net/ipv6/route.c

   1 /*
   2  *      Linux INET6 implementation
   3  *      FIB front-end.
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  */
  13
  14 /*      Changes:
  15  *
  16  *      YOSHIFUJI Hideaki @USAGI
  17  *              reworked default router selection.
  18  *              - respect outgoing interface
  19  *              - select from (probably) reachable routers (i.e.
  20  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  21  *              - always select the same router if it is (probably)
  22  *              reachable.  otherwise, round-robin the list.
  23  *      Ville Nuorvala
  24  *              Fixed routing subtrees.
  25  */
  26
  27 #define pr_fmt(fmt) "IPv6: " fmt
  28
  29 #include <linux/capability.h>
  30 #include <linux/errno.h>
  31 #include <linux/export.h>
  32 #include <linux/types.h>
  33 #include <linux/times.h>
  34 #include <linux/socket.h>
  35 #include <linux/sockios.h>
  36 #include <linux/net.h>
  37 #include <linux/route.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/in6.h>
  40 #include <linux/mroute6.h>
  41 #include <linux/init.h>
  42 #include <linux/if_arp.h>
  43 #include <linux/proc_fs.h>
  44 #include <linux/seq_file.h>
  45 #include <linux/nsproxy.h>
  46 #include <linux/slab.h>
  47 #include <linux/jhash.h>
  48 #include <net/net_namespace.h>
  49 #include <net/snmp.h>
  50 #include <net/ipv6.h>
  51 #include <net/ip6_fib.h>
  52 #include <net/ip6_route.h>
  53 #include <net/ndisc.h>
  54 #include <net/addrconf.h>
  55 #include <net/tcp.h>
  56 #include <linux/rtnetlink.h>
  57 #include <net/dst.h>
  58 #include <net/dst_metadata.h>
  59 #include <net/xfrm.h>
  60 #include <net/netevent.h>
  61 #include <net/netlink.h>
  62 #include <net/rtnh.h>
  63 #include <net/lwtunnel.h>
  64 #include <net/ip_tunnels.h>
  65 #include <net/l3mdev.h>
  66 #include <net/ip.h>
  67 #include <linux/uaccess.h>
  68
  69 #ifdef CONFIG_SYSCTL
  70 #include <linux/sysctl.h>
  71 #endif
  72
  73 static int ip6_rt_type_to_error(u8 fib6_type);
  74
  75 #define CREATE_TRACE_POINTS
  76 #include <trace/events/fib6.h>
  77 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  78 #undef CREATE_TRACE_POINTS
  79
  80 enum rt6_nud_state {
  81         RT6_NUD_FAIL_HARD = -3,
  82         RT6_NUD_FAIL_PROBE = -2,
  83         RT6_NUD_FAIL_DO_RR = -1,
  84         RT6_NUD_SUCCEED = 1
  85 };
  86
  87 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  88 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  89 static unsigned int      ip6_mtu(const struct dst_entry *dst);
  90 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  91 static void             ip6_dst_destroy(struct dst_entry *);
  92 static void             ip6_dst_ifdown(struct dst_entry *,
  93                                        struct net_device *dev, int how);
  94 static int               ip6_dst_gc(struct dst_ops *ops);
  95
  96 static int              ip6_pkt_discard(struct sk_buff *skb);
  97 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  98 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  99 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
 100 static void             ip6_link_failure(struct sk_buff *skb);
 101 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 102                                            struct sk_buff *skb, u32 mtu);
 103 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 *rt);
 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 = loopback_dev;
 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(&rt->rt6i_gateway, dst->dev, skb, daddr);
 226 }
 227
 228 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 229 {
 230         struct net_device *dev = dst->dev;
 231         struct rt6_info *rt = (struct rt6_info *)dst;
 232
 233         daddr = choose_neigh_daddr(&rt->rt6i_gateway, NULL, daddr);
 234         if (!daddr)
 235                 return;
 236         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 237                 return;
 238         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 239                 return;
 240         __ipv6_confirm_neigh(dev, daddr);
 241 }
 242
 243 static struct dst_ops ip6_dst_ops_template = {
 244         .family                 =       AF_INET6,
 245         .gc                     =       ip6_dst_gc,
 246         .gc_thresh              =       1024,
 247         .check                  =       ip6_dst_check,
 248         .default_advmss         =       ip6_default_advmss,
 249         .mtu                    =       ip6_mtu,
 250         .cow_metrics            =       dst_cow_metrics_generic,
 251         .destroy                =       ip6_dst_destroy,
 252         .ifdown                 =       ip6_dst_ifdown,
 253         .negative_advice        =       ip6_negative_advice,
 254         .link_failure           =       ip6_link_failure,
 255         .update_pmtu            =       ip6_rt_update_pmtu,
 256         .redirect               =       rt6_do_redirect,
 257         .local_out              =       __ip6_local_out,
 258         .neigh_lookup           =       ip6_dst_neigh_lookup,
 259         .confirm_neigh          =       ip6_confirm_neigh,
 260 };
 261
 262 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 263 {
 264         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 265
 266         return mtu ? : dst->dev->mtu;
 267 }
 268
 269 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 270                                          struct sk_buff *skb, u32 mtu)
 271 {
 272 }
 273
 274 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 275                                       struct sk_buff *skb)
 276 {
 277 }
 278
 279 static struct dst_ops ip6_dst_blackhole_ops = {
 280         .family                 =       AF_INET6,
 281         .destroy                =       ip6_dst_destroy,
 282         .check                  =       ip6_dst_check,
 283         .mtu                    =       ip6_blackhole_mtu,
 284         .default_advmss         =       ip6_default_advmss,
 285         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 286         .redirect               =       ip6_rt_blackhole_redirect,
 287         .cow_metrics            =       dst_cow_metrics_generic,
 288         .neigh_lookup           =       ip6_dst_neigh_lookup,
 289 };
 290
 291 static const u32 ip6_template_metrics[RTAX_MAX] = {
 292         [RTAX_HOPLIMIT - 1] = 0,
 293 };
 294
 295 static const struct fib6_info fib6_null_entry_template = {
 296         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 297         .fib6_protocol  = RTPROT_KERNEL,
 298         .fib6_metric    = ~(u32)0,
 299         .fib6_ref       = REFCOUNT_INIT(1),
 300         .fib6_type      = RTN_UNREACHABLE,
 301         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 302 };
 303
 304 static const struct rt6_info ip6_null_entry_template = {
 305         .dst = {
 306                 .__refcnt       = ATOMIC_INIT(1),
 307                 .__use          = 1,
 308                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 309                 .error          = -ENETUNREACH,
 310                 .input          = ip6_pkt_discard,
 311                 .output         = ip6_pkt_discard_out,
 312         },
 313         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 314 };
 315
 316 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 317
 318 static const struct rt6_info ip6_prohibit_entry_template = {
 319         .dst = {
 320                 .__refcnt       = ATOMIC_INIT(1),
 321                 .__use          = 1,
 322                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 323                 .error          = -EACCES,
 324                 .input          = ip6_pkt_prohibit,
 325                 .output         = ip6_pkt_prohibit_out,
 326         },
 327         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 328 };
 329
 330 static const struct rt6_info ip6_blk_hole_entry_template = {
 331         .dst = {
 332                 .__refcnt       = ATOMIC_INIT(1),
 333                 .__use          = 1,
 334                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 335                 .error          = -EINVAL,
 336                 .input          = dst_discard,
 337                 .output         = dst_discard_out,
 338         },
 339         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 340 };
 341
 342 #endif
 343
 344 static void rt6_info_init(struct rt6_info *rt)
 345 {
 346         struct dst_entry *dst = &rt->dst;
 347
 348         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 349         INIT_LIST_HEAD(&rt->rt6i_uncached);
 350 }
 351
 352 /* allocate dst with ip6_dst_ops */
 353 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 354                                int flags)
 355 {
 356         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 357                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 358
 359         if (rt) {
 360                 rt6_info_init(rt);
 361                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 362         }
 363
 364         return rt;
 365 }
 366 EXPORT_SYMBOL(ip6_dst_alloc);
 367
 368 static void ip6_dst_destroy(struct dst_entry *dst)
 369 {
 370         struct rt6_info *rt = (struct rt6_info *)dst;
 371         struct fib6_info *from;
 372         struct inet6_dev *idev;
 373
 374         ip_dst_metrics_put(dst);
 375         rt6_uncached_list_del(rt);
 376
 377         idev = rt->rt6i_idev;
 378         if (idev) {
 379                 rt->rt6i_idev = NULL;
 380                 in6_dev_put(idev);
 381         }
 382
 383         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 384         fib6_info_release(from);
 385 }
 386
 387 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 388                            int how)
 389 {
 390         struct rt6_info *rt = (struct rt6_info *)dst;
 391         struct inet6_dev *idev = rt->rt6i_idev;
 392         struct net_device *loopback_dev =
 393                 dev_net(dev)->loopback_dev;
 394
 395         if (idev && idev->dev != loopback_dev) {
 396                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 397                 if (loopback_idev) {
 398                         rt->rt6i_idev = loopback_idev;
 399                         in6_dev_put(idev);
 400                 }
 401         }
 402 }
 403
 404 static bool __rt6_check_expired(const struct rt6_info *rt)
 405 {
 406         if (rt->rt6i_flags & RTF_EXPIRES)
 407                 return time_after(jiffies, rt->dst.expires);
 408         else
 409                 return false;
 410 }
 411
 412 static bool rt6_check_expired(const struct rt6_info *rt)
 413 {
 414         struct fib6_info *from;
 415
 416         from = rcu_dereference(rt->from);
 417
 418         if (rt->rt6i_flags & RTF_EXPIRES) {
 419                 if (time_after(jiffies, rt->dst.expires))
 420                         return true;
 421         } else if (from) {
 422                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 423                         fib6_check_expired(from);
 424         }
 425         return false;
 426 }
 427
 428 void fib6_select_path(const struct net *net, struct fib6_result *res,
 429                       struct flowi6 *fl6, int oif, bool have_oif_match,
 430                       const struct sk_buff *skb, int strict)
 431 {
 432         struct fib6_info *sibling, *next_sibling;
 433         struct fib6_info *match = res->f6i;
 434
 435         if (!match->fib6_nsiblings || have_oif_match)
 436                 goto out;
 437
 438         /* We might have already computed the hash for ICMPv6 errors. In such
 439          * case it will always be non-zero. Otherwise now is the time to do it.
 440          */
 441         if (!fl6->mp_hash)
 442                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 443
 444         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 445                 goto out;
 446
 447         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 448                                  fib6_siblings) {
 449                 const struct fib6_nh *nh = sibling->fib6_nh;
 450                 int nh_upper_bound;
 451
 452                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 453                 if (fl6->mp_hash > nh_upper_bound)
 454                         continue;
 455                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 456                         break;
 457                 match = sibling;
 458                 break;
 459         }
 460
 461 out:
 462         res->f6i = match;
 463         res->nh = match->fib6_nh;
 464 }
 465
 466 /*
 467  *      Route lookup. rcu_read_lock() should be held.
 468  */
 469
 470 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 471                                const struct in6_addr *saddr, int oif, int flags)
 472 {
 473         const struct net_device *dev;
 474
 475         if (nh->fib_nh_flags & RTNH_F_DEAD)
 476                 return false;
 477
 478         dev = nh->fib_nh_dev;
 479         if (oif) {
 480                 if (dev->ifindex == oif)
 481                         return true;
 482         } else {
 483                 if (ipv6_chk_addr(net, saddr, dev,
 484                                   flags & RT6_LOOKUP_F_IFACE))
 485                         return true;
 486         }
 487
 488         return false;
 489 }
 490
 491 static void rt6_device_match(struct net *net, struct fib6_result *res,
 492                              const struct in6_addr *saddr, int oif, int flags)
 493 {
 494         struct fib6_info *f6i = res->f6i;
 495         struct fib6_info *spf6i;
 496         struct fib6_nh *nh;
 497
 498         if (!oif && ipv6_addr_any(saddr)) {
 499                 nh = f6i->fib6_nh;
 500                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 501                         goto out;
 502         }
 503
 504         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 505                 nh = spf6i->fib6_nh;
 506                 if (__rt6_device_match(net, nh, saddr, oif, flags)) {
 507                         res->f6i = spf6i;
 508                         goto out;
 509                 }
 510         }
 511
 512         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 513                 res->f6i = net->ipv6.fib6_null_entry;
 514                 nh = res->f6i->fib6_nh;
 515                 goto out;
 516         }
 517
 518         nh = f6i->fib6_nh;
 519         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 520                 res->f6i = net->ipv6.fib6_null_entry;
 521                 nh = res->f6i->fib6_nh;
 522         }
 523 out:
 524         res->nh = nh;
 525         res->fib6_type = res->f6i->fib6_type;
 526         res->fib6_flags = res->f6i->fib6_flags;
 527 }
 528
 529 #ifdef CONFIG_IPV6_ROUTER_PREF
 530 struct __rt6_probe_work {
 531         struct work_struct work;
 532         struct in6_addr target;
 533         struct net_device *dev;
 534 };
 535
 536 static void rt6_probe_deferred(struct work_struct *w)
 537 {
 538         struct in6_addr mcaddr;
 539         struct __rt6_probe_work *work =
 540                 container_of(w, struct __rt6_probe_work, work);
 541
 542         addrconf_addr_solict_mult(&work->target, &mcaddr);
 543         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 544         dev_put(work->dev);
 545         kfree(work);
 546 }
 547
 548 static void rt6_probe(struct fib6_nh *fib6_nh)
 549 {
 550         struct __rt6_probe_work *work = NULL;
 551         const struct in6_addr *nh_gw;
 552         struct neighbour *neigh;
 553         struct net_device *dev;
 554         struct inet6_dev *idev;
 555
 556         /*
 557          * Okay, this does not seem to be appropriate
 558          * for now, however, we need to check if it
 559          * is really so; aka Router Reachability Probing.
 560          *
 561          * Router Reachability Probe MUST be rate-limited
 562          * to no more than one per minute.
 563          */
 564         if (fib6_nh->fib_nh_gw_family)
 565                 return;
 566
 567         nh_gw = &fib6_nh->fib_nh_gw6;
 568         dev = fib6_nh->fib_nh_dev;
 569         rcu_read_lock_bh();
 570         idev = __in6_dev_get(dev);
 571         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 572         if (neigh) {
 573                 if (neigh->nud_state & NUD_VALID)
 574                         goto out;
 575
 576                 write_lock(&neigh->lock);
 577                 if (!(neigh->nud_state & NUD_VALID) &&
 578                     time_after(jiffies,
 579                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 580                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 581                         if (work)
 582                                 __neigh_set_probe_once(neigh);
 583                 }
 584                 write_unlock(&neigh->lock);
 585         } else if (time_after(jiffies, fib6_nh->last_probe +
 586                                        idev->cnf.rtr_probe_interval)) {
 587                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 588         }
 589
 590         if (work) {
 591                 fib6_nh->last_probe = jiffies;
 592                 INIT_WORK(&work->work, rt6_probe_deferred);
 593                 work->target = *nh_gw;
 594                 dev_hold(dev);
 595                 work->dev = dev;
 596                 schedule_work(&work->work);
 597         }
 598
 599 out:
 600         rcu_read_unlock_bh();
 601 }
 602 #else
 603 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 604 {
 605 }
 606 #endif
 607
 608 /*
 609  * Default Router Selection (RFC 2461 6.3.6)
 610  */
 611 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 612 {
 613         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 614         struct neighbour *neigh;
 615
 616         rcu_read_lock_bh();
 617         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 618                                           &fib6_nh->fib_nh_gw6);
 619         if (neigh) {
 620                 read_lock(&neigh->lock);
 621                 if (neigh->nud_state & NUD_VALID)
 622                         ret = RT6_NUD_SUCCEED;
 623 #ifdef CONFIG_IPV6_ROUTER_PREF
 624                 else if (!(neigh->nud_state & NUD_FAILED))
 625                         ret = RT6_NUD_SUCCEED;
 626                 else
 627                         ret = RT6_NUD_FAIL_PROBE;
 628 #endif
 629                 read_unlock(&neigh->lock);
 630         } else {
 631                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 632                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 633         }
 634         rcu_read_unlock_bh();
 635
 636         return ret;
 637 }
 638
 639 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 640                            int strict)
 641 {
 642         int m = 0;
 643
 644         if (!oif || nh->fib_nh_dev->ifindex == oif)
 645                 m = 2;
 646
 647         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 648                 return RT6_NUD_FAIL_HARD;
 649 #ifdef CONFIG_IPV6_ROUTER_PREF
 650         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 651 #endif
 652         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 653             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 654                 int n = rt6_check_neigh(nh);
 655                 if (n < 0)
 656                         return n;
 657         }
 658         return m;
 659 }
 660
 661 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 662                        int oif, int strict, int *mpri, bool *do_rr)
 663 {
 664         bool match_do_rr = false;
 665         bool rc = false;
 666         int m;
 667
 668         if (nh->fib_nh_flags & RTNH_F_DEAD)
 669                 goto out;
 670
 671         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 672             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 673             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 674                 goto out;
 675
 676         m = rt6_score_route(nh, fib6_flags, oif, strict);
 677         if (m == RT6_NUD_FAIL_DO_RR) {
 678                 match_do_rr = true;
 679                 m = 0; /* lowest valid score */
 680         } else if (m == RT6_NUD_FAIL_HARD) {
 681                 goto out;
 682         }
 683
 684         if (strict & RT6_LOOKUP_F_REACHABLE)
 685                 rt6_probe(nh);
 686
 687         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 688         if (m > *mpri) {
 689                 *do_rr = match_do_rr;
 690                 *mpri = m;
 691                 rc = true;
 692         }
 693 out:
 694         return rc;
 695 }
 696
 697 static void __find_rr_leaf(struct fib6_info *f6i_start,
 698                            struct fib6_info *nomatch, u32 metric,
 699                            struct fib6_result *res, struct fib6_info **cont,
 700                            int oif, int strict, bool *do_rr, int *mpri)
 701 {
 702         struct fib6_info *f6i;
 703
 704         for (f6i = f6i_start;
 705              f6i && f6i != nomatch;
 706              f6i = rcu_dereference(f6i->fib6_next)) {
 707                 struct fib6_nh *nh;
 708
 709                 if (cont && f6i->fib6_metric != metric) {
 710                         *cont = f6i;
 711                         return;
 712                 }
 713
 714                 if (fib6_check_expired(f6i))
 715                         continue;
 716
 717                 nh = f6i->fib6_nh;
 718                 if (find_match(nh, f6i->fib6_flags, oif, strict, mpri, do_rr)) {
 719                         res->f6i = f6i;
 720                         res->nh = nh;
 721                         res->fib6_flags = f6i->fib6_flags;
 722                         res->fib6_type = f6i->fib6_type;
 723                 }
 724         }
 725 }
 726
 727 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 728                          struct fib6_info *rr_head, int oif, int strict,
 729                          bool *do_rr, struct fib6_result *res)
 730 {
 731         u32 metric = rr_head->fib6_metric;
 732         struct fib6_info *cont = NULL;
 733         int mpri = -1;
 734
 735         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 736                        oif, strict, do_rr, &mpri);
 737
 738         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 739                        oif, strict, do_rr, &mpri);
 740
 741         if (res->f6i || !cont)
 742                 return;
 743
 744         __find_rr_leaf(cont, NULL, metric, res, NULL,
 745                        oif, strict, do_rr, &mpri);
 746 }
 747
 748 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 749                        struct fib6_result *res, int strict)
 750 {
 751         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 752         struct fib6_info *rt0;
 753         bool do_rr = false;
 754         int key_plen;
 755
 756         /* make sure this function or its helpers sets f6i */
 757         res->f6i = NULL;
 758
 759         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 760                 goto out;
 761
 762         rt0 = rcu_dereference(fn->rr_ptr);
 763         if (!rt0)
 764                 rt0 = leaf;
 765
 766         /* Double check to make sure fn is not an intermediate node
 767          * and fn->leaf does not points to its child's leaf
 768          * (This might happen if all routes under fn are deleted from
 769          * the tree and fib6_repair_tree() is called on the node.)
 770          */
 771         key_plen = rt0->fib6_dst.plen;
 772 #ifdef CONFIG_IPV6_SUBTREES
 773         if (rt0->fib6_src.plen)
 774                 key_plen = rt0->fib6_src.plen;
 775 #endif
 776         if (fn->fn_bit != key_plen)
 777                 goto out;
 778
 779         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 780         if (do_rr) {
 781                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 782
 783                 /* no entries matched; do round-robin */
 784                 if (!next || next->fib6_metric != rt0->fib6_metric)
 785                         next = leaf;
 786
 787                 if (next != rt0) {
 788                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 789                         /* make sure next is not being deleted from the tree */
 790                         if (next->fib6_node)
 791                                 rcu_assign_pointer(fn->rr_ptr, next);
 792                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 793                 }
 794         }
 795
 796 out:
 797         if (!res->f6i) {
 798                 res->f6i = net->ipv6.fib6_null_entry;
 799                 res->nh = res->f6i->fib6_nh;
 800                 res->fib6_flags = res->f6i->fib6_flags;
 801                 res->fib6_type = res->f6i->fib6_type;
 802         }
 803 }
 804
 805 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 806 {
 807         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 808                res->nh->fib_nh_gw_family;
 809 }
 810
 811 #ifdef CONFIG_IPV6_ROUTE_INFO
 812 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 813                   const struct in6_addr *gwaddr)
 814 {
 815         struct net *net = dev_net(dev);
 816         struct route_info *rinfo = (struct route_info *) opt;
 817         struct in6_addr prefix_buf, *prefix;
 818         unsigned int pref;
 819         unsigned long lifetime;
 820         struct fib6_info *rt;
 821
 822         if (len < sizeof(struct route_info)) {
 823                 return -EINVAL;
 824         }
 825
 826         /* Sanity check for prefix_len and length */
 827         if (rinfo->length > 3) {
 828                 return -EINVAL;
 829         } else if (rinfo->prefix_len > 128) {
 830                 return -EINVAL;
 831         } else if (rinfo->prefix_len > 64) {
 832                 if (rinfo->length < 2) {
 833                         return -EINVAL;
 834                 }
 835         } else if (rinfo->prefix_len > 0) {
 836                 if (rinfo->length < 1) {
 837                         return -EINVAL;
 838                 }
 839         }
 840
 841         pref = rinfo->route_pref;
 842         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 843                 return -EINVAL;
 844
 845         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 846
 847         if (rinfo->length == 3)
 848                 prefix = (struct in6_addr *)rinfo->prefix;
 849         else {
 850                 /* this function is safe */
 851                 ipv6_addr_prefix(&prefix_buf,
 852                                  (struct in6_addr *)rinfo->prefix,
 853                                  rinfo->prefix_len);
 854                 prefix = &prefix_buf;
 855         }
 856
 857         if (rinfo->prefix_len == 0)
 858                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 859         else
 860                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 861                                         gwaddr, dev);
 862
 863         if (rt && !lifetime) {
 864                 ip6_del_rt(net, rt);
 865                 rt = NULL;
 866         }
 867
 868         if (!rt && lifetime)
 869                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 870                                         dev, pref);
 871         else if (rt)
 872                 rt->fib6_flags = RTF_ROUTEINFO |
 873                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 874
 875         if (rt) {
 876                 if (!addrconf_finite_timeout(lifetime))
 877                         fib6_clean_expires(rt);
 878                 else
 879                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 880
 881                 fib6_info_release(rt);
 882         }
 883         return 0;
 884 }
 885 #endif
 886
 887 /*
 888  *      Misc support functions
 889  */
 890
 891 /* called with rcu_lock held */
 892 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
 893 {
 894         struct net_device *dev = res->nh->fib_nh_dev;
 895
 896         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
 897                 /* for copies of local routes, dst->dev needs to be the
 898                  * device if it is a master device, the master device if
 899                  * device is enslaved, and the loopback as the default
 900                  */
 901                 if (netif_is_l3_slave(dev) &&
 902                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
 903                         dev = l3mdev_master_dev_rcu(dev);
 904                 else if (!netif_is_l3_master(dev))
 905                         dev = dev_net(dev)->loopback_dev;
 906                 /* last case is netif_is_l3_master(dev) is true in which
 907                  * case we want dev returned to be dev
 908                  */
 909         }
 910
 911         return dev;
 912 }
 913
 914 static const int fib6_prop[RTN_MAX + 1] = {
 915         [RTN_UNSPEC]    = 0,
 916         [RTN_UNICAST]   = 0,
 917         [RTN_LOCAL]     = 0,
 918         [RTN_BROADCAST] = 0,
 919         [RTN_ANYCAST]   = 0,
 920         [RTN_MULTICAST] = 0,
 921         [RTN_BLACKHOLE] = -EINVAL,
 922         [RTN_UNREACHABLE] = -EHOSTUNREACH,
 923         [RTN_PROHIBIT]  = -EACCES,
 924         [RTN_THROW]     = -EAGAIN,
 925         [RTN_NAT]       = -EINVAL,
 926         [RTN_XRESOLVE]  = -EINVAL,
 927 };
 928
 929 static int ip6_rt_type_to_error(u8 fib6_type)
 930 {
 931         return fib6_prop[fib6_type];
 932 }
 933
 934 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
 935 {
 936         unsigned short flags = 0;
 937
 938         if (rt->dst_nocount)
 939                 flags |= DST_NOCOUNT;
 940         if (rt->dst_nopolicy)
 941                 flags |= DST_NOPOLICY;
 942         if (rt->dst_host)
 943                 flags |= DST_HOST;
 944
 945         return flags;
 946 }
 947
 948 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
 949 {
 950         rt->dst.error = ip6_rt_type_to_error(fib6_type);
 951
 952         switch (fib6_type) {
 953         case RTN_BLACKHOLE:
 954                 rt->dst.output = dst_discard_out;
 955                 rt->dst.input = dst_discard;
 956                 break;
 957         case RTN_PROHIBIT:
 958                 rt->dst.output = ip6_pkt_prohibit_out;
 959                 rt->dst.input = ip6_pkt_prohibit;
 960                 break;
 961         case RTN_THROW:
 962         case RTN_UNREACHABLE:
 963         default:
 964                 rt->dst.output = ip6_pkt_discard_out;
 965                 rt->dst.input = ip6_pkt_discard;
 966                 break;
 967         }
 968 }
 969
 970 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
 971 {
 972         struct fib6_info *f6i = res->f6i;
 973
 974         if (res->fib6_flags & RTF_REJECT) {
 975                 ip6_rt_init_dst_reject(rt, res->fib6_type);
 976                 return;
 977         }
 978
 979         rt->dst.error = 0;
 980         rt->dst.output = ip6_output;
 981
 982         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
 983                 rt->dst.input = ip6_input;
 984         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
 985                 rt->dst.input = ip6_mc_input;
 986         } else {
 987                 rt->dst.input = ip6_forward;
 988         }
 989
 990         if (res->nh->fib_nh_lws) {
 991                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
 992                 lwtunnel_set_redirect(&rt->dst);
 993         }
 994
 995         rt->dst.lastuse = jiffies;
 996 }
 997
 998 /* Caller must already hold reference to @from */
 999 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1000 {
1001         rt->rt6i_flags &= ~RTF_EXPIRES;
1002         rcu_assign_pointer(rt->from, from);
1003         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1004 }
1005
1006 /* Caller must already hold reference to f6i in result */
1007 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1008 {
1009         const struct fib6_nh *nh = res->nh;
1010         const struct net_device *dev = nh->fib_nh_dev;
1011         struct fib6_info *f6i = res->f6i;
1012
1013         ip6_rt_init_dst(rt, res);
1014
1015         rt->rt6i_dst = f6i->fib6_dst;
1016         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1017         rt->rt6i_flags = res->fib6_flags;
1018         if (nh->fib_nh_gw_family) {
1019                 rt->rt6i_gateway = nh->fib_nh_gw6;
1020                 rt->rt6i_flags |= RTF_GATEWAY;
1021         }
1022         rt6_set_from(rt, f6i);
1023 #ifdef CONFIG_IPV6_SUBTREES
1024         rt->rt6i_src = f6i->fib6_src;
1025 #endif
1026 }
1027
1028 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1029                                         struct in6_addr *saddr)
1030 {
1031         struct fib6_node *pn, *sn;
1032         while (1) {
1033                 if (fn->fn_flags & RTN_TL_ROOT)
1034                         return NULL;
1035                 pn = rcu_dereference(fn->parent);
1036                 sn = FIB6_SUBTREE(pn);
1037                 if (sn && sn != fn)
1038                         fn = fib6_node_lookup(sn, NULL, saddr);
1039                 else
1040                         fn = pn;
1041                 if (fn->fn_flags & RTN_RTINFO)
1042                         return fn;
1043         }
1044 }
1045
1046 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1047 {
1048         struct rt6_info *rt = *prt;
1049
1050         if (dst_hold_safe(&rt->dst))
1051                 return true;
1052         if (net) {
1053                 rt = net->ipv6.ip6_null_entry;
1054                 dst_hold(&rt->dst);
1055         } else {
1056                 rt = NULL;
1057         }
1058         *prt = rt;
1059         return false;
1060 }
1061
1062 /* called with rcu_lock held */
1063 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1064 {
1065         struct net_device *dev = res->nh->fib_nh_dev;
1066         struct fib6_info *f6i = res->f6i;
1067         unsigned short flags;
1068         struct rt6_info *nrt;
1069
1070         if (!fib6_info_hold_safe(f6i))
1071                 goto fallback;
1072
1073         flags = fib6_info_dst_flags(f6i);
1074         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1075         if (!nrt) {
1076                 fib6_info_release(f6i);
1077                 goto fallback;
1078         }
1079
1080         ip6_rt_copy_init(nrt, res);
1081         return nrt;
1082
1083 fallback:
1084         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1085         dst_hold(&nrt->dst);
1086         return nrt;
1087 }
1088
1089 static struct rt6_info *ip6_pol_route_lookup(struct net *net,
1090                                              struct fib6_table *table,
1091                                              struct flowi6 *fl6,
1092                                              const struct sk_buff *skb,
1093                                              int flags)
1094 {
1095         struct fib6_result res = {};
1096         struct fib6_node *fn;
1097         struct rt6_info *rt;
1098
1099         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1100                 flags &= ~RT6_LOOKUP_F_IFACE;
1101
1102         rcu_read_lock();
1103         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1104 restart:
1105         res.f6i = rcu_dereference(fn->leaf);
1106         if (!res.f6i)
1107                 res.f6i = net->ipv6.fib6_null_entry;
1108         else
1109                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1110                                  flags);
1111
1112         if (res.f6i == net->ipv6.fib6_null_entry) {
1113                 fn = fib6_backtrack(fn, &fl6->saddr);
1114                 if (fn)
1115                         goto restart;
1116
1117                 rt = net->ipv6.ip6_null_entry;
1118                 dst_hold(&rt->dst);
1119                 goto out;
1120         }
1121
1122         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1123                          fl6->flowi6_oif != 0, skb, flags);
1124
1125         /* Search through exception table */
1126         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1127         if (rt) {
1128                 if (ip6_hold_safe(net, &rt))
1129                         dst_use_noref(&rt->dst, jiffies);
1130         } else {
1131                 rt = ip6_create_rt_rcu(&res);
1132         }
1133
1134 out:
1135         trace_fib6_table_lookup(net, &res, table, fl6);
1136
1137         rcu_read_unlock();
1138
1139         return rt;
1140 }
1141
1142 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1143                                    const struct sk_buff *skb, int flags)
1144 {
1145         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1146 }
1147 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1148
1149 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1150                             const struct in6_addr *saddr, int oif,
1151                             const struct sk_buff *skb, int strict)
1152 {
1153         struct flowi6 fl6 = {
1154                 .flowi6_oif = oif,
1155                 .daddr = *daddr,
1156         };
1157         struct dst_entry *dst;
1158         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1159
1160         if (saddr) {
1161                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1162                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1163         }
1164
1165         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1166         if (dst->error == 0)
1167                 return (struct rt6_info *) dst;
1168
1169         dst_release(dst);
1170
1171         return NULL;
1172 }
1173 EXPORT_SYMBOL(rt6_lookup);
1174
1175 /* ip6_ins_rt is called with FREE table->tb6_lock.
1176  * It takes new route entry, the addition fails by any reason the
1177  * route is released.
1178  * Caller must hold dst before calling it.
1179  */
1180
1181 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1182                         struct netlink_ext_ack *extack)
1183 {
1184         int err;
1185         struct fib6_table *table;
1186
1187         table = rt->fib6_table;
1188         spin_lock_bh(&table->tb6_lock);
1189         err = fib6_add(&table->tb6_root, rt, info, extack);
1190         spin_unlock_bh(&table->tb6_lock);
1191
1192         return err;
1193 }
1194
1195 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1196 {
1197         struct nl_info info = { .nl_net = net, };
1198
1199         return __ip6_ins_rt(rt, &info, NULL);
1200 }
1201
1202 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1203                                            const struct in6_addr *daddr,
1204                                            const struct in6_addr *saddr)
1205 {
1206         struct fib6_info *f6i = res->f6i;
1207         struct net_device *dev;
1208         struct rt6_info *rt;
1209
1210         /*
1211          *      Clone the route.
1212          */
1213
1214         if (!fib6_info_hold_safe(f6i))
1215                 return NULL;
1216
1217         dev = ip6_rt_get_dev_rcu(res);
1218         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1219         if (!rt) {
1220                 fib6_info_release(f6i);
1221                 return NULL;
1222         }
1223
1224         ip6_rt_copy_init(rt, res);
1225         rt->rt6i_flags |= RTF_CACHE;
1226         rt->dst.flags |= DST_HOST;
1227         rt->rt6i_dst.addr = *daddr;
1228         rt->rt6i_dst.plen = 128;
1229
1230         if (!rt6_is_gw_or_nonexthop(res)) {
1231                 if (f6i->fib6_dst.plen != 128 &&
1232                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1233                         rt->rt6i_flags |= RTF_ANYCAST;
1234 #ifdef CONFIG_IPV6_SUBTREES
1235                 if (rt->rt6i_src.plen && saddr) {
1236                         rt->rt6i_src.addr = *saddr;
1237                         rt->rt6i_src.plen = 128;
1238                 }
1239 #endif
1240         }
1241
1242         return rt;
1243 }
1244
1245 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1246 {
1247         struct fib6_info *f6i = res->f6i;
1248         unsigned short flags = fib6_info_dst_flags(f6i);
1249         struct net_device *dev;
1250         struct rt6_info *pcpu_rt;
1251
1252         if (!fib6_info_hold_safe(f6i))
1253                 return NULL;
1254
1255         rcu_read_lock();
1256         dev = ip6_rt_get_dev_rcu(res);
1257         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags);
1258         rcu_read_unlock();
1259         if (!pcpu_rt) {
1260                 fib6_info_release(f6i);
1261                 return NULL;
1262         }
1263         ip6_rt_copy_init(pcpu_rt, res);
1264         pcpu_rt->rt6i_flags |= RTF_PCPU;
1265         return pcpu_rt;
1266 }
1267
1268 /* It should be called with rcu_read_lock() acquired */
1269 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1270 {
1271         struct rt6_info *pcpu_rt, **p;
1272
1273         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1274         pcpu_rt = *p;
1275
1276         if (pcpu_rt)
1277                 ip6_hold_safe(NULL, &pcpu_rt);
1278
1279         return pcpu_rt;
1280 }
1281
1282 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1283                                             const struct fib6_result *res)
1284 {
1285         struct rt6_info *pcpu_rt, *prev, **p;
1286
1287         pcpu_rt = ip6_rt_pcpu_alloc(res);
1288         if (!pcpu_rt) {
1289                 dst_hold(&net->ipv6.ip6_null_entry->dst);
1290                 return net->ipv6.ip6_null_entry;
1291         }
1292
1293         dst_hold(&pcpu_rt->dst);
1294         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1295         prev = cmpxchg(p, NULL, pcpu_rt);
1296         BUG_ON(prev);
1297
1298         if (res->f6i->fib6_destroying) {
1299                 struct fib6_info *from;
1300
1301                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1302                 fib6_info_release(from);
1303         }
1304
1305         return pcpu_rt;
1306 }
1307
1308 /* exception hash table implementation
1309  */
1310 static DEFINE_SPINLOCK(rt6_exception_lock);
1311
1312 /* Remove rt6_ex from hash table and free the memory
1313  * Caller must hold rt6_exception_lock
1314  */
1315 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1316                                  struct rt6_exception *rt6_ex)
1317 {
1318         struct fib6_info *from;
1319         struct net *net;
1320
1321         if (!bucket || !rt6_ex)
1322                 return;
1323
1324         net = dev_net(rt6_ex->rt6i->dst.dev);
1325         net->ipv6.rt6_stats->fib_rt_cache--;
1326
1327         /* purge completely the exception to allow releasing the held resources:
1328          * some [sk] cache may keep the dst around for unlimited time
1329          */
1330         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1331         fib6_info_release(from);
1332         dst_dev_put(&rt6_ex->rt6i->dst);
1333
1334         hlist_del_rcu(&rt6_ex->hlist);
1335         dst_release(&rt6_ex->rt6i->dst);
1336         kfree_rcu(rt6_ex, rcu);
1337         WARN_ON_ONCE(!bucket->depth);
1338         bucket->depth--;
1339 }
1340
1341 /* Remove oldest rt6_ex in bucket and free the memory
1342  * Caller must hold rt6_exception_lock
1343  */
1344 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1345 {
1346         struct rt6_exception *rt6_ex, *oldest = NULL;
1347
1348         if (!bucket)
1349                 return;
1350
1351         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1352                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1353                         oldest = rt6_ex;
1354         }
1355         rt6_remove_exception(bucket, oldest);
1356 }
1357
1358 static u32 rt6_exception_hash(const struct in6_addr *dst,
1359                               const struct in6_addr *src)
1360 {
1361         static u32 seed __read_mostly;
1362         u32 val;
1363
1364         net_get_random_once(&seed, sizeof(seed));
1365         val = jhash(dst, sizeof(*dst), seed);
1366
1367 #ifdef CONFIG_IPV6_SUBTREES
1368         if (src)
1369                 val = jhash(src, sizeof(*src), val);
1370 #endif
1371         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1372 }
1373
1374 /* Helper function to find the cached rt in the hash table
1375  * and update bucket pointer to point to the bucket for this
1376  * (daddr, saddr) pair
1377  * Caller must hold rt6_exception_lock
1378  */
1379 static struct rt6_exception *
1380 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1381                               const struct in6_addr *daddr,
1382                               const struct in6_addr *saddr)
1383 {
1384         struct rt6_exception *rt6_ex;
1385         u32 hval;
1386
1387         if (!(*bucket) || !daddr)
1388                 return NULL;
1389
1390         hval = rt6_exception_hash(daddr, saddr);
1391         *bucket += hval;
1392
1393         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1394                 struct rt6_info *rt6 = rt6_ex->rt6i;
1395                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1396
1397 #ifdef CONFIG_IPV6_SUBTREES
1398                 if (matched && saddr)
1399                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1400 #endif
1401                 if (matched)
1402                         return rt6_ex;
1403         }
1404         return NULL;
1405 }
1406
1407 /* Helper function to find the cached rt in the hash table
1408  * and update bucket pointer to point to the bucket for this
1409  * (daddr, saddr) pair
1410  * Caller must hold rcu_read_lock()
1411  */
1412 static struct rt6_exception *
1413 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1414                          const struct in6_addr *daddr,
1415                          const struct in6_addr *saddr)
1416 {
1417         struct rt6_exception *rt6_ex;
1418         u32 hval;
1419
1420         WARN_ON_ONCE(!rcu_read_lock_held());
1421
1422         if (!(*bucket) || !daddr)
1423                 return NULL;
1424
1425         hval = rt6_exception_hash(daddr, saddr);
1426         *bucket += hval;
1427
1428         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1429                 struct rt6_info *rt6 = rt6_ex->rt6i;
1430                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1431
1432 #ifdef CONFIG_IPV6_SUBTREES
1433                 if (matched && saddr)
1434                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1435 #endif
1436                 if (matched)
1437                         return rt6_ex;
1438         }
1439         return NULL;
1440 }
1441
1442 static unsigned int fib6_mtu(const struct fib6_result *res)
1443 {
1444         const struct fib6_nh *nh = res->nh;
1445         unsigned int mtu;
1446
1447         if (res->f6i->fib6_pmtu) {
1448                 mtu = res->f6i->fib6_pmtu;
1449         } else {
1450                 struct net_device *dev = nh->fib_nh_dev;
1451                 struct inet6_dev *idev;
1452
1453                 rcu_read_lock();
1454                 idev = __in6_dev_get(dev);
1455                 mtu = idev->cnf.mtu6;
1456                 rcu_read_unlock();
1457         }
1458
1459         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1460
1461         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1462 }
1463
1464 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1465
1466 /* used when the flushed bit is not relevant, only access to the bucket
1467  * (ie., all bucket users except rt6_insert_exception);
1468  *
1469  * called under rcu lock; sometimes called with rt6_exception_lock held
1470  */
1471 static
1472 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1473                                                        spinlock_t *lock)
1474 {
1475         struct rt6_exception_bucket *bucket;
1476
1477         if (lock)
1478                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1479                                                    lockdep_is_held(lock));
1480         else
1481                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1482
1483         /* remove bucket flushed bit if set */
1484         if (bucket) {
1485                 unsigned long p = (unsigned long)bucket;
1486
1487                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1488                 bucket = (struct rt6_exception_bucket *)p;
1489         }
1490
1491         return bucket;
1492 }
1493
1494 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1495 {
1496         unsigned long p = (unsigned long)bucket;
1497
1498         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1499 }
1500
1501 /* called with rt6_exception_lock held */
1502 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1503                                               spinlock_t *lock)
1504 {
1505         struct rt6_exception_bucket *bucket;
1506         unsigned long p;
1507
1508         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1509                                            lockdep_is_held(lock));
1510
1511         p = (unsigned long)bucket;
1512         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1513         bucket = (struct rt6_exception_bucket *)p;
1514         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1515 }
1516
1517 static int rt6_insert_exception(struct rt6_info *nrt,
1518                                 const struct fib6_result *res)
1519 {
1520         struct net *net = dev_net(nrt->dst.dev);
1521         struct rt6_exception_bucket *bucket;
1522         struct fib6_info *f6i = res->f6i;
1523         struct in6_addr *src_key = NULL;
1524         struct rt6_exception *rt6_ex;
1525         struct fib6_nh *nh = res->nh;
1526         int err = 0;
1527
1528         spin_lock_bh(&rt6_exception_lock);
1529
1530         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1531                                           lockdep_is_held(&rt6_exception_lock));
1532         if (!bucket) {
1533                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1534                                  GFP_ATOMIC);
1535                 if (!bucket) {
1536                         err = -ENOMEM;
1537                         goto out;
1538                 }
1539                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1540         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1541                 err = -EINVAL;
1542                 goto out;
1543         }
1544
1545 #ifdef CONFIG_IPV6_SUBTREES
1546         /* fib6_src.plen != 0 indicates f6i is in subtree
1547          * and exception table is indexed by a hash of
1548          * both fib6_dst and fib6_src.
1549          * Otherwise, the exception table is indexed by
1550          * a hash of only fib6_dst.
1551          */
1552         if (f6i->fib6_src.plen)
1553                 src_key = &nrt->rt6i_src.addr;
1554 #endif
1555         /* rt6_mtu_change() might lower mtu on f6i.
1556          * Only insert this exception route if its mtu
1557          * is less than f6i's mtu value.
1558          */
1559         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1560                 err = -EINVAL;
1561                 goto out;
1562         }
1563
1564         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1565                                                src_key);
1566         if (rt6_ex)
1567                 rt6_remove_exception(bucket, rt6_ex);
1568
1569         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1570         if (!rt6_ex) {
1571                 err = -ENOMEM;
1572                 goto out;
1573         }
1574         rt6_ex->rt6i = nrt;
1575         rt6_ex->stamp = jiffies;
1576         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1577         bucket->depth++;
1578         net->ipv6.rt6_stats->fib_rt_cache++;
1579
1580         if (bucket->depth > FIB6_MAX_DEPTH)
1581                 rt6_exception_remove_oldest(bucket);
1582
1583 out:
1584         spin_unlock_bh(&rt6_exception_lock);
1585
1586         /* Update fn->fn_sernum to invalidate all cached dst */
1587         if (!err) {
1588                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1589                 fib6_update_sernum(net, f6i);
1590                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1591                 fib6_force_start_gc(net);
1592         }
1593
1594         return err;
1595 }
1596
1597 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1598 {
1599         struct rt6_exception_bucket *bucket;
1600         struct rt6_exception *rt6_ex;
1601         struct hlist_node *tmp;
1602         int i;
1603
1604         spin_lock_bh(&rt6_exception_lock);
1605
1606         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1607         if (!bucket)
1608                 goto out;
1609
1610         /* Prevent rt6_insert_exception() to recreate the bucket list */
1611         if (!from)
1612                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1613
1614         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1615                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1616                         if (!from ||
1617                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1618                                 rt6_remove_exception(bucket, rt6_ex);
1619                 }
1620                 WARN_ON_ONCE(!from && bucket->depth);
1621                 bucket++;
1622         }
1623 out:
1624         spin_unlock_bh(&rt6_exception_lock);
1625 }
1626
1627 void rt6_flush_exceptions(struct fib6_info *f6i)
1628 {
1629         fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1630 }
1631
1632 /* Find cached rt in the hash table inside passed in rt
1633  * Caller has to hold rcu_read_lock()
1634  */
1635 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1636                                            const struct in6_addr *daddr,
1637                                            const struct in6_addr *saddr)
1638 {
1639         const struct in6_addr *src_key = NULL;
1640         struct rt6_exception_bucket *bucket;
1641         struct rt6_exception *rt6_ex;
1642         struct rt6_info *ret = NULL;
1643
1644 #ifdef CONFIG_IPV6_SUBTREES
1645         /* fib6i_src.plen != 0 indicates f6i is in subtree
1646          * and exception table is indexed by a hash of
1647          * both fib6_dst and fib6_src.
1648          * However, the src addr used to create the hash
1649          * might not be exactly the passed in saddr which
1650          * is a /128 addr from the flow.
1651          * So we need to use f6i->fib6_src to redo lookup
1652          * if the passed in saddr does not find anything.
1653          * (See the logic in ip6_rt_cache_alloc() on how
1654          * rt->rt6i_src is updated.)
1655          */
1656         if (res->f6i->fib6_src.plen)
1657                 src_key = saddr;
1658 find_ex:
1659 #endif
1660         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1661         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1662
1663         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1664                 ret = rt6_ex->rt6i;
1665
1666 #ifdef CONFIG_IPV6_SUBTREES
1667         /* Use fib6_src as src_key and redo lookup */
1668         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1669                 src_key = &res->f6i->fib6_src.addr;
1670                 goto find_ex;
1671         }
1672 #endif
1673
1674         return ret;
1675 }
1676
1677 /* Remove the passed in cached rt from the hash table that contains it */
1678 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1679                                     const struct rt6_info *rt)
1680 {
1681         const struct in6_addr *src_key = NULL;
1682         struct rt6_exception_bucket *bucket;
1683         struct rt6_exception *rt6_ex;
1684         int err;
1685
1686         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1687                 return -ENOENT;
1688
1689         spin_lock_bh(&rt6_exception_lock);
1690         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1691
1692 #ifdef CONFIG_IPV6_SUBTREES
1693         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1694          * and exception table is indexed by a hash of
1695          * both rt6i_dst and rt6i_src.
1696          * Otherwise, the exception table is indexed by
1697          * a hash of only rt6i_dst.
1698          */
1699         if (plen)
1700                 src_key = &rt->rt6i_src.addr;
1701 #endif
1702         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1703                                                &rt->rt6i_dst.addr,
1704                                                src_key);
1705         if (rt6_ex) {
1706                 rt6_remove_exception(bucket, rt6_ex);
1707                 err = 0;
1708         } else {
1709                 err = -ENOENT;
1710         }
1711
1712         spin_unlock_bh(&rt6_exception_lock);
1713         return err;
1714 }
1715
1716 static int rt6_remove_exception_rt(struct rt6_info *rt)
1717 {
1718         struct fib6_info *from;
1719
1720         from = rcu_dereference(rt->from);
1721         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1722                 return -EINVAL;
1723
1724         return fib6_nh_remove_exception(from->fib6_nh,
1725                                         from->fib6_src.plen, rt);
1726 }
1727
1728 /* Find rt6_ex which contains the passed in rt cache and
1729  * refresh its stamp
1730  */
1731 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1732                                      const struct rt6_info *rt)
1733 {
1734         const struct in6_addr *src_key = NULL;
1735         struct rt6_exception_bucket *bucket;
1736         struct rt6_exception *rt6_ex;
1737
1738         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1739 #ifdef CONFIG_IPV6_SUBTREES
1740         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1741          * and exception table is indexed by a hash of
1742          * both rt6i_dst and rt6i_src.
1743          * Otherwise, the exception table is indexed by
1744          * a hash of only rt6i_dst.
1745          */
1746         if (plen)
1747                 src_key = &rt->rt6i_src.addr;
1748 #endif
1749         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1750         if (rt6_ex)
1751                 rt6_ex->stamp = jiffies;
1752 }
1753
1754 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1755 {
1756         struct fib6_info *from;
1757
1758         rcu_read_lock();
1759
1760         from = rcu_dereference(rt->from);
1761         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1762                 goto unlock;
1763
1764         fib6_nh_update_exception(from->fib6_nh, from->fib6_src.plen, rt);
1765 unlock:
1766         rcu_read_unlock();
1767 }
1768
1769 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1770                                          struct rt6_info *rt, int mtu)
1771 {
1772         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1773          * lowest MTU in the path: always allow updating the route PMTU to
1774          * reflect PMTU decreases.
1775          *
1776          * If the new MTU is higher, and the route PMTU is equal to the local
1777          * MTU, this means the old MTU is the lowest in the path, so allow
1778          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1779          * handle this.
1780          */
1781
1782         if (dst_mtu(&rt->dst) >= mtu)
1783                 return true;
1784
1785         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1786                 return true;
1787
1788         return false;
1789 }
1790
1791 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
1792                                        const struct fib6_nh *nh, int mtu)
1793 {
1794         struct rt6_exception_bucket *bucket;
1795         struct rt6_exception *rt6_ex;
1796         int i;
1797
1798         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1799         if (!bucket)
1800                 return;
1801
1802         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1803                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1804                         struct rt6_info *entry = rt6_ex->rt6i;
1805
1806                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
1807                          * route), the metrics of its rt->from have already
1808                          * been updated.
1809                          */
1810                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
1811                             rt6_mtu_change_route_allowed(idev, entry, mtu))
1812                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
1813                 }
1814                 bucket++;
1815         }
1816 }
1817
1818 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
1819
1820 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
1821                                             const struct in6_addr *gateway)
1822 {
1823         struct rt6_exception_bucket *bucket;
1824         struct rt6_exception *rt6_ex;
1825         struct hlist_node *tmp;
1826         int i;
1827
1828         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1829                 return;
1830
1831         spin_lock_bh(&rt6_exception_lock);
1832         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1833         if (bucket) {
1834                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1835                         hlist_for_each_entry_safe(rt6_ex, tmp,
1836                                                   &bucket->chain, hlist) {
1837                                 struct rt6_info *entry = rt6_ex->rt6i;
1838
1839                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
1840                                     RTF_CACHE_GATEWAY &&
1841                                     ipv6_addr_equal(gateway,
1842                                                     &entry->rt6i_gateway)) {
1843                                         rt6_remove_exception(bucket, rt6_ex);
1844                                 }
1845                         }
1846                         bucket++;
1847                 }
1848         }
1849
1850         spin_unlock_bh(&rt6_exception_lock);
1851 }
1852
1853 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
1854                                       struct rt6_exception *rt6_ex,
1855                                       struct fib6_gc_args *gc_args,
1856                                       unsigned long now)
1857 {
1858         struct rt6_info *rt = rt6_ex->rt6i;
1859
1860         /* we are pruning and obsoleting aged-out and non gateway exceptions
1861          * even if others have still references to them, so that on next
1862          * dst_check() such references can be dropped.
1863          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
1864          * expired, independently from their aging, as per RFC 8201 section 4
1865          */
1866         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
1867                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
1868                         RT6_TRACE("aging clone %p\n", rt);
1869                         rt6_remove_exception(bucket, rt6_ex);
1870                         return;
1871                 }
1872         } else if (time_after(jiffies, rt->dst.expires)) {
1873                 RT6_TRACE("purging expired route %p\n", rt);
1874                 rt6_remove_exception(bucket, rt6_ex);
1875                 return;
1876         }
1877
1878         if (rt->rt6i_flags & RTF_GATEWAY) {
1879                 struct neighbour *neigh;
1880                 __u8 neigh_flags = 0;
1881
1882                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
1883                 if (neigh)
1884                         neigh_flags = neigh->flags;
1885
1886                 if (!(neigh_flags & NTF_ROUTER)) {
1887                         RT6_TRACE("purging route %p via non-router but gateway\n",
1888                                   rt);
1889                         rt6_remove_exception(bucket, rt6_ex);
1890                         return;
1891                 }
1892         }
1893
1894         gc_args->more++;
1895 }
1896
1897 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
1898                                    struct fib6_gc_args *gc_args,
1899                                    unsigned long now)
1900 {
1901         struct rt6_exception_bucket *bucket;
1902         struct rt6_exception *rt6_ex;
1903         struct hlist_node *tmp;
1904         int i;
1905
1906         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1907                 return;
1908
1909         rcu_read_lock_bh();
1910         spin_lock(&rt6_exception_lock);
1911         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1912         if (bucket) {
1913                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1914                         hlist_for_each_entry_safe(rt6_ex, tmp,
1915                                                   &bucket->chain, hlist) {
1916                                 rt6_age_examine_exception(bucket, rt6_ex,
1917                                                           gc_args, now);
1918                         }
1919                         bucket++;
1920                 }
1921         }
1922         spin_unlock(&rt6_exception_lock);
1923         rcu_read_unlock_bh();
1924 }
1925
1926 void rt6_age_exceptions(struct fib6_info *f6i,
1927                         struct fib6_gc_args *gc_args,
1928                         unsigned long now)
1929 {
1930         fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
1931 }
1932
1933 /* must be called with rcu lock held */
1934 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
1935                       struct flowi6 *fl6, struct fib6_result *res, int strict)
1936 {
1937         struct fib6_node *fn, *saved_fn;
1938
1939         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1940         saved_fn = fn;
1941
1942         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1943                 oif = 0;
1944
1945 redo_rt6_select:
1946         rt6_select(net, fn, oif, res, strict);
1947         if (res->f6i == net->ipv6.fib6_null_entry) {
1948                 fn = fib6_backtrack(fn, &fl6->saddr);
1949                 if (fn)
1950                         goto redo_rt6_select;
1951                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
1952                         /* also consider unreachable route */
1953                         strict &= ~RT6_LOOKUP_F_REACHABLE;
1954                         fn = saved_fn;
1955                         goto redo_rt6_select;
1956                 }
1957         }
1958
1959         trace_fib6_table_lookup(net, res, table, fl6);
1960
1961         return 0;
1962 }
1963
1964 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
1965                                int oif, struct flowi6 *fl6,
1966                                const struct sk_buff *skb, int flags)
1967 {
1968         struct fib6_result res = {};
1969         struct rt6_info *rt;
1970         int strict = 0;
1971
1972         strict |= flags & RT6_LOOKUP_F_IFACE;
1973         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
1974         if (net->ipv6.devconf_all->forwarding == 0)
1975                 strict |= RT6_LOOKUP_F_REACHABLE;
1976
1977         rcu_read_lock();
1978
1979         fib6_table_lookup(net, table, oif, fl6, &res, strict);
1980         if (res.f6i == net->ipv6.fib6_null_entry) {
1981                 rt = net->ipv6.ip6_null_entry;
1982                 rcu_read_unlock();
1983                 dst_hold(&rt->dst);
1984                 return rt;
1985         }
1986
1987         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
1988
1989         /*Search through exception table */
1990         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1991         if (rt) {
1992                 if (ip6_hold_safe(net, &rt))
1993                         dst_use_noref(&rt->dst, jiffies);
1994
1995                 rcu_read_unlock();
1996                 return rt;
1997         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
1998                             !res.nh->fib_nh_gw_family)) {
1999                 /* Create a RTF_CACHE clone which will not be
2000                  * owned by the fib6 tree.  It is for the special case where
2001                  * the daddr in the skb during the neighbor look-up is different
2002                  * from the fl6->daddr used to look-up route here.
2003                  */
2004                 struct rt6_info *uncached_rt;
2005
2006                 uncached_rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2007
2008                 rcu_read_unlock();
2009
2010                 if (uncached_rt) {
2011                         /* Uncached_rt's refcnt is taken during ip6_rt_cache_alloc()
2012                          * No need for another dst_hold()
2013                          */
2014                         rt6_uncached_list_add(uncached_rt);
2015                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2016                 } else {
2017                         uncached_rt = net->ipv6.ip6_null_entry;
2018                         dst_hold(&uncached_rt->dst);
2019                 }
2020
2021                 return uncached_rt;
2022         } else {
2023                 /* Get a percpu copy */
2024
2025                 struct rt6_info *pcpu_rt;
2026
2027                 local_bh_disable();
2028                 pcpu_rt = rt6_get_pcpu_route(&res);
2029
2030                 if (!pcpu_rt)
2031                         pcpu_rt = rt6_make_pcpu_route(net, &res);
2032
2033                 local_bh_enable();
2034                 rcu_read_unlock();
2035
2036                 return pcpu_rt;
2037         }
2038 }
2039 EXPORT_SYMBOL_GPL(ip6_pol_route);
2040
2041 static struct rt6_info *ip6_pol_route_input(struct net *net,
2042                                             struct fib6_table *table,
2043                                             struct flowi6 *fl6,
2044                                             const struct sk_buff *skb,
2045                                             int flags)
2046 {
2047         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2048 }
2049
2050 struct dst_entry *ip6_route_input_lookup(struct net *net,
2051                                          struct net_device *dev,
2052                                          struct flowi6 *fl6,
2053                                          const struct sk_buff *skb,
2054                                          int flags)
2055 {
2056         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2057                 flags |= RT6_LOOKUP_F_IFACE;
2058
2059         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2060 }
2061 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2062
2063 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2064                                   struct flow_keys *keys,
2065                                   struct flow_keys *flkeys)
2066 {
2067         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2068         const struct ipv6hdr *key_iph = outer_iph;
2069         struct flow_keys *_flkeys = flkeys;
2070         const struct ipv6hdr *inner_iph;
2071         const struct icmp6hdr *icmph;
2072         struct ipv6hdr _inner_iph;
2073         struct icmp6hdr _icmph;
2074
2075         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2076                 goto out;
2077
2078         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2079                                    sizeof(_icmph), &_icmph);
2080         if (!icmph)
2081                 goto out;
2082
2083         if (icmph->icmp6_type != ICMPV6_DEST_UNREACH &&
2084             icmph->icmp6_type != ICMPV6_PKT_TOOBIG &&
2085             icmph->icmp6_type != ICMPV6_TIME_EXCEED &&
2086             icmph->icmp6_type != ICMPV6_PARAMPROB)
2087                 goto out;
2088
2089         inner_iph = skb_header_pointer(skb,
2090                                        skb_transport_offset(skb) + sizeof(*icmph),
2091                                        sizeof(_inner_iph), &_inner_iph);
2092         if (!inner_iph)
2093                 goto out;
2094
2095         key_iph = inner_iph;
2096         _flkeys = NULL;
2097 out:
2098         if (_flkeys) {
2099                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2100                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2101                 keys->tags.flow_label = _flkeys->tags.flow_label;
2102                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2103         } else {
2104                 keys->addrs.v6addrs.src = key_iph->saddr;
2105                 keys->addrs.v6addrs.dst = key_iph->daddr;
2106                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2107                 keys->basic.ip_proto = key_iph->nexthdr;
2108         }
2109 }
2110
2111 /* if skb is set it will be used and fl6 can be NULL */
2112 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2113                        const struct sk_buff *skb, struct flow_keys *flkeys)
2114 {
2115         struct flow_keys hash_keys;
2116         u32 mhash;
2117
2118         switch (ip6_multipath_hash_policy(net)) {
2119         case 0:
2120                 memset(&hash_keys, 0, sizeof(hash_keys));
2121                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2122                 if (skb) {
2123                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2124                 } else {
2125                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2126                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2127                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2128                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2129                 }
2130                 break;
2131         case 1:
2132                 if (skb) {
2133                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2134                         struct flow_keys keys;
2135
2136                         /* short-circuit if we already have L4 hash present */
2137                         if (skb->l4_hash)
2138                                 return skb_get_hash_raw(skb) >> 1;
2139
2140                         memset(&hash_keys, 0, sizeof(hash_keys));
2141
2142                         if (!flkeys) {
2143                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2144                                 flkeys = &keys;
2145                         }
2146                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2147                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2148                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2149                         hash_keys.ports.src = flkeys->ports.src;
2150                         hash_keys.ports.dst = flkeys->ports.dst;
2151                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2152                 } else {
2153                         memset(&hash_keys, 0, sizeof(hash_keys));
2154                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2155                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2156                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2157                         hash_keys.ports.src = fl6->fl6_sport;
2158                         hash_keys.ports.dst = fl6->fl6_dport;
2159                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2160                 }
2161                 break;
2162         }
2163         mhash = flow_hash_from_keys(&hash_keys);
2164
2165         return mhash >> 1;
2166 }
2167
2168 void ip6_route_input(struct sk_buff *skb)
2169 {
2170         const struct ipv6hdr *iph = ipv6_hdr(skb);
2171         struct net *net = dev_net(skb->dev);
2172         int flags = RT6_LOOKUP_F_HAS_SADDR;
2173         struct ip_tunnel_info *tun_info;
2174         struct flowi6 fl6 = {
2175                 .flowi6_iif = skb->dev->ifindex,
2176                 .daddr = iph->daddr,
2177                 .saddr = iph->saddr,
2178                 .flowlabel = ip6_flowinfo(iph),
2179                 .flowi6_mark = skb->mark,
2180                 .flowi6_proto = iph->nexthdr,
2181         };
2182         struct flow_keys *flkeys = NULL, _flkeys;
2183
2184         tun_info = skb_tunnel_info(skb);
2185         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2186                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2187
2188         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2189                 flkeys = &_flkeys;
2190
2191         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2192                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2193         skb_dst_drop(skb);
2194         skb_dst_set(skb,
2195                     ip6_route_input_lookup(net, skb->dev, &fl6, skb, flags));
2196 }
2197
2198 static struct rt6_info *ip6_pol_route_output(struct net *net,
2199                                              struct fib6_table *table,
2200                                              struct flowi6 *fl6,
2201                                              const struct sk_buff *skb,
2202                                              int flags)
2203 {
2204         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2205 }
2206
2207 struct dst_entry *ip6_route_output_flags(struct net *net, const struct sock *sk,
2208                                          struct flowi6 *fl6, int flags)
2209 {
2210         bool any_src;
2211
2212         if (ipv6_addr_type(&fl6->daddr) &
2213             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2214                 struct dst_entry *dst;
2215
2216                 dst = l3mdev_link_scope_lookup(net, fl6);
2217                 if (dst)
2218                         return dst;
2219         }
2220
2221         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2222
2223         any_src = ipv6_addr_any(&fl6->saddr);
2224         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2225             (fl6->flowi6_oif && any_src))
2226                 flags |= RT6_LOOKUP_F_IFACE;
2227
2228         if (!any_src)
2229                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2230         else if (sk)
2231                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2232
2233         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2234 }
2235 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2236
2237 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2238 {
2239         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2240         struct net_device *loopback_dev = net->loopback_dev;
2241         struct dst_entry *new = NULL;
2242
2243         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2244                        DST_OBSOLETE_DEAD, 0);
2245         if (rt) {
2246                 rt6_info_init(rt);
2247                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2248
2249                 new = &rt->dst;
2250                 new->__use = 1;
2251                 new->input = dst_discard;
2252                 new->output = dst_discard_out;
2253
2254                 dst_copy_metrics(new, &ort->dst);
2255
2256                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2257                 rt->rt6i_gateway = ort->rt6i_gateway;
2258                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2259
2260                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2261 #ifdef CONFIG_IPV6_SUBTREES
2262                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2263 #endif
2264         }
2265
2266         dst_release(dst_orig);
2267         return new ? new : ERR_PTR(-ENOMEM);
2268 }
2269
2270 /*
2271  *      Destination cache support functions
2272  */
2273
2274 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2275 {
2276         u32 rt_cookie = 0;
2277
2278         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2279                 return false;
2280
2281         if (fib6_check_expired(f6i))
2282                 return false;
2283
2284         return true;
2285 }
2286
2287 static struct dst_entry *rt6_check(struct rt6_info *rt,
2288                                    struct fib6_info *from,
2289                                    u32 cookie)
2290 {
2291         u32 rt_cookie = 0;
2292
2293         if ((from && !fib6_get_cookie_safe(from, &rt_cookie)) ||
2294             rt_cookie != cookie)
2295                 return NULL;
2296
2297         if (rt6_check_expired(rt))
2298                 return NULL;
2299
2300         return &rt->dst;
2301 }
2302
2303 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2304                                             struct fib6_info *from,
2305                                             u32 cookie)
2306 {
2307         if (!__rt6_check_expired(rt) &&
2308             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2309             fib6_check(from, cookie))
2310                 return &rt->dst;
2311         else
2312                 return NULL;
2313 }
2314
2315 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2316 {
2317         struct dst_entry *dst_ret;
2318         struct fib6_info *from;
2319         struct rt6_info *rt;
2320
2321         rt = container_of(dst, struct rt6_info, dst);
2322
2323         rcu_read_lock();
2324
2325         /* All IPV6 dsts are created with ->obsolete set to the value
2326          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2327          * into this function always.
2328          */
2329
2330         from = rcu_dereference(rt->from);
2331
2332         if (from && (rt->rt6i_flags & RTF_PCPU ||
2333             unlikely(!list_empty(&rt->rt6i_uncached))))
2334                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2335         else
2336                 dst_ret = rt6_check(rt, from, cookie);
2337
2338         rcu_read_unlock();
2339
2340         return dst_ret;
2341 }
2342
2343 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2344 {
2345         struct rt6_info *rt = (struct rt6_info *) dst;
2346
2347         if (rt) {
2348                 if (rt->rt6i_flags & RTF_CACHE) {
2349                         rcu_read_lock();
2350                         if (rt6_check_expired(rt)) {
2351                                 rt6_remove_exception_rt(rt);
2352                                 dst = NULL;
2353                         }
2354                         rcu_read_unlock();
2355                 } else {
2356                         dst_release(dst);
2357                         dst = NULL;
2358                 }
2359         }
2360         return dst;
2361 }
2362
2363 static void ip6_link_failure(struct sk_buff *skb)
2364 {
2365         struct rt6_info *rt;
2366
2367         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2368
2369         rt = (struct rt6_info *) skb_dst(skb);
2370         if (rt) {
2371                 rcu_read_lock();
2372                 if (rt->rt6i_flags & RTF_CACHE) {
2373                         rt6_remove_exception_rt(rt);
2374                 } else {
2375                         struct fib6_info *from;
2376                         struct fib6_node *fn;
2377
2378                         from = rcu_dereference(rt->from);
2379                         if (from) {
2380                                 fn = rcu_dereference(from->fib6_node);
2381                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2382                                         fn->fn_sernum = -1;
2383                         }
2384                 }
2385                 rcu_read_unlock();
2386         }
2387 }
2388
2389 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2390 {
2391         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2392                 struct fib6_info *from;
2393
2394                 rcu_read_lock();
2395                 from = rcu_dereference(rt0->from);
2396                 if (from)
2397                         rt0->dst.expires = from->expires;
2398                 rcu_read_unlock();
2399         }
2400
2401         dst_set_expires(&rt0->dst, timeout);
2402         rt0->rt6i_flags |= RTF_EXPIRES;
2403 }
2404
2405 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2406 {
2407         struct net *net = dev_net(rt->dst.dev);
2408
2409         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2410         rt->rt6i_flags |= RTF_MODIFIED;
2411         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2412 }
2413
2414 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2415 {
2416         return !(rt->rt6i_flags & RTF_CACHE) &&
2417                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2418 }
2419
2420 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2421                                  const struct ipv6hdr *iph, u32 mtu)
2422 {
2423         const struct in6_addr *daddr, *saddr;
2424         struct rt6_info *rt6 = (struct rt6_info *)dst;
2425
2426         if (dst_metric_locked(dst, RTAX_MTU))
2427                 return;
2428
2429         if (iph) {
2430                 daddr = &iph->daddr;
2431                 saddr = &iph->saddr;
2432         } else if (sk) {
2433                 daddr = &sk->sk_v6_daddr;
2434                 saddr = &inet6_sk(sk)->saddr;
2435         } else {
2436                 daddr = NULL;
2437                 saddr = NULL;
2438         }
2439         dst_confirm_neigh(dst, daddr);
2440         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2441         if (mtu >= dst_mtu(dst))
2442                 return;
2443
2444         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2445                 rt6_do_update_pmtu(rt6, mtu);
2446                 /* update rt6_ex->stamp for cache */
2447                 if (rt6->rt6i_flags & RTF_CACHE)
2448                         rt6_update_exception_stamp_rt(rt6);
2449         } else if (daddr) {
2450                 struct fib6_result res = {};
2451                 struct rt6_info *nrt6;
2452
2453                 rcu_read_lock();
2454                 res.f6i = rcu_dereference(rt6->from);
2455                 if (!res.f6i) {
2456                         rcu_read_unlock();
2457                         return;
2458                 }
2459                 res.nh = res.f6i->fib6_nh;
2460                 res.fib6_flags = res.f6i->fib6_flags;
2461                 res.fib6_type = res.f6i->fib6_type;
2462
2463                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2464                 if (nrt6) {
2465                         rt6_do_update_pmtu(nrt6, mtu);
2466                         if (rt6_insert_exception(nrt6, &res))
2467                                 dst_release_immediate(&nrt6->dst);
2468                 }
2469                 rcu_read_unlock();
2470         }
2471 }
2472
2473 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2474                                struct sk_buff *skb, u32 mtu)
2475 {
2476         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu);
2477 }
2478
2479 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2480                      int oif, u32 mark, kuid_t uid)
2481 {
2482         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2483         struct dst_entry *dst;
2484         struct flowi6 fl6 = {
2485                 .flowi6_oif = oif,
2486                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2487                 .daddr = iph->daddr,
2488                 .saddr = iph->saddr,
2489                 .flowlabel = ip6_flowinfo(iph),
2490                 .flowi6_uid = uid,
2491         };
2492
2493         dst = ip6_route_output(net, NULL, &fl6);
2494         if (!dst->error)
2495                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu));
2496         dst_release(dst);
2497 }
2498 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2499
2500 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2501 {
2502         int oif = sk->sk_bound_dev_if;
2503         struct dst_entry *dst;
2504
2505         if (!oif && skb->dev)
2506                 oif = l3mdev_master_ifindex(skb->dev);
2507
2508         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2509
2510         dst = __sk_dst_get(sk);
2511         if (!dst || !dst->obsolete ||
2512             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2513                 return;
2514
2515         bh_lock_sock(sk);
2516         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2517                 ip6_datagram_dst_update(sk, false);
2518         bh_unlock_sock(sk);
2519 }
2520 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2521
2522 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2523                            const struct flowi6 *fl6)
2524 {
2525 #ifdef CONFIG_IPV6_SUBTREES
2526         struct ipv6_pinfo *np = inet6_sk(sk);
2527 #endif
2528
2529         ip6_dst_store(sk, dst,
2530                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2531                       &sk->sk_v6_daddr : NULL,
2532 #ifdef CONFIG_IPV6_SUBTREES
2533                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2534                       &np->saddr :
2535 #endif
2536                       NULL);
2537 }
2538
2539 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2540                                   struct flowi6 *fl6,
2541                                   const struct in6_addr *gw,
2542                                   struct rt6_info **ret)
2543 {
2544         const struct fib6_nh *nh = res->nh;
2545
2546         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2547             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2548                 return false;
2549
2550         /* rt_cache's gateway might be different from its 'parent'
2551          * in the case of an ip redirect.
2552          * So we keep searching in the exception table if the gateway
2553          * is different.
2554          */
2555         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2556                 struct rt6_info *rt_cache;
2557
2558                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2559                 if (rt_cache &&
2560                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2561                         *ret = rt_cache;
2562                         return true;
2563                 }
2564                 return false;
2565         }
2566         return true;
2567 }
2568
2569 /* Handle redirects */
2570 struct ip6rd_flowi {
2571         struct flowi6 fl6;
2572         struct in6_addr gateway;
2573 };
2574
2575 static struct rt6_info *__ip6_route_redirect(struct net *net,
2576                                              struct fib6_table *table,
2577                                              struct flowi6 *fl6,
2578                                              const struct sk_buff *skb,
2579                                              int flags)
2580 {
2581         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2582         struct rt6_info *ret = NULL;
2583         struct fib6_result res = {};
2584         struct fib6_info *rt;
2585         struct fib6_node *fn;
2586
2587         /* Get the "current" route for this destination and
2588          * check if the redirect has come from appropriate router.
2589          *
2590          * RFC 4861 specifies that redirects should only be
2591          * accepted if they come from the nexthop to the target.
2592          * Due to the way the routes are chosen, this notion
2593          * is a bit fuzzy and one might need to check all possible
2594          * routes.
2595          */
2596
2597         rcu_read_lock();
2598         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2599 restart:
2600         for_each_fib6_node_rt_rcu(fn) {
2601                 res.f6i = rt;
2602                 res.nh = rt->fib6_nh;
2603
2604                 if (fib6_check_expired(rt))
2605                         continue;
2606                 if (rt->fib6_flags & RTF_REJECT)
2607                         break;
2608                 if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway, &ret))
2609                         goto out;
2610         }
2611
2612         if (!rt)
2613                 rt = net->ipv6.fib6_null_entry;
2614         else if (rt->fib6_flags & RTF_REJECT) {
2615                 ret = net->ipv6.ip6_null_entry;
2616                 goto out;
2617         }
2618
2619         if (rt == net->ipv6.fib6_null_entry) {
2620                 fn = fib6_backtrack(fn, &fl6->saddr);
2621                 if (fn)
2622                         goto restart;
2623         }
2624
2625         res.f6i = rt;
2626         res.nh = rt->fib6_nh;
2627 out:
2628         if (ret) {
2629                 ip6_hold_safe(net, &ret);
2630         } else {
2631                 res.fib6_flags = res.f6i->fib6_flags;
2632                 res.fib6_type = res.f6i->fib6_type;
2633                 ret = ip6_create_rt_rcu(&res);
2634         }
2635
2636         rcu_read_unlock();
2637
2638         trace_fib6_table_lookup(net, &res, table, fl6);
2639         return ret;
2640 };
2641
2642 static struct dst_entry *ip6_route_redirect(struct net *net,
2643                                             const struct flowi6 *fl6,
2644                                             const struct sk_buff *skb,
2645                                             const struct in6_addr *gateway)
2646 {
2647         int flags = RT6_LOOKUP_F_HAS_SADDR;
2648         struct ip6rd_flowi rdfl;
2649
2650         rdfl.fl6 = *fl6;
2651         rdfl.gateway = *gateway;
2652
2653         return fib6_rule_lookup(net, &rdfl.fl6, skb,
2654                                 flags, __ip6_route_redirect);
2655 }
2656
2657 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
2658                   kuid_t uid)
2659 {
2660         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2661         struct dst_entry *dst;
2662         struct flowi6 fl6 = {
2663                 .flowi6_iif = LOOPBACK_IFINDEX,
2664                 .flowi6_oif = oif,
2665                 .flowi6_mark = mark,
2666                 .daddr = iph->daddr,
2667                 .saddr = iph->saddr,
2668                 .flowlabel = ip6_flowinfo(iph),
2669                 .flowi6_uid = uid,
2670         };
2671
2672         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
2673         rt6_do_redirect(dst, NULL, skb);
2674         dst_release(dst);
2675 }
2676 EXPORT_SYMBOL_GPL(ip6_redirect);
2677
2678 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
2679 {
2680         const struct ipv6hdr *iph = ipv6_hdr(skb);
2681         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
2682         struct dst_entry *dst;
2683         struct flowi6 fl6 = {
2684                 .flowi6_iif = LOOPBACK_IFINDEX,
2685                 .flowi6_oif = oif,
2686                 .daddr = msg->dest,
2687                 .saddr = iph->daddr,
2688                 .flowi6_uid = sock_net_uid(net, NULL),
2689         };
2690
2691         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
2692         rt6_do_redirect(dst, NULL, skb);
2693         dst_release(dst);
2694 }
2695
2696 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
2697 {
2698         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
2699                      sk->sk_uid);
2700 }
2701 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
2702
2703 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
2704 {
2705         struct net_device *dev = dst->dev;
2706         unsigned int mtu = dst_mtu(dst);
2707         struct net *net = dev_net(dev);
2708
2709         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
2710
2711         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
2712                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
2713
2714         /*
2715          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
2716          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
2717          * IPV6_MAXPLEN is also valid and means: "any MSS,
2718          * rely only on pmtu discovery"
2719          */
2720         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
2721                 mtu = IPV6_MAXPLEN;
2722         return mtu;
2723 }
2724
2725 static unsigned int ip6_mtu(const struct dst_entry *dst)
2726 {
2727         struct inet6_dev *idev;
2728         unsigned int mtu;
2729
2730         mtu = dst_metric_raw(dst, RTAX_MTU);
2731         if (mtu)
2732                 goto out;
2733
2734         mtu = IPV6_MIN_MTU;
2735
2736         rcu_read_lock();
2737         idev = __in6_dev_get(dst->dev);
2738         if (idev)
2739                 mtu = idev->cnf.mtu6;
2740         rcu_read_unlock();
2741
2742 out:
2743         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
2744
2745         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
2746 }
2747
2748 /* MTU selection:
2749  * 1. mtu on route is locked - use it
2750  * 2. mtu from nexthop exception
2751  * 3. mtu from egress device
2752  *
2753  * based on ip6_dst_mtu_forward and exception logic of
2754  * rt6_find_cached_rt; called with rcu_read_lock
2755  */
2756 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
2757                       const struct in6_addr *daddr,
2758                       const struct in6_addr *saddr)
2759 {
2760         const struct fib6_nh *nh = res->nh;
2761         struct fib6_info *f6i = res->f6i;
2762         struct inet6_dev *idev;
2763         struct rt6_info *rt;
2764         u32 mtu = 0;
2765
2766         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
2767                 mtu = f6i->fib6_pmtu;
2768                 if (mtu)
2769                         goto out;
2770         }
2771
2772         rt = rt6_find_cached_rt(res, daddr, saddr);
2773         if (unlikely(rt)) {
2774                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
2775         } else {
2776                 struct net_device *dev = nh->fib_nh_dev;
2777
2778                 mtu = IPV6_MIN_MTU;
2779                 idev = __in6_dev_get(dev);
2780                 if (idev && idev->cnf.mtu6 > mtu)
2781                         mtu = idev->cnf.mtu6;
2782         }
2783
2784         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
2785 out:
2786         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
2787 }
2788
2789 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
2790                                   struct flowi6 *fl6)
2791 {
2792         struct dst_entry *dst;
2793         struct rt6_info *rt;
2794         struct inet6_dev *idev = in6_dev_get(dev);
2795         struct net *net = dev_net(dev);
2796
2797         if (unlikely(!idev))
2798                 return ERR_PTR(-ENODEV);
2799
2800         rt = ip6_dst_alloc(net, dev, 0);
2801         if (unlikely(!rt)) {
2802                 in6_dev_put(idev);
2803                 dst = ERR_PTR(-ENOMEM);
2804                 goto out;
2805         }
2806
2807         rt->dst.flags |= DST_HOST;
2808         rt->dst.input = ip6_input;
2809         rt->dst.output  = ip6_output;
2810         rt->rt6i_gateway  = fl6->daddr;
2811         rt->rt6i_dst.addr = fl6->daddr;
2812         rt->rt6i_dst.plen = 128;
2813         rt->rt6i_idev     = idev;
2814         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
2815
2816         /* Add this dst into uncached_list so that rt6_disable_ip() can
2817          * do proper release of the net_device
2818          */
2819         rt6_uncached_list_add(rt);
2820         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2821
2822         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
2823
2824 out:
2825         return dst;
2826 }
2827
2828 static int ip6_dst_gc(struct dst_ops *ops)
2829 {
2830         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
2831         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
2832         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
2833         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
2834         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
2835         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
2836         int entries;
2837
2838         entries = dst_entries_get_fast(ops);
2839         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
2840             entries <= rt_max_size)
2841                 goto out;
2842
2843         net->ipv6.ip6_rt_gc_expire++;
2844         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
2845         entries = dst_entries_get_slow(ops);
2846         if (entries < ops->gc_thresh)
2847                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
2848 out:
2849         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
2850         return entries > rt_max_size;
2851 }
2852
2853 static struct rt6_info *ip6_nh_lookup_table(struct net *net,
2854                                             struct fib6_config *cfg,
2855                                             const struct in6_addr *gw_addr,
2856                                             u32 tbid, int flags)
2857 {
2858         struct flowi6 fl6 = {
2859                 .flowi6_oif = cfg->fc_ifindex,
2860                 .daddr = *gw_addr,
2861                 .saddr = cfg->fc_prefsrc,
2862         };
2863         struct fib6_table *table;
2864         struct rt6_info *rt;
2865
2866         table = fib6_get_table(net, tbid);
2867         if (!table)
2868                 return NULL;
2869
2870         if (!ipv6_addr_any(&cfg->fc_prefsrc))
2871                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2872
2873         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
2874         rt = ip6_pol_route(net, table, cfg->fc_ifindex, &fl6, NULL, flags);
2875
2876         /* if table lookup failed, fall back to full lookup */
2877         if (rt == net->ipv6.ip6_null_entry) {
2878                 ip6_rt_put(rt);
2879                 rt = NULL;
2880         }
2881
2882         return rt;
2883 }
2884
2885 static int ip6_route_check_nh_onlink(struct net *net,
2886                                      struct fib6_config *cfg,
2887                                      const struct net_device *dev,
2888                                      struct netlink_ext_ack *extack)
2889 {
2890         u32 tbid = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
2891         const struct in6_addr *gw_addr = &cfg->fc_gateway;
2892         u32 flags = RTF_LOCAL | RTF_ANYCAST | RTF_REJECT;
2893         struct fib6_info *from;
2894         struct rt6_info *grt;
2895         int err;
2896
2897         err = 0;
2898         grt = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0);
2899         if (grt) {
2900                 rcu_read_lock();
2901                 from = rcu_dereference(grt->from);
2902                 if (!grt->dst.error &&
2903                     /* ignore match if it is the default route */
2904                     from && !ipv6_addr_any(&from->fib6_dst.addr) &&
2905                     (grt->rt6i_flags & flags || dev != grt->dst.dev)) {
2906                         NL_SET_ERR_MSG(extack,
2907                                        "Nexthop has invalid gateway or device mismatch");
2908                         err = -EINVAL;
2909                 }
2910                 rcu_read_unlock();
2911
2912                 ip6_rt_put(grt);
2913         }
2914
2915         return err;
2916 }
2917
2918 static int ip6_route_check_nh(struct net *net,
2919                               struct fib6_config *cfg,
2920                               struct net_device **_dev,
2921                               struct inet6_dev **idev)
2922 {
2923         const struct in6_addr *gw_addr = &cfg->fc_gateway;
2924         struct net_device *dev = _dev ? *_dev : NULL;
2925         struct rt6_info *grt = NULL;
2926         int err = -EHOSTUNREACH;
2927
2928         if (cfg->fc_table) {
2929                 int flags = RT6_LOOKUP_F_IFACE;
2930
2931                 grt = ip6_nh_lookup_table(net, cfg, gw_addr,
2932                                           cfg->fc_table, flags);
2933                 if (grt) {
2934                         if (grt->rt6i_flags & RTF_GATEWAY ||
2935                             (dev && dev != grt->dst.dev)) {
2936                                 ip6_rt_put(grt);
2937                                 grt = NULL;
2938                         }
2939                 }
2940         }
2941
2942         if (!grt)
2943                 grt = rt6_lookup(net, gw_addr, NULL, cfg->fc_ifindex, NULL, 1);
2944
2945         if (!grt)
2946                 goto out;
2947
2948         if (dev) {
2949                 if (dev != grt->dst.dev) {
2950                         ip6_rt_put(grt);
2951                         goto out;
2952                 }
2953         } else {
2954                 *_dev = dev = grt->dst.dev;
2955                 *idev = grt->rt6i_idev;
2956                 dev_hold(dev);
2957                 in6_dev_hold(grt->rt6i_idev);
2958         }
2959
2960         if (!(grt->rt6i_flags & RTF_GATEWAY))
2961                 err = 0;
2962
2963         ip6_rt_put(grt);
2964
2965 out:
2966         return err;
2967 }
2968
2969 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
2970                            struct net_device **_dev, struct inet6_dev **idev,
2971                            struct netlink_ext_ack *extack)
2972 {
2973         const struct in6_addr *gw_addr = &cfg->fc_gateway;
2974         int gwa_type = ipv6_addr_type(gw_addr);
2975         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
2976         const struct net_device *dev = *_dev;
2977         bool need_addr_check = !dev;
2978         int err = -EINVAL;
2979
2980         /* if gw_addr is local we will fail to detect this in case
2981          * address is still TENTATIVE (DAD in progress). rt6_lookup()
2982          * will return already-added prefix route via interface that
2983          * prefix route was assigned to, which might be non-loopback.
2984          */
2985         if (dev &&
2986             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
2987                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
2988                 goto out;
2989         }
2990
2991         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
2992                 /* IPv6 strictly inhibits using not link-local
2993                  * addresses as nexthop address.
2994                  * Otherwise, router will not able to send redirects.
2995                  * It is very good, but in some (rare!) circumstances
2996                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
2997                  * some exceptions. --ANK
2998                  * We allow IPv4-mapped nexthops to support RFC4798-type
2999                  * addressing
3000                  */
3001                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3002                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3003                         goto out;
3004                 }
3005
3006                 if (cfg->fc_flags & RTNH_F_ONLINK)
3007                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3008                 else
3009                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3010
3011                 if (err)
3012                         goto out;
3013         }
3014
3015         /* reload in case device was changed */
3016         dev = *_dev;
3017
3018         err = -EINVAL;
3019         if (!dev) {
3020                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3021                 goto out;
3022         } else if (dev->flags & IFF_LOOPBACK) {
3023                 NL_SET_ERR_MSG(extack,
3024                                "Egress device can not be loopback device for this route");
3025                 goto out;
3026         }
3027
3028         /* if we did not check gw_addr above, do so now that the
3029          * egress device has been resolved.
3030          */
3031         if (need_addr_check &&
3032             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3033                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3034                 goto out;
3035         }
3036
3037         err = 0;
3038 out:
3039         return err;
3040 }
3041
3042 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3043 {
3044         if ((flags & RTF_REJECT) ||
3045             (dev && (dev->flags & IFF_LOOPBACK) &&
3046              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3047              !(flags & RTF_LOCAL)))
3048                 return true;
3049
3050         return false;
3051 }
3052
3053 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3054                  struct fib6_config *cfg, gfp_t gfp_flags,
3055                  struct netlink_ext_ack *extack)
3056 {
3057         struct net_device *dev = NULL;
3058         struct inet6_dev *idev = NULL;
3059         int addr_type;
3060         int err;
3061
3062         fib6_nh->fib_nh_family = AF_INET6;
3063
3064         err = -ENODEV;
3065         if (cfg->fc_ifindex) {
3066                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3067                 if (!dev)
3068                         goto out;
3069                 idev = in6_dev_get(dev);
3070                 if (!idev)
3071                         goto out;
3072         }
3073
3074         if (cfg->fc_flags & RTNH_F_ONLINK) {
3075                 if (!dev) {
3076                         NL_SET_ERR_MSG(extack,
3077                                        "Nexthop device required for onlink");
3078                         goto out;
3079                 }
3080
3081                 if (!(dev->flags & IFF_UP)) {
3082                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3083                         err = -ENETDOWN;
3084                         goto out;
3085                 }
3086
3087                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3088         }
3089
3090         fib6_nh->fib_nh_weight = 1;
3091
3092         /* We cannot add true routes via loopback here,
3093          * they would result in kernel looping; promote them to reject routes
3094          */
3095         addr_type = ipv6_addr_type(&cfg->fc_dst);
3096         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3097                 /* hold loopback dev/idev if we haven't done so. */
3098                 if (dev != net->loopback_dev) {
3099                         if (dev) {
3100                                 dev_put(dev);
3101                                 in6_dev_put(idev);
3102                         }
3103                         dev = net->loopback_dev;
3104                         dev_hold(dev);
3105                         idev = in6_dev_get(dev);
3106                         if (!idev) {
3107                                 err = -ENODEV;
3108                                 goto out;
3109                         }
3110                 }
3111                 goto set_dev;
3112         }
3113
3114         if (cfg->fc_flags & RTF_GATEWAY) {
3115                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3116                 if (err)
3117                         goto out;
3118
3119                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3120                 fib6_nh->fib_nh_gw_family = AF_INET6;
3121         }
3122
3123         err = -ENODEV;
3124         if (!dev)
3125                 goto out;
3126
3127         if (idev->cnf.disable_ipv6) {
3128                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3129                 err = -EACCES;
3130                 goto out;
3131         }
3132
3133         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3134                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3135                 err = -ENETDOWN;
3136                 goto out;
3137         }
3138
3139         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3140             !netif_carrier_ok(dev))
3141                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3142
3143         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3144         if (!fib6_nh->rt6i_pcpu) {
3145                 err = -ENOMEM;
3146                 goto out;
3147         }
3148
3149         err = fib_nh_common_init(&fib6_nh->nh_common, cfg->fc_encap,
3150                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3151         if (err)
3152                 goto out;
3153 set_dev:
3154         fib6_nh->fib_nh_dev = dev;
3155         fib6_nh->fib_nh_oif = dev->ifindex;
3156         err = 0;
3157 out:
3158         if (idev)
3159                 in6_dev_put(idev);
3160
3161         if (err) {
3162                 lwtstate_put(fib6_nh->fib_nh_lws);
3163                 fib6_nh->fib_nh_lws = NULL;
3164                 if (dev)
3165                         dev_put(dev);
3166         }
3167
3168         return err;
3169 }
3170
3171 void fib6_nh_release(struct fib6_nh *fib6_nh)
3172 {
3173         struct rt6_exception_bucket *bucket;
3174
3175         rcu_read_lock();
3176
3177         fib6_nh_flush_exceptions(fib6_nh, NULL);
3178         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3179         if (bucket) {
3180                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3181                 kfree(bucket);
3182         }
3183
3184         rcu_read_unlock();
3185
3186         if (fib6_nh->rt6i_pcpu) {
3187                 int cpu;
3188
3189                 for_each_possible_cpu(cpu) {
3190                         struct rt6_info **ppcpu_rt;
3191                         struct rt6_info *pcpu_rt;
3192
3193                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3194                         pcpu_rt = *ppcpu_rt;
3195                         if (pcpu_rt) {
3196                                 dst_dev_put(&pcpu_rt->dst);
3197                                 dst_release(&pcpu_rt->dst);
3198                                 *ppcpu_rt = NULL;
3199                         }
3200                 }
3201
3202                 free_percpu(fib6_nh->rt6i_pcpu);
3203         }
3204
3205         fib_nh_common_release(&fib6_nh->nh_common);
3206 }
3207
3208 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3209                                               gfp_t gfp_flags,
3210                                               struct netlink_ext_ack *extack)
3211 {
3212         struct net *net = cfg->fc_nlinfo.nl_net;
3213         struct fib6_info *rt = NULL;
3214         struct fib6_table *table;
3215         int err = -EINVAL;
3216         int addr_type;
3217
3218         /* RTF_PCPU is an internal flag; can not be set by userspace */
3219         if (cfg->fc_flags & RTF_PCPU) {
3220                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3221                 goto out;
3222         }
3223
3224         /* RTF_CACHE is an internal flag; can not be set by userspace */
3225         if (cfg->fc_flags & RTF_CACHE) {
3226                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3227                 goto out;
3228         }
3229
3230         if (cfg->fc_type > RTN_MAX) {
3231                 NL_SET_ERR_MSG(extack, "Invalid route type");
3232                 goto out;
3233         }
3234
3235         if (cfg->fc_dst_len > 128) {
3236                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3237                 goto out;
3238         }
3239         if (cfg->fc_src_len > 128) {
3240                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3241                 goto out;
3242         }
3243 #ifndef CONFIG_IPV6_SUBTREES
3244         if (cfg->fc_src_len) {
3245                 NL_SET_ERR_MSG(extack,
3246                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3247                 goto out;
3248         }
3249 #endif
3250
3251         err = -ENOBUFS;
3252         if (cfg->fc_nlinfo.nlh &&
3253             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3254                 table = fib6_get_table(net, cfg->fc_table);
3255                 if (!table) {
3256                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3257                         table = fib6_new_table(net, cfg->fc_table);
3258                 }
3259         } else {
3260                 table = fib6_new_table(net, cfg->fc_table);
3261         }
3262
3263         if (!table)
3264                 goto out;
3265
3266         err = -ENOMEM;
3267         rt = fib6_info_alloc(gfp_flags, true);
3268         if (!rt)
3269                 goto out;
3270
3271         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3272                                                extack);
3273         if (IS_ERR(rt->fib6_metrics)) {
3274                 err = PTR_ERR(rt->fib6_metrics);
3275                 /* Do not leave garbage there. */
3276                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3277                 goto out;
3278         }
3279
3280         if (cfg->fc_flags & RTF_ADDRCONF)
3281                 rt->dst_nocount = true;
3282
3283         if (cfg->fc_flags & RTF_EXPIRES)
3284                 fib6_set_expires(rt, jiffies +
3285                                 clock_t_to_jiffies(cfg->fc_expires));
3286         else
3287                 fib6_clean_expires(rt);
3288
3289         if (cfg->fc_protocol == RTPROT_UNSPEC)
3290                 cfg->fc_protocol = RTPROT_BOOT;
3291         rt->fib6_protocol = cfg->fc_protocol;
3292
3293         rt->fib6_table = table;
3294         rt->fib6_metric = cfg->fc_metric;
3295         rt->fib6_type = cfg->fc_type;
3296         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3297
3298         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3299         rt->fib6_dst.plen = cfg->fc_dst_len;
3300         if (rt->fib6_dst.plen == 128)
3301                 rt->dst_host = true;
3302
3303 #ifdef CONFIG_IPV6_SUBTREES
3304         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3305         rt->fib6_src.plen = cfg->fc_src_len;
3306 #endif
3307         err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3308         if (err)
3309                 goto out;
3310
3311         /* We cannot add true routes via loopback here,
3312          * they would result in kernel looping; promote them to reject routes
3313          */
3314         addr_type = ipv6_addr_type(&cfg->fc_dst);
3315         if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev, addr_type))
3316                 rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3317
3318         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3319                 struct net_device *dev = fib6_info_nh_dev(rt);
3320
3321                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3322                         NL_SET_ERR_MSG(extack, "Invalid source address");
3323                         err = -EINVAL;
3324                         goto out;
3325                 }
3326                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3327                 rt->fib6_prefsrc.plen = 128;
3328         } else
3329                 rt->fib6_prefsrc.plen = 0;
3330
3331         return rt;
3332 out:
3333         fib6_info_release(rt);
3334         return ERR_PTR(err);
3335 }
3336
3337 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3338                   struct netlink_ext_ack *extack)
3339 {
3340         struct fib6_info *rt;
3341         int err;
3342
3343         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3344         if (IS_ERR(rt))
3345                 return PTR_ERR(rt);
3346
3347         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3348         fib6_info_release(rt);
3349
3350         return err;
3351 }
3352
3353 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3354 {
3355         struct net *net = info->nl_net;
3356         struct fib6_table *table;
3357         int err;
3358
3359         if (rt == net->ipv6.fib6_null_entry) {
3360                 err = -ENOENT;
3361                 goto out;
3362         }
3363
3364         table = rt->fib6_table;
3365         spin_lock_bh(&table->tb6_lock);
3366         err = fib6_del(rt, info);
3367         spin_unlock_bh(&table->tb6_lock);
3368
3369 out:
3370         fib6_info_release(rt);
3371         return err;
3372 }
3373
3374 int ip6_del_rt(struct net *net, struct fib6_info *rt)
3375 {
3376         struct nl_info info = { .nl_net = net };
3377
3378         return __ip6_del_rt(rt, &info);
3379 }
3380
3381 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3382 {
3383         struct nl_info *info = &cfg->fc_nlinfo;
3384         struct net *net = info->nl_net;
3385         struct sk_buff *skb = NULL;
3386         struct fib6_table *table;
3387         int err = -ENOENT;
3388
3389         if (rt == net->ipv6.fib6_null_entry)
3390                 goto out_put;
3391         table = rt->fib6_table;
3392         spin_lock_bh(&table->tb6_lock);
3393
3394         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3395                 struct fib6_info *sibling, *next_sibling;
3396
3397                 /* prefer to send a single notification with all hops */
3398                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3399                 if (skb) {
3400                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3401
3402                         if (rt6_fill_node(net, skb, rt, NULL,
3403                                           NULL, NULL, 0, RTM_DELROUTE,
3404                                           info->portid, seq, 0) < 0) {
3405                                 kfree_skb(skb);
3406                                 skb = NULL;
3407                         } else
3408                                 info->skip_notify = 1;
3409                 }
3410
3411                 list_for_each_entry_safe(sibling, next_sibling,
3412                                          &rt->fib6_siblings,
3413                                          fib6_siblings) {
3414                         err = fib6_del(sibling, info);
3415                         if (err)
3416                                 goto out_unlock;
3417                 }
3418         }
3419
3420         err = fib6_del(rt, info);
3421 out_unlock:
3422         spin_unlock_bh(&table->tb6_lock);
3423 out_put:
3424         fib6_info_release(rt);
3425
3426         if (skb) {
3427                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3428                             info->nlh, gfp_any());
3429         }
3430         return err;
3431 }
3432
3433 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3434 {
3435         int rc = -ESRCH;
3436
3437         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3438                 goto out;
3439
3440         if (cfg->fc_flags & RTF_GATEWAY &&
3441             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3442                 goto out;
3443
3444         rc = rt6_remove_exception_rt(rt);
3445 out:
3446         return rc;
3447 }
3448
3449 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3450                              struct fib6_nh *nh)
3451 {
3452         struct fib6_result res = {
3453                 .f6i = rt,
3454                 .nh = nh,
3455         };
3456         struct rt6_info *rt_cache;
3457
3458         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3459         if (rt_cache)
3460                 return __ip6_del_cached_rt(rt_cache, cfg);
3461
3462         return 0;
3463 }
3464
3465 static int ip6_route_del(struct fib6_config *cfg,
3466                          struct netlink_ext_ack *extack)
3467 {
3468         struct fib6_table *table;
3469         struct fib6_info *rt;
3470         struct fib6_node *fn;
3471         int err = -ESRCH;
3472
3473         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3474         if (!table) {
3475                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3476                 return err;
3477         }
3478
3479         rcu_read_lock();
3480
3481         fn = fib6_locate(&table->tb6_root,
3482                          &cfg->fc_dst, cfg->fc_dst_len,
3483                          &cfg->fc_src, cfg->fc_src_len,
3484                          !(cfg->fc_flags & RTF_CACHE));
3485
3486         if (fn) {
3487                 for_each_fib6_node_rt_rcu(fn) {
3488                         struct fib6_nh *nh;
3489
3490                         nh = rt->fib6_nh;
3491                         if (cfg->fc_flags & RTF_CACHE) {
3492                                 int rc;
3493
3494                                 rc = ip6_del_cached_rt(cfg, rt, nh);
3495                                 if (rc != -ESRCH) {
3496                                         rcu_read_unlock();
3497                                         return rc;
3498                                 }
3499                                 continue;
3500                         }
3501
3502                         if (cfg->fc_ifindex &&
3503                             (!nh->fib_nh_dev ||
3504                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3505                                 continue;
3506                         if (cfg->fc_flags & RTF_GATEWAY &&
3507                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3508                                 continue;
3509                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3510                                 continue;
3511                         if (cfg->fc_protocol && cfg->fc_protocol != rt->fib6_protocol)
3512                                 continue;
3513                         if (!fib6_info_hold_safe(rt))
3514                                 continue;
3515                         rcu_read_unlock();
3516
3517                         /* if gateway was specified only delete the one hop */
3518                         if (cfg->fc_flags & RTF_GATEWAY)
3519                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3520
3521                         return __ip6_del_rt_siblings(rt, cfg);
3522                 }
3523         }
3524         rcu_read_unlock();
3525
3526         return err;
3527 }
3528
3529 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3530 {
3531         struct netevent_redirect netevent;
3532         struct rt6_info *rt, *nrt = NULL;
3533         struct fib6_result res = {};
3534         struct ndisc_options ndopts;
3535         struct inet6_dev *in6_dev;
3536         struct neighbour *neigh;
3537         struct rd_msg *msg;
3538         int optlen, on_link;
3539         u8 *lladdr;
3540
3541         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
3542         optlen -= sizeof(*msg);
3543
3544         if (optlen < 0) {
3545                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
3546                 return;
3547         }
3548
3549         msg = (struct rd_msg *)icmp6_hdr(skb);
3550
3551         if (ipv6_addr_is_multicast(&msg->dest)) {
3552                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
3553                 return;
3554         }
3555
3556         on_link = 0;
3557         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
3558                 on_link = 1;
3559         } else if (ipv6_addr_type(&msg->target) !=
3560                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
3561                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
3562                 return;
3563         }
3564
3565         in6_dev = __in6_dev_get(skb->dev);
3566         if (!in6_dev)
3567                 return;
3568         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
3569                 return;
3570
3571         /* RFC2461 8.1:
3572          *      The IP source address of the Redirect MUST be the same as the current
3573          *      first-hop router for the specified ICMP Destination Address.
3574          */
3575
3576         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
3577                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
3578                 return;
3579         }
3580
3581         lladdr = NULL;
3582         if (ndopts.nd_opts_tgt_lladdr) {
3583                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
3584                                              skb->dev);
3585                 if (!lladdr) {
3586                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
3587                         return;
3588                 }
3589         }
3590
3591         rt = (struct rt6_info *) dst;
3592         if (rt->rt6i_flags & RTF_REJECT) {
3593                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
3594                 return;
3595         }
3596
3597         /* Redirect received -> path was valid.
3598          * Look, redirects are sent only in response to data packets,
3599          * so that this nexthop apparently is reachable. --ANK
3600          */
3601         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
3602
3603         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
3604         if (!neigh)
3605                 return;
3606
3607         /*
3608          *      We have finally decided to accept it.
3609          */
3610
3611         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
3612                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
3613                      NEIGH_UPDATE_F_OVERRIDE|
3614                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
3615                                      NEIGH_UPDATE_F_ISROUTER)),
3616                      NDISC_REDIRECT, &ndopts);
3617
3618         rcu_read_lock();
3619         res.f6i = rcu_dereference(rt->from);
3620         if (!res.f6i)
3621                 goto out;
3622
3623         res.nh = res.f6i->fib6_nh;
3624         res.fib6_flags = res.f6i->fib6_flags;
3625         res.fib6_type = res.f6i->fib6_type;
3626         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
3627         if (!nrt)
3628                 goto out;
3629
3630         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
3631         if (on_link)
3632                 nrt->rt6i_flags &= ~RTF_GATEWAY;
3633
3634         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
3635
3636         /* rt6_insert_exception() will take care of duplicated exceptions */
3637         if (rt6_insert_exception(nrt, &res)) {
3638                 dst_release_immediate(&nrt->dst);
3639                 goto out;
3640         }
3641
3642         netevent.old = &rt->dst;
3643         netevent.new = &nrt->dst;
3644         netevent.daddr = &msg->dest;
3645         netevent.neigh = neigh;
3646         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
3647
3648 out:
3649         rcu_read_unlock();
3650         neigh_release(neigh);
3651 }
3652
3653 #ifdef CONFIG_IPV6_ROUTE_INFO
3654 static struct fib6_info *rt6_get_route_info(struct net *net,
3655                                            const struct in6_addr *prefix, int prefixlen,
3656                                            const struct in6_addr *gwaddr,
3657                                            struct net_device *dev)
3658 {
3659         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
3660         int ifindex = dev->ifindex;
3661         struct fib6_node *fn;
3662         struct fib6_info *rt = NULL;
3663         struct fib6_table *table;
3664
3665         table = fib6_get_table(net, tb_id);
3666         if (!table)
3667                 return NULL;
3668
3669         rcu_read_lock();
3670         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
3671         if (!fn)
3672                 goto out;
3673
3674         for_each_fib6_node_rt_rcu(fn) {
3675                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
3676                         continue;
3677                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
3678                     !rt->fib6_nh->fib_nh_gw_family)
3679                         continue;
3680                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
3681                         continue;
3682                 if (!fib6_info_hold_safe(rt))
3683                         continue;
3684                 break;
3685         }
3686 out:
3687         rcu_read_unlock();
3688         return rt;
3689 }
3690
3691 static struct fib6_info *rt6_add_route_info(struct net *net,
3692                                            const struct in6_addr *prefix, int prefixlen,
3693                                            const struct in6_addr *gwaddr,
3694                                            struct net_device *dev,
3695                                            unsigned int pref)
3696 {
3697         struct fib6_config cfg = {
3698                 .fc_metric      = IP6_RT_PRIO_USER,
3699                 .fc_ifindex     = dev->ifindex,
3700                 .fc_dst_len     = prefixlen,
3701                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
3702                                   RTF_UP | RTF_PREF(pref),
3703                 .fc_protocol = RTPROT_RA,
3704                 .fc_type = RTN_UNICAST,
3705                 .fc_nlinfo.portid = 0,
3706                 .fc_nlinfo.nlh = NULL,
3707                 .fc_nlinfo.nl_net = net,
3708         };
3709
3710         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
3711         cfg.fc_dst = *prefix;
3712         cfg.fc_gateway = *gwaddr;
3713
3714         /* We should treat it as a default route if prefix length is 0. */
3715         if (!prefixlen)
3716                 cfg.fc_flags |= RTF_DEFAULT;
3717
3718         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
3719
3720         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
3721 }
3722 #endif
3723
3724 struct fib6_info *rt6_get_dflt_router(struct net *net,
3725                                      const struct in6_addr *addr,
3726                                      struct net_device *dev)
3727 {
3728         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
3729         struct fib6_info *rt;
3730         struct fib6_table *table;
3731
3732         table = fib6_get_table(net, tb_id);
3733         if (!table)
3734                 return NULL;
3735
3736         rcu_read_lock();
3737         for_each_fib6_node_rt_rcu(&table->tb6_root) {
3738                 struct fib6_nh *nh = rt->fib6_nh;
3739
3740                 if (dev == nh->fib_nh_dev &&
3741                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
3742                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
3743                         break;
3744         }
3745         if (rt && !fib6_info_hold_safe(rt))
3746                 rt = NULL;
3747         rcu_read_unlock();
3748         return rt;
3749 }
3750
3751 struct fib6_info *rt6_add_dflt_router(struct net *net,
3752                                      const struct in6_addr *gwaddr,
3753                                      struct net_device *dev,
3754                                      unsigned int pref)
3755 {
3756         struct fib6_config cfg = {
3757                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
3758                 .fc_metric      = IP6_RT_PRIO_USER,
3759                 .fc_ifindex     = dev->ifindex,
3760                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
3761                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
3762                 .fc_protocol = RTPROT_RA,
3763                 .fc_type = RTN_UNICAST,
3764                 .fc_nlinfo.portid = 0,
3765                 .fc_nlinfo.nlh = NULL,
3766                 .fc_nlinfo.nl_net = net,
3767         };
3768
3769         cfg.fc_gateway = *gwaddr;
3770
3771         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
3772                 struct fib6_table *table;
3773
3774                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
3775                 if (table)
3776                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
3777         }
3778
3779         return rt6_get_dflt_router(net, gwaddr, dev);
3780 }
3781
3782 static void __rt6_purge_dflt_routers(struct net *net,
3783                                      struct fib6_table *table)
3784 {
3785         struct fib6_info *rt;
3786
3787 restart:
3788         rcu_read_lock();
3789         for_each_fib6_node_rt_rcu(&table->tb6_root) {
3790                 struct net_device *dev = fib6_info_nh_dev(rt);
3791                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
3792
3793                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
3794                     (!idev || idev->cnf.accept_ra != 2) &&
3795                     fib6_info_hold_safe(rt)) {
3796                         rcu_read_unlock();
3797                         ip6_del_rt(net, rt);
3798                         goto restart;
3799                 }
3800         }
3801         rcu_read_unlock();
3802
3803         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
3804 }
3805
3806 void rt6_purge_dflt_routers(struct net *net)
3807 {
3808         struct fib6_table *table;
3809         struct hlist_head *head;
3810         unsigned int h;
3811
3812         rcu_read_lock();
3813
3814         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
3815                 head = &net->ipv6.fib_table_hash[h];
3816                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
3817                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
3818                                 __rt6_purge_dflt_routers(net, table);
3819                 }
3820         }
3821
3822         rcu_read_unlock();
3823 }
3824
3825 static void rtmsg_to_fib6_config(struct net *net,
3826                                  struct in6_rtmsg *rtmsg,
3827                                  struct fib6_config *cfg)
3828 {
3829         *cfg = (struct fib6_config){
3830                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
3831                          : RT6_TABLE_MAIN,
3832                 .fc_ifindex = rtmsg->rtmsg_ifindex,
3833                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
3834                 .fc_expires = rtmsg->rtmsg_info,
3835                 .fc_dst_len = rtmsg->rtmsg_dst_len,
3836                 .fc_src_len = rtmsg->rtmsg_src_len,
3837                 .fc_flags = rtmsg->rtmsg_flags,
3838                 .fc_type = rtmsg->rtmsg_type,
3839
3840                 .fc_nlinfo.nl_net = net,
3841
3842                 .fc_dst = rtmsg->rtmsg_dst,
3843                 .fc_src = rtmsg->rtmsg_src,
3844                 .fc_gateway = rtmsg->rtmsg_gateway,
3845         };
3846 }
3847
3848 int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3849 {
3850         struct fib6_config cfg;
3851         struct in6_rtmsg rtmsg;
3852         int err;
3853
3854         switch (cmd) {
3855         case SIOCADDRT:         /* Add a route */
3856         case SIOCDELRT:         /* Delete a route */
3857                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3858                         return -EPERM;
3859                 err = copy_from_user(&rtmsg, arg,
3860                                      sizeof(struct in6_rtmsg));
3861                 if (err)
3862                         return -EFAULT;
3863
3864                 rtmsg_to_fib6_config(net, &rtmsg, &cfg);
3865
3866                 rtnl_lock();
3867                 switch (cmd) {
3868                 case SIOCADDRT:
3869                         err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
3870                         break;
3871                 case SIOCDELRT:
3872                         err = ip6_route_del(&cfg, NULL);
3873                         break;
3874                 default:
3875                         err = -EINVAL;
3876                 }
3877                 rtnl_unlock();
3878
3879                 return err;
3880         }
3881
3882         return -EINVAL;
3883 }
3884
3885 /*
3886  *      Drop the packet on the floor
3887  */
3888
3889 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
3890 {
3891         struct dst_entry *dst = skb_dst(skb);
3892         struct net *net = dev_net(dst->dev);
3893         struct inet6_dev *idev;
3894         int type;
3895
3896         if (netif_is_l3_master(skb->dev) &&
3897             dst->dev == net->loopback_dev)
3898                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
3899         else
3900                 idev = ip6_dst_idev(dst);
3901
3902         switch (ipstats_mib_noroutes) {
3903         case IPSTATS_MIB_INNOROUTES:
3904                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
3905                 if (type == IPV6_ADDR_ANY) {
3906                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
3907                         break;
3908                 }
3909                 /* FALLTHROUGH */
3910         case IPSTATS_MIB_OUTNOROUTES:
3911                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
3912                 break;
3913         }
3914
3915         /* Start over by dropping the dst for l3mdev case */
3916         if (netif_is_l3_master(skb->dev))
3917                 skb_dst_drop(skb);
3918
3919         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
3920         kfree_skb(skb);
3921         return 0;
3922 }
3923
3924 static int ip6_pkt_discard(struct sk_buff *skb)
3925 {
3926         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
3927 }
3928
3929 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
3930 {
3931         skb->dev = skb_dst(skb)->dev;
3932         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
3933 }
3934
3935 static int ip6_pkt_prohibit(struct sk_buff *skb)
3936 {
3937         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
3938 }
3939
3940 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
3941 {
3942         skb->dev = skb_dst(skb)->dev;
3943         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
3944 }
3945
3946 /*
3947  *      Allocate a dst for local (unicast / anycast) address.
3948  */
3949
3950 struct fib6_info *addrconf_f6i_alloc(struct net *net,
3951                                      struct inet6_dev *idev,
3952                                      const struct in6_addr *addr,
3953                                      bool anycast, gfp_t gfp_flags)
3954 {
3955         struct fib6_config cfg = {
3956                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
3957                 .fc_ifindex = idev->dev->ifindex,
3958                 .fc_flags = RTF_UP | RTF_ADDRCONF | RTF_NONEXTHOP,
3959                 .fc_dst = *addr,
3960                 .fc_dst_len = 128,
3961                 .fc_protocol = RTPROT_KERNEL,
3962                 .fc_nlinfo.nl_net = net,
3963                 .fc_ignore_dev_down = true,
3964         };
3965
3966         if (anycast) {
3967                 cfg.fc_type = RTN_ANYCAST;
3968                 cfg.fc_flags |= RTF_ANYCAST;
3969         } else {
3970                 cfg.fc_type = RTN_LOCAL;
3971                 cfg.fc_flags |= RTF_LOCAL;
3972         }
3973
3974         return ip6_route_info_create(&cfg, gfp_flags, NULL);
3975 }
3976
3977 /* remove deleted ip from prefsrc entries */
3978 struct arg_dev_net_ip {
3979         struct net_device *dev;
3980         struct net *net;
3981         struct in6_addr *addr;
3982 };
3983
3984 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
3985 {
3986         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
3987         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
3988         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
3989
3990         if (((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
3991             rt != net->ipv6.fib6_null_entry &&
3992             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
3993                 spin_lock_bh(&rt6_exception_lock);
3994                 /* remove prefsrc entry */
3995                 rt->fib6_prefsrc.plen = 0;
3996                 spin_unlock_bh(&rt6_exception_lock);
3997         }
3998         return 0;
3999 }
4000
4001 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4002 {
4003         struct net *net = dev_net(ifp->idev->dev);
4004         struct arg_dev_net_ip adni = {
4005                 .dev = ifp->idev->dev,
4006                 .net = net,
4007                 .addr = &ifp->addr,
4008         };
4009         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4010 }
4011
4012 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4013
4014 /* Remove routers and update dst entries when gateway turn into host. */
4015 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4016 {
4017         struct in6_addr *gateway = (struct in6_addr *)arg;
4018         struct fib6_nh *nh = rt->fib6_nh;
4019
4020         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4021             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4022                 return -1;
4023
4024         /* Further clean up cached routes in exception table.
4025          * This is needed because cached route may have a different
4026          * gateway than its 'parent' in the case of an ip redirect.
4027          */
4028         fib6_nh_exceptions_clean_tohost(nh, gateway);
4029
4030         return 0;
4031 }
4032
4033 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4034 {
4035         fib6_clean_all(net, fib6_clean_tohost, gateway);
4036 }
4037
4038 struct arg_netdev_event {
4039         const struct net_device *dev;
4040         union {
4041                 unsigned char nh_flags;
4042                 unsigned long event;
4043         };
4044 };
4045
4046 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4047 {
4048         struct fib6_info *iter;
4049         struct fib6_node *fn;
4050
4051         fn = rcu_dereference_protected(rt->fib6_node,
4052                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4053         iter = rcu_dereference_protected(fn->leaf,
4054                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4055         while (iter) {
4056                 if (iter->fib6_metric == rt->fib6_metric &&
4057                     rt6_qualify_for_ecmp(iter))
4058                         return iter;
4059                 iter = rcu_dereference_protected(iter->fib6_next,
4060                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4061         }
4062
4063         return NULL;
4064 }
4065
4066 static bool rt6_is_dead(const struct fib6_info *rt)
4067 {
4068         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4069             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4070              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4071                 return true;
4072
4073         return false;
4074 }
4075
4076 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4077 {
4078         struct fib6_info *iter;
4079         int total = 0;
4080
4081         if (!rt6_is_dead(rt))
4082                 total += rt->fib6_nh->fib_nh_weight;
4083
4084         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4085                 if (!rt6_is_dead(iter))
4086                         total += iter->fib6_nh->fib_nh_weight;
4087         }
4088
4089         return total;
4090 }
4091
4092 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4093 {
4094         int upper_bound = -1;
4095
4096         if (!rt6_is_dead(rt)) {
4097                 *weight += rt->fib6_nh->fib_nh_weight;
4098                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4099                                                     total) - 1;
4100         }
4101         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4102 }
4103
4104 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4105 {
4106         struct fib6_info *iter;
4107         int weight = 0;
4108
4109         rt6_upper_bound_set(rt, &weight, total);
4110
4111         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4112                 rt6_upper_bound_set(iter, &weight, total);
4113 }
4114
4115 void rt6_multipath_rebalance(struct fib6_info *rt)
4116 {
4117         struct fib6_info *first;
4118         int total;
4119
4120         /* In case the entire multipath route was marked for flushing,
4121          * then there is no need to rebalance upon the removal of every
4122          * sibling route.
4123          */
4124         if (!rt->fib6_nsiblings || rt->should_flush)
4125                 return;
4126
4127         /* During lookup routes are evaluated in order, so we need to
4128          * make sure upper bounds are assigned from the first sibling
4129          * onwards.
4130          */
4131         first = rt6_multipath_first_sibling(rt);
4132         if (WARN_ON_ONCE(!first))
4133                 return;
4134
4135         total = rt6_multipath_total_weight(first);
4136         rt6_multipath_upper_bound_set(first, total);
4137 }
4138
4139 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4140 {
4141         const struct arg_netdev_event *arg = p_arg;
4142         struct net *net = dev_net(arg->dev);
4143
4144         if (rt != net->ipv6.fib6_null_entry &&
4145             rt->fib6_nh->fib_nh_dev == arg->dev) {
4146                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4147                 fib6_update_sernum_upto_root(net, rt);
4148                 rt6_multipath_rebalance(rt);
4149         }
4150
4151         return 0;
4152 }
4153
4154 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4155 {
4156         struct arg_netdev_event arg = {
4157                 .dev = dev,
4158                 {
4159                         .nh_flags = nh_flags,
4160                 },
4161         };
4162
4163         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4164                 arg.nh_flags |= RTNH_F_LINKDOWN;
4165
4166         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4167 }
4168
4169 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4170                                    const struct net_device *dev)
4171 {
4172         struct fib6_info *iter;
4173
4174         if (rt->fib6_nh->fib_nh_dev == dev)
4175                 return true;
4176         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4177                 if (iter->fib6_nh->fib_nh_dev == dev)
4178                         return true;
4179
4180         return false;
4181 }
4182
4183 static void rt6_multipath_flush(struct fib6_info *rt)
4184 {
4185         struct fib6_info *iter;
4186
4187         rt->should_flush = 1;
4188         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4189                 iter->should_flush = 1;
4190 }
4191
4192 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4193                                              const struct net_device *down_dev)
4194 {
4195         struct fib6_info *iter;
4196         unsigned int dead = 0;
4197
4198         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4199             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4200                 dead++;
4201         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4202                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4203                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4204                         dead++;
4205
4206         return dead;
4207 }
4208
4209 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4210                                        const struct net_device *dev,
4211                                        unsigned char nh_flags)
4212 {
4213         struct fib6_info *iter;
4214
4215         if (rt->fib6_nh->fib_nh_dev == dev)
4216                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4217         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4218                 if (iter->fib6_nh->fib_nh_dev == dev)
4219                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4220 }
4221
4222 /* called with write lock held for table with rt */
4223 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4224 {
4225         const struct arg_netdev_event *arg = p_arg;
4226         const struct net_device *dev = arg->dev;
4227         struct net *net = dev_net(dev);
4228
4229         if (rt == net->ipv6.fib6_null_entry)
4230                 return 0;
4231
4232         switch (arg->event) {
4233         case NETDEV_UNREGISTER:
4234                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4235         case NETDEV_DOWN:
4236                 if (rt->should_flush)
4237                         return -1;
4238                 if (!rt->fib6_nsiblings)
4239                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4240                 if (rt6_multipath_uses_dev(rt, dev)) {
4241                         unsigned int count;
4242
4243                         count = rt6_multipath_dead_count(rt, dev);
4244                         if (rt->fib6_nsiblings + 1 == count) {
4245                                 rt6_multipath_flush(rt);
4246                                 return -1;
4247                         }
4248                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4249                                                    RTNH_F_LINKDOWN);
4250                         fib6_update_sernum(net, rt);
4251                         rt6_multipath_rebalance(rt);
4252                 }
4253                 return -2;
4254         case NETDEV_CHANGE:
4255                 if (rt->fib6_nh->fib_nh_dev != dev ||
4256                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4257                         break;
4258                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4259                 rt6_multipath_rebalance(rt);
4260                 break;
4261         }
4262
4263         return 0;
4264 }
4265
4266 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4267 {
4268         struct arg_netdev_event arg = {
4269                 .dev = dev,
4270                 {
4271                         .event = event,
4272                 },
4273         };
4274         struct net *net = dev_net(dev);
4275
4276         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4277                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4278         else
4279                 fib6_clean_all(net, fib6_ifdown, &arg);
4280 }
4281
4282 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4283 {
4284         rt6_sync_down_dev(dev, event);
4285         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4286         neigh_ifdown(&nd_tbl, dev);
4287 }
4288
4289 struct rt6_mtu_change_arg {
4290         struct net_device *dev;
4291         unsigned int mtu;
4292         struct fib6_info *f6i;
4293 };
4294
4295 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4296 {
4297         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4298         struct fib6_info *f6i = arg->f6i;
4299
4300         /* For administrative MTU increase, there is no way to discover
4301          * IPv6 PMTU increase, so PMTU increase should be updated here.
4302          * Since RFC 1981 doesn't include administrative MTU increase
4303          * update PMTU increase is a MUST. (i.e. jumbo frame)
4304          */
4305         if (nh->fib_nh_dev == arg->dev) {
4306                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4307                 u32 mtu = f6i->fib6_pmtu;
4308
4309                 if (mtu >= arg->mtu ||
4310                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4311                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4312
4313                 spin_lock_bh(&rt6_exception_lock);
4314                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4315                 spin_unlock_bh(&rt6_exception_lock);
4316         }
4317
4318         return 0;
4319 }
4320
4321 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4322 {
4323         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4324         struct inet6_dev *idev;
4325
4326         /* In IPv6 pmtu discovery is not optional,
4327            so that RTAX_MTU lock cannot disable it.
4328            We still use this lock to block changes
4329            caused by addrconf/ndisc.
4330         */
4331
4332         idev = __in6_dev_get(arg->dev);
4333         if (!idev)
4334                 return 0;
4335
4336         if (fib6_metric_locked(f6i, RTAX_MTU))
4337                 return 0;
4338
4339         arg->f6i = f6i;
4340         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4341 }
4342
4343 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4344 {
4345         struct rt6_mtu_change_arg arg = {
4346                 .dev = dev,
4347                 .mtu = mtu,
4348         };
4349
4350         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4351 }
4352
4353 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4354         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4355         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4356         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4357         [RTA_OIF]               = { .type = NLA_U32 },
4358         [RTA_IIF]               = { .type = NLA_U32 },
4359         [RTA_PRIORITY]          = { .type = NLA_U32 },
4360         [RTA_METRICS]           = { .type = NLA_NESTED },
4361         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4362         [RTA_PREF]              = { .type = NLA_U8 },
4363         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4364         [RTA_ENCAP]             = { .type = NLA_NESTED },
4365         [RTA_EXPIRES]           = { .type = NLA_U32 },
4366         [RTA_UID]               = { .type = NLA_U32 },
4367         [RTA_MARK]              = { .type = NLA_U32 },
4368         [RTA_TABLE]             = { .type = NLA_U32 },
4369         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4370         [RTA_SPORT]             = { .type = NLA_U16 },
4371         [RTA_DPORT]             = { .type = NLA_U16 },
4372 };
4373
4374 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4375                               struct fib6_config *cfg,
4376                               struct netlink_ext_ack *extack)
4377 {
4378         struct rtmsg *rtm;
4379         struct nlattr *tb[RTA_MAX+1];
4380         unsigned int pref;
4381         int err;
4382
4383         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4384                                      rtm_ipv6_policy, extack);
4385         if (err < 0)
4386                 goto errout;
4387
4388         err = -EINVAL;
4389         rtm = nlmsg_data(nlh);
4390
4391         *cfg = (struct fib6_config){
4392                 .fc_table = rtm->rtm_table,
4393                 .fc_dst_len = rtm->rtm_dst_len,
4394                 .fc_src_len = rtm->rtm_src_len,
4395                 .fc_flags = RTF_UP,
4396                 .fc_protocol = rtm->rtm_protocol,
4397                 .fc_type = rtm->rtm_type,
4398
4399                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4400                 .fc_nlinfo.nlh = nlh,
4401                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4402         };
4403
4404         if (rtm->rtm_type == RTN_UNREACHABLE ||
4405             rtm->rtm_type == RTN_BLACKHOLE ||
4406             rtm->rtm_type == RTN_PROHIBIT ||
4407             rtm->rtm_type == RTN_THROW)
4408                 cfg->fc_flags |= RTF_REJECT;
4409
4410         if (rtm->rtm_type == RTN_LOCAL)
4411                 cfg->fc_flags |= RTF_LOCAL;
4412
4413         if (rtm->rtm_flags & RTM_F_CLONED)
4414                 cfg->fc_flags |= RTF_CACHE;
4415
4416         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4417
4418         if (tb[RTA_GATEWAY]) {
4419                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4420                 cfg->fc_flags |= RTF_GATEWAY;
4421         }
4422         if (tb[RTA_VIA]) {
4423                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4424                 goto errout;
4425         }
4426
4427         if (tb[RTA_DST]) {
4428                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4429
4430                 if (nla_len(tb[RTA_DST]) < plen)
4431                         goto errout;
4432
4433                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4434         }
4435
4436         if (tb[RTA_SRC]) {
4437                 int plen = (rtm->rtm_src_len + 7) >> 3;
4438
4439                 if (nla_len(tb[RTA_SRC]) < plen)
4440                         goto errout;
4441
4442                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4443         }
4444
4445         if (tb[RTA_PREFSRC])
4446                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4447
4448         if (tb[RTA_OIF])
4449                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4450
4451         if (tb[RTA_PRIORITY])
4452                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4453
4454         if (tb[RTA_METRICS]) {
4455                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4456                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4457         }
4458
4459         if (tb[RTA_TABLE])
4460                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4461
4462         if (tb[RTA_MULTIPATH]) {
4463                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4464                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4465
4466                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4467                                                      cfg->fc_mp_len, extack);
4468                 if (err < 0)
4469                         goto errout;
4470         }
4471
4472         if (tb[RTA_PREF]) {
4473                 pref = nla_get_u8(tb[RTA_PREF]);
4474                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4475                     pref != ICMPV6_ROUTER_PREF_HIGH)
4476                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4477                 cfg->fc_flags |= RTF_PREF(pref);
4478         }
4479
4480         if (tb[RTA_ENCAP])
4481                 cfg->fc_encap = tb[RTA_ENCAP];
4482
4483         if (tb[RTA_ENCAP_TYPE]) {
4484                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4485
4486                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4487                 if (err < 0)
4488                         goto errout;
4489         }
4490
4491         if (tb[RTA_EXPIRES]) {
4492                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4493
4494                 if (addrconf_finite_timeout(timeout)) {
4495                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
4496                         cfg->fc_flags |= RTF_EXPIRES;
4497                 }
4498         }
4499
4500         err = 0;
4501 errout:
4502         return err;
4503 }
4504
4505 struct rt6_nh {
4506         struct fib6_info *fib6_info;
4507         struct fib6_config r_cfg;
4508         struct list_head next;
4509 };
4510
4511 static int ip6_route_info_append(struct net *net,
4512                                  struct list_head *rt6_nh_list,
4513                                  struct fib6_info *rt,
4514                                  struct fib6_config *r_cfg)
4515 {
4516         struct rt6_nh *nh;
4517         int err = -EEXIST;
4518
4519         list_for_each_entry(nh, rt6_nh_list, next) {
4520                 /* check if fib6_info already exists */
4521                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
4522                         return err;
4523         }
4524
4525         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
4526         if (!nh)
4527                 return -ENOMEM;
4528         nh->fib6_info = rt;
4529         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
4530         list_add_tail(&nh->next, rt6_nh_list);
4531
4532         return 0;
4533 }
4534
4535 static void ip6_route_mpath_notify(struct fib6_info *rt,
4536                                    struct fib6_info *rt_last,
4537                                    struct nl_info *info,
4538                                    __u16 nlflags)
4539 {
4540         /* if this is an APPEND route, then rt points to the first route
4541          * inserted and rt_last points to last route inserted. Userspace
4542          * wants a consistent dump of the route which starts at the first
4543          * nexthop. Since sibling routes are always added at the end of
4544          * the list, find the first sibling of the last route appended
4545          */
4546         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
4547                 rt = list_first_entry(&rt_last->fib6_siblings,
4548                                       struct fib6_info,
4549                                       fib6_siblings);
4550         }
4551
4552         if (rt)
4553                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
4554 }
4555
4556 static int ip6_route_multipath_add(struct fib6_config *cfg,
4557                                    struct netlink_ext_ack *extack)
4558 {
4559         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
4560         struct nl_info *info = &cfg->fc_nlinfo;
4561         struct fib6_config r_cfg;
4562         struct rtnexthop *rtnh;
4563         struct fib6_info *rt;
4564         struct rt6_nh *err_nh;
4565         struct rt6_nh *nh, *nh_safe;
4566         __u16 nlflags;
4567         int remaining;
4568         int attrlen;
4569         int err = 1;
4570         int nhn = 0;
4571         int replace = (cfg->fc_nlinfo.nlh &&
4572                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
4573         LIST_HEAD(rt6_nh_list);
4574
4575         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
4576         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
4577                 nlflags |= NLM_F_APPEND;
4578
4579         remaining = cfg->fc_mp_len;
4580         rtnh = (struct rtnexthop *)cfg->fc_mp;
4581
4582         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
4583          * fib6_info structs per nexthop
4584          */
4585         while (rtnh_ok(rtnh, remaining)) {
4586                 memcpy(&r_cfg, cfg, sizeof(*cfg));
4587                 if (rtnh->rtnh_ifindex)
4588                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
4589
4590                 attrlen = rtnh_attrlen(rtnh);
4591                 if (attrlen > 0) {
4592                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
4593
4594                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
4595                         if (nla) {
4596                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
4597                                 r_cfg.fc_flags |= RTF_GATEWAY;
4598                         }
4599                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
4600                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
4601                         if (nla)
4602                                 r_cfg.fc_encap_type = nla_get_u16(nla);
4603                 }
4604
4605                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
4606                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
4607                 if (IS_ERR(rt)) {
4608                         err = PTR_ERR(rt);
4609                         rt = NULL;
4610                         goto cleanup;
4611                 }
4612                 if (!rt6_qualify_for_ecmp(rt)) {
4613                         err = -EINVAL;
4614                         NL_SET_ERR_MSG(extack,
4615                                        "Device only routes can not be added for IPv6 using the multipath API.");
4616                         fib6_info_release(rt);
4617                         goto cleanup;
4618                 }
4619
4620                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
4621
4622                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
4623                                             rt, &r_cfg);
4624                 if (err) {
4625                         fib6_info_release(rt);
4626                         goto cleanup;
4627                 }
4628
4629                 rtnh = rtnh_next(rtnh, &remaining);
4630         }
4631
4632         /* for add and replace send one notification with all nexthops.
4633          * Skip the notification in fib6_add_rt2node and send one with
4634          * the full route when done
4635          */
4636         info->skip_notify = 1;
4637
4638         err_nh = NULL;
4639         list_for_each_entry(nh, &rt6_nh_list, next) {
4640                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
4641                 fib6_info_release(nh->fib6_info);
4642
4643                 if (!err) {
4644                         /* save reference to last route successfully inserted */
4645                         rt_last = nh->fib6_info;
4646
4647                         /* save reference to first route for notification */
4648                         if (!rt_notif)
4649                                 rt_notif = nh->fib6_info;
4650                 }
4651
4652                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
4653                 nh->fib6_info = NULL;
4654                 if (err) {
4655                         if (replace && nhn)
4656                                 NL_SET_ERR_MSG_MOD(extack,
4657                                                    "multipath route replace failed (check consistency of installed routes)");
4658                         err_nh = nh;
4659                         goto add_errout;
4660                 }
4661
4662                 /* Because each route is added like a single route we remove
4663                  * these flags after the first nexthop: if there is a collision,
4664                  * we have already failed to add the first nexthop:
4665                  * fib6_add_rt2node() has rejected it; when replacing, old
4666                  * nexthops have been replaced by first new, the rest should
4667                  * be added to it.
4668                  */
4669                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
4670                                                      NLM_F_REPLACE);
4671                 nhn++;
4672         }
4673
4674         /* success ... tell user about new route */
4675         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
4676         goto cleanup;
4677
4678 add_errout:
4679         /* send notification for routes that were added so that
4680          * the delete notifications sent by ip6_route_del are
4681          * coherent
4682          */
4683         if (rt_notif)
4684                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
4685
4686         /* Delete routes that were already added */
4687         list_for_each_entry(nh, &rt6_nh_list, next) {
4688                 if (err_nh == nh)
4689                         break;
4690                 ip6_route_del(&nh->r_cfg, extack);
4691         }
4692
4693 cleanup:
4694         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
4695                 if (nh->fib6_info)
4696                         fib6_info_release(nh->fib6_info);
4697                 list_del(&nh->next);
4698                 kfree(nh);
4699         }
4700
4701         return err;
4702 }
4703
4704 static int ip6_route_multipath_del(struct fib6_config *cfg,
4705                                    struct netlink_ext_ack *extack)
4706 {
4707         struct fib6_config r_cfg;
4708         struct rtnexthop *rtnh;
4709         int remaining;
4710         int attrlen;
4711         int err = 1, last_err = 0;
4712
4713         remaining = cfg->fc_mp_len;
4714         rtnh = (struct rtnexthop *)cfg->fc_mp;
4715
4716         /* Parse a Multipath Entry */
4717         while (rtnh_ok(rtnh, remaining)) {
4718                 memcpy(&r_cfg, cfg, sizeof(*cfg));
4719                 if (rtnh->rtnh_ifindex)
4720                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
4721
4722                 attrlen = rtnh_attrlen(rtnh);
4723                 if (attrlen > 0) {
4724                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
4725
4726                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
4727                         if (nla) {
4728                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
4729                                 r_cfg.fc_flags |= RTF_GATEWAY;
4730                         }
4731                 }
4732                 err = ip6_route_del(&r_cfg, extack);
4733                 if (err)
4734                         last_err = err;
4735
4736                 rtnh = rtnh_next(rtnh, &remaining);
4737         }
4738
4739         return last_err;
4740 }
4741
4742 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
4743                               struct netlink_ext_ack *extack)
4744 {
4745         struct fib6_config cfg;
4746         int err;
4747
4748         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
4749         if (err < 0)
4750                 return err;
4751
4752         if (cfg.fc_mp)
4753                 return ip6_route_multipath_del(&cfg, extack);
4754         else {
4755                 cfg.fc_delete_all_nh = 1;
4756                 return ip6_route_del(&cfg, extack);
4757         }
4758 }
4759
4760 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
4761                               struct netlink_ext_ack *extack)
4762 {
4763         struct fib6_config cfg;
4764         int err;
4765
4766         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
4767         if (err < 0)
4768                 return err;
4769
4770         if (cfg.fc_metric == 0)
4771                 cfg.fc_metric = IP6_RT_PRIO_USER;
4772
4773         if (cfg.fc_mp)
4774                 return ip6_route_multipath_add(&cfg, extack);
4775         else
4776                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
4777 }
4778
4779 static size_t rt6_nlmsg_size(struct fib6_info *rt)
4780 {
4781         int nexthop_len = 0;
4782
4783         if (rt->fib6_nsiblings) {
4784                 nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
4785                             + NLA_ALIGN(sizeof(struct rtnexthop))
4786                             + nla_total_size(16) /* RTA_GATEWAY */
4787                             + lwtunnel_get_encap_size(rt->fib6_nh->fib_nh_lws);
4788
4789                 nexthop_len *= rt->fib6_nsiblings;
4790         }
4791
4792         return NLMSG_ALIGN(sizeof(struct rtmsg))
4793                + nla_total_size(16) /* RTA_SRC */
4794                + nla_total_size(16) /* RTA_DST */
4795                + nla_total_size(16) /* RTA_GATEWAY */
4796                + nla_total_size(16) /* RTA_PREFSRC */
4797                + nla_total_size(4) /* RTA_TABLE */
4798                + nla_total_size(4) /* RTA_IIF */
4799                + nla_total_size(4) /* RTA_OIF */
4800                + nla_total_size(4) /* RTA_PRIORITY */
4801                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
4802                + nla_total_size(sizeof(struct rta_cacheinfo))
4803                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
4804                + nla_total_size(1) /* RTA_PREF */
4805                + lwtunnel_get_encap_size(rt->fib6_nh->fib_nh_lws)
4806                + nexthop_len;
4807 }
4808
4809 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
4810                          struct fib6_info *rt, struct dst_entry *dst,
4811                          struct in6_addr *dest, struct in6_addr *src,
4812                          int iif, int type, u32 portid, u32 seq,
4813                          unsigned int flags)
4814 {
4815         struct rt6_info *rt6 = (struct rt6_info *)dst;
4816         struct rt6key *rt6_dst, *rt6_src;
4817         u32 *pmetrics, table, rt6_flags;
4818         struct nlmsghdr *nlh;
4819         struct rtmsg *rtm;
4820         long expires = 0;
4821
4822         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
4823         if (!nlh)
4824                 return -EMSGSIZE;
4825
4826         if (rt6) {
4827                 rt6_dst = &rt6->rt6i_dst;
4828                 rt6_src = &rt6->rt6i_src;
4829                 rt6_flags = rt6->rt6i_flags;
4830         } else {
4831                 rt6_dst = &rt->fib6_dst;
4832                 rt6_src = &rt->fib6_src;
4833                 rt6_flags = rt->fib6_flags;
4834         }
4835
4836         rtm = nlmsg_data(nlh);
4837         rtm->rtm_family = AF_INET6;
4838         rtm->rtm_dst_len = rt6_dst->plen;
4839         rtm->rtm_src_len = rt6_src->plen;
4840         rtm->rtm_tos = 0;
4841         if (rt->fib6_table)
4842                 table = rt->fib6_table->tb6_id;
4843         else
4844                 table = RT6_TABLE_UNSPEC;
4845         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
4846         if (nla_put_u32(skb, RTA_TABLE, table))
4847                 goto nla_put_failure;
4848
4849         rtm->rtm_type = rt->fib6_type;
4850         rtm->rtm_flags = 0;
4851         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
4852         rtm->rtm_protocol = rt->fib6_protocol;
4853
4854         if (rt6_flags & RTF_CACHE)
4855                 rtm->rtm_flags |= RTM_F_CLONED;
4856
4857         if (dest) {
4858                 if (nla_put_in6_addr(skb, RTA_DST, dest))
4859                         goto nla_put_failure;
4860                 rtm->rtm_dst_len = 128;
4861         } else if (rtm->rtm_dst_len)
4862                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
4863                         goto nla_put_failure;
4864 #ifdef CONFIG_IPV6_SUBTREES
4865         if (src) {
4866                 if (nla_put_in6_addr(skb, RTA_SRC, src))
4867                         goto nla_put_failure;
4868                 rtm->rtm_src_len = 128;
4869         } else if (rtm->rtm_src_len &&
4870                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
4871                 goto nla_put_failure;
4872 #endif
4873         if (iif) {
4874 #ifdef CONFIG_IPV6_MROUTE
4875                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
4876                         int err = ip6mr_get_route(net, skb, rtm, portid);
4877
4878                         if (err == 0)
4879                                 return 0;
4880                         if (err < 0)
4881                                 goto nla_put_failure;
4882                 } else
4883 #endif
4884                         if (nla_put_u32(skb, RTA_IIF, iif))
4885                                 goto nla_put_failure;
4886         } else if (dest) {
4887                 struct in6_addr saddr_buf;
4888                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
4889                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
4890                         goto nla_put_failure;
4891         }
4892
4893         if (rt->fib6_prefsrc.plen) {
4894                 struct in6_addr saddr_buf;
4895                 saddr_buf = rt->fib6_prefsrc.addr;
4896                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
4897                         goto nla_put_failure;
4898         }
4899
4900         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
4901         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
4902                 goto nla_put_failure;
4903
4904         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
4905                 goto nla_put_failure;
4906
4907         /* For multipath routes, walk the siblings list and add
4908          * each as a nexthop within RTA_MULTIPATH.
4909          */
4910         if (rt6) {
4911                 if (rt6_flags & RTF_GATEWAY &&
4912                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
4913                         goto nla_put_failure;
4914
4915                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
4916                         goto nla_put_failure;
4917         } else if (rt->fib6_nsiblings) {
4918                 struct fib6_info *sibling, *next_sibling;
4919                 struct nlattr *mp;
4920
4921                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
4922                 if (!mp)
4923                         goto nla_put_failure;
4924
4925                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
4926                                     rt->fib6_nh->fib_nh_weight) < 0)
4927                         goto nla_put_failure;
4928
4929                 list_for_each_entry_safe(sibling, next_sibling,
4930                                          &rt->fib6_siblings, fib6_siblings) {
4931                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
4932                                             sibling->fib6_nh->fib_nh_weight) < 0)
4933                                 goto nla_put_failure;
4934                 }
4935
4936                 nla_nest_end(skb, mp);
4937         } else {
4938                 unsigned char nh_flags = 0;
4939
4940                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common,
4941                                      &nh_flags, false) < 0)
4942                         goto nla_put_failure;
4943
4944                 rtm->rtm_flags |= nh_flags;
4945         }
4946
4947         if (rt6_flags & RTF_EXPIRES) {
4948                 expires = dst ? dst->expires : rt->expires;
4949                 expires -= jiffies;
4950         }
4951
4952         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
4953                 goto nla_put_failure;
4954
4955         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
4956                 goto nla_put_failure;
4957
4958
4959         nlmsg_end(skb, nlh);
4960         return 0;
4961
4962 nla_put_failure:
4963         nlmsg_cancel(skb, nlh);
4964         return -EMSGSIZE;
4965 }
4966
4967 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
4968                                const struct net_device *dev)
4969 {
4970         if (f6i->fib6_nh->fib_nh_dev == dev)
4971                 return true;
4972
4973         if (f6i->fib6_nsiblings) {
4974                 struct fib6_info *sibling, *next_sibling;
4975
4976                 list_for_each_entry_safe(sibling, next_sibling,
4977                                          &f6i->fib6_siblings, fib6_siblings) {
4978                         if (sibling->fib6_nh->fib_nh_dev == dev)
4979                                 return true;
4980                 }
4981         }
4982
4983         return false;
4984 }
4985
4986 int rt6_dump_route(struct fib6_info *rt, void *p_arg)
4987 {
4988         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
4989         struct fib_dump_filter *filter = &arg->filter;
4990         unsigned int flags = NLM_F_MULTI;
4991         struct net *net = arg->net;
4992
4993         if (rt == net->ipv6.fib6_null_entry)
4994                 return 0;
4995
4996         if ((filter->flags & RTM_F_PREFIX) &&
4997             !(rt->fib6_flags & RTF_PREFIX_RT)) {
4998                 /* success since this is not a prefix route */
4999                 return 1;
5000         }
5001         if (filter->filter_set) {
5002                 if ((filter->rt_type && rt->fib6_type != filter->rt_type) ||
5003                     (filter->dev && !fib6_info_uses_dev(rt, filter->dev)) ||
5004                     (filter->protocol && rt->fib6_protocol != filter->protocol)) {
5005                         return 1;
5006                 }
5007                 flags |= NLM_F_DUMP_FILTERED;
5008         }
5009
5010         return rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL, 0,
5011                              RTM_NEWROUTE, NETLINK_CB(arg->cb->skb).portid,
5012                              arg->cb->nlh->nlmsg_seq, flags);
5013 }
5014
5015 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5016                                         const struct nlmsghdr *nlh,
5017                                         struct nlattr **tb,
5018                                         struct netlink_ext_ack *extack)
5019 {
5020         struct rtmsg *rtm;
5021         int i, err;
5022
5023         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5024                 NL_SET_ERR_MSG_MOD(extack,
5025                                    "Invalid header for get route request");
5026                 return -EINVAL;
5027         }
5028
5029         if (!netlink_strict_get_check(skb))
5030                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5031                                               rtm_ipv6_policy, extack);
5032
5033         rtm = nlmsg_data(nlh);
5034         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5035             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5036             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5037             rtm->rtm_type) {
5038                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5039                 return -EINVAL;
5040         }
5041         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5042                 NL_SET_ERR_MSG_MOD(extack,
5043                                    "Invalid flags for get route request");
5044                 return -EINVAL;
5045         }
5046
5047         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5048                                             rtm_ipv6_policy, extack);
5049         if (err)
5050                 return err;
5051
5052         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5053             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5054                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5055                 return -EINVAL;
5056         }
5057
5058         for (i = 0; i <= RTA_MAX; i++) {
5059                 if (!tb[i])
5060                         continue;
5061
5062                 switch (i) {
5063                 case RTA_SRC:
5064                 case RTA_DST:
5065                 case RTA_IIF:
5066                 case RTA_OIF:
5067                 case RTA_MARK:
5068                 case RTA_UID:
5069                 case RTA_SPORT:
5070                 case RTA_DPORT:
5071                 case RTA_IP_PROTO:
5072                         break;
5073                 default:
5074                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5075                         return -EINVAL;
5076                 }
5077         }
5078
5079         return 0;
5080 }
5081
5082 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5083                               struct netlink_ext_ack *extack)
5084 {
5085         struct net *net = sock_net(in_skb->sk);
5086         struct nlattr *tb[RTA_MAX+1];
5087         int err, iif = 0, oif = 0;
5088         struct fib6_info *from;
5089         struct dst_entry *dst;
5090         struct rt6_info *rt;
5091         struct sk_buff *skb;
5092         struct rtmsg *rtm;
5093         struct flowi6 fl6 = {};
5094         bool fibmatch;
5095
5096         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5097         if (err < 0)
5098                 goto errout;
5099
5100         err = -EINVAL;
5101         rtm = nlmsg_data(nlh);
5102         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5103         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5104
5105         if (tb[RTA_SRC]) {
5106                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5107                         goto errout;
5108
5109                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5110         }
5111
5112         if (tb[RTA_DST]) {
5113                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5114                         goto errout;
5115
5116                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5117         }
5118
5119         if (tb[RTA_IIF])
5120                 iif = nla_get_u32(tb[RTA_IIF]);
5121
5122         if (tb[RTA_OIF])
5123                 oif = nla_get_u32(tb[RTA_OIF]);
5124
5125         if (tb[RTA_MARK])
5126                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5127
5128         if (tb[RTA_UID])
5129                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5130                                            nla_get_u32(tb[RTA_UID]));
5131         else
5132                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5133
5134         if (tb[RTA_SPORT])
5135                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5136
5137         if (tb[RTA_DPORT])
5138                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5139
5140         if (tb[RTA_IP_PROTO]) {
5141                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5142                                                   &fl6.flowi6_proto, AF_INET6,
5143                                                   extack);
5144                 if (err)
5145                         goto errout;
5146         }
5147
5148         if (iif) {
5149                 struct net_device *dev;
5150                 int flags = 0;
5151
5152                 rcu_read_lock();
5153
5154                 dev = dev_get_by_index_rcu(net, iif);
5155                 if (!dev) {
5156                         rcu_read_unlock();
5157                         err = -ENODEV;
5158                         goto errout;
5159                 }
5160
5161                 fl6.flowi6_iif = iif;
5162
5163                 if (!ipv6_addr_any(&fl6.saddr))
5164                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5165
5166                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5167
5168                 rcu_read_unlock();
5169         } else {
5170                 fl6.flowi6_oif = oif;
5171
5172                 dst = ip6_route_output(net, NULL, &fl6);
5173         }
5174
5175
5176         rt = container_of(dst, struct rt6_info, dst);
5177         if (rt->dst.error) {
5178                 err = rt->dst.error;
5179                 ip6_rt_put(rt);
5180                 goto errout;
5181         }
5182
5183         if (rt == net->ipv6.ip6_null_entry) {
5184                 err = rt->dst.error;
5185                 ip6_rt_put(rt);
5186                 goto errout;
5187         }
5188
5189         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5190         if (!skb) {
5191                 ip6_rt_put(rt);
5192                 err = -ENOBUFS;
5193                 goto errout;
5194         }
5195
5196         skb_dst_set(skb, &rt->dst);
5197
5198         rcu_read_lock();
5199         from = rcu_dereference(rt->from);
5200         if (from) {
5201                 if (fibmatch)
5202                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5203                                             iif, RTM_NEWROUTE,
5204                                             NETLINK_CB(in_skb).portid,
5205                                             nlh->nlmsg_seq, 0);
5206                 else
5207                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5208                                             &fl6.saddr, iif, RTM_NEWROUTE,
5209                                             NETLINK_CB(in_skb).portid,
5210                                             nlh->nlmsg_seq, 0);
5211         } else {
5212                 err = -ENETUNREACH;
5213         }
5214         rcu_read_unlock();
5215
5216         if (err < 0) {
5217                 kfree_skb(skb);
5218                 goto errout;
5219         }
5220
5221         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5222 errout:
5223         return err;
5224 }
5225
5226 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
5227                      unsigned int nlm_flags)
5228 {
5229         struct sk_buff *skb;
5230         struct net *net = info->nl_net;
5231         u32 seq;
5232         int err;
5233
5234         err = -ENOBUFS;
5235         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5236
5237         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5238         if (!skb)
5239                 goto errout;
5240
5241         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5242                             event, info->portid, seq, nlm_flags);
5243         if (err < 0) {
5244                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5245                 WARN_ON(err == -EMSGSIZE);
5246                 kfree_skb(skb);
5247                 goto errout;
5248         }
5249         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5250                     info->nlh, gfp_any());
5251         return;
5252 errout:
5253         if (err < 0)
5254                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5255 }
5256
5257 void fib6_rt_update(struct net *net, struct fib6_info *rt,
5258                     struct nl_info *info)
5259 {
5260         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
5261         struct sk_buff *skb;
5262         int err = -ENOBUFS;
5263
5264         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
5265          * is implemented and supported for nexthop objects
5266          */
5267         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
5268
5269         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
5270         if (!skb)
5271                 goto errout;
5272
5273         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
5274                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
5275         if (err < 0) {
5276                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
5277                 WARN_ON(err == -EMSGSIZE);
5278                 kfree_skb(skb);
5279                 goto errout;
5280         }
5281         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
5282                     info->nlh, gfp_any());
5283         return;
5284 errout:
5285         if (err < 0)
5286                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
5287 }
5288
5289 static int ip6_route_dev_notify(struct notifier_block *this,
5290                                 unsigned long event, void *ptr)
5291 {
5292         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
5293         struct net *net = dev_net(dev);
5294
5295         if (!(dev->flags & IFF_LOOPBACK))
5296                 return NOTIFY_OK;
5297
5298         if (event == NETDEV_REGISTER) {
5299                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
5300                 net->ipv6.ip6_null_entry->dst.dev = dev;
5301                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
5302 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5303                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
5304                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
5305                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
5306                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
5307 #endif
5308          } else if (event == NETDEV_UNREGISTER &&
5309                     dev->reg_state != NETREG_UNREGISTERED) {
5310                 /* NETDEV_UNREGISTER could be fired for multiple times by
5311                  * netdev_wait_allrefs(). Make sure we only call this once.
5312                  */
5313                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
5314 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5315                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
5316                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
5317 #endif
5318         }
5319
5320         return NOTIFY_OK;
5321 }
5322
5323 /*
5324  *      /proc
5325  */
5326
5327 #ifdef CONFIG_PROC_FS
5328 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
5329 {
5330         struct net *net = (struct net *)seq->private;
5331         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
5332                    net->ipv6.rt6_stats->fib_nodes,
5333                    net->ipv6.rt6_stats->fib_route_nodes,
5334                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
5335                    net->ipv6.rt6_stats->fib_rt_entries,
5336                    net->ipv6.rt6_stats->fib_rt_cache,
5337                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
5338                    net->ipv6.rt6_stats->fib_discarded_routes);
5339
5340         return 0;
5341 }
5342 #endif  /* CONFIG_PROC_FS */
5343
5344 #ifdef CONFIG_SYSCTL
5345
5346 static
5347 int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
5348                               void __user *buffer, size_t *lenp, loff_t *ppos)
5349 {
5350         struct net *net;
5351         int delay;
5352         int ret;
5353         if (!write)
5354                 return -EINVAL;
5355
5356         net = (struct net *)ctl->extra1;
5357         delay = net->ipv6.sysctl.flush_delay;
5358         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
5359         if (ret)
5360                 return ret;
5361
5362         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
5363         return 0;
5364 }
5365
5366 static int zero;
5367 static int one = 1;
5368
5369 static struct ctl_table ipv6_route_table_template[] = {
5370         {
5371                 .procname       =       "flush",
5372                 .data           =       &init_net.ipv6.sysctl.flush_delay,
5373                 .maxlen         =       sizeof(int),
5374                 .mode           =       0200,
5375                 .proc_handler   =       ipv6_sysctl_rtcache_flush
5376         },
5377         {
5378                 .procname       =       "gc_thresh",
5379                 .data           =       &ip6_dst_ops_template.gc_thresh,
5380                 .maxlen         =       sizeof(int),
5381                 .mode           =       0644,
5382                 .proc_handler   =       proc_dointvec,
5383         },
5384         {
5385                 .procname       =       "max_size",
5386                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
5387                 .maxlen         =       sizeof(int),
5388                 .mode           =       0644,
5389                 .proc_handler   =       proc_dointvec,
5390         },
5391         {
5392                 .procname       =       "gc_min_interval",
5393                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
5394                 .maxlen         =       sizeof(int),
5395                 .mode           =       0644,
5396                 .proc_handler   =       proc_dointvec_jiffies,
5397         },
5398         {
5399                 .procname       =       "gc_timeout",
5400                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
5401                 .maxlen         =       sizeof(int),
5402                 .mode           =       0644,
5403                 .proc_handler   =       proc_dointvec_jiffies,
5404         },
5405         {
5406                 .procname       =       "gc_interval",
5407                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
5408                 .maxlen         =       sizeof(int),
5409                 .mode           =       0644,
5410                 .proc_handler   =       proc_dointvec_jiffies,
5411         },
5412         {
5413                 .procname       =       "gc_elasticity",
5414                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
5415                 .maxlen         =       sizeof(int),
5416                 .mode           =       0644,
5417                 .proc_handler   =       proc_dointvec,
5418         },
5419         {
5420                 .procname       =       "mtu_expires",
5421                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
5422                 .maxlen         =       sizeof(int),
5423                 .mode           =       0644,
5424                 .proc_handler   =       proc_dointvec_jiffies,
5425         },
5426         {
5427                 .procname       =       "min_adv_mss",
5428                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
5429                 .maxlen         =       sizeof(int),
5430                 .mode           =       0644,
5431                 .proc_handler   =       proc_dointvec,
5432         },
5433         {
5434                 .procname       =       "gc_min_interval_ms",
5435                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
5436                 .maxlen         =       sizeof(int),
5437                 .mode           =       0644,
5438                 .proc_handler   =       proc_dointvec_ms_jiffies,
5439         },
5440         {
5441                 .procname       =       "skip_notify_on_dev_down",
5442                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
5443                 .maxlen         =       sizeof(int),
5444                 .mode           =       0644,
5445                 .proc_handler   =       proc_dointvec,
5446                 .extra1         =       &zero,
5447                 .extra2         =       &one,
5448         },
5449         { }
5450 };
5451
5452 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
5453 {
5454         struct ctl_table *table;
5455
5456         table = kmemdup(ipv6_route_table_template,
5457                         sizeof(ipv6_route_table_template),
5458                         GFP_KERNEL);
5459
5460         if (table) {
5461                 table[0].data = &net->ipv6.sysctl.flush_delay;
5462                 table[0].extra1 = net;
5463                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
5464                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
5465                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
5466                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
5467                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
5468                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
5469                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
5470                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
5471                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
5472                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
5473
5474                 /* Don't export sysctls to unprivileged users */
5475                 if (net->user_ns != &init_user_ns)
5476                         table[0].procname = NULL;
5477         }
5478
5479         return table;
5480 }
5481 #endif
5482
5483 static int __net_init ip6_route_net_init(struct net *net)
5484 {
5485         int ret = -ENOMEM;
5486
5487         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
5488                sizeof(net->ipv6.ip6_dst_ops));
5489
5490         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
5491                 goto out_ip6_dst_ops;
5492
5493         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
5494         if (!net->ipv6.fib6_null_entry)
5495                 goto out_ip6_dst_entries;
5496         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
5497                sizeof(*net->ipv6.fib6_null_entry));
5498
5499         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
5500                                            sizeof(*net->ipv6.ip6_null_entry),
5501                                            GFP_KERNEL);
5502         if (!net->ipv6.ip6_null_entry)
5503                 goto out_fib6_null_entry;
5504         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5505         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
5506                          ip6_template_metrics, true);
5507
5508 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5509         net->ipv6.fib6_has_custom_rules = false;
5510         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
5511                                                sizeof(*net->ipv6.ip6_prohibit_entry),
5512                                                GFP_KERNEL);
5513         if (!net->ipv6.ip6_prohibit_entry)
5514                 goto out_ip6_null_entry;
5515         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5516         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
5517                          ip6_template_metrics, true);
5518
5519         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
5520                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
5521                                                GFP_KERNEL);
5522         if (!net->ipv6.ip6_blk_hole_entry)
5523                 goto out_ip6_prohibit_entry;
5524         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5525         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
5526                          ip6_template_metrics, true);
5527 #endif
5528
5529         net->ipv6.sysctl.flush_delay = 0;
5530         net->ipv6.sysctl.ip6_rt_max_size = 4096;
5531         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
5532         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
5533         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
5534         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
5535         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
5536         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
5537         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
5538
5539         net->ipv6.ip6_rt_gc_expire = 30*HZ;
5540
5541         ret = 0;
5542 out:
5543         return ret;
5544
5545 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5546 out_ip6_prohibit_entry:
5547         kfree(net->ipv6.ip6_prohibit_entry);
5548 out_ip6_null_entry:
5549         kfree(net->ipv6.ip6_null_entry);
5550 #endif
5551 out_fib6_null_entry:
5552         kfree(net->ipv6.fib6_null_entry);
5553 out_ip6_dst_entries:
5554         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
5555 out_ip6_dst_ops:
5556         goto out;
5557 }
5558
5559 static void __net_exit ip6_route_net_exit(struct net *net)
5560 {
5561         kfree(net->ipv6.fib6_null_entry);
5562         kfree(net->ipv6.ip6_null_entry);
5563 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5564         kfree(net->ipv6.ip6_prohibit_entry);
5565         kfree(net->ipv6.ip6_blk_hole_entry);
5566 #endif
5567         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
5568 }
5569
5570 static int __net_init ip6_route_net_init_late(struct net *net)
5571 {
5572 #ifdef CONFIG_PROC_FS
5573         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
5574                         sizeof(struct ipv6_route_iter));
5575         proc_create_net_single("rt6_stats", 0444, net->proc_net,
5576                         rt6_stats_seq_show, NULL);
5577 #endif
5578         return 0;
5579 }
5580
5581 static void __net_exit ip6_route_net_exit_late(struct net *net)
5582 {
5583 #ifdef CONFIG_PROC_FS
5584         remove_proc_entry("ipv6_route", net->proc_net);
5585         remove_proc_entry("rt6_stats", net->proc_net);
5586 #endif
5587 }
5588
5589 static struct pernet_operations ip6_route_net_ops = {
5590         .init = ip6_route_net_init,
5591         .exit = ip6_route_net_exit,
5592 };
5593
5594 static int __net_init ipv6_inetpeer_init(struct net *net)
5595 {
5596         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
5597
5598         if (!bp)
5599                 return -ENOMEM;
5600         inet_peer_base_init(bp);
5601         net->ipv6.peers = bp;
5602         return 0;
5603 }
5604
5605 static void __net_exit ipv6_inetpeer_exit(struct net *net)
5606 {
5607         struct inet_peer_base *bp = net->ipv6.peers;
5608
5609         net->ipv6.peers = NULL;
5610         inetpeer_invalidate_tree(bp);
5611         kfree(bp);
5612 }
5613
5614 static struct pernet_operations ipv6_inetpeer_ops = {
5615         .init   =       ipv6_inetpeer_init,
5616         .exit   =       ipv6_inetpeer_exit,
5617 };
5618
5619 static struct pernet_operations ip6_route_net_late_ops = {
5620         .init = ip6_route_net_init_late,
5621         .exit = ip6_route_net_exit_late,
5622 };
5623
5624 static struct notifier_block ip6_route_dev_notifier = {
5625         .notifier_call = ip6_route_dev_notify,
5626         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
5627 };
5628
5629 void __init ip6_route_init_special_entries(void)
5630 {
5631         /* Registering of the loopback is done before this portion of code,
5632          * the loopback reference in rt6_info will not be taken, do it
5633          * manually for init_net */
5634         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
5635         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
5636         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5637   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5638         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
5639         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5640         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
5641         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5642   #endif
5643 }
5644
5645 int __init ip6_route_init(void)
5646 {
5647         int ret;
5648         int cpu;
5649
5650         ret = -ENOMEM;
5651         ip6_dst_ops_template.kmem_cachep =
5652                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
5653                                   SLAB_HWCACHE_ALIGN, NULL);
5654         if (!ip6_dst_ops_template.kmem_cachep)
5655                 goto out;
5656
5657         ret = dst_entries_init(&ip6_dst_blackhole_ops);
5658         if (ret)
5659                 goto out_kmem_cache;
5660
5661         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
5662         if (ret)
5663                 goto out_dst_entries;
5664
5665         ret = register_pernet_subsys(&ip6_route_net_ops);
5666         if (ret)
5667                 goto out_register_inetpeer;
5668
5669         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
5670
5671         ret = fib6_init();
5672         if (ret)
5673                 goto out_register_subsys;
5674
5675         ret = xfrm6_init();
5676         if (ret)
5677                 goto out_fib6_init;
5678
5679         ret = fib6_rules_init();
5680         if (ret)
5681                 goto xfrm6_init;
5682
5683         ret = register_pernet_subsys(&ip6_route_net_late_ops);
5684         if (ret)
5685                 goto fib6_rules_init;
5686
5687         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
5688                                    inet6_rtm_newroute, NULL, 0);
5689         if (ret < 0)
5690                 goto out_register_late_subsys;
5691
5692         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
5693                                    inet6_rtm_delroute, NULL, 0);
5694         if (ret < 0)
5695                 goto out_register_late_subsys;
5696
5697         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
5698                                    inet6_rtm_getroute, NULL,
5699                                    RTNL_FLAG_DOIT_UNLOCKED);
5700         if (ret < 0)
5701                 goto out_register_late_subsys;
5702
5703         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
5704         if (ret)
5705                 goto out_register_late_subsys;
5706
5707         for_each_possible_cpu(cpu) {
5708                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
5709
5710                 INIT_LIST_HEAD(&ul->head);
5711                 spin_lock_init(&ul->lock);
5712         }
5713
5714 out:
5715         return ret;
5716
5717 out_register_late_subsys:
5718         rtnl_unregister_all(PF_INET6);
5719         unregister_pernet_subsys(&ip6_route_net_late_ops);
5720 fib6_rules_init:
5721         fib6_rules_cleanup();
5722 xfrm6_init:
5723         xfrm6_fini();
5724 out_fib6_init:
5725         fib6_gc_cleanup();
5726 out_register_subsys:
5727         unregister_pernet_subsys(&ip6_route_net_ops);
5728 out_register_inetpeer:
5729         unregister_pernet_subsys(&ipv6_inetpeer_ops);
5730 out_dst_entries:
5731         dst_entries_destroy(&ip6_dst_blackhole_ops);
5732 out_kmem_cache:
5733         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
5734         goto out;
5735 }
5736
5737 void ip6_route_cleanup(void)
5738 {
5739         unregister_netdevice_notifier(&ip6_route_dev_notifier);
5740         unregister_pernet_subsys(&ip6_route_net_late_ops);
5741         fib6_rules_cleanup();
5742         xfrm6_fini();
5743         fib6_gc_cleanup();
5744         unregister_pernet_subsys(&ipv6_inetpeer_ops);
5745         unregister_pernet_subsys(&ip6_route_net_ops);
5746         dst_entries_destroy(&ip6_dst_blackhole_ops);
5747         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
5748 }