4977         [RTA_PREF]              = { .type = NLA_U8 },
4978         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4979         [RTA_ENCAP]             = { .type = NLA_NESTED },
4980         [RTA_EXPIRES]           = { .type = NLA_U32 },
4981         [RTA_UID]               = { .type = NLA_U32 },
4982         [RTA_MARK]              = { .type = NLA_U32 },
4983         [RTA_TABLE]             = { .type = NLA_U32 },
4984         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4985         [RTA_SPORT]             = { .type = NLA_U16 },
4986         [RTA_DPORT]             = { .type = NLA_U16 },
4987         [RTA_NH_ID]             = { .type = NLA_U32 },
4988 };
4989
4990 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4991                               struct fib6_config *cfg,
4992                               struct netlink_ext_ack *extack)
4993 {
4994         struct rtmsg *rtm;
4995         struct nlattr *tb[RTA_MAX+1];
4996         unsigned int pref;
4997         int err;
4998
4999         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5000                                      rtm_ipv6_policy, extack);
5001         if (err < 0)
5002                 goto errout;
5003
5004         err = -EINVAL;
5005         rtm = nlmsg_data(nlh);
5006
5007         *cfg = (struct fib6_config){
5008                 .fc_table = rtm->rtm_table,
5009                 .fc_dst_len = rtm->rtm_dst_len,
5010                 .fc_src_len = rtm->rtm_src_len,
5011                 .fc_flags = RTF_UP,
5012                 .fc_protocol = rtm->rtm_protocol,
5013                 .fc_type = rtm->rtm_type,
5014
5015                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
5016                 .fc_nlinfo.nlh = nlh,
5017                 .fc_nlinfo.nl_net = sock_net(skb->sk),
5018         };
5019
5020         if (rtm->rtm_type == RTN_UNREACHABLE ||
5021             rtm->rtm_type == RTN_BLACKHOLE ||
5022             rtm->rtm_type == RTN_PROHIBIT ||
5023             rtm->rtm_type == RTN_THROW)
5024                 cfg->fc_flags |= RTF_REJECT;
5025
5026         if (rtm->rtm_type == RTN_LOCAL)
5027                 cfg->fc_flags |= RTF_LOCAL;
5028
5029         if (rtm->rtm_flags & RTM_F_CLONED)
5030                 cfg->fc_flags |= RTF_CACHE;
5031
5032         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
5033
5034         if (tb[RTA_NH_ID]) {
5035                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
5036                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
5037                         NL_SET_ERR_MSG(extack,
5038                                        "Nexthop specification and nexthop id are mutually exclusive");
5039                         goto errout;
5040                 }
5041                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
5042         }
5043
5044         if (tb[RTA_GATEWAY]) {
5045                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
5046                 cfg->fc_flags |= RTF_GATEWAY;
5047         }
5048         if (tb[RTA_VIA]) {
5049                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
5050                 goto errout;
5051         }
5052
5053         if (tb[RTA_DST]) {
5054                 int plen = (rtm->rtm_dst_len + 7) >> 3;
5055
5056                 if (nla_len(tb[RTA_DST]) < plen)
5057                         goto errout;
5058
5059                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
5060         }
5061
5062         if (tb[RTA_SRC]) {
5063                 int plen = (rtm->rtm_src_len + 7) >> 3;
5064
5065                 if (nla_len(tb[RTA_SRC]) < plen)
5066                         goto errout;
5067
5068                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
5069         }
5070
5071         if (tb[RTA_PREFSRC])
5072                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
5073
5074         if (tb[RTA_OIF])
5075                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
5076
5077         if (tb[RTA_PRIORITY])
5078                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
5079
5080         if (tb[RTA_METRICS]) {
5081                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
5082                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
5083         }
5084
5085         if (tb[RTA_TABLE])
5086                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5087
5088         if (tb[RTA_MULTIPATH]) {
5089                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5090                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5091
5092                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5093                                                      cfg->fc_mp_len, extack);
5094                 if (err < 0)
5095                         goto errout;
5096         }
5097
5098         if (tb[RTA_PREF]) {
5099                 pref = nla_get_u8(tb[RTA_PREF]);
5100                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5101                     pref != ICMPV6_ROUTER_PREF_HIGH)
5102                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5103                 cfg->fc_flags |= RTF_PREF(pref);
5104         }
5105
5106         if (tb[RTA_ENCAP])
5107                 cfg->fc_encap = tb[RTA_ENCAP];
5108
5109         if (tb[RTA_ENCAP_TYPE]) {
5110                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5111
5112                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5113                 if (err < 0)
5114                         goto errout;
5115         }
5116
5117         if (tb[RTA_EXPIRES]) {
5118                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5119
5120                 if (addrconf_finite_timeout(timeout)) {
5121                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5122                         cfg->fc_flags |= RTF_EXPIRES;
5123                 }
5124         }
5125
5126         err = 0;
5127 errout:
5128         return err;
5129 }
5130
5131 struct rt6_nh {
5132         struct fib6_info *fib6_info;
5133         struct fib6_config r_cfg;
5134         struct list_head next;
5135 };
5136
5137 static int ip6_route_info_append(struct net *net,
5138                                  struct list_head *rt6_nh_list,
5139                                  struct fib6_info *rt,
5140                                  struct fib6_config *r_cfg)
5141 {
5142         struct rt6_nh *nh;
5143         int err = -EEXIST;
5144
5145         list_for_each_entry(nh, rt6_nh_list, next) {
5146                 /* check if fib6_info already exists */
5147                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5148                         return err;
5149         }
5150
5151         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5152         if (!nh)
5153                 return -ENOMEM;
5154         nh->fib6_info = rt;
5155         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5156         list_add_tail(&nh->next, rt6_nh_list);
5157
5158         return 0;
5159 }
5160
5161 static void ip6_route_mpath_notify(struct fib6_info *rt,
5162                                    struct fib6_info *rt_last,
5163                                    struct nl_info *info,
5164                                    __u16 nlflags)
5165 {
5166         /* if this is an APPEND route, then rt points to the first route
5167          * inserted and rt_last points to last route inserted. Userspace
5168          * wants a consistent dump of the route which starts at the first
5169          * nexthop. Since sibling routes are always added at the end of
5170          * the list, find the first sibling of the last route appended
5171          */
5172         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5173                 rt = list_first_entry(&rt_last->fib6_siblings,
5174                                       struct fib6_info,
5175                                       fib6_siblings);
5176         }
5177
5178         if (rt)
5179                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5180 }
5181
5182 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5183 {
5184         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5185         bool should_notify = false;
5186         struct fib6_info *leaf;
5187         struct fib6_node *fn;
5188
5189         rcu_read_lock();
5190         fn = rcu_dereference(rt->fib6_node);
5191         if (!fn)
5192                 goto out;
5193
5194         leaf = rcu_dereference(fn->leaf);
5195         if (!leaf)
5196                 goto out;
5197
5198         if (rt == leaf ||
5199             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5200              rt6_qualify_for_ecmp(leaf)))
5201                 should_notify = true;
5202 out:
5203         rcu_read_unlock();
5204
5205         return should_notify;
5206 }
5207
5208 static int ip6_route_multipath_add(struct fib6_config *cfg,
5209                                    struct netlink_ext_ack *extack)
5210 {
5211         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5212         struct nl_info *info = &cfg->fc_nlinfo;
5213         struct fib6_config r_cfg;
5214         struct rtnexthop *rtnh;
5215         struct fib6_info *rt;
5216         struct rt6_nh *err_nh;
5217         struct rt6_nh *nh, *nh_safe;
5218         __u16 nlflags;
5219         int remaining;
5220         int attrlen;
5221         int err = 1;
5222         int nhn = 0;
5223         int replace = (cfg->fc_nlinfo.nlh &&
5224                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5225         LIST_HEAD(rt6_nh_list);
5226
5227         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5228         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5229                 nlflags |= NLM_F_APPEND;
5230
5231         remaining = cfg->fc_mp_len;
5232         rtnh = (struct rtnexthop *)cfg->fc_mp;
5233
5234         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5235          * fib6_info structs per nexthop
5236          */
5237         while (rtnh_ok(rtnh, remaining)) {
5238                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5239                 if (rtnh->rtnh_ifindex)
5240                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5241
5242                 attrlen = rtnh_attrlen(rtnh);
5243                 if (attrlen > 0) {
5244                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5245
5246                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5247                         if (nla) {
5248                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5249                                 r_cfg.fc_flags |= RTF_GATEWAY;
5250                         }
5251                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5252                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5253                         if (nla)
5254                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5255                 }
5256
5257                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5258                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5259                 if (IS_ERR(rt)) {
5260                         err = PTR_ERR(rt);
5261                         rt = NULL;
5262                         goto cleanup;
5263                 }
5264                 if (!rt6_qualify_for_ecmp(rt)) {
5265                         err = -EINVAL;
5266                         NL_SET_ERR_MSG(extack,
5267                                        "Device only routes can not be added for IPv6 using the multipath API.");
5268                         fib6_info_release(rt);
5269                         goto cleanup;
5270                 }
5271
5272                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5273
5274                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5275                                             rt, &r_cfg);
5276                 if (err) {
5277                         fib6_info_release(rt);
5278                         goto cleanup;
5279                 }
5280
5281                 rtnh = rtnh_next(rtnh, &remaining);
5282         }
5283
5284         if (list_empty(&rt6_nh_list)) {
5285                 NL_SET_ERR_MSG(extack,
5286                                "Invalid nexthop configuration - no valid nexthops");
5287                 return -EINVAL;
5288         }
5289
5290         /* for add and replace send one notification with all nexthops.
5291          * Skip the notification in fib6_add_rt2node and send one with
5292          * the full route when done
5293          */
5294         info->skip_notify = 1;
5295
5296         /* For add and replace, send one notification with all nexthops. For
5297          * append, send one notification with all appended nexthops.
5298          */
5299         info->skip_notify_kernel = 1;
5300
5301         err_nh = NULL;
5302         list_for_each_entry(nh, &rt6_nh_list, next) {
5303                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5304                 fib6_info_release(nh->fib6_info);
5305
5306                 if (!err) {
5307                         /* save reference to last route successfully inserted */
5308                         rt_last = nh->fib6_info;
5309
5310                         /* save reference to first route for notification */
5311                         if (!rt_notif)
5312                                 rt_notif = nh->fib6_info;
5313                 }
5314
5315                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5316                 nh->fib6_info = NULL;
5317                 if (err) {
5318                         if (replace && nhn)
5319                                 NL_SET_ERR_MSG_MOD(extack,
5320                                                    "multipath route replace failed (check consistency of installed routes)");
5321                         err_nh = nh;
5322                         goto add_errout;
5323                 }
5324
5325                 /* Because each route is added like a single route we remove
5326                  * these flags after the first nexthop: if there is a collision,
5327                  * we have already failed to add the first nexthop:
5328                  * fib6_add_rt2node() has rejected it; when replacing, old
5329                  * nexthops have been replaced by first new, the rest should
5330                  * be added to it.
5331                  */
5332                 if (cfg->fc_nlinfo.nlh) {
5333                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5334                                                              NLM_F_REPLACE);
5335                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5336                 }
5337                 nhn++;
5338         }
5339
5340         /* An in-kernel notification should only be sent in case the new
5341          * multipath route is added as the first route in the node, or if
5342          * it was appended to it. We pass 'rt_notif' since it is the first
5343          * sibling and might allow us to skip some checks in the replace case.
5344          */
5345         if (ip6_route_mpath_should_notify(rt_notif)) {
5346                 enum fib_event_type fib_event;
5347
5348                 if (rt_notif->fib6_nsiblings != nhn - 1)
5349                         fib_event = FIB_EVENT_ENTRY_APPEND;
5350                 else
5351                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5352
5353                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5354                                                           fib_event, rt_notif,
5355                                                           nhn - 1, extack);
5356                 if (err) {
5357                         /* Delete all the siblings that were just added */
5358                         err_nh = NULL;
5359                         goto add_errout;
5360                 }
5361         }
5362
5363         /* success ... tell user about new route */
5364         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5365         goto cleanup;
5366
5367 add_errout:
5368         /* send notification for routes that were added so that
5369          * the delete notifications sent by ip6_route_del are
5370          * coherent
5371          */
5372         if (rt_notif)
5373                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5374
5375         /* Delete routes that were already added */
5376         list_for_each_entry(nh, &rt6_nh_list, next) {
5377                 if (err_nh == nh)
5378                         break;
5379                 ip6_route_del(&nh->r_cfg, extack);
5380         }
5381
5382 cleanup:
5383         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5384                 if (nh->fib6_info)
5385                         fib6_info_release(nh->fib6_info);
5386                 list_del(&nh->next);
5387                 kfree(nh);
5388         }
5389
5390         return err;
5391 }
5392
5393 static int ip6_route_multipath_del(struct fib6_config *cfg,
5394                                    struct netlink_ext_ack *extack)
5395 {
5396         struct fib6_config r_cfg;
5397         struct rtnexthop *rtnh;
5398         int last_err = 0;
5399         int remaining;
5400         int attrlen;
5401         int err;
5402
5403         remaining = cfg->fc_mp_len;
5404         rtnh = (struct rtnexthop *)cfg->fc_mp;
5405
5406         /* Parse a Multipath Entry */
5407         while (rtnh_ok(rtnh, remaining)) {
5408                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5409                 if (rtnh->rtnh_ifindex)
5410                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5411
5412                 attrlen = rtnh_attrlen(rtnh);
5413                 if (attrlen > 0) {
5414                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5415
5416                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5417                         if (nla) {
5418                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5419                                 r_cfg.fc_flags |= RTF_GATEWAY;
5420                         }
5421                 }
5422                 err = ip6_route_del(&r_cfg, extack);
5423                 if (err)
5424                         last_err = err;
5425
5426                 rtnh = rtnh_next(rtnh, &remaining);
5427         }
5428
5429         return last_err;
5430 }
5431
5432 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5433                               struct netlink_ext_ack *extack)
5434 {
5435         struct fib6_config cfg;
5436         int err;
5437
5438         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5439         if (err < 0)
5440                 return err;
5441
5442         if (cfg.fc_nh_id &&
5443             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5444                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5445                 return -EINVAL;
5446         }
5447
5448         if (cfg.fc_mp)
5449                 return ip6_route_multipath_del(&cfg, extack);
5450         else {
5451                 cfg.fc_delete_all_nh = 1;
5452                 return ip6_route_del(&cfg, extack);
5453         }
5454 }
5455
5456 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5457                               struct netlink_ext_ack *extack)
5458 {
5459         struct fib6_config cfg;
5460         int err;
5461
5462         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5463         if (err < 0)
5464                 return err;
5465
5466         if (cfg.fc_metric == 0)
5467                 cfg.fc_metric = IP6_RT_PRIO_USER;
5468
5469         if (cfg.fc_mp)
5470                 return ip6_route_multipath_add(&cfg, extack);
5471         else
5472                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5473 }
5474
5475 /* add the overhead of this fib6_nh to nexthop_len */
5476 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5477 {
5478         int *nexthop_len = arg;
5479
5480         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5481                      + NLA_ALIGN(sizeof(struct rtnexthop))
5482                      + nla_total_size(16); /* RTA_GATEWAY */
5483
5484         if (nh->fib_nh_lws) {
5485                 /* RTA_ENCAP_TYPE */
5486                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5487                 /* RTA_ENCAP */
5488                 *nexthop_len += nla_total_size(2);
5489         }
5490
5491         return 0;
5492 }
5493
5494 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5495 {
5496         int nexthop_len;
5497
5498         if (f6i->nh) {
5499                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5500                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5501                                          &nexthop_len);
5502         } else {
5503                 struct fib6_nh *nh = f6i->fib6_nh;
5504
5505                 nexthop_len = 0;
5506                 if (f6i->fib6_nsiblings) {
5507                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5508                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5509                                     + nla_total_size(16) /* RTA_GATEWAY */
5510                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5511
5512                         nexthop_len *= f6i->fib6_nsiblings;
5513                 }
5514                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5515         }
5516
5517         return NLMSG_ALIGN(sizeof(struct rtmsg))
5518                + nla_total_size(16) /* RTA_SRC */
5519                + nla_total_size(16) /* RTA_DST */
5520                + nla_total_size(16) /* RTA_GATEWAY */
5521                + nla_total_size(16) /* RTA_PREFSRC */
5522                + nla_total_size(4) /* RTA_TABLE */
5523                + nla_total_size(4) /* RTA_IIF */
5524                + nla_total_size(4) /* RTA_OIF */
5525                + nla_total_size(4) /* RTA_PRIORITY */
5526                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5527                + nla_total_size(sizeof(struct rta_cacheinfo))
5528                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5529                + nla_total_size(1) /* RTA_PREF */
5530                + nexthop_len;
5531 }
5532
5533 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5534                                  unsigned char *flags)
5535 {
5536         if (nexthop_is_multipath(nh)) {
5537                 struct nlattr *mp;
5538
5539                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5540                 if (!mp)
5541                         goto nla_put_failure;
5542
5543                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5544                         goto nla_put_failure;
5545
5546                 nla_nest_end(skb, mp);
5547         } else {
5548                 struct fib6_nh *fib6_nh;
5549
5550                 fib6_nh = nexthop_fib6_nh(nh);
5551                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5552                                      flags, false) < 0)
5553                         goto nla_put_failure;
5554         }
5555
5556         return 0;
5557
5558 nla_put_failure:
5559         return -EMSGSIZE;
5560 }
5561
5562 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5563                          struct fib6_info *rt, struct dst_entry *dst,
5564                          struct in6_addr *dest, struct in6_addr *src,
5565                          int iif, int type, u32 portid, u32 seq,
5566                          unsigned int flags)
5567 {
5568         struct rt6_info *rt6 = (struct rt6_info *)dst;
5569         struct rt6key *rt6_dst, *rt6_src;
5570         u32 *pmetrics, table, rt6_flags;
5571         unsigned char nh_flags = 0;
5572         struct nlmsghdr *nlh;
5573         struct rtmsg *rtm;
5574         long expires = 0;
5575
5576         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5577         if (!nlh)
5578                 return -EMSGSIZE;
5579
5580         if (rt6) {
5581                 rt6_dst = &rt6->rt6i_dst;
5582                 rt6_src = &rt6->rt6i_src;
5583                 rt6_flags = rt6->rt6i_flags;
5584         } else {
5585                 rt6_dst = &rt->fib6_dst;
5586                 rt6_src = &rt->fib6_src;
5587                 rt6_flags = rt->fib6_flags;
5588         }
5589
5590         rtm = nlmsg_data(nlh);
5591         rtm->rtm_family = AF_INET6;
5592         rtm->rtm_dst_len = rt6_dst->plen;
5593         rtm->rtm_src_len = rt6_src->plen;
5594         rtm->rtm_tos = 0;
5595         if (rt->fib6_table)
5596                 table = rt->fib6_table->tb6_id;
5597         else
5598                 table = RT6_TABLE_UNSPEC;
5599         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5600         if (nla_put_u32(skb, RTA_TABLE, table))
5601                 goto nla_put_failure;
5602
5603         rtm->rtm_type = rt->fib6_type;
5604         rtm->rtm_flags = 0;
5605         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5606         rtm->rtm_protocol = rt->fib6_protocol;
5607
5608         if (rt6_flags & RTF_CACHE)
5609                 rtm->rtm_flags |= RTM_F_CLONED;
5610
5611         if (dest) {
5612                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5613                         goto nla_put_failure;
5614                 rtm->rtm_dst_len = 128;
5615         } else if (rtm->rtm_dst_len)
5616                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5617                         goto nla_put_failure;
5618 #ifdef CONFIG_IPV6_SUBTREES
5619         if (src) {
5620                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5621                         goto nla_put_failure;
5622                 rtm->rtm_src_len = 128;
5623         } else if (rtm->rtm_src_len &&
5624                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5625                 goto nla_put_failure;
5626 #endif
5627         if (iif) {
5628 #ifdef CONFIG_IPV6_MROUTE
5629                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5630                         int err = ip6mr_get_route(net, skb, rtm, portid);
5631
5632                         if (err == 0)
5633                                 return 0;
5634                         if (err < 0)
5635                                 goto nla_put_failure;
5636                 } else
5637 #endif
5638                         if (nla_put_u32(skb, RTA_IIF, iif))
5639                                 goto nla_put_failure;
5640         } else if (dest) {
5641                 struct in6_addr saddr_buf;
5642                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5643                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5644                         goto nla_put_failure;
5645         }
5646
5647         if (rt->fib6_prefsrc.plen) {
5648                 struct in6_addr saddr_buf;
5649                 saddr_buf = rt->fib6_prefsrc.addr;
5650                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5651                         goto nla_put_failure;
5652         }
5653
5654         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5655         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5656                 goto nla_put_failure;
5657
5658         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5659                 goto nla_put_failure;
5660
5661         /* For multipath routes, walk the siblings list and add
5662          * each as a nexthop within RTA_MULTIPATH.
5663          */
5664         if (rt6) {
5665                 if (rt6_flags & RTF_GATEWAY &&
5666                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5667                         goto nla_put_failure;
5668
5669                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5670                         goto nla_put_failure;
5671
5672                 if (dst->lwtstate &&
5673                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5674                         goto nla_put_failure;
5675         } else if (rt->fib6_nsiblings) {
5676                 struct fib6_info *sibling, *next_sibling;
5677                 struct nlattr *mp;
5678
5679                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5680                 if (!mp)
5681                         goto nla_put_failure;
5682
5683                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5684                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5685                         goto nla_put_failure;
5686
5687                 list_for_each_entry_safe(sibling, next_sibling,
5688                                          &rt->fib6_siblings, fib6_siblings) {
5689                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5690                                             sibling->fib6_nh->fib_nh_weight,
5691                                             AF_INET6) < 0)
5692                                 goto nla_put_failure;
5693                 }
5694
5695                 nla_nest_end(skb, mp);
5696         } else if (rt->nh) {
5697                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5698                         goto nla_put_failure;
5699
5700                 if (nexthop_is_blackhole(rt->nh))
5701                         rtm->rtm_type = RTN_BLACKHOLE;
5702
5703                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5704                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5705                         goto nla_put_failure;
5706
5707                 rtm->rtm_flags |= nh_flags;
5708         } else {
5709                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5710                                      &nh_flags, false) < 0)
5711                         goto nla_put_failure;
5712
5713                 rtm->rtm_flags |= nh_flags;
5714         }
5715
5716         if (rt6_flags & RTF_EXPIRES) {
5717                 expires = dst ? dst->expires : rt->expires;
5718                 expires -= jiffies;
5719         }
5720
5721         if (!dst) {
5722                 if (rt->offload)
5723                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5724                 if (rt->trap)
5725                         rtm->rtm_flags |= RTM_F_TRAP;
5726                 if (rt->offload_failed)
5727                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5728         }
5729
5730         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5731                 goto nla_put_failure;
5732
5733         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5734                 goto nla_put_failure;
5735
5736
5737         nlmsg_end(skb, nlh);
5738         return 0;
5739
5740 nla_put_failure:
5741         nlmsg_cancel(skb, nlh);
5742         return -EMSGSIZE;
5743 }
5744
5745 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5746 {
5747         const struct net_device *dev = arg;
5748
5749         if (nh->fib_nh_dev == dev)
5750                 return 1;
5751
5752         return 0;
5753 }
5754
5755 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5756                                const struct net_device *dev)
5757 {
5758         if (f6i->nh) {
5759                 struct net_device *_dev = (struct net_device *)dev;
5760
5761                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5762                                                   fib6_info_nh_uses_dev,
5763                                                   _dev);
5764         }
5765
5766         if (f6i->fib6_nh->fib_nh_dev == dev)
5767                 return true;
5768
5769         if (f6i->fib6_nsiblings) {
5770                 struct fib6_info *sibling, *next_sibling;
5771
5772                 list_for_each_entry_safe(sibling, next_sibling,
5773                                          &f6i->fib6_siblings, fib6_siblings) {
5774                         if (sibling->fib6_nh->fib_nh_dev == dev)
5775                                 return true;
5776                 }
5777         }
5778
5779         return false;
5780 }
5781
5782 struct fib6_nh_exception_dump_walker {
5783         struct rt6_rtnl_dump_arg *dump;
5784         struct fib6_info *rt;
5785         unsigned int flags;
5786         unsigned int skip;
5787         unsigned int count;
5788 };
5789
5790 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5791 {
5792         struct fib6_nh_exception_dump_walker *w = arg;
5793         struct rt6_rtnl_dump_arg *dump = w->dump;
5794         struct rt6_exception_bucket *bucket;
5795         struct rt6_exception *rt6_ex;
5796         int i, err;
5797
5798         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5799         if (!bucket)
5800                 return 0;
5801
5802         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5803                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5804                         if (w->skip) {
5805                                 w->skip--;
5806                                 continue;
5807                         }
5808
5809                         /* Expiration of entries doesn't bump sernum, insertion
5810                          * does. Removal is triggered by insertion, so we can
5811                          * rely on the fact that if entries change between two
5812                          * partial dumps, this node is scanned again completely,
5813                          * see rt6_insert_exception() and fib6_dump_table().
5814                          *
5815                          * Count expired entries we go through as handled
5816                          * entries that we'll skip next time, in case of partial
5817                          * node dump. Otherwise, if entries expire meanwhile,
5818                          * we'll skip the wrong amount.
5819                          */
5820                         if (rt6_check_expired(rt6_ex->rt6i)) {
5821                                 w->count++;
5822                                 continue;
5823                         }
5824
5825                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5826                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5827                                             RTM_NEWROUTE,
5828                                             NETLINK_CB(dump->cb->skb).portid,
5829                                             dump->cb->nlh->nlmsg_seq, w->flags);
5830                         if (err)
5831                                 return err;
5832
5833                         w->count++;
5834                 }
5835                 bucket++;
5836         }
5837
5838         return 0;
5839 }
5840
5841 /* Return -1 if done with node, number of handled routes on partial dump */
5842 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5843 {
5844         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5845         struct fib_dump_filter *filter = &arg->filter;
5846         unsigned int flags = NLM_F_MULTI;
5847         struct net *net = arg->net;
5848         int count = 0;
5849
5850         if (rt == net->ipv6.fib6_null_entry)
5851                 return -1;
5852
5853         if ((filter->flags & RTM_F_PREFIX) &&
5854             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5855                 /* success since this is not a prefix route */
5856                 return -1;
5857         }
5858         if (filter->filter_set &&
5859             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5860              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5861              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5862                 return -1;
5863         }
5864
5865         if (filter->filter_set ||
5866             !filter->dump_routes || !filter->dump_exceptions) {
5867                 flags |= NLM_F_DUMP_FILTERED;
5868         }
5869
5870         if (filter->dump_routes) {
5871                 if (skip) {
5872                         skip--;
5873                 } else {
5874                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5875                                           0, RTM_NEWROUTE,
5876                                           NETLINK_CB(arg->cb->skb).portid,
5877                                           arg->cb->nlh->nlmsg_seq, flags)) {
5878                                 return 0;
5879                         }
5880                         count++;
5881                 }
5882         }
5883
5884         if (filter->dump_exceptions) {
5885                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5886                                                            .rt = rt,
5887                                                            .flags = flags,
5888                                                            .skip = skip,
5889                                                            .count = 0 };
5890                 int err;
5891
5892                 rcu_read_lock();
5893                 if (rt->nh) {
5894                         err = nexthop_for_each_fib6_nh(rt->nh,
5895                                                        rt6_nh_dump_exceptions,
5896                                                        &w);
5897                 } else {
5898                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5899                 }
5900                 rcu_read_unlock();
5901
5902                 if (err)
5903                         return count += w.count;
5904         }
5905
5906         return -1;
5907 }
5908
5909 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5910                                         const struct nlmsghdr *nlh,
5911                                         struct nlattr **tb,
5912                                         struct netlink_ext_ack *extack)
5913 {
5914         struct rtmsg *rtm;
5915         int i, err;
5916
5917         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5918                 NL_SET_ERR_MSG_MOD(extack,
5919                                    "Invalid header for get route request");
5920                 return -EINVAL;
5921         }
5922
5923         if (!netlink_strict_get_check(skb))
5924                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5925                                               rtm_ipv6_policy, extack);
5926
5927         rtm = nlmsg_data(nlh);
5928         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5929             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5930             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5931             rtm->rtm_type) {
5932                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5933                 return -EINVAL;
5934         }
5935         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5936                 NL_SET_ERR_MSG_MOD(extack,
5937                                    "Invalid flags for get route request");
5938                 return -EINVAL;
5939         }
5940
5941         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5942                                             rtm_ipv6_policy, extack);
5943         if (err)
5944                 return err;
5945
5946         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5947             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5948                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5949                 return -EINVAL;
5950         }
5951
5952         for (i = 0; i <= RTA_MAX; i++) {
5953                 if (!tb[i])
5954                         continue;
5955
5956                 switch (i) {
5957                 case RTA_SRC:
5958                 case RTA_DST:
5959                 case RTA_IIF:
5960                 case RTA_OIF:
5961                 case RTA_MARK:
5962                 case RTA_UID:
5963                 case RTA_SPORT:
5964                 case RTA_DPORT:
5965                 case RTA_IP_PROTO:
5966                         break;
5967                 default:
5968                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5969                         return -EINVAL;
5970                 }
5971         }
5972
5973         return 0;
5974 }
5975
5976 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5977                               struct netlink_ext_ack *extack)
5978 {
5979         struct net *net = sock_net(in_skb->sk);
5980         struct nlattr *tb[RTA_MAX+1];
5981         int err, iif = 0, oif = 0;
5982         struct fib6_info *from;
5983         struct dst_entry *dst;
5984         struct rt6_info *rt;
5985         struct sk_buff *skb;
5986         struct rtmsg *rtm;
5987         struct flowi6 fl6 = {};
5988         bool fibmatch;
5989
5990         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5991         if (err < 0)
5992                 goto errout;
5993
5994         err = -EINVAL;
5995         rtm = nlmsg_data(nlh);
5996         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5997         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5998
5999         if (tb[RTA_SRC]) {
6000                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
6001                         goto errout;
6002
6003                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
6004         }
6005
6006         if (tb[RTA_DST]) {
6007                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
6008                         goto errout;
6009
6010                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
6011         }
6012
6013         if (tb[RTA_IIF])
6014                 iif = nla_get_u32(tb[RTA_IIF]);
6015
6016         if (tb[RTA_OIF])
6017                 oif = nla_get_u32(tb[RTA_OIF]);
6018
6019         if (tb[RTA_MARK])
6020                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
6021
6022         if (tb[RTA_UID])
6023                 fl6.flowi6_uid = make_kuid(current_user_ns(),
6024                                            nla_get_u32(tb[RTA_UID]));
6025         else
6026                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
6027
6028         if (tb[RTA_SPORT])
6029                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
6030
6031         if (tb[RTA_DPORT])
6032                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
6033
6034         if (tb[RTA_IP_PROTO]) {
6035                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
6036                                                   &fl6.flowi6_proto, AF_INET6,
6037                                                   extack);
6038                 if (err)
6039                         goto errout;
6040         }
6041
6042         if (iif) {
6043                 struct net_device *dev;
6044                 int flags = 0;
6045
6046                 rcu_read_lock();
6047
6048                 dev = dev_get_by_index_rcu(net, iif);
6049                 if (!dev) {
6050                         rcu_read_unlock();
6051                         err = -ENODEV;
6052                         goto errout;
6053                 }
6054
6055                 fl6.flowi6_iif = iif;
6056
6057                 if (!ipv6_addr_any(&fl6.saddr))
6058                         flags |= RT6_LOOKUP_F_HAS_SADDR;
6059
6060                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6061
6062                 rcu_read_unlock();
6063         } else {
6064                 fl6.flowi6_oif = oif;
6065
6066                 dst = ip6_route_output(net, NULL, &fl6);
6067         }
6068
6069
6070         rt = container_of(dst, struct rt6_info, dst);
6071         if (rt->dst.error) {
6072                 err = rt->dst.error;
6073                 ip6_rt_put(rt);
6074                 goto errout;
6075         }
6076
6077         if (rt == net->ipv6.ip6_null_entry) {
6078                 err = rt->dst.error;
6079                 ip6_rt_put(rt);
6080                 goto errout;
6081         }
6082
6083         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6084         if (!skb) {
6085                 ip6_rt_put(rt);
6086                 err = -ENOBUFS;
6087                 goto errout;
6088         }
6089
6090         skb_dst_set(skb, &rt->dst);
6091
6092         rcu_read_lock();
6093         from = rcu_dereference(rt->from);
6094         if (from) {
6095                 if (fibmatch)
6096                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6097                                             iif, RTM_NEWROUTE,
6098                                             NETLINK_CB(in_skb).portid,
6099                                             nlh->nlmsg_seq, 0);
6100                 else
6101                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6102                                             &fl6.saddr, iif, RTM_NEWROUTE,
6103                                             NETLINK_CB(in_skb).portid,
6104                                             nlh->nlmsg_seq, 0);
6105         } else {
6106                 err = -ENETUNREACH;
6107         }
6108         rcu_read_unlock();
6109
6110         if (err < 0) {
6111                 kfree_skb(skb);
6112                 goto errout;
6113         }
6114
6115         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6116 errout:
6117         return err;
6118 }
6119
6120 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6121                      unsigned int nlm_flags)
6122 {
6123         struct sk_buff *skb;
6124         struct net *net = info->nl_net;
6125         u32 seq;
6126         int err;
6127
6128         err = -ENOBUFS;
6129         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6130
6131         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6132         if (!skb)
6133                 goto errout;
6134
6135         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6136                             event, info->portid, seq, nlm_flags);
6137         if (err < 0) {
6138                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6139                 WARN_ON(err == -EMSGSIZE);
6140                 kfree_skb(skb);
6141                 goto errout;
6142         }
6143         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6144                     info->nlh, gfp_any());
6145         return;
6146 errout:
6147         if (err < 0)
6148                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6149 }
6150
6151 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6152                     struct nl_info *info)
6153 {
6154         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6155         struct sk_buff *skb;
6156         int err = -ENOBUFS;
6157
6158         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6159         if (!skb)
6160                 goto errout;
6161
6162         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6163                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6164         if (err < 0) {
6165                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6166                 WARN_ON(err == -EMSGSIZE);
6167                 kfree_skb(skb);
6168                 goto errout;
6169         }
6170         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6171                     info->nlh, gfp_any());
6172         return;
6173 errout:
6174         if (err < 0)
6175                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6176 }
6177
6178 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6179                             bool offload, bool trap, bool offload_failed)
6180 {
6181         struct sk_buff *skb;
6182         int err;
6183
6184         if (f6i->offload == offload && f6i->trap == trap &&
6185             f6i->offload_failed == offload_failed)
6186                 return;
6187
6188         f6i->offload = offload;
6189         f6i->trap = trap;
6190
6191         /* 2 means send notifications only if offload_failed was changed. */
6192         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6193             f6i->offload_failed == offload_failed)
6194                 return;
6195
6196         f6i->offload_failed = offload_failed;
6197
6198         if (!rcu_access_pointer(f6i->fib6_node))
6199                 /* The route was removed from the tree, do not send
6200                  * notification.
6201                  */
6202                 return;
6203
6204         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6205                 return;
6206
6207         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6208         if (!skb) {
6209                 err = -ENOBUFS;
6210                 goto errout;
6211         }
6212
6213         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6214                             0, 0);
6215         if (err < 0) {
6216                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6217                 WARN_ON(err == -EMSGSIZE);
6218                 kfree_skb(skb);
6219                 goto errout;
6220         }
6221
6222         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6223         return;
6224
6225 errout:
6226         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6227 }
6228 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6229
6230 static int ip6_route_dev_notify(struct notifier_block *this,
6231                                 unsigned long event, void *ptr)
6232 {
6233         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6234         struct net *net = dev_net(dev);
6235
6236         if (!(dev->flags & IFF_LOOPBACK))
6237                 return NOTIFY_OK;
6238
6239         if (event == NETDEV_REGISTER) {
6240                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6241                 net->ipv6.ip6_null_entry->dst.dev = dev;
6242                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6243 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6244                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6245                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6246                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6247                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6248 #endif
6249          } else if (event == NETDEV_UNREGISTER &&
6250                     dev->reg_state != NETREG_UNREGISTERED) {
6251                 /* NETDEV_UNREGISTER could be fired for multiple times by
6252                  * netdev_wait_allrefs(). Make sure we only call this once.
6253                  */
6254                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6255 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6256                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6257                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6258 #endif
6259         }
6260
6261         return NOTIFY_OK;
6262 }
6263
6264 /*
6265  *      /proc
6266  */
6267
6268 #ifdef CONFIG_PROC_FS
6269 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6270 {
6271         struct net *net = (struct net *)seq->private;
6272         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6273                    net->ipv6.rt6_stats->fib_nodes,
6274                    net->ipv6.rt6_stats->fib_route_nodes,
6275                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6276                    net->ipv6.rt6_stats->fib_rt_entries,
6277                    net->ipv6.rt6_stats->fib_rt_cache,
6278                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6279                    net->ipv6.rt6_stats->fib_discarded_routes);
6280
6281         return 0;
6282 }
6283 #endif  /* CONFIG_PROC_FS */
6284
6285 #ifdef CONFIG_SYSCTL
6286
6287 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6288                               void *buffer, size_t *lenp, loff_t *ppos)
6289 {
6290         struct net *net;
6291         int delay;
6292         int ret;
6293         if (!write)
6294                 return -EINVAL;
6295
6296         net = (struct net *)ctl->extra1;
6297         delay = net->ipv6.sysctl.flush_delay;
6298         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6299         if (ret)
6300                 return ret;
6301
6302         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6303         return 0;
6304 }
6305
6306 static struct ctl_table ipv6_route_table_template[] = {
6307         {
6308                 .procname       =       "flush",
6309                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6310                 .maxlen         =       sizeof(int),
6311                 .mode           =       0200,
6312                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6313         },
6314         {
6315                 .procname       =       "gc_thresh",
6316                 .data           =       &ip6_dst_ops_template.gc_thresh,
6317                 .maxlen         =       sizeof(int),
6318                 .mode           =       0644,
6319                 .proc_handler   =       proc_dointvec,
6320         },
6321         {
6322                 .procname       =       "max_size",
6323                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6324                 .maxlen         =       sizeof(int),
6325                 .mode           =       0644,
6326                 .proc_handler   =       proc_dointvec,
6327         },
6328         {
6329                 .procname       =       "gc_min_interval",
6330                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6331                 .maxlen         =       sizeof(int),
6332                 .mode           =       0644,
6333                 .proc_handler   =       proc_dointvec_jiffies,
6334         },
6335         {
6336                 .procname       =       "gc_timeout",
6337                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6338                 .maxlen         =       sizeof(int),
6339                 .mode           =       0644,
6340                 .proc_handler   =       proc_dointvec_jiffies,
6341         },
6342         {
6343                 .procname       =       "gc_interval",
6344                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6345                 .maxlen         =       sizeof(int),
6346                 .mode           =       0644,
6347                 .proc_handler   =       proc_dointvec_jiffies,
6348         },
6349         {
6350                 .procname       =       "gc_elasticity",
6351                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6352                 .maxlen         =       sizeof(int),
6353                 .mode           =       0644,
6354                 .proc_handler   =       proc_dointvec,
6355         },
6356         {
6357                 .procname       =       "mtu_expires",
6358                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6359                 .maxlen         =       sizeof(int),
6360                 .mode           =       0644,
6361                 .proc_handler   =       proc_dointvec_jiffies,
6362         },
6363         {
6364                 .procname       =       "min_adv_mss",
6365                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6366                 .maxlen         =       sizeof(int),
6367                 .mode           =       0644,
6368                 .proc_handler   =       proc_dointvec,
6369         },
6370         {
6371                 .procname       =       "gc_min_interval_ms",
6372                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6373                 .maxlen         =       sizeof(int),
6374                 .mode           =       0644,
6375                 .proc_handler   =       proc_dointvec_ms_jiffies,
6376         },
6377         {
6378                 .procname       =       "skip_notify_on_dev_down",
6379                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6380                 .maxlen         =       sizeof(int),
6381                 .mode           =       0644,
6382                 .proc_handler   =       proc_dointvec_minmax,
6383                 .extra1         =       SYSCTL_ZERO,
6384                 .extra2         =       SYSCTL_ONE,
6385         },
6386         { }
6387 };
6388
6389 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6390 {
6391         struct ctl_table *table;
6392
6393         table = kmemdup(ipv6_route_table_template,
6394                         sizeof(ipv6_route_table_template),
6395                         GFP_KERNEL);
6396
6397         if (table) {
6398                 table[0].data = &net->ipv6.sysctl.flush_delay;
6399                 table[0].extra1 = net;
6400                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6401                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6402                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6403                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6404                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6405                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6406                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6407                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6408                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6409                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6410
6411                 /* Don't export sysctls to unprivileged users */
6412                 if (net->user_ns != &init_user_ns)
6413                         table[0].procname = NULL;
6414         }
6415
6416         return table;
6417 }
6418 #endif
6419
6420 static int __net_init ip6_route_net_init(struct net *net)
6421 {
6422         int ret = -ENOMEM;
6423
6424         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6425                sizeof(net->ipv6.ip6_dst_ops));
6426
6427         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6428                 goto out_ip6_dst_ops;
6429
6430         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6431         if (!net->ipv6.fib6_null_entry)
6432                 goto out_ip6_dst_entries;
6433         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6434                sizeof(*net->ipv6.fib6_null_entry));
6435
6436         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6437                                            sizeof(*net->ipv6.ip6_null_entry),
6438                                            GFP_KERNEL);
6439         if (!net->ipv6.ip6_null_entry)
6440                 goto out_fib6_null_entry;
6441         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6442         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6443                          ip6_template_metrics, true);
6444         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6445
6446 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6447         net->ipv6.fib6_has_custom_rules = false;
6448         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6449                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6450                                                GFP_KERNEL);
6451         if (!net->ipv6.ip6_prohibit_entry)
6452                 goto out_ip6_null_entry;
6453         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6454         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6455                          ip6_template_metrics, true);
6456         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6457
6458         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6459                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6460                                                GFP_KERNEL);
6461         if (!net->ipv6.ip6_blk_hole_entry)
6462                 goto out_ip6_prohibit_entry;
6463         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6464         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6465                          ip6_template_metrics, true);
6466         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6467 #ifdef CONFIG_IPV6_SUBTREES
6468         net->ipv6.fib6_routes_require_src = 0;
6469 #endif
6470 #endif
6471
6472         net->ipv6.sysctl.flush_delay = 0;
6473         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6474         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6475         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6476         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6477         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6478         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6479         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6480         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6481
6482         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6483
6484         ret = 0;
6485 out:
6486         return ret;
6487
6488 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6489 out_ip6_prohibit_entry:
6490         kfree(net->ipv6.ip6_prohibit_entry);
6491 out_ip6_null_entry:
6492         kfree(net->ipv6.ip6_null_entry);
6493 #endif
6494 out_fib6_null_entry:
6495         kfree(net->ipv6.fib6_null_entry);
6496 out_ip6_dst_entries:
6497         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6498 out_ip6_dst_ops:
6499         goto out;
6500 }
6501
6502 static void __net_exit ip6_route_net_exit(struct net *net)
6503 {
6504         kfree(net->ipv6.fib6_null_entry);
6505         kfree(net->ipv6.ip6_null_entry);
6506 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6507         kfree(net->ipv6.ip6_prohibit_entry);
6508         kfree(net->ipv6.ip6_blk_hole_entry);
6509 #endif
6510         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6511 }
6512
6513 static int __net_init ip6_route_net_init_late(struct net *net)
6514 {
6515 #ifdef CONFIG_PROC_FS
6516         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6517                         sizeof(struct ipv6_route_iter));
6518         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6519                         rt6_stats_seq_show, NULL);
6520 #endif
6521         return 0;
6522 }
6523
6524 static void __net_exit ip6_route_net_exit_late(struct net *net)
6525 {
6526 #ifdef CONFIG_PROC_FS
6527         remove_proc_entry("ipv6_route", net->proc_net);
6528         remove_proc_entry("rt6_stats", net->proc_net);
6529 #endif
6530 }
6531
6532 static struct pernet_operations ip6_route_net_ops = {
6533         .init = ip6_route_net_init,
6534         .exit = ip6_route_net_exit,
6535 };
6536
6537 static int __net_init ipv6_inetpeer_init(struct net *net)
6538 {
6539         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6540
6541         if (!bp)
6542                 return -ENOMEM;
6543         inet_peer_base_init(bp);
6544         net->ipv6.peers = bp;
6545         return 0;
6546 }
6547
6548 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6549 {
6550         struct inet_peer_base *bp = net->ipv6.peers;
6551
6552         net->ipv6.peers = NULL;
6553         inetpeer_invalidate_tree(bp);
6554         kfree(bp);
6555 }
6556
6557 static struct pernet_operations ipv6_inetpeer_ops = {
6558         .init   =       ipv6_inetpeer_init,
6559         .exit   =       ipv6_inetpeer_exit,
6560 };
6561
6562 static struct pernet_operations ip6_route_net_late_ops = {
6563         .init = ip6_route_net_init_late,
6564         .exit = ip6_route_net_exit_late,
6565 };
6566
6567 static struct notifier_block ip6_route_dev_notifier = {
6568         .notifier_call = ip6_route_dev_notify,
6569         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6570 };
6571
6572 void __init ip6_route_init_special_entries(void)
6573 {
6574         /* Registering of the loopback is done before this portion of code,
6575          * the loopback reference in rt6_info will not be taken, do it
6576          * manually for init_net */
6577         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6578         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6579         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6580   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6581         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6582         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6583         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6584         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6585   #endif
6586 }
6587
6588 #if IS_BUILTIN(CONFIG_IPV6)
6589 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6590 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6591
6592 BTF_ID_LIST(btf_fib6_info_id)
6593 BTF_ID(struct, fib6_info)
6594
6595 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6596         .seq_ops                = &ipv6_route_seq_ops,
6597         .init_seq_private       = bpf_iter_init_seq_net,
6598         .fini_seq_private       = bpf_iter_fini_seq_net,
6599         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6600 };
6601
6602 static struct bpf_iter_reg ipv6_route_reg_info = {
6603         .target                 = "ipv6_route",
6604         .ctx_arg_info_size      = 1,
6605         .ctx_arg_info           = {
6606                 { offsetof(struct bpf_iter__ipv6_route, rt),
6607                   PTR_TO_BTF_ID_OR_NULL },
6608         },
6609         .seq_info               = &ipv6_route_seq_info,
6610 };
6611
6612 static int __init bpf_iter_register(void)
6613 {
6614         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6615         return bpf_iter_reg_target(&ipv6_route_reg_info);
6616 }
6617
6618 static void bpf_iter_unregister(void)
6619 {
6620         bpf_iter_unreg_target(&ipv6_route_reg_info);
6621 }
6622 #endif
6623 #endif
6624
6625 int __init ip6_route_init(void)
6626 {
6627         int ret;
6628         int cpu;
6629
6630         ret = -ENOMEM;
6631         ip6_dst_ops_template.kmem_cachep =
6632                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6633                                   SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT, NULL);
6634         if (!ip6_dst_ops_template.kmem_cachep)
6635                 goto out;
6636
6637         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6638         if (ret)
6639                 goto out_kmem_cache;
6640
6641         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6642         if (ret)
6643                 goto out_dst_entries;
6644
6645         ret = register_pernet_subsys(&ip6_route_net_ops);
6646         if (ret)
6647                 goto out_register_inetpeer;
6648
6649         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6650
6651         ret = fib6_init();
6652         if (ret)
6653                 goto out_register_subsys;
6654
6655         ret = xfrm6_init();
6656         if (ret)
6657                 goto out_fib6_init;
6658
6659         ret = fib6_rules_init();
6660         if (ret)
6661                 goto xfrm6_init;
6662
6663         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6664         if (ret)
6665                 goto fib6_rules_init;
6666
6667         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6668                                    inet6_rtm_newroute, NULL, 0);
6669         if (ret < 0)
6670                 goto out_register_late_subsys;
6671
6672         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6673                                    inet6_rtm_delroute, NULL, 0);
6674         if (ret < 0)
6675                 goto out_register_late_subsys;
6676
6677         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6678                                    inet6_rtm_getroute, NULL,
6679                                    RTNL_FLAG_DOIT_UNLOCKED);
6680         if (ret < 0)
6681                 goto out_register_late_subsys;
6682
6683         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6684         if (ret)
6685                 goto out_register_late_subsys;
6686
6687 #if IS_BUILTIN(CONFIG_IPV6)
6688 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6689         ret = bpf_iter_register();
6690         if (ret)
6691                 goto out_register_late_subsys;
6692 #endif
6693 #endif
6694
6695         for_each_possible_cpu(cpu) {
6696                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6697
6698                 INIT_LIST_HEAD(&ul->head);
6699                 spin_lock_init(&ul->lock);
6700         }
6701
6702 out:
6703         return ret;
6704
6705 out_register_late_subsys:
6706         rtnl_unregister_all(PF_INET6);
6707         unregister_pernet_subsys(&ip6_route_net_late_ops);
6708 fib6_rules_init:
6709         fib6_rules_cleanup();
6710 xfrm6_init:
6711         xfrm6_fini();
6712 out_fib6_init:
6713         fib6_gc_cleanup();
6714 out_register_subsys:
6715         unregister_pernet_subsys(&ip6_route_net_ops);
6716 out_register_inetpeer:
6717         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6718 out_dst_entries:
6719         dst_entries_destroy(&ip6_dst_blackhole_ops);
6720 out_kmem_cache:
6721         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6722         goto out;
6723 }
6724
6725 void ip6_route_cleanup(void)
6726 {
6727 #if IS_BUILTIN(CONFIG_IPV6)
6728 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6729         bpf_iter_unregister();
6730 #endif
6731 #endif
6732         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6733         unregister_pernet_subsys(&ip6_route_net_late_ops);
6734         fib6_rules_cleanup();
6735         xfrm6_fini();
6736         fib6_gc_cleanup();
6737         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6738         unregister_pernet_subsys(&ip6_route_net_ops);
6739         dst_entries_destroy(&ip6_dst_blackhole_ops);
6740         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6741 }