net/ipv6/route.c

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