net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <[email protected]>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <net/net_namespace.h>
  45 #include <net/snmp.h>
  46 #include <net/ipv6.h>
  47 #include <net/ip6_fib.h>
  48 #include <net/ip6_route.h>
  49 #include <net/ndisc.h>
  50 #include <net/addrconf.h>
  51 #include <net/tcp.h>
  52 #include <linux/rtnetlink.h>
  53 #include <net/dst.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/xfrm.h>
  56 #include <net/netevent.h>
  57 #include <net/netlink.h>
  58 #include <net/rtnh.h>
  59 #include <net/lwtunnel.h>
  60 #include <net/ip_tunnels.h>
  61 #include <net/l3mdev.h>
  62 #include <net/ip.h>
  63 #include <linux/uaccess.h>
  64 #include <linux/btf_ids.h>
  65
  66 #ifdef CONFIG_SYSCTL
  67 #include <linux/sysctl.h>
  68 #endif
  69
  70 static int ip6_rt_type_to_error(u8 fib6_type);
  71
  72 #define CREATE_TRACE_POINTS
  73 #include <trace/events/fib6.h>
  74 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  75 #undef CREATE_TRACE_POINTS
  76
  77 enum rt6_nud_state {
  78         RT6_NUD_FAIL_HARD = -3,
  79         RT6_NUD_FAIL_PROBE = -2,
  80         RT6_NUD_FAIL_DO_RR = -1,
  81         RT6_NUD_SUCCEED = 1
  82 };
  83
  84 INDIRECT_CALLABLE_SCOPE
  85 struct dst_entry        *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  86 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  87 INDIRECT_CALLABLE_SCOPE
  88 unsigned int            ip6_mtu(const struct dst_entry *dst);
  89 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  90 static void             ip6_dst_destroy(struct dst_entry *);
  91 static void             ip6_dst_ifdown(struct dst_entry *,
  92                                        struct net_device *dev, int how);
  93 static int               ip6_dst_gc(struct dst_ops *ops);
  94
  95 static int              ip6_pkt_discard(struct sk_buff *skb);
  96 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  97 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  99 static void             ip6_link_failure(struct sk_buff *skb);
 100 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 101                                            struct sk_buff *skb, u32 mtu,
 102                                            bool confirm_neigh);
 103 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 104                                         struct sk_buff *skb);
 105 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 106                            int strict);
 107 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 108 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 109                          struct fib6_info *rt, struct dst_entry *dst,
 110                          struct in6_addr *dest, struct in6_addr *src,
 111                          int iif, int type, u32 portid, u32 seq,
 112                          unsigned int flags);
 113 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 114                                            const struct in6_addr *daddr,
 115                                            const struct in6_addr *saddr);
 116
 117 #ifdef CONFIG_IPV6_ROUTE_INFO
 118 static struct fib6_info *rt6_add_route_info(struct net *net,
 119                                            const struct in6_addr *prefix, int prefixlen,
 120                                            const struct in6_addr *gwaddr,
 121                                            struct net_device *dev,
 122                                            unsigned int pref);
 123 static struct fib6_info *rt6_get_route_info(struct net *net,
 124                                            const struct in6_addr *prefix, int prefixlen,
 125                                            const struct in6_addr *gwaddr,
 126                                            struct net_device *dev);
 127 #endif
 128
 129 struct uncached_list {
 130         spinlock_t              lock;
 131         struct list_head        head;
 132 };
 133
 134 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 135
 136 void rt6_uncached_list_add(struct rt6_info *rt)
 137 {
 138         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 139
 140         rt->rt6i_uncached_list = ul;
 141
 142         spin_lock_bh(&ul->lock);
 143         list_add_tail(&rt->rt6i_uncached, &ul->head);
 144         spin_unlock_bh(&ul->lock);
 145 }
 146
 147 void rt6_uncached_list_del(struct rt6_info *rt)
 148 {
 149         if (!list_empty(&rt->rt6i_uncached)) {
 150                 struct uncached_list *ul = rt->rt6i_uncached_list;
 151                 struct net *net = dev_net(rt->dst.dev);
 152
 153                 spin_lock_bh(&ul->lock);
 154                 list_del(&rt->rt6i_uncached);
 155                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 156                 spin_unlock_bh(&ul->lock);
 157         }
 158 }
 159
 160 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 161 {
 162         struct net_device *loopback_dev = net->loopback_dev;
 163         int cpu;
 164
 165         if (dev == loopback_dev)
 166                 return;
 167
 168         for_each_possible_cpu(cpu) {
 169                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 170                 struct rt6_info *rt;
 171
 172                 spin_lock_bh(&ul->lock);
 173                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 174                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 175                         struct net_device *rt_dev = rt->dst.dev;
 176
 177                         if (rt_idev->dev == dev) {
 178                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 179                                 in6_dev_put(rt_idev);
 180                         }
 181
 182                         if (rt_dev == dev) {
 183                                 rt->dst.dev = blackhole_netdev;
 184                                 dev_hold(rt->dst.dev);
 185                                 dev_put(rt_dev);
 186                         }
 187                 }
 188                 spin_unlock_bh(&ul->lock);
 189         }
 190 }
 191
 192 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 193                                              struct sk_buff *skb,
 194                                              const void *daddr)
 195 {
 196         if (!ipv6_addr_any(p))
 197                 return (const void *) p;
 198         else if (skb)
 199                 return &ipv6_hdr(skb)->daddr;
 200         return daddr;
 201 }
 202
 203 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 204                                    struct net_device *dev,
 205                                    struct sk_buff *skb,
 206                                    const void *daddr)
 207 {
 208         struct neighbour *n;
 209
 210         daddr = choose_neigh_daddr(gw, skb, daddr);
 211         n = __ipv6_neigh_lookup(dev, daddr);
 212         if (n)
 213                 return n;
 214
 215         n = neigh_create(&nd_tbl, daddr, dev);
 216         return IS_ERR(n) ? NULL : n;
 217 }
 218
 219 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 220                                               struct sk_buff *skb,
 221                                               const void *daddr)
 222 {
 223         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 224
 225         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 226                                 dst->dev, skb, daddr);
 227 }
 228
 229 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 230 {
 231         struct net_device *dev = dst->dev;
 232         struct rt6_info *rt = (struct rt6_info *)dst;
 233
 234         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 235         if (!daddr)
 236                 return;
 237         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 238                 return;
 239         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 240                 return;
 241         __ipv6_confirm_neigh(dev, daddr);
 242 }
 243
 244 static struct dst_ops ip6_dst_ops_template = {
 245         .family                 =       AF_INET6,
 246         .gc                     =       ip6_dst_gc,
 247         .gc_thresh              =       1024,
 248         .check                  =       ip6_dst_check,
 249         .default_advmss         =       ip6_default_advmss,
 250         .mtu                    =       ip6_mtu,
 251         .cow_metrics            =       dst_cow_metrics_generic,
 252         .destroy                =       ip6_dst_destroy,
 253         .ifdown                 =       ip6_dst_ifdown,
 254         .negative_advice        =       ip6_negative_advice,
 255         .link_failure           =       ip6_link_failure,
 256         .update_pmtu            =       ip6_rt_update_pmtu,
 257         .redirect               =       rt6_do_redirect,
 258         .local_out              =       __ip6_local_out,
 259         .neigh_lookup           =       ip6_dst_neigh_lookup,
 260         .confirm_neigh          =       ip6_confirm_neigh,
 261 };
 262
 263 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 264 {
 265         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 266
 267         return mtu ? : dst->dev->mtu;
 268 }
 269
 270 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 271                                          struct sk_buff *skb, u32 mtu,
 272                                          bool confirm_neigh)
 273 {
 274 }
 275
 276 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 277                                       struct sk_buff *skb)
 278 {
 279 }
 280
 281 static struct dst_ops ip6_dst_blackhole_ops = {
 282         .family                 =       AF_INET6,
 283         .destroy                =       ip6_dst_destroy,
 284         .check                  =       ip6_dst_check,
 285         .mtu                    =       ip6_blackhole_mtu,
 286         .default_advmss         =       ip6_default_advmss,
 287         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 288         .redirect               =       ip6_rt_blackhole_redirect,
 289         .cow_metrics            =       dst_cow_metrics_generic,
 290         .neigh_lookup           =       ip6_dst_neigh_lookup,
 291 };
 292
 293 static const u32 ip6_template_metrics[RTAX_MAX] = {
 294         [RTAX_HOPLIMIT - 1] = 0,
 295 };
 296
 297 static const struct fib6_info fib6_null_entry_template = {
 298         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 299         .fib6_protocol  = RTPROT_KERNEL,
 300         .fib6_metric    = ~(u32)0,
 301         .fib6_ref       = REFCOUNT_INIT(1),
 302         .fib6_type      = RTN_UNREACHABLE,
 303         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 304 };
 305
 306 static const struct rt6_info ip6_null_entry_template = {
 307         .dst = {
 308                 .__refcnt       = ATOMIC_INIT(1),
 309                 .__use          = 1,
 310                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 311                 .error          = -ENETUNREACH,
 312                 .input          = ip6_pkt_discard,
 313                 .output         = ip6_pkt_discard_out,
 314         },
 315         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 316 };
 317
 318 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 319
 320 static const struct rt6_info ip6_prohibit_entry_template = {
 321         .dst = {
 322                 .__refcnt       = ATOMIC_INIT(1),
 323                 .__use          = 1,
 324                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 325                 .error          = -EACCES,
 326                 .input          = ip6_pkt_prohibit,
 327                 .output         = ip6_pkt_prohibit_out,
 328         },
 329         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 330 };
 331
 332 static const struct rt6_info ip6_blk_hole_entry_template = {
 333         .dst = {
 334                 .__refcnt       = ATOMIC_INIT(1),
 335                 .__use          = 1,
 336                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 337                 .error          = -EINVAL,
 338                 .input          = dst_discard,
 339                 .output         = dst_discard_out,
 340         },
 341         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 342 };
 343
 344 #endif
 345
 346 static void rt6_info_init(struct rt6_info *rt)
 347 {
 348         struct dst_entry *dst = &rt->dst;
 349
 350         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 351         INIT_LIST_HEAD(&rt->rt6i_uncached);
 352 }
 353
 354 /* allocate dst with ip6_dst_ops */
 355 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 356                                int flags)
 357 {
 358         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 359                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 360
 361         if (rt) {
 362                 rt6_info_init(rt);
 363                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 364         }
 365
 366         return rt;
 367 }
 368 EXPORT_SYMBOL(ip6_dst_alloc);
 369
 370 static void ip6_dst_destroy(struct dst_entry *dst)
 371 {
 372         struct rt6_info *rt = (struct rt6_info *)dst;
 373         struct fib6_info *from;
 374         struct inet6_dev *idev;
 375
 376         ip_dst_metrics_put(dst);
 377         rt6_uncached_list_del(rt);
 378
 379         idev = rt->rt6i_idev;
 380         if (idev) {
 381                 rt->rt6i_idev = NULL;
 382                 in6_dev_put(idev);
 383         }
 384
 385         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 386         fib6_info_release(from);
 387 }
 388
 389 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 390                            int how)
 391 {
 392         struct rt6_info *rt = (struct rt6_info *)dst;
 393         struct inet6_dev *idev = rt->rt6i_idev;
 394         struct net_device *loopback_dev =
 395                 dev_net(dev)->loopback_dev;
 396
 397         if (idev && idev->dev != loopback_dev) {
 398                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 399                 if (loopback_idev) {
 400                         rt->rt6i_idev = loopback_idev;
 401                         in6_dev_put(idev);
 402                 }
 403         }
 404 }
 405
 406 static bool __rt6_check_expired(const struct rt6_info *rt)
 407 {
 408         if (rt->rt6i_flags & RTF_EXPIRES)
 409                 return time_after(jiffies, rt->dst.expires);
 410         else
 411                 return false;
 412 }
 413
 414 static bool rt6_check_expired(const struct rt6_info *rt)
 415 {
 416         struct fib6_info *from;
 417
 418         from = rcu_dereference(rt->from);
 419
 420         if (rt->rt6i_flags & RTF_EXPIRES) {
 421                 if (time_after(jiffies, rt->dst.expires))
 422                         return true;
 423         } else if (from) {
 424                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 425                         fib6_check_expired(from);
 426         }
 427         return false;
 428 }
 429
 430 void fib6_select_path(const struct net *net, struct fib6_result *res,
 431                       struct flowi6 *fl6, int oif, bool have_oif_match,
 432                       const struct sk_buff *skb, int strict)
 433 {
 434         struct fib6_info *sibling, *next_sibling;
 435         struct fib6_info *match = res->f6i;
 436
 437         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 438                 goto out;
 439
 440         if (match->nh && have_oif_match && res->nh)
 441                 return;
 442
 443         /* We might have already computed the hash for ICMPv6 errors. In such
 444          * case it will always be non-zero. Otherwise now is the time to do it.
 445          */
 446         if (!fl6->mp_hash &&
 447             (!match->nh || nexthop_is_multipath(match->nh)))
 448                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 449
 450         if (unlikely(match->nh)) {
 451                 nexthop_path_fib6_result(res, fl6->mp_hash);
 452                 return;
 453         }
 454
 455         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 456                 goto out;
 457
 458         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 459                                  fib6_siblings) {
 460                 const struct fib6_nh *nh = sibling->fib6_nh;
 461                 int nh_upper_bound;
 462
 463                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 464                 if (fl6->mp_hash > nh_upper_bound)
 465                         continue;
 466                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 467                         break;
 468                 match = sibling;
 469                 break;
 470         }
 471
 472 out:
 473         res->f6i = match;
 474         res->nh = match->fib6_nh;
 475 }
 476
 477 /*
 478  *      Route lookup. rcu_read_lock() should be held.
 479  */
 480
 481 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 482                                const struct in6_addr *saddr, int oif, int flags)
 483 {
 484         const struct net_device *dev;
 485
 486         if (nh->fib_nh_flags & RTNH_F_DEAD)
 487                 return false;
 488
 489         dev = nh->fib_nh_dev;
 490         if (oif) {
 491                 if (dev->ifindex == oif)
 492                         return true;
 493         } else {
 494                 if (ipv6_chk_addr(net, saddr, dev,
 495                                   flags & RT6_LOOKUP_F_IFACE))
 496                         return true;
 497         }
 498
 499         return false;
 500 }
 501
 502 struct fib6_nh_dm_arg {
 503         struct net              *net;
 504         const struct in6_addr   *saddr;
 505         int                     oif;
 506         int                     flags;
 507         struct fib6_nh          *nh;
 508 };
 509
 510 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 511 {
 512         struct fib6_nh_dm_arg *arg = _arg;
 513
 514         arg->nh = nh;
 515         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 516                                   arg->flags);
 517 }
 518
 519 /* returns fib6_nh from nexthop or NULL */
 520 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 521                                         struct fib6_result *res,
 522                                         const struct in6_addr *saddr,
 523                                         int oif, int flags)
 524 {
 525         struct fib6_nh_dm_arg arg = {
 526                 .net   = net,
 527                 .saddr = saddr,
 528                 .oif   = oif,
 529                 .flags = flags,
 530         };
 531
 532         if (nexthop_is_blackhole(nh))
 533                 return NULL;
 534
 535         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 536                 return arg.nh;
 537
 538         return NULL;
 539 }
 540
 541 static void rt6_device_match(struct net *net, struct fib6_result *res,
 542                              const struct in6_addr *saddr, int oif, int flags)
 543 {
 544         struct fib6_info *f6i = res->f6i;
 545         struct fib6_info *spf6i;
 546         struct fib6_nh *nh;
 547
 548         if (!oif && ipv6_addr_any(saddr)) {
 549                 if (unlikely(f6i->nh)) {
 550                         nh = nexthop_fib6_nh(f6i->nh);
 551                         if (nexthop_is_blackhole(f6i->nh))
 552                                 goto out_blackhole;
 553                 } else {
 554                         nh = f6i->fib6_nh;
 555                 }
 556                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 557                         goto out;
 558         }
 559
 560         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 561                 bool matched = false;
 562
 563                 if (unlikely(spf6i->nh)) {
 564                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 565                                               oif, flags);
 566                         if (nh)
 567                                 matched = true;
 568                 } else {
 569                         nh = spf6i->fib6_nh;
 570                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 571                                 matched = true;
 572                 }
 573                 if (matched) {
 574                         res->f6i = spf6i;
 575                         goto out;
 576                 }
 577         }
 578
 579         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 580                 res->f6i = net->ipv6.fib6_null_entry;
 581                 nh = res->f6i->fib6_nh;
 582                 goto out;
 583         }
 584
 585         if (unlikely(f6i->nh)) {
 586                 nh = nexthop_fib6_nh(f6i->nh);
 587                 if (nexthop_is_blackhole(f6i->nh))
 588                         goto out_blackhole;
 589         } else {
 590                 nh = f6i->fib6_nh;
 591         }
 592
 593         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 594                 res->f6i = net->ipv6.fib6_null_entry;
 595                 nh = res->f6i->fib6_nh;
 596         }
 597 out:
 598         res->nh = nh;
 599         res->fib6_type = res->f6i->fib6_type;
 600         res->fib6_flags = res->f6i->fib6_flags;
 601         return;
 602
 603 out_blackhole:
 604         res->fib6_flags |= RTF_REJECT;
 605         res->fib6_type = RTN_BLACKHOLE;
 606         res->nh = nh;
 607 }
 608
 609 #ifdef CONFIG_IPV6_ROUTER_PREF
 610 struct __rt6_probe_work {
 611         struct work_struct work;
 612         struct in6_addr target;
 613         struct net_device *dev;
 614 };
 615
 616 static void rt6_probe_deferred(struct work_struct *w)
 617 {
 618         struct in6_addr mcaddr;
 619         struct __rt6_probe_work *work =
 620                 container_of(w, struct __rt6_probe_work, work);
 621
 622         addrconf_addr_solict_mult(&work->target, &mcaddr);
 623         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 624         dev_put(work->dev);
 625         kfree(work);
 626 }
 627
 628 static void rt6_probe(struct fib6_nh *fib6_nh)
 629 {
 630         struct __rt6_probe_work *work = NULL;
 631         const struct in6_addr *nh_gw;
 632         unsigned long last_probe;
 633         struct neighbour *neigh;
 634         struct net_device *dev;
 635         struct inet6_dev *idev;
 636
 637         /*
 638          * Okay, this does not seem to be appropriate
 639          * for now, however, we need to check if it
 640          * is really so; aka Router Reachability Probing.
 641          *
 642          * Router Reachability Probe MUST be rate-limited
 643          * to no more than one per minute.
 644          */
 645         if (!fib6_nh->fib_nh_gw_family)
 646                 return;
 647
 648         nh_gw = &fib6_nh->fib_nh_gw6;
 649         dev = fib6_nh->fib_nh_dev;
 650         rcu_read_lock_bh();
 651         last_probe = READ_ONCE(fib6_nh->last_probe);
 652         idev = __in6_dev_get(dev);
 653         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 654         if (neigh) {
 655                 if (neigh->nud_state & NUD_VALID)
 656                         goto out;
 657
 658                 write_lock(&neigh->lock);
 659                 if (!(neigh->nud_state & NUD_VALID) &&
 660                     time_after(jiffies,
 661                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 662                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 663                         if (work)
 664                                 __neigh_set_probe_once(neigh);
 665                 }
 666                 write_unlock(&neigh->lock);
 667         } else if (time_after(jiffies, last_probe +
 668                                        idev->cnf.rtr_probe_interval)) {
 669                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 670         }
 671
 672         if (!work || cmpxchg(&fib6_nh->last_probe,
 673                              last_probe, jiffies) != last_probe) {
 674                 kfree(work);
 675         } else {
 676                 INIT_WORK(&work->work, rt6_probe_deferred);
 677                 work->target = *nh_gw;
 678                 dev_hold(dev);
 679                 work->dev = dev;
 680                 schedule_work(&work->work);
 681         }
 682
 683 out:
 684         rcu_read_unlock_bh();
 685 }
 686 #else
 687 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 688 {
 689 }
 690 #endif
 691
 692 /*
 693  * Default Router Selection (RFC 2461 6.3.6)
 694  */
 695 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 696 {
 697         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 698         struct neighbour *neigh;
 699
 700         rcu_read_lock_bh();
 701         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 702                                           &fib6_nh->fib_nh_gw6);
 703         if (neigh) {
 704                 read_lock(&neigh->lock);
 705                 if (neigh->nud_state & NUD_VALID)
 706                         ret = RT6_NUD_SUCCEED;
 707 #ifdef CONFIG_IPV6_ROUTER_PREF
 708                 else if (!(neigh->nud_state & NUD_FAILED))
 709                         ret = RT6_NUD_SUCCEED;
 710                 else
 711                         ret = RT6_NUD_FAIL_PROBE;
 712 #endif
 713                 read_unlock(&neigh->lock);
 714         } else {
 715                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 716                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 717         }
 718         rcu_read_unlock_bh();
 719
 720         return ret;
 721 }
 722
 723 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 724                            int strict)
 725 {
 726         int m = 0;
 727
 728         if (!oif || nh->fib_nh_dev->ifindex == oif)
 729                 m = 2;
 730
 731         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 732                 return RT6_NUD_FAIL_HARD;
 733 #ifdef CONFIG_IPV6_ROUTER_PREF
 734         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 735 #endif
 736         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 737             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 738                 int n = rt6_check_neigh(nh);
 739                 if (n < 0)
 740                         return n;
 741         }
 742         return m;
 743 }
 744
 745 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 746                        int oif, int strict, int *mpri, bool *do_rr)
 747 {
 748         bool match_do_rr = false;
 749         bool rc = false;
 750         int m;
 751
 752         if (nh->fib_nh_flags & RTNH_F_DEAD)
 753                 goto out;
 754
 755         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 756             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 757             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 758                 goto out;
 759
 760         m = rt6_score_route(nh, fib6_flags, oif, strict);
 761         if (m == RT6_NUD_FAIL_DO_RR) {
 762                 match_do_rr = true;
 763                 m = 0; /* lowest valid score */
 764         } else if (m == RT6_NUD_FAIL_HARD) {
 765                 goto out;
 766         }
 767
 768         if (strict & RT6_LOOKUP_F_REACHABLE)
 769                 rt6_probe(nh);
 770
 771         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 772         if (m > *mpri) {
 773                 *do_rr = match_do_rr;
 774                 *mpri = m;
 775                 rc = true;
 776         }
 777 out:
 778         return rc;
 779 }
 780
 781 struct fib6_nh_frl_arg {
 782         u32             flags;
 783         int             oif;
 784         int             strict;
 785         int             *mpri;
 786         bool            *do_rr;
 787         struct fib6_nh  *nh;
 788 };
 789
 790 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 791 {
 792         struct fib6_nh_frl_arg *arg = _arg;
 793
 794         arg->nh = nh;
 795         return find_match(nh, arg->flags, arg->oif, arg->strict,
 796                           arg->mpri, arg->do_rr);
 797 }
 798
 799 static void __find_rr_leaf(struct fib6_info *f6i_start,
 800                            struct fib6_info *nomatch, u32 metric,
 801                            struct fib6_result *res, struct fib6_info **cont,
 802                            int oif, int strict, bool *do_rr, int *mpri)
 803 {
 804         struct fib6_info *f6i;
 805
 806         for (f6i = f6i_start;
 807              f6i && f6i != nomatch;
 808              f6i = rcu_dereference(f6i->fib6_next)) {
 809                 bool matched = false;
 810                 struct fib6_nh *nh;
 811
 812                 if (cont && f6i->fib6_metric != metric) {
 813                         *cont = f6i;
 814                         return;
 815                 }
 816
 817                 if (fib6_check_expired(f6i))
 818                         continue;
 819
 820                 if (unlikely(f6i->nh)) {
 821                         struct fib6_nh_frl_arg arg = {
 822                                 .flags  = f6i->fib6_flags,
 823                                 .oif    = oif,
 824                                 .strict = strict,
 825                                 .mpri   = mpri,
 826                                 .do_rr  = do_rr
 827                         };
 828
 829                         if (nexthop_is_blackhole(f6i->nh)) {
 830                                 res->fib6_flags = RTF_REJECT;
 831                                 res->fib6_type = RTN_BLACKHOLE;
 832                                 res->f6i = f6i;
 833                                 res->nh = nexthop_fib6_nh(f6i->nh);
 834                                 return;
 835                         }
 836                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 837                                                      &arg)) {
 838                                 matched = true;
 839                                 nh = arg.nh;
 840                         }
 841                 } else {
 842                         nh = f6i->fib6_nh;
 843                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 844                                        mpri, do_rr))
 845                                 matched = true;
 846                 }
 847                 if (matched) {
 848                         res->f6i = f6i;
 849                         res->nh = nh;
 850                         res->fib6_flags = f6i->fib6_flags;
 851                         res->fib6_type = f6i->fib6_type;
 852                 }
 853         }
 854 }
 855
 856 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 857                          struct fib6_info *rr_head, int oif, int strict,
 858                          bool *do_rr, struct fib6_result *res)
 859 {
 860         u32 metric = rr_head->fib6_metric;
 861         struct fib6_info *cont = NULL;
 862         int mpri = -1;
 863
 864         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 865                        oif, strict, do_rr, &mpri);
 866
 867         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 868                        oif, strict, do_rr, &mpri);
 869
 870         if (res->f6i || !cont)
 871                 return;
 872
 873         __find_rr_leaf(cont, NULL, metric, res, NULL,
 874                        oif, strict, do_rr, &mpri);
 875 }
 876
 877 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 878                        struct fib6_result *res, int strict)
 879 {
 880         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 881         struct fib6_info *rt0;
 882         bool do_rr = false;
 883         int key_plen;
 884
 885         /* make sure this function or its helpers sets f6i */
 886         res->f6i = NULL;
 887
 888         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 889                 goto out;
 890
 891         rt0 = rcu_dereference(fn->rr_ptr);
 892         if (!rt0)
 893                 rt0 = leaf;
 894
 895         /* Double check to make sure fn is not an intermediate node
 896          * and fn->leaf does not points to its child's leaf
 897          * (This might happen if all routes under fn are deleted from
 898          * the tree and fib6_repair_tree() is called on the node.)
 899          */
 900         key_plen = rt0->fib6_dst.plen;
 901 #ifdef CONFIG_IPV6_SUBTREES
 902         if (rt0->fib6_src.plen)
 903                 key_plen = rt0->fib6_src.plen;
 904 #endif
 905         if (fn->fn_bit != key_plen)
 906                 goto out;
 907
 908         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 909         if (do_rr) {
 910                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 911
 912                 /* no entries matched; do round-robin */
 913                 if (!next || next->fib6_metric != rt0->fib6_metric)
 914                         next = leaf;
 915
 916                 if (next != rt0) {
 917                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 918                         /* make sure next is not being deleted from the tree */
 919                         if (next->fib6_node)
 920                                 rcu_assign_pointer(fn->rr_ptr, next);
 921                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 922                 }
 923         }
 924
 925 out:
 926         if (!res->f6i) {
 927                 res->f6i = net->ipv6.fib6_null_entry;
 928                 res->nh = res->f6i->fib6_nh;
 929                 res->fib6_flags = res->f6i->fib6_flags;
 930                 res->fib6_type = res->f6i->fib6_type;
 931         }
 932 }
 933
 934 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 935 {
 936         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 937                res->nh->fib_nh_gw_family;
 938 }
 939
 940 #ifdef CONFIG_IPV6_ROUTE_INFO
 941 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 942                   const struct in6_addr *gwaddr)
 943 {
 944         struct net *net = dev_net(dev);
 945         struct route_info *rinfo = (struct route_info *) opt;
 946         struct in6_addr prefix_buf, *prefix;
 947         unsigned int pref;
 948         unsigned long lifetime;
 949         struct fib6_info *rt;
 950
 951         if (len < sizeof(struct route_info)) {
 952                 return -EINVAL;
 953         }
 954
 955         /* Sanity check for prefix_len and length */
 956         if (rinfo->length > 3) {
 957                 return -EINVAL;
 958         } else if (rinfo->prefix_len > 128) {
 959                 return -EINVAL;
 960         } else if (rinfo->prefix_len > 64) {
 961                 if (rinfo->length < 2) {
 962                         return -EINVAL;
 963                 }
 964         } else if (rinfo->prefix_len > 0) {
 965                 if (rinfo->length < 1) {
 966                         return -EINVAL;
 967                 }
 968         }
 969
 970         pref = rinfo->route_pref;
 971         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 972                 return -EINVAL;
 973
 974         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 975
 976         if (rinfo->length == 3)
 977                 prefix = (struct in6_addr *)rinfo->prefix;
 978         else {
 979                 /* this function is safe */
 980                 ipv6_addr_prefix(&prefix_buf,
 981                                  (struct in6_addr *)rinfo->prefix,
 982                                  rinfo->prefix_len);
 983                 prefix = &prefix_buf;
 984         }
 985
 986         if (rinfo->prefix_len == 0)
 987                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 988         else
 989                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 990                                         gwaddr, dev);
 991
 992         if (rt && !lifetime) {
 993                 ip6_del_rt(net, rt, false);
 994                 rt = NULL;
 995         }
 996
 997         if (!rt && lifetime)
 998                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 999                                         dev, pref);
1000         else if (rt)
1001                 rt->fib6_flags = RTF_ROUTEINFO |
1002                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
1003
1004         if (rt) {
1005                 if (!addrconf_finite_timeout(lifetime))
1006                         fib6_clean_expires(rt);
1007                 else
1008                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1009
1010                 fib6_info_release(rt);
1011         }
1012         return 0;
1013 }
1014 #endif
1015
1016 /*
1017  *      Misc support functions
1018  */
1019
1020 /* called with rcu_lock held */
1021 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1022 {
1023         struct net_device *dev = res->nh->fib_nh_dev;
1024
1025         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1026                 /* for copies of local routes, dst->dev needs to be the
1027                  * device if it is a master device, the master device if
1028                  * device is enslaved, and the loopback as the default
1029                  */
1030                 if (netif_is_l3_slave(dev) &&
1031                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1032                         dev = l3mdev_master_dev_rcu(dev);
1033                 else if (!netif_is_l3_master(dev))
1034                         dev = dev_net(dev)->loopback_dev;
1035                 /* last case is netif_is_l3_master(dev) is true in which
1036                  * case we want dev returned to be dev
1037                  */
1038         }
1039
1040         return dev;
1041 }
1042
1043 static const int fib6_prop[RTN_MAX + 1] = {
1044         [RTN_UNSPEC]    = 0,
1045         [RTN_UNICAST]   = 0,
1046         [RTN_LOCAL]     = 0,
1047         [RTN_BROADCAST] = 0,
1048         [RTN_ANYCAST]   = 0,
1049         [RTN_MULTICAST] = 0,
1050         [RTN_BLACKHOLE] = -EINVAL,
1051         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1052         [RTN_PROHIBIT]  = -EACCES,
1053         [RTN_THROW]     = -EAGAIN,
1054         [RTN_NAT]       = -EINVAL,
1055         [RTN_XRESOLVE]  = -EINVAL,
1056 };
1057
1058 static int ip6_rt_type_to_error(u8 fib6_type)
1059 {
1060         return fib6_prop[fib6_type];
1061 }
1062
1063 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1064 {
1065         unsigned short flags = 0;
1066
1067         if (rt->dst_nocount)
1068                 flags |= DST_NOCOUNT;
1069         if (rt->dst_nopolicy)
1070                 flags |= DST_NOPOLICY;
1071
1072         return flags;
1073 }
1074
1075 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1076 {
1077         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1078
1079         switch (fib6_type) {
1080         case RTN_BLACKHOLE:
1081                 rt->dst.output = dst_discard_out;
1082                 rt->dst.input = dst_discard;
1083                 break;
1084         case RTN_PROHIBIT:
1085                 rt->dst.output = ip6_pkt_prohibit_out;
1086                 rt->dst.input = ip6_pkt_prohibit;
1087                 break;
1088         case RTN_THROW:
1089         case RTN_UNREACHABLE:
1090         default:
1091                 rt->dst.output = ip6_pkt_discard_out;
1092                 rt->dst.input = ip6_pkt_discard;
1093                 break;
1094         }
1095 }
1096
1097 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1098 {
1099         struct fib6_info *f6i = res->f6i;
1100
1101         if (res->fib6_flags & RTF_REJECT) {
1102                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1103                 return;
1104         }
1105
1106         rt->dst.error = 0;
1107         rt->dst.output = ip6_output;
1108
1109         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1110                 rt->dst.input = ip6_input;
1111         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1112                 rt->dst.input = ip6_mc_input;
1113         } else {
1114                 rt->dst.input = ip6_forward;
1115         }
1116
1117         if (res->nh->fib_nh_lws) {
1118                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1119                 lwtunnel_set_redirect(&rt->dst);
1120         }
1121
1122         rt->dst.lastuse = jiffies;
1123 }
1124
1125 /* Caller must already hold reference to @from */
1126 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1127 {
1128         rt->rt6i_flags &= ~RTF_EXPIRES;
1129         rcu_assign_pointer(rt->from, from);
1130         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1131 }
1132
1133 /* Caller must already hold reference to f6i in result */
1134 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1135 {
1136         const struct fib6_nh *nh = res->nh;
1137         const struct net_device *dev = nh->fib_nh_dev;
1138         struct fib6_info *f6i = res->f6i;
1139
1140         ip6_rt_init_dst(rt, res);
1141
1142         rt->rt6i_dst = f6i->fib6_dst;
1143         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1144         rt->rt6i_flags = res->fib6_flags;
1145         if (nh->fib_nh_gw_family) {
1146                 rt->rt6i_gateway = nh->fib_nh_gw6;
1147                 rt->rt6i_flags |= RTF_GATEWAY;
1148         }
1149         rt6_set_from(rt, f6i);
1150 #ifdef CONFIG_IPV6_SUBTREES
1151         rt->rt6i_src = f6i->fib6_src;
1152 #endif
1153 }
1154
1155 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1156                                         struct in6_addr *saddr)
1157 {
1158         struct fib6_node *pn, *sn;
1159         while (1) {
1160                 if (fn->fn_flags & RTN_TL_ROOT)
1161                         return NULL;
1162                 pn = rcu_dereference(fn->parent);
1163                 sn = FIB6_SUBTREE(pn);
1164                 if (sn && sn != fn)
1165                         fn = fib6_node_lookup(sn, NULL, saddr);
1166                 else
1167                         fn = pn;
1168                 if (fn->fn_flags & RTN_RTINFO)
1169                         return fn;
1170         }
1171 }
1172
1173 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1174 {
1175         struct rt6_info *rt = *prt;
1176
1177         if (dst_hold_safe(&rt->dst))
1178                 return true;
1179         if (net) {
1180                 rt = net->ipv6.ip6_null_entry;
1181                 dst_hold(&rt->dst);
1182         } else {
1183                 rt = NULL;
1184         }
1185         *prt = rt;
1186         return false;
1187 }
1188
1189 /* called with rcu_lock held */
1190 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1191 {
1192         struct net_device *dev = res->nh->fib_nh_dev;
1193         struct fib6_info *f6i = res->f6i;
1194         unsigned short flags;
1195         struct rt6_info *nrt;
1196
1197         if (!fib6_info_hold_safe(f6i))
1198                 goto fallback;
1199
1200         flags = fib6_info_dst_flags(f6i);
1201         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1202         if (!nrt) {
1203                 fib6_info_release(f6i);
1204                 goto fallback;
1205         }
1206
1207         ip6_rt_copy_init(nrt, res);
1208         return nrt;
1209
1210 fallback:
1211         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1212         dst_hold(&nrt->dst);
1213         return nrt;
1214 }
1215
1216 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1217                                              struct fib6_table *table,
1218                                              struct flowi6 *fl6,
1219                                              const struct sk_buff *skb,
1220                                              int flags)
1221 {
1222         struct fib6_result res = {};
1223         struct fib6_node *fn;
1224         struct rt6_info *rt;
1225
1226         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1227                 flags &= ~RT6_LOOKUP_F_IFACE;
1228
1229         rcu_read_lock();
1230         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1231 restart:
1232         res.f6i = rcu_dereference(fn->leaf);
1233         if (!res.f6i)
1234                 res.f6i = net->ipv6.fib6_null_entry;
1235         else
1236                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1237                                  flags);
1238
1239         if (res.f6i == net->ipv6.fib6_null_entry) {
1240                 fn = fib6_backtrack(fn, &fl6->saddr);
1241                 if (fn)
1242                         goto restart;
1243
1244                 rt = net->ipv6.ip6_null_entry;
1245                 dst_hold(&rt->dst);
1246                 goto out;
1247         } else if (res.fib6_flags & RTF_REJECT) {
1248                 goto do_create;
1249         }
1250
1251         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1252                          fl6->flowi6_oif != 0, skb, flags);
1253
1254         /* Search through exception table */
1255         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1256         if (rt) {
1257                 if (ip6_hold_safe(net, &rt))
1258                         dst_use_noref(&rt->dst, jiffies);
1259         } else {
1260 do_create:
1261                 rt = ip6_create_rt_rcu(&res);
1262         }
1263
1264 out:
1265         trace_fib6_table_lookup(net, &res, table, fl6);
1266
1267         rcu_read_unlock();
1268
1269         return rt;
1270 }
1271
1272 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1273                                    const struct sk_buff *skb, int flags)
1274 {
1275         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1276 }
1277 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1278
1279 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1280                             const struct in6_addr *saddr, int oif,
1281                             const struct sk_buff *skb, int strict)
1282 {
1283         struct flowi6 fl6 = {
1284                 .flowi6_oif = oif,
1285                 .daddr = *daddr,
1286         };
1287         struct dst_entry *dst;
1288         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1289
1290         if (saddr) {
1291                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1292                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1293         }
1294
1295         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1296         if (dst->error == 0)
1297                 return (struct rt6_info *) dst;
1298
1299         dst_release(dst);
1300
1301         return NULL;
1302 }
1303 EXPORT_SYMBOL(rt6_lookup);
1304
1305 /* ip6_ins_rt is called with FREE table->tb6_lock.
1306  * It takes new route entry, the addition fails by any reason the
1307  * route is released.
1308  * Caller must hold dst before calling it.
1309  */
1310
1311 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1312                         struct netlink_ext_ack *extack)
1313 {
1314         int err;
1315         struct fib6_table *table;
1316
1317         table = rt->fib6_table;
1318         spin_lock_bh(&table->tb6_lock);
1319         err = fib6_add(&table->tb6_root, rt, info, extack);
1320         spin_unlock_bh(&table->tb6_lock);
1321
1322         return err;
1323 }
1324
1325 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1326 {
1327         struct nl_info info = { .nl_net = net, };
1328
1329         return __ip6_ins_rt(rt, &info, NULL);
1330 }
1331
1332 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1333                                            const struct in6_addr *daddr,
1334                                            const struct in6_addr *saddr)
1335 {
1336         struct fib6_info *f6i = res->f6i;
1337         struct net_device *dev;
1338         struct rt6_info *rt;
1339
1340         /*
1341          *      Clone the route.
1342          */
1343
1344         if (!fib6_info_hold_safe(f6i))
1345                 return NULL;
1346
1347         dev = ip6_rt_get_dev_rcu(res);
1348         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1349         if (!rt) {
1350                 fib6_info_release(f6i);
1351                 return NULL;
1352         }
1353
1354         ip6_rt_copy_init(rt, res);
1355         rt->rt6i_flags |= RTF_CACHE;
1356         rt->rt6i_dst.addr = *daddr;
1357         rt->rt6i_dst.plen = 128;
1358
1359         if (!rt6_is_gw_or_nonexthop(res)) {
1360                 if (f6i->fib6_dst.plen != 128 &&
1361                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1362                         rt->rt6i_flags |= RTF_ANYCAST;
1363 #ifdef CONFIG_IPV6_SUBTREES
1364                 if (rt->rt6i_src.plen && saddr) {
1365                         rt->rt6i_src.addr = *saddr;
1366                         rt->rt6i_src.plen = 128;
1367                 }
1368 #endif
1369         }
1370
1371         return rt;
1372 }
1373
1374 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1375 {
1376         struct fib6_info *f6i = res->f6i;
1377         unsigned short flags = fib6_info_dst_flags(f6i);
1378         struct net_device *dev;
1379         struct rt6_info *pcpu_rt;
1380
1381         if (!fib6_info_hold_safe(f6i))
1382                 return NULL;
1383
1384         rcu_read_lock();
1385         dev = ip6_rt_get_dev_rcu(res);
1386         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1387         rcu_read_unlock();
1388         if (!pcpu_rt) {
1389                 fib6_info_release(f6i);
1390                 return NULL;
1391         }
1392         ip6_rt_copy_init(pcpu_rt, res);
1393         pcpu_rt->rt6i_flags |= RTF_PCPU;
1394
1395         if (f6i->nh)
1396                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1397
1398         return pcpu_rt;
1399 }
1400
1401 static bool rt6_is_valid(const struct rt6_info *rt6)
1402 {
1403         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1404 }
1405
1406 /* It should be called with rcu_read_lock() acquired */
1407 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1408 {
1409         struct rt6_info *pcpu_rt;
1410
1411         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1412
1413         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1414                 struct rt6_info *prev, **p;
1415
1416                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1417                 prev = xchg(p, NULL);
1418                 if (prev) {
1419                         dst_dev_put(&prev->dst);
1420                         dst_release(&prev->dst);
1421                 }
1422
1423                 pcpu_rt = NULL;
1424         }
1425
1426         return pcpu_rt;
1427 }
1428
1429 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1430                                             const struct fib6_result *res)
1431 {
1432         struct rt6_info *pcpu_rt, *prev, **p;
1433
1434         pcpu_rt = ip6_rt_pcpu_alloc(res);
1435         if (!pcpu_rt)
1436                 return NULL;
1437
1438         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1439         prev = cmpxchg(p, NULL, pcpu_rt);
1440         BUG_ON(prev);
1441
1442         if (res->f6i->fib6_destroying) {
1443                 struct fib6_info *from;
1444
1445                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1446                 fib6_info_release(from);
1447         }
1448
1449         return pcpu_rt;
1450 }
1451
1452 /* exception hash table implementation
1453  */
1454 static DEFINE_SPINLOCK(rt6_exception_lock);
1455
1456 /* Remove rt6_ex from hash table and free the memory
1457  * Caller must hold rt6_exception_lock
1458  */
1459 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1460                                  struct rt6_exception *rt6_ex)
1461 {
1462         struct fib6_info *from;
1463         struct net *net;
1464
1465         if (!bucket || !rt6_ex)
1466                 return;
1467
1468         net = dev_net(rt6_ex->rt6i->dst.dev);
1469         net->ipv6.rt6_stats->fib_rt_cache--;
1470
1471         /* purge completely the exception to allow releasing the held resources:
1472          * some [sk] cache may keep the dst around for unlimited time
1473          */
1474         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1475         fib6_info_release(from);
1476         dst_dev_put(&rt6_ex->rt6i->dst);
1477
1478         hlist_del_rcu(&rt6_ex->hlist);
1479         dst_release(&rt6_ex->rt6i->dst);
1480         kfree_rcu(rt6_ex, rcu);
1481         WARN_ON_ONCE(!bucket->depth);
1482         bucket->depth--;
1483 }
1484
1485 /* Remove oldest rt6_ex in bucket and free the memory
1486  * Caller must hold rt6_exception_lock
1487  */
1488 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1489 {
1490         struct rt6_exception *rt6_ex, *oldest = NULL;
1491
1492         if (!bucket)
1493                 return;
1494
1495         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1496                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1497                         oldest = rt6_ex;
1498         }
1499         rt6_remove_exception(bucket, oldest);
1500 }
1501
1502 static u32 rt6_exception_hash(const struct in6_addr *dst,
1503                               const struct in6_addr *src)
1504 {
1505         static u32 seed __read_mostly;
1506         u32 val;
1507
1508         net_get_random_once(&seed, sizeof(seed));
1509         val = jhash2((const u32 *)dst, sizeof(*dst)/sizeof(u32), seed);
1510
1511 #ifdef CONFIG_IPV6_SUBTREES
1512         if (src)
1513                 val = jhash2((const u32 *)src, sizeof(*src)/sizeof(u32), val);
1514 #endif
1515         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1516 }
1517
1518 /* Helper function to find the cached rt in the hash table
1519  * and update bucket pointer to point to the bucket for this
1520  * (daddr, saddr) pair
1521  * Caller must hold rt6_exception_lock
1522  */
1523 static struct rt6_exception *
1524 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1525                               const struct in6_addr *daddr,
1526                               const struct in6_addr *saddr)
1527 {
1528         struct rt6_exception *rt6_ex;
1529         u32 hval;
1530
1531         if (!(*bucket) || !daddr)
1532                 return NULL;
1533
1534         hval = rt6_exception_hash(daddr, saddr);
1535         *bucket += hval;
1536
1537         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1538                 struct rt6_info *rt6 = rt6_ex->rt6i;
1539                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1540
1541 #ifdef CONFIG_IPV6_SUBTREES
1542                 if (matched && saddr)
1543                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1544 #endif
1545                 if (matched)
1546                         return rt6_ex;
1547         }
1548         return NULL;
1549 }
1550
1551 /* Helper function to find the cached rt in the hash table
1552  * and update bucket pointer to point to the bucket for this
1553  * (daddr, saddr) pair
1554  * Caller must hold rcu_read_lock()
1555  */
1556 static struct rt6_exception *
1557 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1558                          const struct in6_addr *daddr,
1559                          const struct in6_addr *saddr)
1560 {
1561         struct rt6_exception *rt6_ex;
1562         u32 hval;
1563
1564         WARN_ON_ONCE(!rcu_read_lock_held());
1565
1566         if (!(*bucket) || !daddr)
1567                 return NULL;
1568
1569         hval = rt6_exception_hash(daddr, saddr);
1570         *bucket += hval;
1571
1572         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1573                 struct rt6_info *rt6 = rt6_ex->rt6i;
1574                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1575
1576 #ifdef CONFIG_IPV6_SUBTREES
1577                 if (matched && saddr)
1578                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1579 #endif
1580                 if (matched)
1581                         return rt6_ex;
1582         }
1583         return NULL;
1584 }
1585
1586 static unsigned int fib6_mtu(const struct fib6_result *res)
1587 {
1588         const struct fib6_nh *nh = res->nh;
1589         unsigned int mtu;
1590
1591         if (res->f6i->fib6_pmtu) {
1592                 mtu = res->f6i->fib6_pmtu;
1593         } else {
1594                 struct net_device *dev = nh->fib_nh_dev;
1595                 struct inet6_dev *idev;
1596
1597                 rcu_read_lock();
1598                 idev = __in6_dev_get(dev);
1599                 mtu = idev->cnf.mtu6;
1600                 rcu_read_unlock();
1601         }
1602
1603         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1604
1605         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1606 }
1607
1608 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1609
1610 /* used when the flushed bit is not relevant, only access to the bucket
1611  * (ie., all bucket users except rt6_insert_exception);
1612  *
1613  * called under rcu lock; sometimes called with rt6_exception_lock held
1614  */
1615 static
1616 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1617                                                        spinlock_t *lock)
1618 {
1619         struct rt6_exception_bucket *bucket;
1620
1621         if (lock)
1622                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1623                                                    lockdep_is_held(lock));
1624         else
1625                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1626
1627         /* remove bucket flushed bit if set */
1628         if (bucket) {
1629                 unsigned long p = (unsigned long)bucket;
1630
1631                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1632                 bucket = (struct rt6_exception_bucket *)p;
1633         }
1634
1635         return bucket;
1636 }
1637
1638 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1639 {
1640         unsigned long p = (unsigned long)bucket;
1641
1642         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1643 }
1644
1645 /* called with rt6_exception_lock held */
1646 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1647                                               spinlock_t *lock)
1648 {
1649         struct rt6_exception_bucket *bucket;
1650         unsigned long p;
1651
1652         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1653                                            lockdep_is_held(lock));
1654
1655         p = (unsigned long)bucket;
1656         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1657         bucket = (struct rt6_exception_bucket *)p;
1658         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1659 }
1660
1661 static int rt6_insert_exception(struct rt6_info *nrt,
1662                                 const struct fib6_result *res)
1663 {
1664         struct net *net = dev_net(nrt->dst.dev);
1665         struct rt6_exception_bucket *bucket;
1666         struct fib6_info *f6i = res->f6i;
1667         struct in6_addr *src_key = NULL;
1668         struct rt6_exception *rt6_ex;
1669         struct fib6_nh *nh = res->nh;
1670         int err = 0;
1671
1672         spin_lock_bh(&rt6_exception_lock);
1673
1674         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1675                                           lockdep_is_held(&rt6_exception_lock));
1676         if (!bucket) {
1677                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1678                                  GFP_ATOMIC);
1679                 if (!bucket) {
1680                         err = -ENOMEM;
1681                         goto out;
1682                 }
1683                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1684         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1685                 err = -EINVAL;
1686                 goto out;
1687         }
1688
1689 #ifdef CONFIG_IPV6_SUBTREES
1690         /* fib6_src.plen != 0 indicates f6i is in subtree
1691          * and exception table is indexed by a hash of
1692          * both fib6_dst and fib6_src.
1693          * Otherwise, the exception table is indexed by
1694          * a hash of only fib6_dst.
1695          */
1696         if (f6i->fib6_src.plen)
1697                 src_key = &nrt->rt6i_src.addr;
1698 #endif
1699         /* rt6_mtu_change() might lower mtu on f6i.
1700          * Only insert this exception route if its mtu
1701          * is less than f6i's mtu value.
1702          */
1703         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1704                 err = -EINVAL;
1705                 goto out;
1706         }
1707
1708         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1709                                                src_key);
1710         if (rt6_ex)
1711                 rt6_remove_exception(bucket, rt6_ex);
1712
1713         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1714         if (!rt6_ex) {
1715                 err = -ENOMEM;
1716                 goto out;
1717         }
1718         rt6_ex->rt6i = nrt;
1719         rt6_ex->stamp = jiffies;
1720         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1721         bucket->depth++;
1722         net->ipv6.rt6_stats->fib_rt_cache++;
1723
1724         if (bucket->depth > FIB6_MAX_DEPTH)
1725                 rt6_exception_remove_oldest(bucket);
1726
1727 out:
1728         spin_unlock_bh(&rt6_exception_lock);
1729
1730         /* Update fn->fn_sernum to invalidate all cached dst */
1731         if (!err) {
1732                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1733                 fib6_update_sernum(net, f6i);
1734                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1735                 fib6_force_start_gc(net);
1736         }
1737
1738         return err;
1739 }
1740
1741 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1742 {
1743         struct rt6_exception_bucket *bucket;
1744         struct rt6_exception *rt6_ex;
1745         struct hlist_node *tmp;
1746         int i;
1747
1748         spin_lock_bh(&rt6_exception_lock);
1749
1750         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1751         if (!bucket)
1752                 goto out;
1753
1754         /* Prevent rt6_insert_exception() to recreate the bucket list */
1755         if (!from)
1756                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1757
1758         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1759                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1760                         if (!from ||
1761                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1762                                 rt6_remove_exception(bucket, rt6_ex);
1763                 }
1764                 WARN_ON_ONCE(!from && bucket->depth);
1765                 bucket++;
1766         }
1767 out:
1768         spin_unlock_bh(&rt6_exception_lock);
1769 }
1770
1771 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1772 {
1773         struct fib6_info *f6i = arg;
1774
1775         fib6_nh_flush_exceptions(nh, f6i);
1776
1777         return 0;
1778 }
1779
1780 void rt6_flush_exceptions(struct fib6_info *f6i)
1781 {
1782         if (f6i->nh)
1783                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1784                                          f6i);
1785         else
1786                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1787 }
1788
1789 /* Find cached rt in the hash table inside passed in rt
1790  * Caller has to hold rcu_read_lock()
1791  */
1792 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1793                                            const struct in6_addr *daddr,
1794                                            const struct in6_addr *saddr)
1795 {
1796         const struct in6_addr *src_key = NULL;
1797         struct rt6_exception_bucket *bucket;
1798         struct rt6_exception *rt6_ex;
1799         struct rt6_info *ret = NULL;
1800
1801 #ifdef CONFIG_IPV6_SUBTREES
1802         /* fib6i_src.plen != 0 indicates f6i is in subtree
1803          * and exception table is indexed by a hash of
1804          * both fib6_dst and fib6_src.
1805          * However, the src addr used to create the hash
1806          * might not be exactly the passed in saddr which
1807          * is a /128 addr from the flow.
1808          * So we need to use f6i->fib6_src to redo lookup
1809          * if the passed in saddr does not find anything.
1810          * (See the logic in ip6_rt_cache_alloc() on how
1811          * rt->rt6i_src is updated.)
1812          */
1813         if (res->f6i->fib6_src.plen)
1814                 src_key = saddr;
1815 find_ex:
1816 #endif
1817         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1818         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1819
1820         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1821                 ret = rt6_ex->rt6i;
1822
1823 #ifdef CONFIG_IPV6_SUBTREES
1824         /* Use fib6_src as src_key and redo lookup */
1825         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1826                 src_key = &res->f6i->fib6_src.addr;
1827                 goto find_ex;
1828         }
1829 #endif
1830
1831         return ret;
1832 }
1833
1834 /* Remove the passed in cached rt from the hash table that contains it */
1835 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1836                                     const struct rt6_info *rt)
1837 {
1838         const struct in6_addr *src_key = NULL;
1839         struct rt6_exception_bucket *bucket;
1840         struct rt6_exception *rt6_ex;
1841         int err;
1842
1843         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1844                 return -ENOENT;
1845
1846         spin_lock_bh(&rt6_exception_lock);
1847         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1848
1849 #ifdef CONFIG_IPV6_SUBTREES
1850         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1851          * and exception table is indexed by a hash of
1852          * both rt6i_dst and rt6i_src.
1853          * Otherwise, the exception table is indexed by
1854          * a hash of only rt6i_dst.
1855          */
1856         if (plen)
1857                 src_key = &rt->rt6i_src.addr;
1858 #endif
1859         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1860                                                &rt->rt6i_dst.addr,
1861                                                src_key);
1862         if (rt6_ex) {
1863                 rt6_remove_exception(bucket, rt6_ex);
1864                 err = 0;
1865         } else {
1866                 err = -ENOENT;
1867         }
1868
1869         spin_unlock_bh(&rt6_exception_lock);
1870         return err;
1871 }
1872
1873 struct fib6_nh_excptn_arg {
1874         struct rt6_info *rt;
1875         int             plen;
1876 };
1877
1878 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1879 {
1880         struct fib6_nh_excptn_arg *arg = _arg;
1881         int err;
1882
1883         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1884         if (err == 0)
1885                 return 1;
1886
1887         return 0;
1888 }
1889
1890 static int rt6_remove_exception_rt(struct rt6_info *rt)
1891 {
1892         struct fib6_info *from;
1893
1894         from = rcu_dereference(rt->from);
1895         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1896                 return -EINVAL;
1897
1898         if (from->nh) {
1899                 struct fib6_nh_excptn_arg arg = {
1900                         .rt = rt,
1901                         .plen = from->fib6_src.plen
1902                 };
1903                 int rc;
1904
1905                 /* rc = 1 means an entry was found */
1906                 rc = nexthop_for_each_fib6_nh(from->nh,
1907                                               rt6_nh_remove_exception_rt,
1908                                               &arg);
1909                 return rc ? 0 : -ENOENT;
1910         }
1911
1912         return fib6_nh_remove_exception(from->fib6_nh,
1913                                         from->fib6_src.plen, rt);
1914 }
1915
1916 /* Find rt6_ex which contains the passed in rt cache and
1917  * refresh its stamp
1918  */
1919 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1920                                      const struct rt6_info *rt)
1921 {
1922         const struct in6_addr *src_key = NULL;
1923         struct rt6_exception_bucket *bucket;
1924         struct rt6_exception *rt6_ex;
1925
1926         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1927 #ifdef CONFIG_IPV6_SUBTREES
1928         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1929          * and exception table is indexed by a hash of
1930          * both rt6i_dst and rt6i_src.
1931          * Otherwise, the exception table is indexed by
1932          * a hash of only rt6i_dst.
1933          */
1934         if (plen)
1935                 src_key = &rt->rt6i_src.addr;
1936 #endif
1937         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1938         if (rt6_ex)
1939                 rt6_ex->stamp = jiffies;
1940 }
1941
1942 struct fib6_nh_match_arg {
1943         const struct net_device *dev;
1944         const struct in6_addr   *gw;
1945         struct fib6_nh          *match;
1946 };
1947
1948 /* determine if fib6_nh has given device and gateway */
1949 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1950 {
1951         struct fib6_nh_match_arg *arg = _arg;
1952
1953         if (arg->dev != nh->fib_nh_dev ||
1954             (arg->gw && !nh->fib_nh_gw_family) ||
1955             (!arg->gw && nh->fib_nh_gw_family) ||
1956             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1957                 return 0;
1958
1959         arg->match = nh;
1960
1961         /* found a match, break the loop */
1962         return 1;
1963 }
1964
1965 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1966 {
1967         struct fib6_info *from;
1968         struct fib6_nh *fib6_nh;
1969
1970         rcu_read_lock();
1971
1972         from = rcu_dereference(rt->from);
1973         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1974                 goto unlock;
1975
1976         if (from->nh) {
1977                 struct fib6_nh_match_arg arg = {
1978                         .dev = rt->dst.dev,
1979                         .gw = &rt->rt6i_gateway,
1980                 };
1981
1982                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1983
1984                 if (!arg.match)
1985                         goto unlock;
1986                 fib6_nh = arg.match;
1987         } else {
1988                 fib6_nh = from->fib6_nh;
1989         }
1990         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1991 unlock:
1992         rcu_read_unlock();
1993 }
1994
1995 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1996                                          struct rt6_info *rt, int mtu)
1997 {
1998         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1999          * lowest MTU in the path: always allow updating the route PMTU to
2000          * reflect PMTU decreases.
2001          *
2002          * If the new MTU is higher, and the route PMTU is equal to the local
2003          * MTU, this means the old MTU is the lowest in the path, so allow
2004          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2005          * handle this.
2006          */
2007
2008         if (dst_mtu(&rt->dst) >= mtu)
2009                 return true;
2010
2011         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2012                 return true;
2013
2014         return false;
2015 }
2016
2017 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2018                                        const struct fib6_nh *nh, int mtu)
2019 {
2020         struct rt6_exception_bucket *bucket;
2021         struct rt6_exception *rt6_ex;
2022         int i;
2023
2024         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2025         if (!bucket)
2026                 return;
2027
2028         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2029                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2030                         struct rt6_info *entry = rt6_ex->rt6i;
2031
2032                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2033                          * route), the metrics of its rt->from have already
2034                          * been updated.
2035                          */
2036                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2037                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2038                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2039                 }
2040                 bucket++;
2041         }
2042 }
2043
2044 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2045
2046 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2047                                             const struct in6_addr *gateway)
2048 {
2049         struct rt6_exception_bucket *bucket;
2050         struct rt6_exception *rt6_ex;
2051         struct hlist_node *tmp;
2052         int i;
2053
2054         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2055                 return;
2056
2057         spin_lock_bh(&rt6_exception_lock);
2058         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2059         if (bucket) {
2060                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2061                         hlist_for_each_entry_safe(rt6_ex, tmp,
2062                                                   &bucket->chain, hlist) {
2063                                 struct rt6_info *entry = rt6_ex->rt6i;
2064
2065                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2066                                     RTF_CACHE_GATEWAY &&
2067                                     ipv6_addr_equal(gateway,
2068                                                     &entry->rt6i_gateway)) {
2069                                         rt6_remove_exception(bucket, rt6_ex);
2070                                 }
2071                         }
2072                         bucket++;
2073                 }
2074         }
2075
2076         spin_unlock_bh(&rt6_exception_lock);
2077 }
2078
2079 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2080                                       struct rt6_exception *rt6_ex,
2081                                       struct fib6_gc_args *gc_args,
2082                                       unsigned long now)
2083 {
2084         struct rt6_info *rt = rt6_ex->rt6i;
2085
2086         /* we are pruning and obsoleting aged-out and non gateway exceptions
2087          * even if others have still references to them, so that on next
2088          * dst_check() such references can be dropped.
2089          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2090          * expired, independently from their aging, as per RFC 8201 section 4
2091          */
2092         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2093                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2094                         RT6_TRACE("aging clone %p\n", rt);
2095                         rt6_remove_exception(bucket, rt6_ex);
2096                         return;
2097                 }
2098         } else if (time_after(jiffies, rt->dst.expires)) {
2099                 RT6_TRACE("purging expired route %p\n", rt);
2100                 rt6_remove_exception(bucket, rt6_ex);
2101                 return;
2102         }
2103
2104         if (rt->rt6i_flags & RTF_GATEWAY) {
2105                 struct neighbour *neigh;
2106                 __u8 neigh_flags = 0;
2107
2108                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2109                 if (neigh)
2110                         neigh_flags = neigh->flags;
2111
2112                 if (!(neigh_flags & NTF_ROUTER)) {
2113                         RT6_TRACE("purging route %p via non-router but gateway\n",
2114                                   rt);
2115                         rt6_remove_exception(bucket, rt6_ex);
2116                         return;
2117                 }
2118         }
2119
2120         gc_args->more++;
2121 }
2122
2123 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2124                                    struct fib6_gc_args *gc_args,
2125                                    unsigned long now)
2126 {
2127         struct rt6_exception_bucket *bucket;
2128         struct rt6_exception *rt6_ex;
2129         struct hlist_node *tmp;
2130         int i;
2131
2132         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2133                 return;
2134
2135         rcu_read_lock_bh();
2136         spin_lock(&rt6_exception_lock);
2137         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2138         if (bucket) {
2139                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2140                         hlist_for_each_entry_safe(rt6_ex, tmp,
2141                                                   &bucket->chain, hlist) {
2142                                 rt6_age_examine_exception(bucket, rt6_ex,
2143                                                           gc_args, now);
2144                         }
2145                         bucket++;
2146                 }
2147         }
2148         spin_unlock(&rt6_exception_lock);
2149         rcu_read_unlock_bh();
2150 }
2151
2152 struct fib6_nh_age_excptn_arg {
2153         struct fib6_gc_args     *gc_args;
2154         unsigned long           now;
2155 };
2156
2157 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2158 {
2159         struct fib6_nh_age_excptn_arg *arg = _arg;
2160
2161         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2162         return 0;
2163 }
2164
2165 void rt6_age_exceptions(struct fib6_info *f6i,
2166                         struct fib6_gc_args *gc_args,
2167                         unsigned long now)
2168 {
2169         if (f6i->nh) {
2170                 struct fib6_nh_age_excptn_arg arg = {
2171                         .gc_args = gc_args,
2172                         .now = now
2173                 };
2174
2175                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2176                                          &arg);
2177         } else {
2178                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2179         }
2180 }
2181
2182 /* must be called with rcu lock held */
2183 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2184                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2185 {
2186         struct fib6_node *fn, *saved_fn;
2187
2188         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2189         saved_fn = fn;
2190
2191         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2192                 oif = 0;
2193
2194 redo_rt6_select:
2195         rt6_select(net, fn, oif, res, strict);
2196         if (res->f6i == net->ipv6.fib6_null_entry) {
2197                 fn = fib6_backtrack(fn, &fl6->saddr);
2198                 if (fn)
2199                         goto redo_rt6_select;
2200                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2201                         /* also consider unreachable route */
2202                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2203                         fn = saved_fn;
2204                         goto redo_rt6_select;
2205                 }
2206         }
2207
2208         trace_fib6_table_lookup(net, res, table, fl6);
2209
2210         return 0;
2211 }
2212
2213 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2214                                int oif, struct flowi6 *fl6,
2215                                const struct sk_buff *skb, int flags)
2216 {
2217         struct fib6_result res = {};
2218         struct rt6_info *rt = NULL;
2219         int strict = 0;
2220
2221         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2222                      !rcu_read_lock_held());
2223
2224         strict |= flags & RT6_LOOKUP_F_IFACE;
2225         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2226         if (net->ipv6.devconf_all->forwarding == 0)
2227                 strict |= RT6_LOOKUP_F_REACHABLE;
2228
2229         rcu_read_lock();
2230
2231         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2232         if (res.f6i == net->ipv6.fib6_null_entry)
2233                 goto out;
2234
2235         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2236
2237         /*Search through exception table */
2238         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2239         if (rt) {
2240                 goto out;
2241         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2242                             !res.nh->fib_nh_gw_family)) {
2243                 /* Create a RTF_CACHE clone which will not be
2244                  * owned by the fib6 tree.  It is for the special case where
2245                  * the daddr in the skb during the neighbor look-up is different
2246                  * from the fl6->daddr used to look-up route here.
2247                  */
2248                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2249
2250                 if (rt) {
2251                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2252                          * As rt6_uncached_list_add() does not consume refcnt,
2253                          * this refcnt is always returned to the caller even
2254                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2255                          */
2256                         rt6_uncached_list_add(rt);
2257                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2258                         rcu_read_unlock();
2259
2260                         return rt;
2261                 }
2262         } else {
2263                 /* Get a percpu copy */
2264                 local_bh_disable();
2265                 rt = rt6_get_pcpu_route(&res);
2266
2267                 if (!rt)
2268                         rt = rt6_make_pcpu_route(net, &res);
2269
2270                 local_bh_enable();
2271         }
2272 out:
2273         if (!rt)
2274                 rt = net->ipv6.ip6_null_entry;
2275         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2276                 ip6_hold_safe(net, &rt);
2277         rcu_read_unlock();
2278
2279         return rt;
2280 }
2281 EXPORT_SYMBOL_GPL(ip6_pol_route);
2282
2283 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2284                                             struct fib6_table *table,
2285                                             struct flowi6 *fl6,
2286                                             const struct sk_buff *skb,
2287                                             int flags)
2288 {
2289         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2290 }
2291
2292 struct dst_entry *ip6_route_input_lookup(struct net *net,
2293                                          struct net_device *dev,
2294                                          struct flowi6 *fl6,
2295                                          const struct sk_buff *skb,
2296                                          int flags)
2297 {
2298         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2299                 flags |= RT6_LOOKUP_F_IFACE;
2300
2301         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2302 }
2303 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2304
2305 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2306                                   struct flow_keys *keys,
2307                                   struct flow_keys *flkeys)
2308 {
2309         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2310         const struct ipv6hdr *key_iph = outer_iph;
2311         struct flow_keys *_flkeys = flkeys;
2312         const struct ipv6hdr *inner_iph;
2313         const struct icmp6hdr *icmph;
2314         struct ipv6hdr _inner_iph;
2315         struct icmp6hdr _icmph;
2316
2317         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2318                 goto out;
2319
2320         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2321                                    sizeof(_icmph), &_icmph);
2322         if (!icmph)
2323                 goto out;
2324
2325         if (!icmpv6_is_err(icmph->icmp6_type))
2326                 goto out;
2327
2328         inner_iph = skb_header_pointer(skb,
2329                                        skb_transport_offset(skb) + sizeof(*icmph),
2330                                        sizeof(_inner_iph), &_inner_iph);
2331         if (!inner_iph)
2332                 goto out;
2333
2334         key_iph = inner_iph;
2335         _flkeys = NULL;
2336 out:
2337         if (_flkeys) {
2338                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2339                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2340                 keys->tags.flow_label = _flkeys->tags.flow_label;
2341                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2342         } else {
2343                 keys->addrs.v6addrs.src = key_iph->saddr;
2344                 keys->addrs.v6addrs.dst = key_iph->daddr;
2345                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2346                 keys->basic.ip_proto = key_iph->nexthdr;
2347         }
2348 }
2349
2350 /* if skb is set it will be used and fl6 can be NULL */
2351 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2352                        const struct sk_buff *skb, struct flow_keys *flkeys)
2353 {
2354         struct flow_keys hash_keys;
2355         u32 mhash;
2356
2357         switch (ip6_multipath_hash_policy(net)) {
2358         case 0:
2359                 memset(&hash_keys, 0, sizeof(hash_keys));
2360                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2361                 if (skb) {
2362                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2363                 } else {
2364                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2365                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2366                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2367                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2368                 }
2369                 break;
2370         case 1:
2371                 if (skb) {
2372                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2373                         struct flow_keys keys;
2374
2375                         /* short-circuit if we already have L4 hash present */
2376                         if (skb->l4_hash)
2377                                 return skb_get_hash_raw(skb) >> 1;
2378
2379                         memset(&hash_keys, 0, sizeof(hash_keys));
2380
2381                         if (!flkeys) {
2382                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2383                                 flkeys = &keys;
2384                         }
2385                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2386                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2387                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2388                         hash_keys.ports.src = flkeys->ports.src;
2389                         hash_keys.ports.dst = flkeys->ports.dst;
2390                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2391                 } else {
2392                         memset(&hash_keys, 0, sizeof(hash_keys));
2393                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2394                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2395                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2396                         hash_keys.ports.src = fl6->fl6_sport;
2397                         hash_keys.ports.dst = fl6->fl6_dport;
2398                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2399                 }
2400                 break;
2401         case 2:
2402                 memset(&hash_keys, 0, sizeof(hash_keys));
2403                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2404                 if (skb) {
2405                         struct flow_keys keys;
2406
2407                         if (!flkeys) {
2408                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2409                                 flkeys = &keys;
2410                         }
2411
2412                         /* Inner can be v4 or v6 */
2413                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2414                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2415                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2416                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2417                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2418                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2419                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2420                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2421                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2422                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2423                         } else {
2424                                 /* Same as case 0 */
2425                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2426                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2427                         }
2428                 } else {
2429                         /* Same as case 0 */
2430                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2431                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2432                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2433                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2434                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2435                 }
2436                 break;
2437         }
2438         mhash = flow_hash_from_keys(&hash_keys);
2439
2440         return mhash >> 1;
2441 }
2442
2443 /* Called with rcu held */
2444 void ip6_route_input(struct sk_buff *skb)
2445 {
2446         const struct ipv6hdr *iph = ipv6_hdr(skb);
2447         struct net *net = dev_net(skb->dev);
2448         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2449         struct ip_tunnel_info *tun_info;
2450         struct flowi6 fl6 = {
2451                 .flowi6_iif = skb->dev->ifindex,
2452                 .daddr = iph->daddr,
2453                 .saddr = iph->saddr,
2454                 .flowlabel = ip6_flowinfo(iph),
2455                 .flowi6_mark = skb->mark,
2456                 .flowi6_proto = iph->nexthdr,
2457         };
2458         struct flow_keys *flkeys = NULL, _flkeys;
2459
2460         tun_info = skb_tunnel_info(skb);
2461         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2462                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2463
2464         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2465                 flkeys = &_flkeys;
2466
2467         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2468                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2469         skb_dst_drop(skb);
2470         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2471                                                       &fl6, skb, flags));
2472 }
2473
2474 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2475                                              struct fib6_table *table,
2476                                              struct flowi6 *fl6,
2477                                              const struct sk_buff *skb,
2478                                              int flags)
2479 {
2480         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2481 }
2482
2483 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2484                                                const struct sock *sk,
2485                                                struct flowi6 *fl6, int flags)
2486 {
2487         bool any_src;
2488
2489         if (ipv6_addr_type(&fl6->daddr) &
2490             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2491                 struct dst_entry *dst;
2492
2493                 /* This function does not take refcnt on the dst */
2494                 dst = l3mdev_link_scope_lookup(net, fl6);
2495                 if (dst)
2496                         return dst;
2497         }
2498
2499         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2500
2501         flags |= RT6_LOOKUP_F_DST_NOREF;
2502         any_src = ipv6_addr_any(&fl6->saddr);
2503         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2504             (fl6->flowi6_oif && any_src))
2505                 flags |= RT6_LOOKUP_F_IFACE;
2506
2507         if (!any_src)
2508                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2509         else if (sk)
2510                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2511
2512         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2513 }
2514 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2515
2516 struct dst_entry *ip6_route_output_flags(struct net *net,
2517                                          const struct sock *sk,
2518                                          struct flowi6 *fl6,
2519                                          int flags)
2520 {
2521         struct dst_entry *dst;
2522         struct rt6_info *rt6;
2523
2524         rcu_read_lock();
2525         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2526         rt6 = (struct rt6_info *)dst;
2527         /* For dst cached in uncached_list, refcnt is already taken. */
2528         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2529                 dst = &net->ipv6.ip6_null_entry->dst;
2530                 dst_hold(dst);
2531         }
2532         rcu_read_unlock();
2533
2534         return dst;
2535 }
2536 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2537
2538 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2539 {
2540         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2541         struct net_device *loopback_dev = net->loopback_dev;
2542         struct dst_entry *new = NULL;
2543
2544         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2545                        DST_OBSOLETE_DEAD, 0);
2546         if (rt) {
2547                 rt6_info_init(rt);
2548                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2549
2550                 new = &rt->dst;
2551                 new->__use = 1;
2552                 new->input = dst_discard;
2553                 new->output = dst_discard_out;
2554
2555                 dst_copy_metrics(new, &ort->dst);
2556
2557                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2558                 rt->rt6i_gateway = ort->rt6i_gateway;
2559                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2560
2561                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2562 #ifdef CONFIG_IPV6_SUBTREES
2563                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2564 #endif
2565         }
2566
2567         dst_release(dst_orig);
2568         return new ? new : ERR_PTR(-ENOMEM);
2569 }
2570
2571 /*
2572  *      Destination cache support functions
2573  */
2574
2575 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2576 {
2577         u32 rt_cookie = 0;
2578
2579         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2580                 return false;
2581
2582         if (fib6_check_expired(f6i))
2583                 return false;
2584
2585         return true;
2586 }
2587
2588 static struct dst_entry *rt6_check(struct rt6_info *rt,
2589                                    struct fib6_info *from,
2590                                    u32 cookie)
2591 {
2592         u32 rt_cookie = 0;
2593
2594         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2595             rt_cookie != cookie)
2596                 return NULL;
2597
2598         if (rt6_check_expired(rt))
2599                 return NULL;
2600
2601         return &rt->dst;
2602 }
2603
2604 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2605                                             struct fib6_info *from,
2606                                             u32 cookie)
2607 {
2608         if (!__rt6_check_expired(rt) &&
2609             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2610             fib6_check(from, cookie))
2611                 return &rt->dst;
2612         else
2613                 return NULL;
2614 }
2615
2616 INDIRECT_CALLABLE_SCOPE struct dst_entry *ip6_dst_check(struct dst_entry *dst,
2617                                                         u32 cookie)
2618 {
2619         struct dst_entry *dst_ret;
2620         struct fib6_info *from;
2621         struct rt6_info *rt;
2622
2623         rt = container_of(dst, struct rt6_info, dst);
2624
2625         if (rt->sernum)
2626                 return rt6_is_valid(rt) ? dst : NULL;
2627
2628         rcu_read_lock();
2629
2630         /* All IPV6 dsts are created with ->obsolete set to the value
2631          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2632          * into this function always.
2633          */
2634
2635         from = rcu_dereference(rt->from);
2636
2637         if (from && (rt->rt6i_flags & RTF_PCPU ||
2638             unlikely(!list_empty(&rt->rt6i_uncached))))
2639                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2640         else
2641                 dst_ret = rt6_check(rt, from, cookie);
2642
2643         rcu_read_unlock();
2644
2645         return dst_ret;
2646 }
2647 EXPORT_INDIRECT_CALLABLE(ip6_dst_check);
2648
2649 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2650 {
2651         struct rt6_info *rt = (struct rt6_info *) dst;
2652
2653         if (rt) {
2654                 if (rt->rt6i_flags & RTF_CACHE) {
2655                         rcu_read_lock();
2656                         if (rt6_check_expired(rt)) {
2657                                 rt6_remove_exception_rt(rt);
2658                                 dst = NULL;
2659                         }
2660                         rcu_read_unlock();
2661                 } else {
2662                         dst_release(dst);
2663                         dst = NULL;
2664                 }
2665         }
2666         return dst;
2667 }
2668
2669 static void ip6_link_failure(struct sk_buff *skb)
2670 {
2671         struct rt6_info *rt;
2672
2673         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2674
2675         rt = (struct rt6_info *) skb_dst(skb);
2676         if (rt) {
2677                 rcu_read_lock();
2678                 if (rt->rt6i_flags & RTF_CACHE) {
2679                         rt6_remove_exception_rt(rt);
2680                 } else {
2681                         struct fib6_info *from;
2682                         struct fib6_node *fn;
2683
2684                         from = rcu_dereference(rt->from);
2685                         if (from) {
2686                                 fn = rcu_dereference(from->fib6_node);
2687                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2688                                         fn->fn_sernum = -1;
2689                         }
2690                 }
2691                 rcu_read_unlock();
2692         }
2693 }
2694
2695 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2696 {
2697         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2698                 struct fib6_info *from;
2699
2700                 rcu_read_lock();
2701                 from = rcu_dereference(rt0->from);
2702                 if (from)
2703                         rt0->dst.expires = from->expires;
2704                 rcu_read_unlock();
2705         }
2706
2707         dst_set_expires(&rt0->dst, timeout);
2708         rt0->rt6i_flags |= RTF_EXPIRES;
2709 }
2710
2711 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2712 {
2713         struct net *net = dev_net(rt->dst.dev);
2714
2715         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2716         rt->rt6i_flags |= RTF_MODIFIED;
2717         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2718 }
2719
2720 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2721 {
2722         return !(rt->rt6i_flags & RTF_CACHE) &&
2723                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2724 }
2725
2726 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2727                                  const struct ipv6hdr *iph, u32 mtu,
2728                                  bool confirm_neigh)
2729 {
2730         const struct in6_addr *daddr, *saddr;
2731         struct rt6_info *rt6 = (struct rt6_info *)dst;
2732
2733         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2734          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2735          * [see also comment in rt6_mtu_change_route()]
2736          */
2737
2738         if (iph) {
2739                 daddr = &iph->daddr;
2740                 saddr = &iph->saddr;
2741         } else if (sk) {
2742                 daddr = &sk->sk_v6_daddr;
2743                 saddr = &inet6_sk(sk)->saddr;
2744         } else {
2745                 daddr = NULL;
2746                 saddr = NULL;
2747         }
2748
2749         if (confirm_neigh)
2750                 dst_confirm_neigh(dst, daddr);
2751
2752         if (mtu < IPV6_MIN_MTU)
2753                 return;
2754         if (mtu >= dst_mtu(dst))
2755                 return;
2756
2757         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2758                 rt6_do_update_pmtu(rt6, mtu);
2759                 /* update rt6_ex->stamp for cache */
2760                 if (rt6->rt6i_flags & RTF_CACHE)
2761                         rt6_update_exception_stamp_rt(rt6);
2762         } else if (daddr) {
2763                 struct fib6_result res = {};
2764                 struct rt6_info *nrt6;
2765
2766                 rcu_read_lock();
2767                 res.f6i = rcu_dereference(rt6->from);
2768                 if (!res.f6i)
2769                         goto out_unlock;
2770
2771                 res.fib6_flags = res.f6i->fib6_flags;
2772                 res.fib6_type = res.f6i->fib6_type;
2773
2774                 if (res.f6i->nh) {
2775                         struct fib6_nh_match_arg arg = {
2776                                 .dev = dst->dev,
2777                                 .gw = &rt6->rt6i_gateway,
2778                         };
2779
2780                         nexthop_for_each_fib6_nh(res.f6i->nh,
2781                                                  fib6_nh_find_match, &arg);
2782
2783                         /* fib6_info uses a nexthop that does not have fib6_nh
2784                          * using the dst->dev + gw. Should be impossible.
2785                          */
2786                         if (!arg.match)
2787                                 goto out_unlock;
2788
2789                         res.nh = arg.match;
2790                 } else {
2791                         res.nh = res.f6i->fib6_nh;
2792                 }
2793
2794                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2795                 if (nrt6) {
2796                         rt6_do_update_pmtu(nrt6, mtu);
2797                         if (rt6_insert_exception(nrt6, &res))
2798                                 dst_release_immediate(&nrt6->dst);
2799                 }
2800 out_unlock:
2801                 rcu_read_unlock();
2802         }
2803 }
2804
2805 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2806                                struct sk_buff *skb, u32 mtu,
2807                                bool confirm_neigh)
2808 {
2809         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2810                              confirm_neigh);
2811 }
2812
2813 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2814                      int oif, u32 mark, kuid_t uid)
2815 {
2816         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2817         struct dst_entry *dst;
2818         struct flowi6 fl6 = {
2819                 .flowi6_oif = oif,
2820                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2821                 .daddr = iph->daddr,
2822                 .saddr = iph->saddr,
2823                 .flowlabel = ip6_flowinfo(iph),
2824                 .flowi6_uid = uid,
2825         };
2826
2827         dst = ip6_route_output(net, NULL, &fl6);
2828         if (!dst->error)
2829                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2830         dst_release(dst);
2831 }
2832 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2833
2834 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2835 {
2836         int oif = sk->sk_bound_dev_if;
2837         struct dst_entry *dst;
2838
2839         if (!oif && skb->dev)
2840                 oif = l3mdev_master_ifindex(skb->dev);
2841
2842         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2843
2844         dst = __sk_dst_get(sk);
2845         if (!dst || !dst->obsolete ||
2846             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2847                 return;
2848
2849         bh_lock_sock(sk);
2850         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2851                 ip6_datagram_dst_update(sk, false);
2852         bh_unlock_sock(sk);
2853 }
2854 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2855
2856 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2857                            const struct flowi6 *fl6)
2858 {
2859 #ifdef CONFIG_IPV6_SUBTREES
2860         struct ipv6_pinfo *np = inet6_sk(sk);
2861 #endif
2862
2863         ip6_dst_store(sk, dst,
2864                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2865                       &sk->sk_v6_daddr : NULL,
2866 #ifdef CONFIG_IPV6_SUBTREES
2867                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2868                       &np->saddr :
2869 #endif
2870                       NULL);
2871 }
2872
2873 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2874                                   struct flowi6 *fl6,
2875                                   const struct in6_addr *gw,
2876                                   struct rt6_info **ret)
2877 {
2878         const struct fib6_nh *nh = res->nh;
2879
2880         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2881             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2882                 return false;
2883
2884         /* rt_cache's gateway might be different from its 'parent'
2885          * in the case of an ip redirect.
2886          * So we keep searching in the exception table if the gateway
2887          * is different.
2888          */
2889         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2890                 struct rt6_info *rt_cache;
2891
2892                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2893                 if (rt_cache &&
2894                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2895                         *ret = rt_cache;
2896                         return true;
2897                 }
2898                 return false;
2899         }
2900         return true;
2901 }
2902
2903 struct fib6_nh_rd_arg {
2904         struct fib6_result      *res;
2905         struct flowi6           *fl6;
2906         const struct in6_addr   *gw;
2907         struct rt6_info         **ret;
2908 };
2909
2910 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2911 {
2912         struct fib6_nh_rd_arg *arg = _arg;
2913
2914         arg->res->nh = nh;
2915         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2916 }
2917
2918 /* Handle redirects */
2919 struct ip6rd_flowi {
2920         struct flowi6 fl6;
2921         struct in6_addr gateway;
2922 };
2923
2924 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
2925                                              struct fib6_table *table,
2926                                              struct flowi6 *fl6,
2927                                              const struct sk_buff *skb,
2928                                              int flags)
2929 {
2930         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2931         struct rt6_info *ret = NULL;
2932         struct fib6_result res = {};
2933         struct fib6_nh_rd_arg arg = {
2934                 .res = &res,
2935                 .fl6 = fl6,
2936                 .gw  = &rdfl->gateway,
2937                 .ret = &ret
2938         };
2939         struct fib6_info *rt;
2940         struct fib6_node *fn;
2941
2942         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2943          * this case we must match on the real ingress device, so reset it
2944          */
2945         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2946                 fl6->flowi6_oif = skb->dev->ifindex;
2947
2948         /* Get the "current" route for this destination and
2949          * check if the redirect has come from appropriate router.
2950          *
2951          * RFC 4861 specifies that redirects should only be
2952          * accepted if they come from the nexthop to the target.
2953          * Due to the way the routes are chosen, this notion
2954          * is a bit fuzzy and one might need to check all possible
2955          * routes.
2956          */
2957
2958         rcu_read_lock();
2959         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2960 restart:
2961         for_each_fib6_node_rt_rcu(fn) {
2962                 res.f6i = rt;
2963                 if (fib6_check_expired(rt))
2964                         continue;
2965                 if (rt->fib6_flags & RTF_REJECT)
2966                         break;
2967                 if (unlikely(rt->nh)) {
2968                         if (nexthop_is_blackhole(rt->nh))
2969                                 continue;
2970                         /* on match, res->nh is filled in and potentially ret */
2971                         if (nexthop_for_each_fib6_nh(rt->nh,
2972                                                      fib6_nh_redirect_match,
2973                                                      &arg))
2974                                 goto out;
2975                 } else {
2976                         res.nh = rt->fib6_nh;
2977                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2978                                                   &ret))
2979                                 goto out;
2980                 }
2981         }
2982
2983         if (!rt)
2984                 rt = net->ipv6.fib6_null_entry;
2985         else if (rt->fib6_flags & RTF_REJECT) {
2986                 ret = net->ipv6.ip6_null_entry;
2987                 goto out;
2988         }
2989
2990         if (rt == net->ipv6.fib6_null_entry) {
2991                 fn = fib6_backtrack(fn, &fl6->saddr);
2992                 if (fn)
2993                         goto restart;
2994         }
2995
2996         res.f6i = rt;
2997         res.nh = rt->fib6_nh;
2998 out:
2999         if (ret) {
3000                 ip6_hold_safe(net, &ret);
3001         } else {
3002                 res.fib6_flags = res.f6i->fib6_flags;
3003                 res.fib6_type = res.f6i->fib6_type;
3004                 ret = ip6_create_rt_rcu(&res);
3005         }
3006
3007         rcu_read_unlock();
3008
3009         trace_fib6_table_lookup(net, &res, table, fl6);
3010         return ret;
3011 };
3012
3013 static struct dst_entry *ip6_route_redirect(struct net *net,
3014                                             const struct flowi6 *fl6,
3015                                             const struct sk_buff *skb,
3016                                             const struct in6_addr *gateway)
3017 {
3018         int flags = RT6_LOOKUP_F_HAS_SADDR;
3019         struct ip6rd_flowi rdfl;
3020
3021         rdfl.fl6 = *fl6;
3022         rdfl.gateway = *gateway;
3023
3024         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3025                                 flags, __ip6_route_redirect);
3026 }
3027
3028 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3029                   kuid_t uid)
3030 {
3031         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3032         struct dst_entry *dst;
3033         struct flowi6 fl6 = {
3034                 .flowi6_iif = LOOPBACK_IFINDEX,
3035                 .flowi6_oif = oif,
3036                 .flowi6_mark = mark,
3037                 .daddr = iph->daddr,
3038                 .saddr = iph->saddr,
3039                 .flowlabel = ip6_flowinfo(iph),
3040                 .flowi6_uid = uid,
3041         };
3042
3043         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3044         rt6_do_redirect(dst, NULL, skb);
3045         dst_release(dst);
3046 }
3047 EXPORT_SYMBOL_GPL(ip6_redirect);
3048
3049 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3050 {
3051         const struct ipv6hdr *iph = ipv6_hdr(skb);
3052         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3053         struct dst_entry *dst;
3054         struct flowi6 fl6 = {
3055                 .flowi6_iif = LOOPBACK_IFINDEX,
3056                 .flowi6_oif = oif,
3057                 .daddr = msg->dest,
3058                 .saddr = iph->daddr,
3059                 .flowi6_uid = sock_net_uid(net, NULL),
3060         };
3061
3062         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3063         rt6_do_redirect(dst, NULL, skb);
3064         dst_release(dst);
3065 }
3066
3067 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3068 {
3069         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3070                      sk->sk_uid);
3071 }
3072 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3073
3074 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3075 {
3076         struct net_device *dev = dst->dev;
3077         unsigned int mtu = dst_mtu(dst);
3078         struct net *net = dev_net(dev);
3079
3080         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3081
3082         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3083                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3084
3085         /*
3086          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3087          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3088          * IPV6_MAXPLEN is also valid and means: "any MSS,
3089          * rely only on pmtu discovery"
3090          */
3091         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3092                 mtu = IPV6_MAXPLEN;
3093         return mtu;
3094 }
3095
3096 INDIRECT_CALLABLE_SCOPE unsigned int ip6_mtu(const struct dst_entry *dst)
3097 {
3098         struct inet6_dev *idev;
3099         unsigned int mtu;
3100
3101         mtu = dst_metric_raw(dst, RTAX_MTU);
3102         if (mtu)
3103                 goto out;
3104
3105         mtu = IPV6_MIN_MTU;
3106
3107         rcu_read_lock();
3108         idev = __in6_dev_get(dst->dev);
3109         if (idev)
3110                 mtu = idev->cnf.mtu6;
3111         rcu_read_unlock();
3112
3113 out:
3114         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3115
3116         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3117 }
3118 EXPORT_INDIRECT_CALLABLE(ip6_mtu);
3119
3120 /* MTU selection:
3121  * 1. mtu on route is locked - use it
3122  * 2. mtu from nexthop exception
3123  * 3. mtu from egress device
3124  *
3125  * based on ip6_dst_mtu_forward and exception logic of
3126  * rt6_find_cached_rt; called with rcu_read_lock
3127  */
3128 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3129                       const struct in6_addr *daddr,
3130                       const struct in6_addr *saddr)
3131 {
3132         const struct fib6_nh *nh = res->nh;
3133         struct fib6_info *f6i = res->f6i;
3134         struct inet6_dev *idev;
3135         struct rt6_info *rt;
3136         u32 mtu = 0;
3137
3138         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3139                 mtu = f6i->fib6_pmtu;
3140                 if (mtu)
3141                         goto out;
3142         }
3143
3144         rt = rt6_find_cached_rt(res, daddr, saddr);
3145         if (unlikely(rt)) {
3146                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3147         } else {
3148                 struct net_device *dev = nh->fib_nh_dev;
3149
3150                 mtu = IPV6_MIN_MTU;
3151                 idev = __in6_dev_get(dev);
3152                 if (idev && idev->cnf.mtu6 > mtu)
3153                         mtu = idev->cnf.mtu6;
3154         }
3155
3156         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3157 out:
3158         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3159 }
3160
3161 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3162                                   struct flowi6 *fl6)
3163 {
3164         struct dst_entry *dst;
3165         struct rt6_info *rt;
3166         struct inet6_dev *idev = in6_dev_get(dev);
3167         struct net *net = dev_net(dev);
3168
3169         if (unlikely(!idev))
3170                 return ERR_PTR(-ENODEV);
3171
3172         rt = ip6_dst_alloc(net, dev, 0);
3173         if (unlikely(!rt)) {
3174                 in6_dev_put(idev);
3175                 dst = ERR_PTR(-ENOMEM);
3176                 goto out;
3177         }
3178
3179         rt->dst.input = ip6_input;
3180         rt->dst.output  = ip6_output;
3181         rt->rt6i_gateway  = fl6->daddr;
3182         rt->rt6i_dst.addr = fl6->daddr;
3183         rt->rt6i_dst.plen = 128;
3184         rt->rt6i_idev     = idev;
3185         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3186
3187         /* Add this dst into uncached_list so that rt6_disable_ip() can
3188          * do proper release of the net_device
3189          */
3190         rt6_uncached_list_add(rt);
3191         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3192
3193         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3194
3195 out:
3196         return dst;
3197 }
3198
3199 static int ip6_dst_gc(struct dst_ops *ops)
3200 {
3201         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3202         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3203         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3204         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3205         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3206         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3207         int entries;
3208
3209         entries = dst_entries_get_fast(ops);
3210         if (entries > rt_max_size)
3211                 entries = dst_entries_get_slow(ops);
3212
3213         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3214             entries <= rt_max_size)
3215                 goto out;
3216
3217         net->ipv6.ip6_rt_gc_expire++;
3218         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3219         entries = dst_entries_get_slow(ops);
3220         if (entries < ops->gc_thresh)
3221                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3222 out:
3223         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3224         return entries > rt_max_size;
3225 }
3226
3227 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3228                                const struct in6_addr *gw_addr, u32 tbid,
3229                                int flags, struct fib6_result *res)
3230 {
3231         struct flowi6 fl6 = {
3232                 .flowi6_oif = cfg->fc_ifindex,
3233                 .daddr = *gw_addr,
3234                 .saddr = cfg->fc_prefsrc,
3235         };
3236         struct fib6_table *table;
3237         int err;
3238
3239         table = fib6_get_table(net, tbid);
3240         if (!table)
3241                 return -EINVAL;
3242
3243         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3244                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3245
3246         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3247
3248         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3249         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3250                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3251                                  cfg->fc_ifindex != 0, NULL, flags);
3252
3253         return err;
3254 }
3255
3256 static int ip6_route_check_nh_onlink(struct net *net,
3257                                      struct fib6_config *cfg,
3258                                      const struct net_device *dev,
3259                                      struct netlink_ext_ack *extack)
3260 {
3261         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3262         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3263         struct fib6_result res = {};
3264         int err;
3265
3266         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3267         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3268             /* ignore match if it is the default route */
3269             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3270             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3271                 NL_SET_ERR_MSG(extack,
3272                                "Nexthop has invalid gateway or device mismatch");
3273                 err = -EINVAL;
3274         }
3275
3276         return err;
3277 }
3278
3279 static int ip6_route_check_nh(struct net *net,
3280                               struct fib6_config *cfg,
3281                               struct net_device **_dev,
3282                               struct inet6_dev **idev)
3283 {
3284         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3285         struct net_device *dev = _dev ? *_dev : NULL;
3286         int flags = RT6_LOOKUP_F_IFACE;
3287         struct fib6_result res = {};
3288         int err = -EHOSTUNREACH;
3289
3290         if (cfg->fc_table) {
3291                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3292                                           cfg->fc_table, flags, &res);
3293                 /* gw_addr can not require a gateway or resolve to a reject
3294                  * route. If a device is given, it must match the result.
3295                  */
3296                 if (err || res.fib6_flags & RTF_REJECT ||
3297                     res.nh->fib_nh_gw_family ||
3298                     (dev && dev != res.nh->fib_nh_dev))
3299                         err = -EHOSTUNREACH;
3300         }
3301
3302         if (err < 0) {
3303                 struct flowi6 fl6 = {
3304                         .flowi6_oif = cfg->fc_ifindex,
3305                         .daddr = *gw_addr,
3306                 };
3307
3308                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3309                 if (err || res.fib6_flags & RTF_REJECT ||
3310                     res.nh->fib_nh_gw_family)
3311                         err = -EHOSTUNREACH;
3312
3313                 if (err)
3314                         return err;
3315
3316                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3317                                  cfg->fc_ifindex != 0, NULL, flags);
3318         }
3319
3320         err = 0;
3321         if (dev) {
3322                 if (dev != res.nh->fib_nh_dev)
3323                         err = -EHOSTUNREACH;
3324         } else {
3325                 *_dev = dev = res.nh->fib_nh_dev;
3326                 dev_hold(dev);
3327                 *idev = in6_dev_get(dev);
3328         }
3329
3330         return err;
3331 }
3332
3333 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3334                            struct net_device **_dev, struct inet6_dev **idev,
3335                            struct netlink_ext_ack *extack)
3336 {
3337         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3338         int gwa_type = ipv6_addr_type(gw_addr);
3339         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3340         const struct net_device *dev = *_dev;
3341         bool need_addr_check = !dev;
3342         int err = -EINVAL;
3343
3344         /* if gw_addr is local we will fail to detect this in case
3345          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3346          * will return already-added prefix route via interface that
3347          * prefix route was assigned to, which might be non-loopback.
3348          */
3349         if (dev &&
3350             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3351                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3352                 goto out;
3353         }
3354
3355         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3356                 /* IPv6 strictly inhibits using not link-local
3357                  * addresses as nexthop address.
3358                  * Otherwise, router will not able to send redirects.
3359                  * It is very good, but in some (rare!) circumstances
3360                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3361                  * some exceptions. --ANK
3362                  * We allow IPv4-mapped nexthops to support RFC4798-type
3363                  * addressing
3364                  */
3365                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3366                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3367                         goto out;
3368                 }
3369
3370                 rcu_read_lock();
3371
3372                 if (cfg->fc_flags & RTNH_F_ONLINK)
3373                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3374                 else
3375                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3376
3377                 rcu_read_unlock();
3378
3379                 if (err)
3380                         goto out;
3381         }
3382
3383         /* reload in case device was changed */
3384         dev = *_dev;
3385
3386         err = -EINVAL;
3387         if (!dev) {
3388                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3389                 goto out;
3390         } else if (dev->flags & IFF_LOOPBACK) {
3391                 NL_SET_ERR_MSG(extack,
3392                                "Egress device can not be loopback device for this route");
3393                 goto out;
3394         }
3395
3396         /* if we did not check gw_addr above, do so now that the
3397          * egress device has been resolved.
3398          */
3399         if (need_addr_check &&
3400             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3401                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3402                 goto out;
3403         }
3404
3405         err = 0;
3406 out:
3407         return err;
3408 }
3409
3410 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3411 {
3412         if ((flags & RTF_REJECT) ||
3413             (dev && (dev->flags & IFF_LOOPBACK) &&
3414              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3415              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3416                 return true;
3417
3418         return false;
3419 }
3420
3421 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3422                  struct fib6_config *cfg, gfp_t gfp_flags,
3423                  struct netlink_ext_ack *extack)
3424 {
3425         struct net_device *dev = NULL;
3426         struct inet6_dev *idev = NULL;
3427         int addr_type;
3428         int err;
3429
3430         fib6_nh->fib_nh_family = AF_INET6;
3431 #ifdef CONFIG_IPV6_ROUTER_PREF
3432         fib6_nh->last_probe = jiffies;
3433 #endif
3434         if (cfg->fc_is_fdb) {
3435                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3436                 fib6_nh->fib_nh_gw_family = AF_INET6;
3437                 return 0;
3438         }
3439
3440         err = -ENODEV;
3441         if (cfg->fc_ifindex) {
3442                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3443                 if (!dev)
3444                         goto out;
3445                 idev = in6_dev_get(dev);
3446                 if (!idev)
3447                         goto out;
3448         }
3449
3450         if (cfg->fc_flags & RTNH_F_ONLINK) {
3451                 if (!dev) {
3452                         NL_SET_ERR_MSG(extack,
3453                                        "Nexthop device required for onlink");
3454                         goto out;
3455                 }
3456
3457                 if (!(dev->flags & IFF_UP)) {
3458                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3459                         err = -ENETDOWN;
3460                         goto out;
3461                 }
3462
3463                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3464         }
3465
3466         fib6_nh->fib_nh_weight = 1;
3467
3468         /* We cannot add true routes via loopback here,
3469          * they would result in kernel looping; promote them to reject routes
3470          */
3471         addr_type = ipv6_addr_type(&cfg->fc_dst);
3472         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3473                 /* hold loopback dev/idev if we haven't done so. */
3474                 if (dev != net->loopback_dev) {
3475                         if (dev) {
3476                                 dev_put(dev);
3477                                 in6_dev_put(idev);
3478                         }
3479                         dev = net->loopback_dev;
3480                         dev_hold(dev);
3481                         idev = in6_dev_get(dev);
3482                         if (!idev) {
3483                                 err = -ENODEV;
3484                                 goto out;
3485                         }
3486                 }
3487                 goto pcpu_alloc;
3488         }
3489
3490         if (cfg->fc_flags & RTF_GATEWAY) {
3491                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3492                 if (err)
3493                         goto out;
3494
3495                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3496                 fib6_nh->fib_nh_gw_family = AF_INET6;
3497         }
3498
3499         err = -ENODEV;
3500         if (!dev)
3501                 goto out;
3502
3503         if (idev->cnf.disable_ipv6) {
3504                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3505                 err = -EACCES;
3506                 goto out;
3507         }
3508
3509         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3510                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3511                 err = -ENETDOWN;
3512                 goto out;
3513         }
3514
3515         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3516             !netif_carrier_ok(dev))
3517                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3518
3519         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3520                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3521         if (err)
3522                 goto out;
3523
3524 pcpu_alloc:
3525         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3526         if (!fib6_nh->rt6i_pcpu) {
3527                 err = -ENOMEM;
3528                 goto out;
3529         }
3530
3531         fib6_nh->fib_nh_dev = dev;
3532         fib6_nh->fib_nh_oif = dev->ifindex;
3533         err = 0;
3534 out:
3535         if (idev)
3536                 in6_dev_put(idev);
3537
3538         if (err) {
3539                 lwtstate_put(fib6_nh->fib_nh_lws);
3540                 fib6_nh->fib_nh_lws = NULL;
3541                 if (dev)
3542                         dev_put(dev);
3543         }
3544
3545         return err;
3546 }
3547
3548 void fib6_nh_release(struct fib6_nh *fib6_nh)
3549 {
3550         struct rt6_exception_bucket *bucket;
3551
3552         rcu_read_lock();
3553
3554         fib6_nh_flush_exceptions(fib6_nh, NULL);
3555         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3556         if (bucket) {
3557                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3558                 kfree(bucket);
3559         }
3560
3561         rcu_read_unlock();
3562
3563         if (fib6_nh->rt6i_pcpu) {
3564                 int cpu;
3565
3566                 for_each_possible_cpu(cpu) {
3567                         struct rt6_info **ppcpu_rt;
3568                         struct rt6_info *pcpu_rt;
3569
3570                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3571                         pcpu_rt = *ppcpu_rt;
3572                         if (pcpu_rt) {
3573                                 dst_dev_put(&pcpu_rt->dst);
3574                                 dst_release(&pcpu_rt->dst);
3575                                 *ppcpu_rt = NULL;
3576                         }
3577                 }
3578
3579                 free_percpu(fib6_nh->rt6i_pcpu);
3580         }
3581
3582         fib_nh_common_release(&fib6_nh->nh_common);
3583 }
3584
3585 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3586                                               gfp_t gfp_flags,
3587                                               struct netlink_ext_ack *extack)
3588 {
3589         struct net *net = cfg->fc_nlinfo.nl_net;
3590         struct fib6_info *rt = NULL;
3591         struct nexthop *nh = NULL;
3592         struct fib6_table *table;
3593         struct fib6_nh *fib6_nh;
3594         int err = -EINVAL;
3595         int addr_type;
3596
3597         /* RTF_PCPU is an internal flag; can not be set by userspace */
3598         if (cfg->fc_flags & RTF_PCPU) {
3599                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3600                 goto out;
3601         }
3602
3603         /* RTF_CACHE is an internal flag; can not be set by userspace */
3604         if (cfg->fc_flags & RTF_CACHE) {
3605                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3606                 goto out;
3607         }
3608
3609         if (cfg->fc_type > RTN_MAX) {
3610                 NL_SET_ERR_MSG(extack, "Invalid route type");
3611                 goto out;
3612         }
3613
3614         if (cfg->fc_dst_len > 128) {
3615                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3616                 goto out;
3617         }
3618         if (cfg->fc_src_len > 128) {
3619                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3620                 goto out;
3621         }
3622 #ifndef CONFIG_IPV6_SUBTREES
3623         if (cfg->fc_src_len) {
3624                 NL_SET_ERR_MSG(extack,
3625                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3626                 goto out;
3627         }
3628 #endif
3629         if (cfg->fc_nh_id) {
3630                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3631                 if (!nh) {
3632                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3633                         goto out;
3634                 }
3635                 err = fib6_check_nexthop(nh, cfg, extack);
3636                 if (err)
3637                         goto out;
3638         }
3639
3640         err = -ENOBUFS;
3641         if (cfg->fc_nlinfo.nlh &&
3642             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3643                 table = fib6_get_table(net, cfg->fc_table);
3644                 if (!table) {
3645                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3646                         table = fib6_new_table(net, cfg->fc_table);
3647                 }
3648         } else {
3649                 table = fib6_new_table(net, cfg->fc_table);
3650         }
3651
3652         if (!table)
3653                 goto out;
3654
3655         err = -ENOMEM;
3656         rt = fib6_info_alloc(gfp_flags, !nh);
3657         if (!rt)
3658                 goto out;
3659
3660         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3661                                                extack);
3662         if (IS_ERR(rt->fib6_metrics)) {
3663                 err = PTR_ERR(rt->fib6_metrics);
3664                 /* Do not leave garbage there. */
3665                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3666                 goto out;
3667         }
3668
3669         if (cfg->fc_flags & RTF_ADDRCONF)
3670                 rt->dst_nocount = true;
3671
3672         if (cfg->fc_flags & RTF_EXPIRES)
3673                 fib6_set_expires(rt, jiffies +
3674                                 clock_t_to_jiffies(cfg->fc_expires));
3675         else
3676                 fib6_clean_expires(rt);
3677
3678         if (cfg->fc_protocol == RTPROT_UNSPEC)
3679                 cfg->fc_protocol = RTPROT_BOOT;
3680         rt->fib6_protocol = cfg->fc_protocol;
3681
3682         rt->fib6_table = table;
3683         rt->fib6_metric = cfg->fc_metric;
3684         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3685         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3686
3687         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3688         rt->fib6_dst.plen = cfg->fc_dst_len;
3689
3690 #ifdef CONFIG_IPV6_SUBTREES
3691         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3692         rt->fib6_src.plen = cfg->fc_src_len;
3693 #endif
3694         if (nh) {
3695                 if (rt->fib6_src.plen) {
3696                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3697                         goto out;
3698                 }
3699                 if (!nexthop_get(nh)) {
3700                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3701                         goto out;
3702                 }
3703                 rt->nh = nh;
3704                 fib6_nh = nexthop_fib6_nh(rt->nh);
3705         } else {
3706                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3707                 if (err)
3708                         goto out;
3709
3710                 fib6_nh = rt->fib6_nh;
3711
3712                 /* We cannot add true routes via loopback here, they would
3713                  * result in kernel looping; promote them to reject routes
3714                  */
3715                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3716                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3717                                    addr_type))
3718                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3719         }
3720
3721         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3722                 struct net_device *dev = fib6_nh->fib_nh_dev;
3723
3724                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3725                         NL_SET_ERR_MSG(extack, "Invalid source address");
3726                         err = -EINVAL;
3727                         goto out;
3728                 }
3729                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3730                 rt->fib6_prefsrc.plen = 128;
3731         } else
3732                 rt->fib6_prefsrc.plen = 0;
3733
3734         return rt;
3735 out:
3736         fib6_info_release(rt);
3737         return ERR_PTR(err);
3738 }
3739
3740 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3741                   struct netlink_ext_ack *extack)
3742 {
3743         struct fib6_info *rt;
3744         int err;
3745
3746         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3747         if (IS_ERR(rt))
3748                 return PTR_ERR(rt);
3749
3750         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3751         fib6_info_release(rt);
3752
3753         return err;
3754 }
3755
3756 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3757 {
3758         struct net *net = info->nl_net;
3759         struct fib6_table *table;
3760         int err;
3761
3762         if (rt == net->ipv6.fib6_null_entry) {
3763                 err = -ENOENT;
3764                 goto out;
3765         }
3766
3767         table = rt->fib6_table;
3768         spin_lock_bh(&table->tb6_lock);
3769         err = fib6_del(rt, info);
3770         spin_unlock_bh(&table->tb6_lock);
3771
3772 out:
3773         fib6_info_release(rt);
3774         return err;
3775 }
3776
3777 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3778 {
3779         struct nl_info info = {
3780                 .nl_net = net,
3781                 .skip_notify = skip_notify
3782         };
3783
3784         return __ip6_del_rt(rt, &info);
3785 }
3786
3787 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3788 {
3789         struct nl_info *info = &cfg->fc_nlinfo;
3790         struct net *net = info->nl_net;
3791         struct sk_buff *skb = NULL;
3792         struct fib6_table *table;
3793         int err = -ENOENT;
3794
3795         if (rt == net->ipv6.fib6_null_entry)
3796                 goto out_put;
3797         table = rt->fib6_table;
3798         spin_lock_bh(&table->tb6_lock);
3799
3800         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3801                 struct fib6_info *sibling, *next_sibling;
3802                 struct fib6_node *fn;
3803
3804                 /* prefer to send a single notification with all hops */
3805                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3806                 if (skb) {
3807                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3808
3809                         if (rt6_fill_node(net, skb, rt, NULL,
3810                                           NULL, NULL, 0, RTM_DELROUTE,
3811                                           info->portid, seq, 0) < 0) {
3812                                 kfree_skb(skb);
3813                                 skb = NULL;
3814                         } else
3815                                 info->skip_notify = 1;
3816                 }
3817
3818                 /* 'rt' points to the first sibling route. If it is not the
3819                  * leaf, then we do not need to send a notification. Otherwise,
3820                  * we need to check if the last sibling has a next route or not
3821                  * and emit a replace or delete notification, respectively.
3822                  */
3823                 info->skip_notify_kernel = 1;
3824                 fn = rcu_dereference_protected(rt->fib6_node,
3825                                             lockdep_is_held(&table->tb6_lock));
3826                 if (rcu_access_pointer(fn->leaf) == rt) {
3827                         struct fib6_info *last_sibling, *replace_rt;
3828
3829                         last_sibling = list_last_entry(&rt->fib6_siblings,
3830                                                        struct fib6_info,
3831                                                        fib6_siblings);
3832                         replace_rt = rcu_dereference_protected(
3833                                             last_sibling->fib6_next,
3834                                             lockdep_is_held(&table->tb6_lock));
3835                         if (replace_rt)
3836                                 call_fib6_entry_notifiers_replace(net,
3837                                                                   replace_rt);
3838                         else
3839                                 call_fib6_multipath_entry_notifiers(net,
3840                                                        FIB_EVENT_ENTRY_DEL,
3841                                                        rt, rt->fib6_nsiblings,
3842                                                        NULL);
3843                 }
3844                 list_for_each_entry_safe(sibling, next_sibling,
3845                                          &rt->fib6_siblings,
3846                                          fib6_siblings) {
3847                         err = fib6_del(sibling, info);
3848                         if (err)
3849                                 goto out_unlock;
3850                 }
3851         }
3852
3853         err = fib6_del(rt, info);
3854 out_unlock:
3855         spin_unlock_bh(&table->tb6_lock);
3856 out_put:
3857         fib6_info_release(rt);
3858
3859         if (skb) {
3860                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3861                             info->nlh, gfp_any());
3862         }
3863         return err;
3864 }
3865
3866 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3867 {
3868         int rc = -ESRCH;
3869
3870         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3871                 goto out;
3872
3873         if (cfg->fc_flags & RTF_GATEWAY &&
3874             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3875                 goto out;
3876
3877         rc = rt6_remove_exception_rt(rt);
3878 out:
3879         return rc;
3880 }
3881
3882 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3883                              struct fib6_nh *nh)
3884 {
3885         struct fib6_result res = {
3886                 .f6i = rt,
3887                 .nh = nh,
3888         };
3889         struct rt6_info *rt_cache;
3890
3891         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3892         if (rt_cache)
3893                 return __ip6_del_cached_rt(rt_cache, cfg);
3894
3895         return 0;
3896 }
3897
3898 struct fib6_nh_del_cached_rt_arg {
3899         struct fib6_config *cfg;
3900         struct fib6_info *f6i;
3901 };
3902
3903 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3904 {
3905         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3906         int rc;
3907
3908         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3909         return rc != -ESRCH ? rc : 0;
3910 }
3911
3912 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3913 {
3914         struct fib6_nh_del_cached_rt_arg arg = {
3915                 .cfg = cfg,
3916                 .f6i = f6i
3917         };
3918
3919         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3920 }
3921
3922 static int ip6_route_del(struct fib6_config *cfg,
3923                          struct netlink_ext_ack *extack)
3924 {
3925         struct fib6_table *table;
3926         struct fib6_info *rt;
3927         struct fib6_node *fn;
3928         int err = -ESRCH;
3929
3930         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3931         if (!table) {
3932                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3933                 return err;
3934         }
3935
3936         rcu_read_lock();
3937
3938         fn = fib6_locate(&table->tb6_root,
3939                          &cfg->fc_dst, cfg->fc_dst_len,
3940                          &cfg->fc_src, cfg->fc_src_len,
3941                          !(cfg->fc_flags & RTF_CACHE));
3942
3943         if (fn) {
3944                 for_each_fib6_node_rt_rcu(fn) {
3945                         struct fib6_nh *nh;
3946
3947                         if (rt->nh && cfg->fc_nh_id &&
3948                             rt->nh->id != cfg->fc_nh_id)
3949                                 continue;
3950
3951                         if (cfg->fc_flags & RTF_CACHE) {
3952                                 int rc = 0;
3953
3954                                 if (rt->nh) {
3955                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3956                                 } else if (cfg->fc_nh_id) {
3957                                         continue;
3958                                 } else {
3959                                         nh = rt->fib6_nh;
3960                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3961                                 }
3962                                 if (rc != -ESRCH) {
3963                                         rcu_read_unlock();
3964                                         return rc;
3965                                 }
3966                                 continue;
3967                         }
3968
3969                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3970                                 continue;
3971                         if (cfg->fc_protocol &&
3972                             cfg->fc_protocol != rt->fib6_protocol)
3973                                 continue;
3974
3975                         if (rt->nh) {
3976                                 if (!fib6_info_hold_safe(rt))
3977                                         continue;
3978                                 rcu_read_unlock();
3979
3980                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3981                         }
3982                         if (cfg->fc_nh_id)
3983                                 continue;
3984
3985                         nh = rt->fib6_nh;
3986                         if (cfg->fc_ifindex &&
3987                             (!nh->fib_nh_dev ||
3988                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3989                                 continue;
3990                         if (cfg->fc_flags & RTF_GATEWAY &&
3991                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3992                                 continue;
3993                         if (!fib6_info_hold_safe(rt))
3994                                 continue;
3995                         rcu_read_unlock();
3996
3997                         /* if gateway was specified only delete the one hop */
3998                         if (cfg->fc_flags & RTF_GATEWAY)
3999                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4000
4001                         return __ip6_del_rt_siblings(rt, cfg);
4002                 }
4003         }
4004         rcu_read_unlock();
4005
4006         return err;
4007 }
4008
4009 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4010 {
4011         struct netevent_redirect netevent;
4012         struct rt6_info *rt, *nrt = NULL;
4013         struct fib6_result res = {};
4014         struct ndisc_options ndopts;
4015         struct inet6_dev *in6_dev;
4016         struct neighbour *neigh;
4017         struct rd_msg *msg;
4018         int optlen, on_link;
4019         u8 *lladdr;
4020
4021         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4022         optlen -= sizeof(*msg);
4023
4024         if (optlen < 0) {
4025                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4026                 return;
4027         }
4028
4029         msg = (struct rd_msg *)icmp6_hdr(skb);
4030
4031         if (ipv6_addr_is_multicast(&msg->dest)) {
4032                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4033                 return;
4034         }
4035
4036         on_link = 0;
4037         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4038                 on_link = 1;
4039         } else if (ipv6_addr_type(&msg->target) !=
4040                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4041                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4042                 return;
4043         }
4044
4045         in6_dev = __in6_dev_get(skb->dev);
4046         if (!in6_dev)
4047                 return;
4048         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4049                 return;
4050
4051         /* RFC2461 8.1:
4052          *      The IP source address of the Redirect MUST be the same as the current
4053          *      first-hop router for the specified ICMP Destination Address.
4054          */
4055
4056         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4057                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4058                 return;
4059         }
4060
4061         lladdr = NULL;
4062         if (ndopts.nd_opts_tgt_lladdr) {
4063                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4064                                              skb->dev);
4065                 if (!lladdr) {
4066                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4067                         return;
4068                 }
4069         }
4070
4071         rt = (struct rt6_info *) dst;
4072         if (rt->rt6i_flags & RTF_REJECT) {
4073                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4074                 return;
4075         }
4076
4077         /* Redirect received -> path was valid.
4078          * Look, redirects are sent only in response to data packets,
4079          * so that this nexthop apparently is reachable. --ANK
4080          */
4081         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4082
4083         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4084         if (!neigh)
4085                 return;
4086
4087         /*
4088          *      We have finally decided to accept it.
4089          */
4090
4091         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4092                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4093                      NEIGH_UPDATE_F_OVERRIDE|
4094                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4095                                      NEIGH_UPDATE_F_ISROUTER)),
4096                      NDISC_REDIRECT, &ndopts);
4097
4098         rcu_read_lock();
4099         res.f6i = rcu_dereference(rt->from);
4100         if (!res.f6i)
4101                 goto out;
4102
4103         if (res.f6i->nh) {
4104                 struct fib6_nh_match_arg arg = {
4105                         .dev = dst->dev,
4106                         .gw = &rt->rt6i_gateway,
4107                 };
4108
4109                 nexthop_for_each_fib6_nh(res.f6i->nh,
4110                                          fib6_nh_find_match, &arg);
4111
4112                 /* fib6_info uses a nexthop that does not have fib6_nh
4113                  * using the dst->dev. Should be impossible
4114                  */
4115                 if (!arg.match)
4116                         goto out;
4117                 res.nh = arg.match;
4118         } else {
4119                 res.nh = res.f6i->fib6_nh;
4120         }
4121
4122         res.fib6_flags = res.f6i->fib6_flags;
4123         res.fib6_type = res.f6i->fib6_type;
4124         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4125         if (!nrt)
4126                 goto out;
4127
4128         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4129         if (on_link)
4130                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4131
4132         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4133
4134         /* rt6_insert_exception() will take care of duplicated exceptions */
4135         if (rt6_insert_exception(nrt, &res)) {
4136                 dst_release_immediate(&nrt->dst);
4137                 goto out;
4138         }
4139
4140         netevent.old = &rt->dst;
4141         netevent.new = &nrt->dst;
4142         netevent.daddr = &msg->dest;
4143         netevent.neigh = neigh;
4144         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4145
4146 out:
4147         rcu_read_unlock();
4148         neigh_release(neigh);
4149 }
4150
4151 #ifdef CONFIG_IPV6_ROUTE_INFO
4152 static struct fib6_info *rt6_get_route_info(struct net *net,
4153                                            const struct in6_addr *prefix, int prefixlen,
4154                                            const struct in6_addr *gwaddr,
4155                                            struct net_device *dev)
4156 {
4157         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4158         int ifindex = dev->ifindex;
4159         struct fib6_node *fn;
4160         struct fib6_info *rt = NULL;
4161         struct fib6_table *table;
4162
4163         table = fib6_get_table(net, tb_id);
4164         if (!table)
4165                 return NULL;
4166
4167         rcu_read_lock();
4168         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4169         if (!fn)
4170                 goto out;
4171
4172         for_each_fib6_node_rt_rcu(fn) {
4173                 /* these routes do not use nexthops */
4174                 if (rt->nh)
4175                         continue;
4176                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4177                         continue;
4178                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4179                     !rt->fib6_nh->fib_nh_gw_family)
4180                         continue;
4181                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4182                         continue;
4183                 if (!fib6_info_hold_safe(rt))
4184                         continue;
4185                 break;
4186         }
4187 out:
4188         rcu_read_unlock();
4189         return rt;
4190 }
4191
4192 static struct fib6_info *rt6_add_route_info(struct net *net,
4193                                            const struct in6_addr *prefix, int prefixlen,
4194                                            const struct in6_addr *gwaddr,
4195                                            struct net_device *dev,
4196                                            unsigned int pref)
4197 {
4198         struct fib6_config cfg = {
4199                 .fc_metric      = IP6_RT_PRIO_USER,
4200                 .fc_ifindex     = dev->ifindex,
4201                 .fc_dst_len     = prefixlen,
4202                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4203                                   RTF_UP | RTF_PREF(pref),
4204                 .fc_protocol = RTPROT_RA,
4205                 .fc_type = RTN_UNICAST,
4206                 .fc_nlinfo.portid = 0,
4207                 .fc_nlinfo.nlh = NULL,
4208                 .fc_nlinfo.nl_net = net,
4209         };
4210
4211         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4212         cfg.fc_dst = *prefix;
4213         cfg.fc_gateway = *gwaddr;
4214
4215         /* We should treat it as a default route if prefix length is 0. */
4216         if (!prefixlen)
4217                 cfg.fc_flags |= RTF_DEFAULT;
4218
4219         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4220
4221         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4222 }
4223 #endif
4224
4225 struct fib6_info *rt6_get_dflt_router(struct net *net,
4226                                      const struct in6_addr *addr,
4227                                      struct net_device *dev)
4228 {
4229         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4230         struct fib6_info *rt;
4231         struct fib6_table *table;
4232
4233         table = fib6_get_table(net, tb_id);
4234         if (!table)
4235                 return NULL;
4236
4237         rcu_read_lock();
4238         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4239                 struct fib6_nh *nh;
4240
4241                 /* RA routes do not use nexthops */
4242                 if (rt->nh)
4243                         continue;
4244
4245                 nh = rt->fib6_nh;
4246                 if (dev == nh->fib_nh_dev &&
4247                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4248                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4249                         break;
4250         }
4251         if (rt && !fib6_info_hold_safe(rt))
4252                 rt = NULL;
4253         rcu_read_unlock();
4254         return rt;
4255 }
4256
4257 struct fib6_info *rt6_add_dflt_router(struct net *net,
4258                                      const struct in6_addr *gwaddr,
4259                                      struct net_device *dev,
4260                                      unsigned int pref,
4261                                      u32 defrtr_usr_metric)
4262 {
4263         struct fib6_config cfg = {
4264                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4265                 .fc_metric      = defrtr_usr_metric,
4266                 .fc_ifindex     = dev->ifindex,
4267                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4268                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4269                 .fc_protocol = RTPROT_RA,
4270                 .fc_type = RTN_UNICAST,
4271                 .fc_nlinfo.portid = 0,
4272                 .fc_nlinfo.nlh = NULL,
4273                 .fc_nlinfo.nl_net = net,
4274         };
4275
4276         cfg.fc_gateway = *gwaddr;
4277
4278         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4279                 struct fib6_table *table;
4280
4281                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4282                 if (table)
4283                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4284         }
4285
4286         return rt6_get_dflt_router(net, gwaddr, dev);
4287 }
4288
4289 static void __rt6_purge_dflt_routers(struct net *net,
4290                                      struct fib6_table *table)
4291 {
4292         struct fib6_info *rt;
4293
4294 restart:
4295         rcu_read_lock();
4296         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4297                 struct net_device *dev = fib6_info_nh_dev(rt);
4298                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4299
4300                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4301                     (!idev || idev->cnf.accept_ra != 2) &&
4302                     fib6_info_hold_safe(rt)) {
4303                         rcu_read_unlock();
4304                         ip6_del_rt(net, rt, false);
4305                         goto restart;
4306                 }
4307         }
4308         rcu_read_unlock();
4309
4310         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4311 }
4312
4313 void rt6_purge_dflt_routers(struct net *net)
4314 {
4315         struct fib6_table *table;
4316         struct hlist_head *head;
4317         unsigned int h;
4318
4319         rcu_read_lock();
4320
4321         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4322                 head = &net->ipv6.fib_table_hash[h];
4323                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4324                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4325                                 __rt6_purge_dflt_routers(net, table);
4326                 }
4327         }
4328
4329         rcu_read_unlock();
4330 }
4331
4332 static void rtmsg_to_fib6_config(struct net *net,
4333                                  struct in6_rtmsg *rtmsg,
4334                                  struct fib6_config *cfg)
4335 {
4336         *cfg = (struct fib6_config){
4337                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4338                          : RT6_TABLE_MAIN,
4339                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4340                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4341                 .fc_expires = rtmsg->rtmsg_info,
4342                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4343                 .fc_src_len = rtmsg->rtmsg_src_len,
4344                 .fc_flags = rtmsg->rtmsg_flags,
4345                 .fc_type = rtmsg->rtmsg_type,
4346
4347                 .fc_nlinfo.nl_net = net,
4348
4349                 .fc_dst = rtmsg->rtmsg_dst,
4350                 .fc_src = rtmsg->rtmsg_src,
4351                 .fc_gateway = rtmsg->rtmsg_gateway,
4352         };
4353 }
4354
4355 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4356 {
4357         struct fib6_config cfg;
4358         int err;
4359
4360         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4361                 return -EINVAL;
4362         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4363                 return -EPERM;
4364
4365         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4366
4367         rtnl_lock();
4368         switch (cmd) {
4369         case SIOCADDRT:
4370                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4371                 break;
4372         case SIOCDELRT:
4373                 err = ip6_route_del(&cfg, NULL);
4374                 break;
4375         }
4376         rtnl_unlock();
4377         return err;
4378 }
4379
4380 /*
4381  *      Drop the packet on the floor
4382  */
4383
4384 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4385 {
4386         struct dst_entry *dst = skb_dst(skb);
4387         struct net *net = dev_net(dst->dev);
4388         struct inet6_dev *idev;
4389         int type;
4390
4391         if (netif_is_l3_master(skb->dev) &&
4392             dst->dev == net->loopback_dev)
4393                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4394         else
4395                 idev = ip6_dst_idev(dst);
4396
4397         switch (ipstats_mib_noroutes) {
4398         case IPSTATS_MIB_INNOROUTES:
4399                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4400                 if (type == IPV6_ADDR_ANY) {
4401                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4402                         break;
4403                 }
4404                 fallthrough;
4405         case IPSTATS_MIB_OUTNOROUTES:
4406                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4407                 break;
4408         }
4409
4410         /* Start over by dropping the dst for l3mdev case */
4411         if (netif_is_l3_master(skb->dev))
4412                 skb_dst_drop(skb);
4413
4414         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4415         kfree_skb(skb);
4416         return 0;
4417 }
4418
4419 static int ip6_pkt_discard(struct sk_buff *skb)
4420 {
4421         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4422 }
4423
4424 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4425 {
4426         skb->dev = skb_dst(skb)->dev;
4427         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4428 }
4429
4430 static int ip6_pkt_prohibit(struct sk_buff *skb)
4431 {
4432         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4433 }
4434
4435 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4436 {
4437         skb->dev = skb_dst(skb)->dev;
4438         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4439 }
4440
4441 /*
4442  *      Allocate a dst for local (unicast / anycast) address.
4443  */
4444
4445 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4446                                      struct inet6_dev *idev,
4447                                      const struct in6_addr *addr,
4448                                      bool anycast, gfp_t gfp_flags)
4449 {
4450         struct fib6_config cfg = {
4451                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4452                 .fc_ifindex = idev->dev->ifindex,
4453                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4454                 .fc_dst = *addr,
4455                 .fc_dst_len = 128,
4456                 .fc_protocol = RTPROT_KERNEL,
4457                 .fc_nlinfo.nl_net = net,
4458                 .fc_ignore_dev_down = true,
4459         };
4460         struct fib6_info *f6i;
4461
4462         if (anycast) {
4463                 cfg.fc_type = RTN_ANYCAST;
4464                 cfg.fc_flags |= RTF_ANYCAST;
4465         } else {
4466                 cfg.fc_type = RTN_LOCAL;
4467                 cfg.fc_flags |= RTF_LOCAL;
4468         }
4469
4470         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4471         if (!IS_ERR(f6i))
4472                 f6i->dst_nocount = true;
4473         return f6i;
4474 }
4475
4476 /* remove deleted ip from prefsrc entries */
4477 struct arg_dev_net_ip {
4478         struct net_device *dev;
4479         struct net *net;
4480         struct in6_addr *addr;
4481 };
4482
4483 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4484 {
4485         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4486         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4487         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4488
4489         if (!rt->nh &&
4490             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4491             rt != net->ipv6.fib6_null_entry &&
4492             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4493                 spin_lock_bh(&rt6_exception_lock);
4494                 /* remove prefsrc entry */
4495                 rt->fib6_prefsrc.plen = 0;
4496                 spin_unlock_bh(&rt6_exception_lock);
4497         }
4498         return 0;
4499 }
4500
4501 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4502 {
4503         struct net *net = dev_net(ifp->idev->dev);
4504         struct arg_dev_net_ip adni = {
4505                 .dev = ifp->idev->dev,
4506                 .net = net,
4507                 .addr = &ifp->addr,
4508         };
4509         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4510 }
4511
4512 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4513
4514 /* Remove routers and update dst entries when gateway turn into host. */
4515 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4516 {
4517         struct in6_addr *gateway = (struct in6_addr *)arg;
4518         struct fib6_nh *nh;
4519
4520         /* RA routes do not use nexthops */
4521         if (rt->nh)
4522                 return 0;
4523
4524         nh = rt->fib6_nh;
4525         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4526             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4527                 return -1;
4528
4529         /* Further clean up cached routes in exception table.
4530          * This is needed because cached route may have a different
4531          * gateway than its 'parent' in the case of an ip redirect.
4532          */
4533         fib6_nh_exceptions_clean_tohost(nh, gateway);
4534
4535         return 0;
4536 }
4537
4538 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4539 {
4540         fib6_clean_all(net, fib6_clean_tohost, gateway);
4541 }
4542
4543 struct arg_netdev_event {
4544         const struct net_device *dev;
4545         union {
4546                 unsigned char nh_flags;
4547                 unsigned long event;
4548         };
4549 };
4550
4551 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4552 {
4553         struct fib6_info *iter;
4554         struct fib6_node *fn;
4555
4556         fn = rcu_dereference_protected(rt->fib6_node,
4557                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4558         iter = rcu_dereference_protected(fn->leaf,
4559                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4560         while (iter) {
4561                 if (iter->fib6_metric == rt->fib6_metric &&
4562                     rt6_qualify_for_ecmp(iter))
4563                         return iter;
4564                 iter = rcu_dereference_protected(iter->fib6_next,
4565                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4566         }
4567
4568         return NULL;
4569 }
4570
4571 /* only called for fib entries with builtin fib6_nh */
4572 static bool rt6_is_dead(const struct fib6_info *rt)
4573 {
4574         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4575             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4576              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4577                 return true;
4578
4579         return false;
4580 }
4581
4582 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4583 {
4584         struct fib6_info *iter;
4585         int total = 0;
4586
4587         if (!rt6_is_dead(rt))
4588                 total += rt->fib6_nh->fib_nh_weight;
4589
4590         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4591                 if (!rt6_is_dead(iter))
4592                         total += iter->fib6_nh->fib_nh_weight;
4593         }
4594
4595         return total;
4596 }
4597
4598 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4599 {
4600         int upper_bound = -1;
4601
4602         if (!rt6_is_dead(rt)) {
4603                 *weight += rt->fib6_nh->fib_nh_weight;
4604                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4605                                                     total) - 1;
4606         }
4607         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4608 }
4609
4610 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4611 {
4612         struct fib6_info *iter;
4613         int weight = 0;
4614
4615         rt6_upper_bound_set(rt, &weight, total);
4616
4617         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4618                 rt6_upper_bound_set(iter, &weight, total);
4619 }
4620
4621 void rt6_multipath_rebalance(struct fib6_info *rt)
4622 {
4623         struct fib6_info *first;
4624         int total;
4625
4626         /* In case the entire multipath route was marked for flushing,
4627          * then there is no need to rebalance upon the removal of every
4628          * sibling route.
4629          */
4630         if (!rt->fib6_nsiblings || rt->should_flush)
4631                 return;
4632
4633         /* During lookup routes are evaluated in order, so we need to
4634          * make sure upper bounds are assigned from the first sibling
4635          * onwards.
4636          */
4637         first = rt6_multipath_first_sibling(rt);
4638         if (WARN_ON_ONCE(!first))
4639                 return;
4640
4641         total = rt6_multipath_total_weight(first);
4642         rt6_multipath_upper_bound_set(first, total);
4643 }
4644
4645 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4646 {
4647         const struct arg_netdev_event *arg = p_arg;
4648         struct net *net = dev_net(arg->dev);
4649
4650         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4651             rt->fib6_nh->fib_nh_dev == arg->dev) {
4652                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4653                 fib6_update_sernum_upto_root(net, rt);
4654                 rt6_multipath_rebalance(rt);
4655         }
4656
4657         return 0;
4658 }
4659
4660 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4661 {
4662         struct arg_netdev_event arg = {
4663                 .dev = dev,
4664                 {
4665                         .nh_flags = nh_flags,
4666                 },
4667         };
4668
4669         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4670                 arg.nh_flags |= RTNH_F_LINKDOWN;
4671
4672         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4673 }
4674
4675 /* only called for fib entries with inline fib6_nh */
4676 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4677                                    const struct net_device *dev)
4678 {
4679         struct fib6_info *iter;
4680
4681         if (rt->fib6_nh->fib_nh_dev == dev)
4682                 return true;
4683         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4684                 if (iter->fib6_nh->fib_nh_dev == dev)
4685                         return true;
4686
4687         return false;
4688 }
4689
4690 static void rt6_multipath_flush(struct fib6_info *rt)
4691 {
4692         struct fib6_info *iter;
4693
4694         rt->should_flush = 1;
4695         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4696                 iter->should_flush = 1;
4697 }
4698
4699 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4700                                              const struct net_device *down_dev)
4701 {
4702         struct fib6_info *iter;
4703         unsigned int dead = 0;
4704
4705         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4706             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4707                 dead++;
4708         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4709                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4710                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4711                         dead++;
4712
4713         return dead;
4714 }
4715
4716 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4717                                        const struct net_device *dev,
4718                                        unsigned char nh_flags)
4719 {
4720         struct fib6_info *iter;
4721
4722         if (rt->fib6_nh->fib_nh_dev == dev)
4723                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4724         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4725                 if (iter->fib6_nh->fib_nh_dev == dev)
4726                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4727 }
4728
4729 /* called with write lock held for table with rt */
4730 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4731 {
4732         const struct arg_netdev_event *arg = p_arg;
4733         const struct net_device *dev = arg->dev;
4734         struct net *net = dev_net(dev);
4735
4736         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4737                 return 0;
4738
4739         switch (arg->event) {
4740         case NETDEV_UNREGISTER:
4741                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4742         case NETDEV_DOWN:
4743                 if (rt->should_flush)
4744                         return -1;
4745                 if (!rt->fib6_nsiblings)
4746                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4747                 if (rt6_multipath_uses_dev(rt, dev)) {
4748                         unsigned int count;
4749
4750                         count = rt6_multipath_dead_count(rt, dev);
4751                         if (rt->fib6_nsiblings + 1 == count) {
4752                                 rt6_multipath_flush(rt);
4753                                 return -1;
4754                         }
4755                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4756                                                    RTNH_F_LINKDOWN);
4757                         fib6_update_sernum(net, rt);
4758                         rt6_multipath_rebalance(rt);
4759                 }
4760                 return -2;
4761         case NETDEV_CHANGE:
4762                 if (rt->fib6_nh->fib_nh_dev != dev ||
4763                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4764                         break;
4765                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4766                 rt6_multipath_rebalance(rt);
4767                 break;
4768         }
4769
4770         return 0;
4771 }
4772
4773 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4774 {
4775         struct arg_netdev_event arg = {
4776                 .dev = dev,
4777                 {
4778                         .event = event,
4779                 },
4780         };
4781         struct net *net = dev_net(dev);
4782
4783         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4784                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4785         else
4786                 fib6_clean_all(net, fib6_ifdown, &arg);
4787 }
4788
4789 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4790 {
4791         rt6_sync_down_dev(dev, event);
4792         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4793         neigh_ifdown(&nd_tbl, dev);
4794 }
4795
4796 struct rt6_mtu_change_arg {
4797         struct net_device *dev;
4798         unsigned int mtu;
4799         struct fib6_info *f6i;
4800 };
4801
4802 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4803 {
4804         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4805         struct fib6_info *f6i = arg->f6i;
4806
4807         /* For administrative MTU increase, there is no way to discover
4808          * IPv6 PMTU increase, so PMTU increase should be updated here.
4809          * Since RFC 1981 doesn't include administrative MTU increase
4810          * update PMTU increase is a MUST. (i.e. jumbo frame)
4811          */
4812         if (nh->fib_nh_dev == arg->dev) {
4813                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4814                 u32 mtu = f6i->fib6_pmtu;
4815
4816                 if (mtu >= arg->mtu ||
4817                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4818                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4819
4820                 spin_lock_bh(&rt6_exception_lock);
4821                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4822                 spin_unlock_bh(&rt6_exception_lock);
4823         }
4824
4825         return 0;
4826 }
4827
4828 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4829 {
4830         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4831         struct inet6_dev *idev;
4832
4833         /* In IPv6 pmtu discovery is not optional,
4834            so that RTAX_MTU lock cannot disable it.
4835            We still use this lock to block changes
4836            caused by addrconf/ndisc.
4837         */
4838
4839         idev = __in6_dev_get(arg->dev);
4840         if (!idev)
4841                 return 0;
4842
4843         if (fib6_metric_locked(f6i, RTAX_MTU))
4844                 return 0;
4845
4846         arg->f6i = f6i;
4847         if (f6i->nh) {
4848                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4849                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4850                                                 arg);
4851         }
4852
4853         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4854 }
4855
4856 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4857 {
4858         struct rt6_mtu_change_arg arg = {
4859                 .dev = dev,
4860                 .mtu = mtu,
4861         };
4862
4863         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4864 }
4865
4866 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4867         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4868         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4869         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4870         [RTA_OIF]               = { .type = NLA_U32 },
4871         [RTA_IIF]               = { .type = NLA_U32 },
4872         [RTA_PRIORITY]          = { .type = NLA_U32 },
4873         [RTA_METRICS]           = { .type = NLA_NESTED },
4874         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4875         [RTA_PREF]              = { .type = NLA_U8 },
4876         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4877         [RTA_ENCAP]             = { .type = NLA_NESTED },
4878         [RTA_EXPIRES]           = { .type = NLA_U32 },
4879         [RTA_UID]               = { .type = NLA_U32 },
4880         [RTA_MARK]              = { .type = NLA_U32 },
4881         [RTA_TABLE]             = { .type = NLA_U32 },
4882         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4883         [RTA_SPORT]             = { .type = NLA_U16 },
4884         [RTA_DPORT]             = { .type = NLA_U16 },
4885         [RTA_NH_ID]             = { .type = NLA_U32 },
4886 };
4887
4888 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4889                               struct fib6_config *cfg,
4890                               struct netlink_ext_ack *extack)
4891 {
4892         struct rtmsg *rtm;
4893         struct nlattr *tb[RTA_MAX+1];
4894         unsigned int pref;
4895         int err;
4896
4897         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4898                                      rtm_ipv6_policy, extack);
4899         if (err < 0)
4900                 goto errout;
4901
4902         err = -EINVAL;
4903         rtm = nlmsg_data(nlh);
4904
4905         *cfg = (struct fib6_config){
4906                 .fc_table = rtm->rtm_table,
4907                 .fc_dst_len = rtm->rtm_dst_len,
4908                 .fc_src_len = rtm->rtm_src_len,
4909                 .fc_flags = RTF_UP,
4910                 .fc_protocol = rtm->rtm_protocol,
4911                 .fc_type = rtm->rtm_type,
4912
4913                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4914                 .fc_nlinfo.nlh = nlh,
4915                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4916         };
4917
4918         if (rtm->rtm_type == RTN_UNREACHABLE ||
4919             rtm->rtm_type == RTN_BLACKHOLE ||
4920             rtm->rtm_type == RTN_PROHIBIT ||
4921             rtm->rtm_type == RTN_THROW)
4922                 cfg->fc_flags |= RTF_REJECT;
4923
4924         if (rtm->rtm_type == RTN_LOCAL)
4925                 cfg->fc_flags |= RTF_LOCAL;
4926
4927         if (rtm->rtm_flags & RTM_F_CLONED)
4928                 cfg->fc_flags |= RTF_CACHE;
4929
4930         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4931
4932         if (tb[RTA_NH_ID]) {
4933                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4934                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4935                         NL_SET_ERR_MSG(extack,
4936                                        "Nexthop specification and nexthop id are mutually exclusive");
4937                         goto errout;
4938                 }
4939                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4940         }
4941
4942         if (tb[RTA_GATEWAY]) {
4943                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4944                 cfg->fc_flags |= RTF_GATEWAY;
4945         }
4946         if (tb[RTA_VIA]) {
4947                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4948                 goto errout;
4949         }
4950
4951         if (tb[RTA_DST]) {
4952                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4953
4954                 if (nla_len(tb[RTA_DST]) < plen)
4955                         goto errout;
4956
4957                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4958         }
4959
4960         if (tb[RTA_SRC]) {
4961                 int plen = (rtm->rtm_src_len + 7) >> 3;
4962
4963                 if (nla_len(tb[RTA_SRC]) < plen)
4964                         goto errout;
4965
4966                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4967         }
4968
4969         if (tb[RTA_PREFSRC])
4970                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4971
4972         if (tb[RTA_OIF])
4973                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4974
4975         if (tb[RTA_PRIORITY])
4976                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4977
4978         if (tb[RTA_METRICS]) {
4979                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4980                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4981         }
4982
4983         if (tb[RTA_TABLE])
4984                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4985
4986         if (tb[RTA_MULTIPATH]) {
4987                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4988                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4989
4990                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4991                                                      cfg->fc_mp_len, extack);
4992                 if (err < 0)
4993                         goto errout;
4994         }
4995
4996         if (tb[RTA_PREF]) {
4997                 pref = nla_get_u8(tb[RTA_PREF]);
4998                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4999                     pref != ICMPV6_ROUTER_PREF_HIGH)
5000                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5001                 cfg->fc_flags |= RTF_PREF(pref);
5002         }
5003
5004         if (tb[RTA_ENCAP])
5005                 cfg->fc_encap = tb[RTA_ENCAP];
5006
5007         if (tb[RTA_ENCAP_TYPE]) {
5008                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5009
5010                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5011                 if (err < 0)
5012                         goto errout;
5013         }
5014
5015         if (tb[RTA_EXPIRES]) {
5016                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5017
5018                 if (addrconf_finite_timeout(timeout)) {
5019                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5020                         cfg->fc_flags |= RTF_EXPIRES;
5021                 }
5022         }
5023
5024         err = 0;
5025 errout:
5026         return err;
5027 }
5028
5029 struct rt6_nh {
5030         struct fib6_info *fib6_info;
5031         struct fib6_config r_cfg;
5032         struct list_head next;
5033 };
5034
5035 static int ip6_route_info_append(struct net *net,
5036                                  struct list_head *rt6_nh_list,
5037                                  struct fib6_info *rt,
5038                                  struct fib6_config *r_cfg)
5039 {
5040         struct rt6_nh *nh;
5041         int err = -EEXIST;
5042
5043         list_for_each_entry(nh, rt6_nh_list, next) {
5044                 /* check if fib6_info already exists */
5045                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5046                         return err;
5047         }
5048
5049         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5050         if (!nh)
5051                 return -ENOMEM;
5052         nh->fib6_info = rt;
5053         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5054         list_add_tail(&nh->next, rt6_nh_list);
5055
5056         return 0;
5057 }
5058
5059 static void ip6_route_mpath_notify(struct fib6_info *rt,
5060                                    struct fib6_info *rt_last,
5061                                    struct nl_info *info,
5062                                    __u16 nlflags)
5063 {
5064         /* if this is an APPEND route, then rt points to the first route
5065          * inserted and rt_last points to last route inserted. Userspace
5066          * wants a consistent dump of the route which starts at the first
5067          * nexthop. Since sibling routes are always added at the end of
5068          * the list, find the first sibling of the last route appended
5069          */
5070         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5071                 rt = list_first_entry(&rt_last->fib6_siblings,
5072                                       struct fib6_info,
5073                                       fib6_siblings);
5074         }
5075
5076         if (rt)
5077                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5078 }
5079
5080 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5081 {
5082         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5083         bool should_notify = false;
5084         struct fib6_info *leaf;
5085         struct fib6_node *fn;
5086
5087         rcu_read_lock();
5088         fn = rcu_dereference(rt->fib6_node);
5089         if (!fn)
5090                 goto out;
5091
5092         leaf = rcu_dereference(fn->leaf);
5093         if (!leaf)
5094                 goto out;
5095
5096         if (rt == leaf ||
5097             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5098              rt6_qualify_for_ecmp(leaf)))
5099                 should_notify = true;
5100 out:
5101         rcu_read_unlock();
5102
5103         return should_notify;
5104 }
5105
5106 static int ip6_route_multipath_add(struct fib6_config *cfg,
5107                                    struct netlink_ext_ack *extack)
5108 {
5109         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5110         struct nl_info *info = &cfg->fc_nlinfo;
5111         struct fib6_config r_cfg;
5112         struct rtnexthop *rtnh;
5113         struct fib6_info *rt;
5114         struct rt6_nh *err_nh;
5115         struct rt6_nh *nh, *nh_safe;
5116         __u16 nlflags;
5117         int remaining;
5118         int attrlen;
5119         int err = 1;
5120         int nhn = 0;
5121         int replace = (cfg->fc_nlinfo.nlh &&
5122                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5123         LIST_HEAD(rt6_nh_list);
5124
5125         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5126         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5127                 nlflags |= NLM_F_APPEND;
5128
5129         remaining = cfg->fc_mp_len;
5130         rtnh = (struct rtnexthop *)cfg->fc_mp;
5131
5132         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5133          * fib6_info structs per nexthop
5134          */
5135         while (rtnh_ok(rtnh, remaining)) {
5136                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5137                 if (rtnh->rtnh_ifindex)
5138                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5139
5140                 attrlen = rtnh_attrlen(rtnh);
5141                 if (attrlen > 0) {
5142                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5143
5144                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5145                         if (nla) {
5146                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5147                                 r_cfg.fc_flags |= RTF_GATEWAY;
5148                         }
5149                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5150                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5151                         if (nla)
5152                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5153                 }
5154
5155                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5156                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5157                 if (IS_ERR(rt)) {
5158                         err = PTR_ERR(rt);
5159                         rt = NULL;
5160                         goto cleanup;
5161                 }
5162                 if (!rt6_qualify_for_ecmp(rt)) {
5163                         err = -EINVAL;
5164                         NL_SET_ERR_MSG(extack,
5165                                        "Device only routes can not be added for IPv6 using the multipath API.");
5166                         fib6_info_release(rt);
5167                         goto cleanup;
5168                 }
5169
5170                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5171
5172                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5173                                             rt, &r_cfg);
5174                 if (err) {
5175                         fib6_info_release(rt);
5176                         goto cleanup;
5177                 }
5178
5179                 rtnh = rtnh_next(rtnh, &remaining);
5180         }
5181
5182         if (list_empty(&rt6_nh_list)) {
5183                 NL_SET_ERR_MSG(extack,
5184                                "Invalid nexthop configuration - no valid nexthops");
5185                 return -EINVAL;
5186         }
5187
5188         /* for add and replace send one notification with all nexthops.
5189          * Skip the notification in fib6_add_rt2node and send one with
5190          * the full route when done
5191          */
5192         info->skip_notify = 1;
5193
5194         /* For add and replace, send one notification with all nexthops. For
5195          * append, send one notification with all appended nexthops.
5196          */
5197         info->skip_notify_kernel = 1;
5198
5199         err_nh = NULL;
5200         list_for_each_entry(nh, &rt6_nh_list, next) {
5201                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5202                 fib6_info_release(nh->fib6_info);
5203
5204                 if (!err) {
5205                         /* save reference to last route successfully inserted */
5206                         rt_last = nh->fib6_info;
5207
5208                         /* save reference to first route for notification */
5209                         if (!rt_notif)
5210                                 rt_notif = nh->fib6_info;
5211                 }
5212
5213                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5214                 nh->fib6_info = NULL;
5215                 if (err) {
5216                         if (replace && nhn)
5217                                 NL_SET_ERR_MSG_MOD(extack,
5218                                                    "multipath route replace failed (check consistency of installed routes)");
5219                         err_nh = nh;
5220                         goto add_errout;
5221                 }
5222
5223                 /* Because each route is added like a single route we remove
5224                  * these flags after the first nexthop: if there is a collision,
5225                  * we have already failed to add the first nexthop:
5226                  * fib6_add_rt2node() has rejected it; when replacing, old
5227                  * nexthops have been replaced by first new, the rest should
5228                  * be added to it.
5229                  */
5230                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5231                                                      NLM_F_REPLACE);
5232                 cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5233                 nhn++;
5234         }
5235
5236         /* An in-kernel notification should only be sent in case the new
5237          * multipath route is added as the first route in the node, or if
5238          * it was appended to it. We pass 'rt_notif' since it is the first
5239          * sibling and might allow us to skip some checks in the replace case.
5240          */
5241         if (ip6_route_mpath_should_notify(rt_notif)) {
5242                 enum fib_event_type fib_event;
5243
5244                 if (rt_notif->fib6_nsiblings != nhn - 1)
5245                         fib_event = FIB_EVENT_ENTRY_APPEND;
5246                 else
5247                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5248
5249                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5250                                                           fib_event, rt_notif,
5251                                                           nhn - 1, extack);
5252                 if (err) {
5253                         /* Delete all the siblings that were just added */
5254                         err_nh = NULL;
5255                         goto add_errout;
5256                 }
5257         }
5258
5259         /* success ... tell user about new route */
5260         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5261         goto cleanup;
5262
5263 add_errout:
5264         /* send notification for routes that were added so that
5265          * the delete notifications sent by ip6_route_del are
5266          * coherent
5267          */
5268         if (rt_notif)
5269                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5270
5271         /* Delete routes that were already added */
5272         list_for_each_entry(nh, &rt6_nh_list, next) {
5273                 if (err_nh == nh)
5274                         break;
5275                 ip6_route_del(&nh->r_cfg, extack);
5276         }
5277
5278 cleanup:
5279         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5280                 if (nh->fib6_info)
5281                         fib6_info_release(nh->fib6_info);
5282                 list_del(&nh->next);
5283                 kfree(nh);
5284         }
5285
5286         return err;
5287 }
5288
5289 static int ip6_route_multipath_del(struct fib6_config *cfg,
5290                                    struct netlink_ext_ack *extack)
5291 {
5292         struct fib6_config r_cfg;
5293         struct rtnexthop *rtnh;
5294         int last_err = 0;
5295         int remaining;
5296         int attrlen;
5297         int err;
5298
5299         remaining = cfg->fc_mp_len;
5300         rtnh = (struct rtnexthop *)cfg->fc_mp;
5301
5302         /* Parse a Multipath Entry */
5303         while (rtnh_ok(rtnh, remaining)) {
5304                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5305                 if (rtnh->rtnh_ifindex)
5306                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5307
5308                 attrlen = rtnh_attrlen(rtnh);
5309                 if (attrlen > 0) {
5310                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5311
5312                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5313                         if (nla) {
5314                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5315                                 r_cfg.fc_flags |= RTF_GATEWAY;
5316                         }
5317                 }
5318                 err = ip6_route_del(&r_cfg, extack);
5319                 if (err)
5320                         last_err = err;
5321
5322                 rtnh = rtnh_next(rtnh, &remaining);
5323         }
5324
5325         return last_err;
5326 }
5327
5328 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5329                               struct netlink_ext_ack *extack)
5330 {
5331         struct fib6_config cfg;
5332         int err;
5333
5334         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5335         if (err < 0)
5336                 return err;
5337
5338         if (cfg.fc_nh_id &&
5339             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5340                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5341                 return -EINVAL;
5342         }
5343
5344         if (cfg.fc_mp)
5345                 return ip6_route_multipath_del(&cfg, extack);
5346         else {
5347                 cfg.fc_delete_all_nh = 1;
5348                 return ip6_route_del(&cfg, extack);
5349         }
5350 }
5351
5352 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5353                               struct netlink_ext_ack *extack)
5354 {
5355         struct fib6_config cfg;
5356         int err;
5357
5358         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5359         if (err < 0)
5360                 return err;
5361
5362         if (cfg.fc_metric == 0)
5363                 cfg.fc_metric = IP6_RT_PRIO_USER;
5364
5365         if (cfg.fc_mp)
5366                 return ip6_route_multipath_add(&cfg, extack);
5367         else
5368                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5369 }
5370
5371 /* add the overhead of this fib6_nh to nexthop_len */
5372 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5373 {
5374         int *nexthop_len = arg;
5375
5376         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5377                      + NLA_ALIGN(sizeof(struct rtnexthop))
5378                      + nla_total_size(16); /* RTA_GATEWAY */
5379
5380         if (nh->fib_nh_lws) {
5381                 /* RTA_ENCAP_TYPE */
5382                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5383                 /* RTA_ENCAP */
5384                 *nexthop_len += nla_total_size(2);
5385         }
5386
5387         return 0;
5388 }
5389
5390 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5391 {
5392         int nexthop_len;
5393
5394         if (f6i->nh) {
5395                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5396                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5397                                          &nexthop_len);
5398         } else {
5399                 struct fib6_nh *nh = f6i->fib6_nh;
5400
5401                 nexthop_len = 0;
5402                 if (f6i->fib6_nsiblings) {
5403                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5404                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5405                                     + nla_total_size(16) /* RTA_GATEWAY */
5406                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5407
5408                         nexthop_len *= f6i->fib6_nsiblings;
5409                 }
5410                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5411         }
5412
5413         return NLMSG_ALIGN(sizeof(struct rtmsg))
5414                + nla_total_size(16) /* RTA_SRC */
5415                + nla_total_size(16) /* RTA_DST */
5416                + nla_total_size(16) /* RTA_GATEWAY */
5417                + nla_total_size(16) /* RTA_PREFSRC */
5418                + nla_total_size(4) /* RTA_TABLE */
5419                + nla_total_size(4) /* RTA_IIF */
5420                + nla_total_size(4) /* RTA_OIF */
5421                + nla_total_size(4) /* RTA_PRIORITY */
5422                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5423                + nla_total_size(sizeof(struct rta_cacheinfo))
5424                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5425                + nla_total_size(1) /* RTA_PREF */
5426                + nexthop_len;
5427 }
5428
5429 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5430                                  unsigned char *flags)
5431 {
5432         if (nexthop_is_multipath(nh)) {
5433                 struct nlattr *mp;
5434
5435                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5436                 if (!mp)
5437                         goto nla_put_failure;
5438
5439                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5440                         goto nla_put_failure;
5441
5442                 nla_nest_end(skb, mp);
5443         } else {
5444                 struct fib6_nh *fib6_nh;
5445
5446                 fib6_nh = nexthop_fib6_nh(nh);
5447                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5448                                      flags, false) < 0)
5449                         goto nla_put_failure;
5450         }
5451
5452         return 0;
5453
5454 nla_put_failure:
5455         return -EMSGSIZE;
5456 }
5457
5458 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5459                          struct fib6_info *rt, struct dst_entry *dst,
5460                          struct in6_addr *dest, struct in6_addr *src,
5461                          int iif, int type, u32 portid, u32 seq,
5462                          unsigned int flags)
5463 {
5464         struct rt6_info *rt6 = (struct rt6_info *)dst;
5465         struct rt6key *rt6_dst, *rt6_src;
5466         u32 *pmetrics, table, rt6_flags;
5467         unsigned char nh_flags = 0;
5468         struct nlmsghdr *nlh;
5469         struct rtmsg *rtm;
5470         long expires = 0;
5471
5472         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5473         if (!nlh)
5474                 return -EMSGSIZE;
5475
5476         if (rt6) {
5477                 rt6_dst = &rt6->rt6i_dst;
5478                 rt6_src = &rt6->rt6i_src;
5479                 rt6_flags = rt6->rt6i_flags;
5480         } else {
5481                 rt6_dst = &rt->fib6_dst;
5482                 rt6_src = &rt->fib6_src;
5483                 rt6_flags = rt->fib6_flags;
5484         }
5485
5486         rtm = nlmsg_data(nlh);
5487         rtm->rtm_family = AF_INET6;
5488         rtm->rtm_dst_len = rt6_dst->plen;
5489         rtm->rtm_src_len = rt6_src->plen;
5490         rtm->rtm_tos = 0;
5491         if (rt->fib6_table)
5492                 table = rt->fib6_table->tb6_id;
5493         else
5494                 table = RT6_TABLE_UNSPEC;
5495         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5496         if (nla_put_u32(skb, RTA_TABLE, table))
5497                 goto nla_put_failure;
5498
5499         rtm->rtm_type = rt->fib6_type;
5500         rtm->rtm_flags = 0;
5501         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5502         rtm->rtm_protocol = rt->fib6_protocol;
5503
5504         if (rt6_flags & RTF_CACHE)
5505                 rtm->rtm_flags |= RTM_F_CLONED;
5506
5507         if (dest) {
5508                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5509                         goto nla_put_failure;
5510                 rtm->rtm_dst_len = 128;
5511         } else if (rtm->rtm_dst_len)
5512                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5513                         goto nla_put_failure;
5514 #ifdef CONFIG_IPV6_SUBTREES
5515         if (src) {
5516                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5517                         goto nla_put_failure;
5518                 rtm->rtm_src_len = 128;
5519         } else if (rtm->rtm_src_len &&
5520                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5521                 goto nla_put_failure;
5522 #endif
5523         if (iif) {
5524 #ifdef CONFIG_IPV6_MROUTE
5525                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5526                         int err = ip6mr_get_route(net, skb, rtm, portid);
5527
5528                         if (err == 0)
5529                                 return 0;
5530                         if (err < 0)
5531                                 goto nla_put_failure;
5532                 } else
5533 #endif
5534                         if (nla_put_u32(skb, RTA_IIF, iif))
5535                                 goto nla_put_failure;
5536         } else if (dest) {
5537                 struct in6_addr saddr_buf;
5538                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5539                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5540                         goto nla_put_failure;
5541         }
5542
5543         if (rt->fib6_prefsrc.plen) {
5544                 struct in6_addr saddr_buf;
5545                 saddr_buf = rt->fib6_prefsrc.addr;
5546                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5547                         goto nla_put_failure;
5548         }
5549
5550         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5551         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5552                 goto nla_put_failure;
5553
5554         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5555                 goto nla_put_failure;
5556
5557         /* For multipath routes, walk the siblings list and add
5558          * each as a nexthop within RTA_MULTIPATH.
5559          */
5560         if (rt6) {
5561                 if (rt6_flags & RTF_GATEWAY &&
5562                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5563                         goto nla_put_failure;
5564
5565                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5566                         goto nla_put_failure;
5567
5568                 if (dst->lwtstate &&
5569                     lwtunnel_fill_encap(skb, dst->lwtstate, RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
5570                         goto nla_put_failure;
5571         } else if (rt->fib6_nsiblings) {
5572                 struct fib6_info *sibling, *next_sibling;
5573                 struct nlattr *mp;
5574
5575                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5576                 if (!mp)
5577                         goto nla_put_failure;
5578
5579                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5580                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5581                         goto nla_put_failure;
5582
5583                 list_for_each_entry_safe(sibling, next_sibling,
5584                                          &rt->fib6_siblings, fib6_siblings) {
5585                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5586                                             sibling->fib6_nh->fib_nh_weight,
5587                                             AF_INET6) < 0)
5588                                 goto nla_put_failure;
5589                 }
5590
5591                 nla_nest_end(skb, mp);
5592         } else if (rt->nh) {
5593                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5594                         goto nla_put_failure;
5595
5596                 if (nexthop_is_blackhole(rt->nh))
5597                         rtm->rtm_type = RTN_BLACKHOLE;
5598
5599                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5600                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5601                         goto nla_put_failure;
5602
5603                 rtm->rtm_flags |= nh_flags;
5604         } else {
5605                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5606                                      &nh_flags, false) < 0)
5607                         goto nla_put_failure;
5608
5609                 rtm->rtm_flags |= nh_flags;
5610         }
5611
5612         if (rt6_flags & RTF_EXPIRES) {
5613                 expires = dst ? dst->expires : rt->expires;
5614                 expires -= jiffies;
5615         }
5616
5617         if (!dst) {
5618                 if (rt->offload)
5619                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5620                 if (rt->trap)
5621                         rtm->rtm_flags |= RTM_F_TRAP;
5622                 if (rt->offload_failed)
5623                         rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
5624         }
5625
5626         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5627                 goto nla_put_failure;
5628
5629         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5630                 goto nla_put_failure;
5631
5632
5633         nlmsg_end(skb, nlh);
5634         return 0;
5635
5636 nla_put_failure:
5637         nlmsg_cancel(skb, nlh);
5638         return -EMSGSIZE;
5639 }
5640
5641 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5642 {
5643         const struct net_device *dev = arg;
5644
5645         if (nh->fib_nh_dev == dev)
5646                 return 1;
5647
5648         return 0;
5649 }
5650
5651 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5652                                const struct net_device *dev)
5653 {
5654         if (f6i->nh) {
5655                 struct net_device *_dev = (struct net_device *)dev;
5656
5657                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5658                                                   fib6_info_nh_uses_dev,
5659                                                   _dev);
5660         }
5661
5662         if (f6i->fib6_nh->fib_nh_dev == dev)
5663                 return true;
5664
5665         if (f6i->fib6_nsiblings) {
5666                 struct fib6_info *sibling, *next_sibling;
5667
5668                 list_for_each_entry_safe(sibling, next_sibling,
5669                                          &f6i->fib6_siblings, fib6_siblings) {
5670                         if (sibling->fib6_nh->fib_nh_dev == dev)
5671                                 return true;
5672                 }
5673         }
5674
5675         return false;
5676 }
5677
5678 struct fib6_nh_exception_dump_walker {
5679         struct rt6_rtnl_dump_arg *dump;
5680         struct fib6_info *rt;
5681         unsigned int flags;
5682         unsigned int skip;
5683         unsigned int count;
5684 };
5685
5686 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5687 {
5688         struct fib6_nh_exception_dump_walker *w = arg;
5689         struct rt6_rtnl_dump_arg *dump = w->dump;
5690         struct rt6_exception_bucket *bucket;
5691         struct rt6_exception *rt6_ex;
5692         int i, err;
5693
5694         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5695         if (!bucket)
5696                 return 0;
5697
5698         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5699                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5700                         if (w->skip) {
5701                                 w->skip--;
5702                                 continue;
5703                         }
5704
5705                         /* Expiration of entries doesn't bump sernum, insertion
5706                          * does. Removal is triggered by insertion, so we can
5707                          * rely on the fact that if entries change between two
5708                          * partial dumps, this node is scanned again completely,
5709                          * see rt6_insert_exception() and fib6_dump_table().
5710                          *
5711                          * Count expired entries we go through as handled
5712                          * entries that we'll skip next time, in case of partial
5713                          * node dump. Otherwise, if entries expire meanwhile,
5714                          * we'll skip the wrong amount.
5715                          */
5716                         if (rt6_check_expired(rt6_ex->rt6i)) {
5717                                 w->count++;
5718                                 continue;
5719                         }
5720
5721                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5722                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5723                                             RTM_NEWROUTE,
5724                                             NETLINK_CB(dump->cb->skb).portid,
5725                                             dump->cb->nlh->nlmsg_seq, w->flags);
5726                         if (err)
5727                                 return err;
5728
5729                         w->count++;
5730                 }
5731                 bucket++;
5732         }
5733
5734         return 0;
5735 }
5736
5737 /* Return -1 if done with node, number of handled routes on partial dump */
5738 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5739 {
5740         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5741         struct fib_dump_filter *filter = &arg->filter;
5742         unsigned int flags = NLM_F_MULTI;
5743         struct net *net = arg->net;
5744         int count = 0;
5745
5746         if (rt == net->ipv6.fib6_null_entry)
5747                 return -1;
5748
5749         if ((filter->flags & RTM_F_PREFIX) &&
5750             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5751                 /* success since this is not a prefix route */
5752                 return -1;
5753         }
5754         if (filter->filter_set &&
5755             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5756              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5757              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5758                 return -1;
5759         }
5760
5761         if (filter->filter_set ||
5762             !filter->dump_routes || !filter->dump_exceptions) {
5763                 flags |= NLM_F_DUMP_FILTERED;
5764         }
5765
5766         if (filter->dump_routes) {
5767                 if (skip) {
5768                         skip--;
5769                 } else {
5770                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5771                                           0, RTM_NEWROUTE,
5772                                           NETLINK_CB(arg->cb->skb).portid,
5773                                           arg->cb->nlh->nlmsg_seq, flags)) {
5774                                 return 0;
5775                         }
5776                         count++;
5777                 }
5778         }
5779
5780         if (filter->dump_exceptions) {
5781                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5782                                                            .rt = rt,
5783                                                            .flags = flags,
5784                                                            .skip = skip,
5785                                                            .count = 0 };
5786                 int err;
5787
5788                 rcu_read_lock();
5789                 if (rt->nh) {
5790                         err = nexthop_for_each_fib6_nh(rt->nh,
5791                                                        rt6_nh_dump_exceptions,
5792                                                        &w);
5793                 } else {
5794                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5795                 }
5796                 rcu_read_unlock();
5797
5798                 if (err)
5799                         return count += w.count;
5800         }
5801
5802         return -1;
5803 }
5804
5805 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5806                                         const struct nlmsghdr *nlh,
5807                                         struct nlattr **tb,
5808                                         struct netlink_ext_ack *extack)
5809 {
5810         struct rtmsg *rtm;
5811         int i, err;
5812
5813         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5814                 NL_SET_ERR_MSG_MOD(extack,
5815                                    "Invalid header for get route request");
5816                 return -EINVAL;
5817         }
5818
5819         if (!netlink_strict_get_check(skb))
5820                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5821                                               rtm_ipv6_policy, extack);
5822
5823         rtm = nlmsg_data(nlh);
5824         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5825             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5826             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5827             rtm->rtm_type) {
5828                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5829                 return -EINVAL;
5830         }
5831         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5832                 NL_SET_ERR_MSG_MOD(extack,
5833                                    "Invalid flags for get route request");
5834                 return -EINVAL;
5835         }
5836
5837         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5838                                             rtm_ipv6_policy, extack);
5839         if (err)
5840                 return err;
5841
5842         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5843             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5844                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5845                 return -EINVAL;
5846         }
5847
5848         for (i = 0; i <= RTA_MAX; i++) {
5849                 if (!tb[i])
5850                         continue;
5851
5852                 switch (i) {
5853                 case RTA_SRC:
5854                 case RTA_DST:
5855                 case RTA_IIF:
5856                 case RTA_OIF:
5857                 case RTA_MARK:
5858                 case RTA_UID:
5859                 case RTA_SPORT:
5860                 case RTA_DPORT:
5861                 case RTA_IP_PROTO:
5862                         break;
5863                 default:
5864                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5865                         return -EINVAL;
5866                 }
5867         }
5868
5869         return 0;
5870 }
5871
5872 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5873                               struct netlink_ext_ack *extack)
5874 {
5875         struct net *net = sock_net(in_skb->sk);
5876         struct nlattr *tb[RTA_MAX+1];
5877         int err, iif = 0, oif = 0;
5878         struct fib6_info *from;
5879         struct dst_entry *dst;
5880         struct rt6_info *rt;
5881         struct sk_buff *skb;
5882         struct rtmsg *rtm;
5883         struct flowi6 fl6 = {};
5884         bool fibmatch;
5885
5886         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5887         if (err < 0)
5888                 goto errout;
5889
5890         err = -EINVAL;
5891         rtm = nlmsg_data(nlh);
5892         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5893         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5894
5895         if (tb[RTA_SRC]) {
5896                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5897                         goto errout;
5898
5899                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5900         }
5901
5902         if (tb[RTA_DST]) {
5903                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5904                         goto errout;
5905
5906                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5907         }
5908
5909         if (tb[RTA_IIF])
5910                 iif = nla_get_u32(tb[RTA_IIF]);
5911
5912         if (tb[RTA_OIF])
5913                 oif = nla_get_u32(tb[RTA_OIF]);
5914
5915         if (tb[RTA_MARK])
5916                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5917
5918         if (tb[RTA_UID])
5919                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5920                                            nla_get_u32(tb[RTA_UID]));
5921         else
5922                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5923
5924         if (tb[RTA_SPORT])
5925                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5926
5927         if (tb[RTA_DPORT])
5928                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5929
5930         if (tb[RTA_IP_PROTO]) {
5931                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5932                                                   &fl6.flowi6_proto, AF_INET6,
5933                                                   extack);
5934                 if (err)
5935                         goto errout;
5936         }
5937
5938         if (iif) {
5939                 struct net_device *dev;
5940                 int flags = 0;
5941
5942                 rcu_read_lock();
5943
5944                 dev = dev_get_by_index_rcu(net, iif);
5945                 if (!dev) {
5946                         rcu_read_unlock();
5947                         err = -ENODEV;
5948                         goto errout;
5949                 }
5950
5951                 fl6.flowi6_iif = iif;
5952
5953                 if (!ipv6_addr_any(&fl6.saddr))
5954                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5955
5956                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5957
5958                 rcu_read_unlock();
5959         } else {
5960                 fl6.flowi6_oif = oif;
5961
5962                 dst = ip6_route_output(net, NULL, &fl6);
5963         }
5964
5965
5966         rt = container_of(dst, struct rt6_info, dst);
5967         if (rt->dst.error) {
5968                 err = rt->dst.error;
5969                 ip6_rt_put(rt);
5970                 goto errout;
5971         }
5972
5973         if (rt == net->ipv6.ip6_null_entry) {
5974                 err = rt->dst.error;
5975                 ip6_rt_put(rt);
5976                 goto errout;
5977         }
5978
5979         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5980         if (!skb) {
5981                 ip6_rt_put(rt);
5982                 err = -ENOBUFS;
5983                 goto errout;
5984         }
5985
5986         skb_dst_set(skb, &rt->dst);
5987
5988         rcu_read_lock();
5989         from = rcu_dereference(rt->from);
5990         if (from) {
5991                 if (fibmatch)
5992                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5993                                             iif, RTM_NEWROUTE,
5994                                             NETLINK_CB(in_skb).portid,
5995                                             nlh->nlmsg_seq, 0);
5996                 else
5997                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5998                                             &fl6.saddr, iif, RTM_NEWROUTE,
5999                                             NETLINK_CB(in_skb).portid,
6000                                             nlh->nlmsg_seq, 0);
6001         } else {
6002                 err = -ENETUNREACH;
6003         }
6004         rcu_read_unlock();
6005
6006         if (err < 0) {
6007                 kfree_skb(skb);
6008                 goto errout;
6009         }
6010
6011         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6012 errout:
6013         return err;
6014 }
6015
6016 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6017                      unsigned int nlm_flags)
6018 {
6019         struct sk_buff *skb;
6020         struct net *net = info->nl_net;
6021         u32 seq;
6022         int err;
6023
6024         err = -ENOBUFS;
6025         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6026
6027         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6028         if (!skb)
6029                 goto errout;
6030
6031         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6032                             event, info->portid, seq, nlm_flags);
6033         if (err < 0) {
6034                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6035                 WARN_ON(err == -EMSGSIZE);
6036                 kfree_skb(skb);
6037                 goto errout;
6038         }
6039         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6040                     info->nlh, gfp_any());
6041         return;
6042 errout:
6043         if (err < 0)
6044                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6045 }
6046
6047 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6048                     struct nl_info *info)
6049 {
6050         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6051         struct sk_buff *skb;
6052         int err = -ENOBUFS;
6053
6054         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6055         if (!skb)
6056                 goto errout;
6057
6058         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6059                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6060         if (err < 0) {
6061                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6062                 WARN_ON(err == -EMSGSIZE);
6063                 kfree_skb(skb);
6064                 goto errout;
6065         }
6066         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6067                     info->nlh, gfp_any());
6068         return;
6069 errout:
6070         if (err < 0)
6071                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6072 }
6073
6074 void fib6_info_hw_flags_set(struct net *net, struct fib6_info *f6i,
6075                             bool offload, bool trap, bool offload_failed)
6076 {
6077         struct sk_buff *skb;
6078         int err;
6079
6080         if (f6i->offload == offload && f6i->trap == trap &&
6081             f6i->offload_failed == offload_failed)
6082                 return;
6083
6084         f6i->offload = offload;
6085         f6i->trap = trap;
6086
6087         /* 2 means send notifications only if offload_failed was changed. */
6088         if (net->ipv6.sysctl.fib_notify_on_flag_change == 2 &&
6089             f6i->offload_failed == offload_failed)
6090                 return;
6091
6092         f6i->offload_failed = offload_failed;
6093
6094         if (!rcu_access_pointer(f6i->fib6_node))
6095                 /* The route was removed from the tree, do not send
6096                  * notfication.
6097                  */
6098                 return;
6099
6100         if (!net->ipv6.sysctl.fib_notify_on_flag_change)
6101                 return;
6102
6103         skb = nlmsg_new(rt6_nlmsg_size(f6i), GFP_KERNEL);
6104         if (!skb) {
6105                 err = -ENOBUFS;
6106                 goto errout;
6107         }
6108
6109         err = rt6_fill_node(net, skb, f6i, NULL, NULL, NULL, 0, RTM_NEWROUTE, 0,
6110                             0, 0);
6111         if (err < 0) {
6112                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6113                 WARN_ON(err == -EMSGSIZE);
6114                 kfree_skb(skb);
6115                 goto errout;
6116         }
6117
6118         rtnl_notify(skb, net, 0, RTNLGRP_IPV6_ROUTE, NULL, GFP_KERNEL);
6119         return;
6120
6121 errout:
6122         rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6123 }
6124 EXPORT_SYMBOL(fib6_info_hw_flags_set);
6125
6126 static int ip6_route_dev_notify(struct notifier_block *this,
6127                                 unsigned long event, void *ptr)
6128 {
6129         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6130         struct net *net = dev_net(dev);
6131
6132         if (!(dev->flags & IFF_LOOPBACK))
6133                 return NOTIFY_OK;
6134
6135         if (event == NETDEV_REGISTER) {
6136                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6137                 net->ipv6.ip6_null_entry->dst.dev = dev;
6138                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6139 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6140                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6141                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6142                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6143                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6144 #endif
6145          } else if (event == NETDEV_UNREGISTER &&
6146                     dev->reg_state != NETREG_UNREGISTERED) {
6147                 /* NETDEV_UNREGISTER could be fired for multiple times by
6148                  * netdev_wait_allrefs(). Make sure we only call this once.
6149                  */
6150                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6151 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6152                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6153                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6154 #endif
6155         }
6156
6157         return NOTIFY_OK;
6158 }
6159
6160 /*
6161  *      /proc
6162  */
6163
6164 #ifdef CONFIG_PROC_FS
6165 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6166 {
6167         struct net *net = (struct net *)seq->private;
6168         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6169                    net->ipv6.rt6_stats->fib_nodes,
6170                    net->ipv6.rt6_stats->fib_route_nodes,
6171                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6172                    net->ipv6.rt6_stats->fib_rt_entries,
6173                    net->ipv6.rt6_stats->fib_rt_cache,
6174                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6175                    net->ipv6.rt6_stats->fib_discarded_routes);
6176
6177         return 0;
6178 }
6179 #endif  /* CONFIG_PROC_FS */
6180
6181 #ifdef CONFIG_SYSCTL
6182
6183 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6184                               void *buffer, size_t *lenp, loff_t *ppos)
6185 {
6186         struct net *net;
6187         int delay;
6188         int ret;
6189         if (!write)
6190                 return -EINVAL;
6191
6192         net = (struct net *)ctl->extra1;
6193         delay = net->ipv6.sysctl.flush_delay;
6194         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6195         if (ret)
6196                 return ret;
6197
6198         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6199         return 0;
6200 }
6201
6202 static struct ctl_table ipv6_route_table_template[] = {
6203         {
6204                 .procname       =       "flush",
6205                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6206                 .maxlen         =       sizeof(int),
6207                 .mode           =       0200,
6208                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6209         },
6210         {
6211                 .procname       =       "gc_thresh",
6212                 .data           =       &ip6_dst_ops_template.gc_thresh,
6213                 .maxlen         =       sizeof(int),
6214                 .mode           =       0644,
6215                 .proc_handler   =       proc_dointvec,
6216         },
6217         {
6218                 .procname       =       "max_size",
6219                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6220                 .maxlen         =       sizeof(int),
6221                 .mode           =       0644,
6222                 .proc_handler   =       proc_dointvec,
6223         },
6224         {
6225                 .procname       =       "gc_min_interval",
6226                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6227                 .maxlen         =       sizeof(int),
6228                 .mode           =       0644,
6229                 .proc_handler   =       proc_dointvec_jiffies,
6230         },
6231         {
6232                 .procname       =       "gc_timeout",
6233                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6234                 .maxlen         =       sizeof(int),
6235                 .mode           =       0644,
6236                 .proc_handler   =       proc_dointvec_jiffies,
6237         },
6238         {
6239                 .procname       =       "gc_interval",
6240                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6241                 .maxlen         =       sizeof(int),
6242                 .mode           =       0644,
6243                 .proc_handler   =       proc_dointvec_jiffies,
6244         },
6245         {
6246                 .procname       =       "gc_elasticity",
6247                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6248                 .maxlen         =       sizeof(int),
6249                 .mode           =       0644,
6250                 .proc_handler   =       proc_dointvec,
6251         },
6252         {
6253                 .procname       =       "mtu_expires",
6254                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6255                 .maxlen         =       sizeof(int),
6256                 .mode           =       0644,
6257                 .proc_handler   =       proc_dointvec_jiffies,
6258         },
6259         {
6260                 .procname       =       "min_adv_mss",
6261                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6262                 .maxlen         =       sizeof(int),
6263                 .mode           =       0644,
6264                 .proc_handler   =       proc_dointvec,
6265         },
6266         {
6267                 .procname       =       "gc_min_interval_ms",
6268                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6269                 .maxlen         =       sizeof(int),
6270                 .mode           =       0644,
6271                 .proc_handler   =       proc_dointvec_ms_jiffies,
6272         },
6273         {
6274                 .procname       =       "skip_notify_on_dev_down",
6275                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6276                 .maxlen         =       sizeof(int),
6277                 .mode           =       0644,
6278                 .proc_handler   =       proc_dointvec_minmax,
6279                 .extra1         =       SYSCTL_ZERO,
6280                 .extra2         =       SYSCTL_ONE,
6281         },
6282         { }
6283 };
6284
6285 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6286 {
6287         struct ctl_table *table;
6288
6289         table = kmemdup(ipv6_route_table_template,
6290                         sizeof(ipv6_route_table_template),
6291                         GFP_KERNEL);
6292
6293         if (table) {
6294                 table[0].data = &net->ipv6.sysctl.flush_delay;
6295                 table[0].extra1 = net;
6296                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6297                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6298                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6299                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6300                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6301                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6302                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6303                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6304                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6305                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6306
6307                 /* Don't export sysctls to unprivileged users */
6308                 if (net->user_ns != &init_user_ns)
6309                         table[0].procname = NULL;
6310         }
6311
6312         return table;
6313 }
6314 #endif
6315
6316 static int __net_init ip6_route_net_init(struct net *net)
6317 {
6318         int ret = -ENOMEM;
6319
6320         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6321                sizeof(net->ipv6.ip6_dst_ops));
6322
6323         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6324                 goto out_ip6_dst_ops;
6325
6326         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6327         if (!net->ipv6.fib6_null_entry)
6328                 goto out_ip6_dst_entries;
6329         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6330                sizeof(*net->ipv6.fib6_null_entry));
6331
6332         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6333                                            sizeof(*net->ipv6.ip6_null_entry),
6334                                            GFP_KERNEL);
6335         if (!net->ipv6.ip6_null_entry)
6336                 goto out_fib6_null_entry;
6337         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6338         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6339                          ip6_template_metrics, true);
6340         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6341
6342 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6343         net->ipv6.fib6_has_custom_rules = false;
6344         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6345                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6346                                                GFP_KERNEL);
6347         if (!net->ipv6.ip6_prohibit_entry)
6348                 goto out_ip6_null_entry;
6349         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6350         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6351                          ip6_template_metrics, true);
6352         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6353
6354         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6355                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6356                                                GFP_KERNEL);
6357         if (!net->ipv6.ip6_blk_hole_entry)
6358                 goto out_ip6_prohibit_entry;
6359         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6360         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6361                          ip6_template_metrics, true);
6362         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6363 #ifdef CONFIG_IPV6_SUBTREES
6364         net->ipv6.fib6_routes_require_src = 0;
6365 #endif
6366 #endif
6367
6368         net->ipv6.sysctl.flush_delay = 0;
6369         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6370         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6371         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6372         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6373         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6374         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6375         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6376         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6377
6378         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6379
6380         ret = 0;
6381 out:
6382         return ret;
6383
6384 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6385 out_ip6_prohibit_entry:
6386         kfree(net->ipv6.ip6_prohibit_entry);
6387 out_ip6_null_entry:
6388         kfree(net->ipv6.ip6_null_entry);
6389 #endif
6390 out_fib6_null_entry:
6391         kfree(net->ipv6.fib6_null_entry);
6392 out_ip6_dst_entries:
6393         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6394 out_ip6_dst_ops:
6395         goto out;
6396 }
6397
6398 static void __net_exit ip6_route_net_exit(struct net *net)
6399 {
6400         kfree(net->ipv6.fib6_null_entry);
6401         kfree(net->ipv6.ip6_null_entry);
6402 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6403         kfree(net->ipv6.ip6_prohibit_entry);
6404         kfree(net->ipv6.ip6_blk_hole_entry);
6405 #endif
6406         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6407 }
6408
6409 static int __net_init ip6_route_net_init_late(struct net *net)
6410 {
6411 #ifdef CONFIG_PROC_FS
6412         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6413                         sizeof(struct ipv6_route_iter));
6414         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6415                         rt6_stats_seq_show, NULL);
6416 #endif
6417         return 0;
6418 }
6419
6420 static void __net_exit ip6_route_net_exit_late(struct net *net)
6421 {
6422 #ifdef CONFIG_PROC_FS
6423         remove_proc_entry("ipv6_route", net->proc_net);
6424         remove_proc_entry("rt6_stats", net->proc_net);
6425 #endif
6426 }
6427
6428 static struct pernet_operations ip6_route_net_ops = {
6429         .init = ip6_route_net_init,
6430         .exit = ip6_route_net_exit,
6431 };
6432
6433 static int __net_init ipv6_inetpeer_init(struct net *net)
6434 {
6435         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6436
6437         if (!bp)
6438                 return -ENOMEM;
6439         inet_peer_base_init(bp);
6440         net->ipv6.peers = bp;
6441         return 0;
6442 }
6443
6444 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6445 {
6446         struct inet_peer_base *bp = net->ipv6.peers;
6447
6448         net->ipv6.peers = NULL;
6449         inetpeer_invalidate_tree(bp);
6450         kfree(bp);
6451 }
6452
6453 static struct pernet_operations ipv6_inetpeer_ops = {
6454         .init   =       ipv6_inetpeer_init,
6455         .exit   =       ipv6_inetpeer_exit,
6456 };
6457
6458 static struct pernet_operations ip6_route_net_late_ops = {
6459         .init = ip6_route_net_init_late,
6460         .exit = ip6_route_net_exit_late,
6461 };
6462
6463 static struct notifier_block ip6_route_dev_notifier = {
6464         .notifier_call = ip6_route_dev_notify,
6465         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6466 };
6467
6468 void __init ip6_route_init_special_entries(void)
6469 {
6470         /* Registering of the loopback is done before this portion of code,
6471          * the loopback reference in rt6_info will not be taken, do it
6472          * manually for init_net */
6473         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6474         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6475         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6476   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6477         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6478         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6479         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6480         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6481   #endif
6482 }
6483
6484 #if IS_BUILTIN(CONFIG_IPV6)
6485 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6486 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6487
6488 BTF_ID_LIST(btf_fib6_info_id)
6489 BTF_ID(struct, fib6_info)
6490
6491 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6492         .seq_ops                = &ipv6_route_seq_ops,
6493         .init_seq_private       = bpf_iter_init_seq_net,
6494         .fini_seq_private       = bpf_iter_fini_seq_net,
6495         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6496 };
6497
6498 static struct bpf_iter_reg ipv6_route_reg_info = {
6499         .target                 = "ipv6_route",
6500         .ctx_arg_info_size      = 1,
6501         .ctx_arg_info           = {
6502                 { offsetof(struct bpf_iter__ipv6_route, rt),
6503                   PTR_TO_BTF_ID_OR_NULL },
6504         },
6505         .seq_info               = &ipv6_route_seq_info,
6506 };
6507
6508 static int __init bpf_iter_register(void)
6509 {
6510         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6511         return bpf_iter_reg_target(&ipv6_route_reg_info);
6512 }
6513
6514 static void bpf_iter_unregister(void)
6515 {
6516         bpf_iter_unreg_target(&ipv6_route_reg_info);
6517 }
6518 #endif
6519 #endif
6520
6521 int __init ip6_route_init(void)
6522 {
6523         int ret;
6524         int cpu;
6525
6526         ret = -ENOMEM;
6527         ip6_dst_ops_template.kmem_cachep =
6528                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6529                                   SLAB_HWCACHE_ALIGN, NULL);
6530         if (!ip6_dst_ops_template.kmem_cachep)
6531                 goto out;
6532
6533         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6534         if (ret)
6535                 goto out_kmem_cache;
6536
6537         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6538         if (ret)
6539                 goto out_dst_entries;
6540
6541         ret = register_pernet_subsys(&ip6_route_net_ops);
6542         if (ret)
6543                 goto out_register_inetpeer;
6544
6545         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6546
6547         ret = fib6_init();
6548         if (ret)
6549                 goto out_register_subsys;
6550
6551         ret = xfrm6_init();
6552         if (ret)
6553                 goto out_fib6_init;
6554
6555         ret = fib6_rules_init();
6556         if (ret)
6557                 goto xfrm6_init;
6558
6559         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6560         if (ret)
6561                 goto fib6_rules_init;
6562
6563         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6564                                    inet6_rtm_newroute, NULL, 0);
6565         if (ret < 0)
6566                 goto out_register_late_subsys;
6567
6568         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6569                                    inet6_rtm_delroute, NULL, 0);
6570         if (ret < 0)
6571                 goto out_register_late_subsys;
6572
6573         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6574                                    inet6_rtm_getroute, NULL,
6575                                    RTNL_FLAG_DOIT_UNLOCKED);
6576         if (ret < 0)
6577                 goto out_register_late_subsys;
6578
6579         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6580         if (ret)
6581                 goto out_register_late_subsys;
6582
6583 #if IS_BUILTIN(CONFIG_IPV6)
6584 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6585         ret = bpf_iter_register();
6586         if (ret)
6587                 goto out_register_late_subsys;
6588 #endif
6589 #endif
6590
6591         for_each_possible_cpu(cpu) {
6592                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6593
6594                 INIT_LIST_HEAD(&ul->head);
6595                 spin_lock_init(&ul->lock);
6596         }
6597
6598 out:
6599         return ret;
6600
6601 out_register_late_subsys:
6602         rtnl_unregister_all(PF_INET6);
6603         unregister_pernet_subsys(&ip6_route_net_late_ops);
6604 fib6_rules_init:
6605         fib6_rules_cleanup();
6606 xfrm6_init:
6607         xfrm6_fini();
6608 out_fib6_init:
6609         fib6_gc_cleanup();
6610 out_register_subsys:
6611         unregister_pernet_subsys(&ip6_route_net_ops);
6612 out_register_inetpeer:
6613         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6614 out_dst_entries:
6615         dst_entries_destroy(&ip6_dst_blackhole_ops);
6616 out_kmem_cache:
6617         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6618         goto out;
6619 }
6620
6621 void ip6_route_cleanup(void)
6622 {
6623 #if IS_BUILTIN(CONFIG_IPV6)
6624 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6625         bpf_iter_unregister();
6626 #endif
6627 #endif
6628         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6629         unregister_pernet_subsys(&ip6_route_net_late_ops);
6630         fib6_rules_cleanup();
6631         xfrm6_fini();
6632         fib6_gc_cleanup();
6633         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6634         unregister_pernet_subsys(&ip6_route_net_ops);
6635         dst_entries_destroy(&ip6_dst_blackhole_ops);
6636         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6637 }