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/nexthop.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(struct fib6_info *rt, int oif, int strict);
 106 static size_t rt6_nlmsg_size(struct fib6_info *rt);
 107 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 108                          struct fib6_info *rt, struct dst_entry *dst,
 109                          struct in6_addr *dest, struct in6_addr *src,
 110                          int iif, int type, u32 portid, u32 seq,
 111                          unsigned int flags);
 112 static struct rt6_info *rt6_find_cached_rt(struct fib6_info *rt,
 113                                            struct in6_addr *daddr,
 114                                            struct in6_addr *saddr);
 115
 116 #ifdef CONFIG_IPV6_ROUTE_INFO
 117 static struct fib6_info *rt6_add_route_info(struct net *net,
 118                                            const struct in6_addr *prefix, int prefixlen,
 119                                            const struct in6_addr *gwaddr,
 120                                            struct net_device *dev,
 121                                            unsigned int pref);
 122 static struct fib6_info *rt6_get_route_info(struct net *net,
 123                                            const struct in6_addr *prefix, int prefixlen,
 124                                            const struct in6_addr *gwaddr,
 125                                            struct net_device *dev);
 126 #endif
 127
 128 struct uncached_list {
 129         spinlock_t              lock;
 130         struct list_head        head;
 131 };
 132
 133 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 134
 135 void rt6_uncached_list_add(struct rt6_info *rt)
 136 {
 137         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 138
 139         rt->rt6i_uncached_list = ul;
 140
 141         spin_lock_bh(&ul->lock);
 142         list_add_tail(&rt->rt6i_uncached, &ul->head);
 143         spin_unlock_bh(&ul->lock);
 144 }
 145
 146 void rt6_uncached_list_del(struct rt6_info *rt)
 147 {
 148         if (!list_empty(&rt->rt6i_uncached)) {
 149                 struct uncached_list *ul = rt->rt6i_uncached_list;
 150                 struct net *net = dev_net(rt->dst.dev);
 151
 152                 spin_lock_bh(&ul->lock);
 153                 list_del(&rt->rt6i_uncached);
 154                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 155                 spin_unlock_bh(&ul->lock);
 156         }
 157 }
 158
 159 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 160 {
 161         struct net_device *loopback_dev = net->loopback_dev;
 162         int cpu;
 163
 164         if (dev == loopback_dev)
 165                 return;
 166
 167         for_each_possible_cpu(cpu) {
 168                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 169                 struct rt6_info *rt;
 170
 171                 spin_lock_bh(&ul->lock);
 172                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 173                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 174                         struct net_device *rt_dev = rt->dst.dev;
 175
 176                         if (rt_idev->dev == dev) {
 177                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 178                                 in6_dev_put(rt_idev);
 179                         }
 180
 181                         if (rt_dev == dev) {
 182                                 rt->dst.dev = loopback_dev;
 183                                 dev_hold(rt->dst.dev);
 184                                 dev_put(rt_dev);
 185                         }
 186                 }
 187                 spin_unlock_bh(&ul->lock);
 188         }
 189 }
 190
 191 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 192                                              struct sk_buff *skb,
 193                                              const void *daddr)
 194 {
 195         if (!ipv6_addr_any(p))
 196                 return (const void *) p;
 197         else if (skb)
 198                 return &ipv6_hdr(skb)->daddr;
 199         return daddr;
 200 }
 201
 202 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 203                                    struct net_device *dev,
 204                                    struct sk_buff *skb,
 205                                    const void *daddr)
 206 {
 207         struct neighbour *n;
 208
 209         daddr = choose_neigh_daddr(gw, skb, daddr);
 210         n = __ipv6_neigh_lookup(dev, daddr);
 211         if (n)
 212                 return n;
 213         return neigh_create(&nd_tbl, daddr, dev);
 214 }
 215
 216 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 217                                               struct sk_buff *skb,
 218                                               const void *daddr)
 219 {
 220         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 221
 222         return ip6_neigh_lookup(&rt->rt6i_gateway, dst->dev, skb, daddr);
 223 }
 224
 225 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 226 {
 227         struct net_device *dev = dst->dev;
 228         struct rt6_info *rt = (struct rt6_info *)dst;
 229
 230         daddr = choose_neigh_daddr(&rt->rt6i_gateway, NULL, daddr);
 231         if (!daddr)
 232                 return;
 233         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 234                 return;
 235         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 236                 return;
 237         __ipv6_confirm_neigh(dev, daddr);
 238 }
 239
 240 static struct dst_ops ip6_dst_ops_template = {
 241         .family                 =       AF_INET6,
 242         .gc                     =       ip6_dst_gc,
 243         .gc_thresh              =       1024,
 244         .check                  =       ip6_dst_check,
 245         .default_advmss         =       ip6_default_advmss,
 246         .mtu                    =       ip6_mtu,
 247         .cow_metrics            =       dst_cow_metrics_generic,
 248         .destroy                =       ip6_dst_destroy,
 249         .ifdown                 =       ip6_dst_ifdown,
 250         .negative_advice        =       ip6_negative_advice,
 251         .link_failure           =       ip6_link_failure,
 252         .update_pmtu            =       ip6_rt_update_pmtu,
 253         .redirect               =       rt6_do_redirect,
 254         .local_out              =       __ip6_local_out,
 255         .neigh_lookup           =       ip6_dst_neigh_lookup,
 256         .confirm_neigh          =       ip6_confirm_neigh,
 257 };
 258
 259 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 260 {
 261         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 262
 263         return mtu ? : dst->dev->mtu;
 264 }
 265
 266 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 267                                          struct sk_buff *skb, u32 mtu)
 268 {
 269 }
 270
 271 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 272                                       struct sk_buff *skb)
 273 {
 274 }
 275
 276 static struct dst_ops ip6_dst_blackhole_ops = {
 277         .family                 =       AF_INET6,
 278         .destroy                =       ip6_dst_destroy,
 279         .check                  =       ip6_dst_check,
 280         .mtu                    =       ip6_blackhole_mtu,
 281         .default_advmss         =       ip6_default_advmss,
 282         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 283         .redirect               =       ip6_rt_blackhole_redirect,
 284         .cow_metrics            =       dst_cow_metrics_generic,
 285         .neigh_lookup           =       ip6_dst_neigh_lookup,
 286 };
 287
 288 static const u32 ip6_template_metrics[RTAX_MAX] = {
 289         [RTAX_HOPLIMIT - 1] = 0,
 290 };
 291
 292 static const struct fib6_info fib6_null_entry_template = {
 293         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 294         .fib6_protocol  = RTPROT_KERNEL,
 295         .fib6_metric    = ~(u32)0,
 296         .fib6_ref       = ATOMIC_INIT(1),
 297         .fib6_type      = RTN_UNREACHABLE,
 298         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 299 };
 300
 301 static const struct rt6_info ip6_null_entry_template = {
 302         .dst = {
 303                 .__refcnt       = ATOMIC_INIT(1),
 304                 .__use          = 1,
 305                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 306                 .error          = -ENETUNREACH,
 307                 .input          = ip6_pkt_discard,
 308                 .output         = ip6_pkt_discard_out,
 309         },
 310         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 311 };
 312
 313 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 314
 315 static const struct rt6_info ip6_prohibit_entry_template = {
 316         .dst = {
 317                 .__refcnt       = ATOMIC_INIT(1),
 318                 .__use          = 1,
 319                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 320                 .error          = -EACCES,
 321                 .input          = ip6_pkt_prohibit,
 322                 .output         = ip6_pkt_prohibit_out,
 323         },
 324         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 325 };
 326
 327 static const struct rt6_info ip6_blk_hole_entry_template = {
 328         .dst = {
 329                 .__refcnt       = ATOMIC_INIT(1),
 330                 .__use          = 1,
 331                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 332                 .error          = -EINVAL,
 333                 .input          = dst_discard,
 334                 .output         = dst_discard_out,
 335         },
 336         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 337 };
 338
 339 #endif
 340
 341 static void rt6_info_init(struct rt6_info *rt)
 342 {
 343         struct dst_entry *dst = &rt->dst;
 344
 345         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 346         INIT_LIST_HEAD(&rt->rt6i_uncached);
 347 }
 348
 349 /* allocate dst with ip6_dst_ops */
 350 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 351                                int flags)
 352 {
 353         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 354                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 355
 356         if (rt) {
 357                 rt6_info_init(rt);
 358                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 359         }
 360
 361         return rt;
 362 }
 363 EXPORT_SYMBOL(ip6_dst_alloc);
 364
 365 static void ip6_dst_destroy(struct dst_entry *dst)
 366 {
 367         struct rt6_info *rt = (struct rt6_info *)dst;
 368         struct fib6_info *from;
 369         struct inet6_dev *idev;
 370
 371         ip_dst_metrics_put(dst);
 372         rt6_uncached_list_del(rt);
 373
 374         idev = rt->rt6i_idev;
 375         if (idev) {
 376                 rt->rt6i_idev = NULL;
 377                 in6_dev_put(idev);
 378         }
 379
 380         rcu_read_lock();
 381         from = rcu_dereference(rt->from);
 382         rcu_assign_pointer(rt->from, NULL);
 383         fib6_info_release(from);
 384         rcu_read_unlock();
 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 struct fib6_info *fib6_multipath_select(const struct net *net,
 429                                         struct fib6_info *match,
 430                                         struct flowi6 *fl6, int oif,
 431                                         const struct sk_buff *skb,
 432                                         int strict)
 433 {
 434         struct fib6_info *sibling, *next_sibling;
 435
 436         /* We might have already computed the hash for ICMPv6 errors. In such
 437          * case it will always be non-zero. Otherwise now is the time to do it.
 438          */
 439         if (!fl6->mp_hash)
 440                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 441
 442         if (fl6->mp_hash <= atomic_read(&match->fib6_nh.nh_upper_bound))
 443                 return match;
 444
 445         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 446                                  fib6_siblings) {
 447                 int nh_upper_bound;
 448
 449                 nh_upper_bound = atomic_read(&sibling->fib6_nh.nh_upper_bound);
 450                 if (fl6->mp_hash > nh_upper_bound)
 451                         continue;
 452                 if (rt6_score_route(sibling, oif, strict) < 0)
 453                         break;
 454                 match = sibling;
 455                 break;
 456         }
 457
 458         return match;
 459 }
 460
 461 /*
 462  *      Route lookup. rcu_read_lock() should be held.
 463  */
 464
 465 static inline struct fib6_info *rt6_device_match(struct net *net,
 466                                                  struct fib6_info *rt,
 467                                                     const struct in6_addr *saddr,
 468                                                     int oif,
 469                                                     int flags)
 470 {
 471         struct fib6_info *sprt;
 472
 473         if (!oif && ipv6_addr_any(saddr) &&
 474             !(rt->fib6_nh.nh_flags & RTNH_F_DEAD))
 475                 return rt;
 476
 477         for (sprt = rt; sprt; sprt = rcu_dereference(sprt->fib6_next)) {
 478                 const struct net_device *dev = sprt->fib6_nh.nh_dev;
 479
 480                 if (sprt->fib6_nh.nh_flags & RTNH_F_DEAD)
 481                         continue;
 482
 483                 if (oif) {
 484                         if (dev->ifindex == oif)
 485                                 return sprt;
 486                 } else {
 487                         if (ipv6_chk_addr(net, saddr, dev,
 488                                           flags & RT6_LOOKUP_F_IFACE))
 489                                 return sprt;
 490                 }
 491         }
 492
 493         if (oif && flags & RT6_LOOKUP_F_IFACE)
 494                 return net->ipv6.fib6_null_entry;
 495
 496         return rt->fib6_nh.nh_flags & RTNH_F_DEAD ? net->ipv6.fib6_null_entry : rt;
 497 }
 498
 499 #ifdef CONFIG_IPV6_ROUTER_PREF
 500 struct __rt6_probe_work {
 501         struct work_struct work;
 502         struct in6_addr target;
 503         struct net_device *dev;
 504 };
 505
 506 static void rt6_probe_deferred(struct work_struct *w)
 507 {
 508         struct in6_addr mcaddr;
 509         struct __rt6_probe_work *work =
 510                 container_of(w, struct __rt6_probe_work, work);
 511
 512         addrconf_addr_solict_mult(&work->target, &mcaddr);
 513         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 514         dev_put(work->dev);
 515         kfree(work);
 516 }
 517
 518 static void rt6_probe(struct fib6_info *rt)
 519 {
 520         struct __rt6_probe_work *work = NULL;
 521         const struct in6_addr *nh_gw;
 522         struct neighbour *neigh;
 523         struct net_device *dev;
 524         struct inet6_dev *idev;
 525
 526         /*
 527          * Okay, this does not seem to be appropriate
 528          * for now, however, we need to check if it
 529          * is really so; aka Router Reachability Probing.
 530          *
 531          * Router Reachability Probe MUST be rate-limited
 532          * to no more than one per minute.
 533          */
 534         if (!rt || !(rt->fib6_flags & RTF_GATEWAY))
 535                 return;
 536
 537         nh_gw = &rt->fib6_nh.nh_gw;
 538         dev = rt->fib6_nh.nh_dev;
 539         rcu_read_lock_bh();
 540         idev = __in6_dev_get(dev);
 541         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 542         if (neigh) {
 543                 if (neigh->nud_state & NUD_VALID)
 544                         goto out;
 545
 546                 write_lock(&neigh->lock);
 547                 if (!(neigh->nud_state & NUD_VALID) &&
 548                     time_after(jiffies,
 549                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 550                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 551                         if (work)
 552                                 __neigh_set_probe_once(neigh);
 553                 }
 554                 write_unlock(&neigh->lock);
 555         } else if (time_after(jiffies, rt->last_probe +
 556                                        idev->cnf.rtr_probe_interval)) {
 557                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 558         }
 559
 560         if (work) {
 561                 rt->last_probe = jiffies;
 562                 INIT_WORK(&work->work, rt6_probe_deferred);
 563                 work->target = *nh_gw;
 564                 dev_hold(dev);
 565                 work->dev = dev;
 566                 schedule_work(&work->work);
 567         }
 568
 569 out:
 570         rcu_read_unlock_bh();
 571 }
 572 #else
 573 static inline void rt6_probe(struct fib6_info *rt)
 574 {
 575 }
 576 #endif
 577
 578 /*
 579  * Default Router Selection (RFC 2461 6.3.6)
 580  */
 581 static inline int rt6_check_dev(struct fib6_info *rt, int oif)
 582 {
 583         const struct net_device *dev = rt->fib6_nh.nh_dev;
 584
 585         if (!oif || dev->ifindex == oif)
 586                 return 2;
 587         return 0;
 588 }
 589
 590 static inline enum rt6_nud_state rt6_check_neigh(struct fib6_info *rt)
 591 {
 592         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 593         struct neighbour *neigh;
 594
 595         if (rt->fib6_flags & RTF_NONEXTHOP ||
 596             !(rt->fib6_flags & RTF_GATEWAY))
 597                 return RT6_NUD_SUCCEED;
 598
 599         rcu_read_lock_bh();
 600         neigh = __ipv6_neigh_lookup_noref(rt->fib6_nh.nh_dev,
 601                                           &rt->fib6_nh.nh_gw);
 602         if (neigh) {
 603                 read_lock(&neigh->lock);
 604                 if (neigh->nud_state & NUD_VALID)
 605                         ret = RT6_NUD_SUCCEED;
 606 #ifdef CONFIG_IPV6_ROUTER_PREF
 607                 else if (!(neigh->nud_state & NUD_FAILED))
 608                         ret = RT6_NUD_SUCCEED;
 609                 else
 610                         ret = RT6_NUD_FAIL_PROBE;
 611 #endif
 612                 read_unlock(&neigh->lock);
 613         } else {
 614                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 615                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 616         }
 617         rcu_read_unlock_bh();
 618
 619         return ret;
 620 }
 621
 622 static int rt6_score_route(struct fib6_info *rt, int oif, int strict)
 623 {
 624         int m;
 625
 626         m = rt6_check_dev(rt, oif);
 627         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 628                 return RT6_NUD_FAIL_HARD;
 629 #ifdef CONFIG_IPV6_ROUTER_PREF
 630         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(rt->fib6_flags)) << 2;
 631 #endif
 632         if (strict & RT6_LOOKUP_F_REACHABLE) {
 633                 int n = rt6_check_neigh(rt);
 634                 if (n < 0)
 635                         return n;
 636         }
 637         return m;
 638 }
 639
 640 /* called with rc_read_lock held */
 641 static inline bool fib6_ignore_linkdown(const struct fib6_info *f6i)
 642 {
 643         const struct net_device *dev = fib6_info_nh_dev(f6i);
 644         bool rc = false;
 645
 646         if (dev) {
 647                 const struct inet6_dev *idev = __in6_dev_get(dev);
 648
 649                 rc = !!idev->cnf.ignore_routes_with_linkdown;
 650         }
 651
 652         return rc;
 653 }
 654
 655 static struct fib6_info *find_match(struct fib6_info *rt, int oif, int strict,
 656                                    int *mpri, struct fib6_info *match,
 657                                    bool *do_rr)
 658 {
 659         int m;
 660         bool match_do_rr = false;
 661
 662         if (rt->fib6_nh.nh_flags & RTNH_F_DEAD)
 663                 goto out;
 664
 665         if (fib6_ignore_linkdown(rt) &&
 666             rt->fib6_nh.nh_flags & RTNH_F_LINKDOWN &&
 667             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 668                 goto out;
 669
 670         if (fib6_check_expired(rt))
 671                 goto out;
 672
 673         m = rt6_score_route(rt, oif, strict);
 674         if (m == RT6_NUD_FAIL_DO_RR) {
 675                 match_do_rr = true;
 676                 m = 0; /* lowest valid score */
 677         } else if (m == RT6_NUD_FAIL_HARD) {
 678                 goto out;
 679         }
 680
 681         if (strict & RT6_LOOKUP_F_REACHABLE)
 682                 rt6_probe(rt);
 683
 684         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 685         if (m > *mpri) {
 686                 *do_rr = match_do_rr;
 687                 *mpri = m;
 688                 match = rt;
 689         }
 690 out:
 691         return match;
 692 }
 693
 694 static struct fib6_info *find_rr_leaf(struct fib6_node *fn,
 695                                      struct fib6_info *leaf,
 696                                      struct fib6_info *rr_head,
 697                                      u32 metric, int oif, int strict,
 698                                      bool *do_rr)
 699 {
 700         struct fib6_info *rt, *match, *cont;
 701         int mpri = -1;
 702
 703         match = NULL;
 704         cont = NULL;
 705         for (rt = rr_head; rt; rt = rcu_dereference(rt->fib6_next)) {
 706                 if (rt->fib6_metric != metric) {
 707                         cont = rt;
 708                         break;
 709                 }
 710
 711                 match = find_match(rt, oif, strict, &mpri, match, do_rr);
 712         }
 713
 714         for (rt = leaf; rt && rt != rr_head;
 715              rt = rcu_dereference(rt->fib6_next)) {
 716                 if (rt->fib6_metric != metric) {
 717                         cont = rt;
 718                         break;
 719                 }
 720
 721                 match = find_match(rt, oif, strict, &mpri, match, do_rr);
 722         }
 723
 724         if (match || !cont)
 725                 return match;
 726
 727         for (rt = cont; rt; rt = rcu_dereference(rt->fib6_next))
 728                 match = find_match(rt, oif, strict, &mpri, match, do_rr);
 729
 730         return match;
 731 }
 732
 733 static struct fib6_info *rt6_select(struct net *net, struct fib6_node *fn,
 734                                    int oif, int strict)
 735 {
 736         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 737         struct fib6_info *match, *rt0;
 738         bool do_rr = false;
 739         int key_plen;
 740
 741         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 742                 return net->ipv6.fib6_null_entry;
 743
 744         rt0 = rcu_dereference(fn->rr_ptr);
 745         if (!rt0)
 746                 rt0 = leaf;
 747
 748         /* Double check to make sure fn is not an intermediate node
 749          * and fn->leaf does not points to its child's leaf
 750          * (This might happen if all routes under fn are deleted from
 751          * the tree and fib6_repair_tree() is called on the node.)
 752          */
 753         key_plen = rt0->fib6_dst.plen;
 754 #ifdef CONFIG_IPV6_SUBTREES
 755         if (rt0->fib6_src.plen)
 756                 key_plen = rt0->fib6_src.plen;
 757 #endif
 758         if (fn->fn_bit != key_plen)
 759                 return net->ipv6.fib6_null_entry;
 760
 761         match = find_rr_leaf(fn, leaf, rt0, rt0->fib6_metric, oif, strict,
 762                              &do_rr);
 763
 764         if (do_rr) {
 765                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 766
 767                 /* no entries matched; do round-robin */
 768                 if (!next || next->fib6_metric != rt0->fib6_metric)
 769                         next = leaf;
 770
 771                 if (next != rt0) {
 772                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 773                         /* make sure next is not being deleted from the tree */
 774                         if (next->fib6_node)
 775                                 rcu_assign_pointer(fn->rr_ptr, next);
 776                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 777                 }
 778         }
 779
 780         return match ? match : net->ipv6.fib6_null_entry;
 781 }
 782
 783 static bool rt6_is_gw_or_nonexthop(const struct fib6_info *rt)
 784 {
 785         return (rt->fib6_flags & (RTF_NONEXTHOP | RTF_GATEWAY));
 786 }
 787
 788 #ifdef CONFIG_IPV6_ROUTE_INFO
 789 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 790                   const struct in6_addr *gwaddr)
 791 {
 792         struct net *net = dev_net(dev);
 793         struct route_info *rinfo = (struct route_info *) opt;
 794         struct in6_addr prefix_buf, *prefix;
 795         unsigned int pref;
 796         unsigned long lifetime;
 797         struct fib6_info *rt;
 798
 799         if (len < sizeof(struct route_info)) {
 800                 return -EINVAL;
 801         }
 802
 803         /* Sanity check for prefix_len and length */
 804         if (rinfo->length > 3) {
 805                 return -EINVAL;
 806         } else if (rinfo->prefix_len > 128) {
 807                 return -EINVAL;
 808         } else if (rinfo->prefix_len > 64) {
 809                 if (rinfo->length < 2) {
 810                         return -EINVAL;
 811                 }
 812         } else if (rinfo->prefix_len > 0) {
 813                 if (rinfo->length < 1) {
 814                         return -EINVAL;
 815                 }
 816         }
 817
 818         pref = rinfo->route_pref;
 819         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 820                 return -EINVAL;
 821
 822         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 823
 824         if (rinfo->length == 3)
 825                 prefix = (struct in6_addr *)rinfo->prefix;
 826         else {
 827                 /* this function is safe */
 828                 ipv6_addr_prefix(&prefix_buf,
 829                                  (struct in6_addr *)rinfo->prefix,
 830                                  rinfo->prefix_len);
 831                 prefix = &prefix_buf;
 832         }
 833
 834         if (rinfo->prefix_len == 0)
 835                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 836         else
 837                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 838                                         gwaddr, dev);
 839
 840         if (rt && !lifetime) {
 841                 ip6_del_rt(net, rt);
 842                 rt = NULL;
 843         }
 844
 845         if (!rt && lifetime)
 846                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 847                                         dev, pref);
 848         else if (rt)
 849                 rt->fib6_flags = RTF_ROUTEINFO |
 850                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 851
 852         if (rt) {
 853                 if (!addrconf_finite_timeout(lifetime))
 854                         fib6_clean_expires(rt);
 855                 else
 856                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 857
 858                 fib6_info_release(rt);
 859         }
 860         return 0;
 861 }
 862 #endif
 863
 864 /*
 865  *      Misc support functions
 866  */
 867
 868 /* called with rcu_lock held */
 869 static struct net_device *ip6_rt_get_dev_rcu(struct fib6_info *rt)
 870 {
 871         struct net_device *dev = rt->fib6_nh.nh_dev;
 872
 873         if (rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
 874                 /* for copies of local routes, dst->dev needs to be the
 875                  * device if it is a master device, the master device if
 876                  * device is enslaved, and the loopback as the default
 877                  */
 878                 if (netif_is_l3_slave(dev) &&
 879                     !rt6_need_strict(&rt->fib6_dst.addr))
 880                         dev = l3mdev_master_dev_rcu(dev);
 881                 else if (!netif_is_l3_master(dev))
 882                         dev = dev_net(dev)->loopback_dev;
 883                 /* last case is netif_is_l3_master(dev) is true in which
 884                  * case we want dev returned to be dev
 885                  */
 886         }
 887
 888         return dev;
 889 }
 890
 891 static const int fib6_prop[RTN_MAX + 1] = {
 892         [RTN_UNSPEC]    = 0,
 893         [RTN_UNICAST]   = 0,
 894         [RTN_LOCAL]     = 0,
 895         [RTN_BROADCAST] = 0,
 896         [RTN_ANYCAST]   = 0,
 897         [RTN_MULTICAST] = 0,
 898         [RTN_BLACKHOLE] = -EINVAL,
 899         [RTN_UNREACHABLE] = -EHOSTUNREACH,
 900         [RTN_PROHIBIT]  = -EACCES,
 901         [RTN_THROW]     = -EAGAIN,
 902         [RTN_NAT]       = -EINVAL,
 903         [RTN_XRESOLVE]  = -EINVAL,
 904 };
 905
 906 static int ip6_rt_type_to_error(u8 fib6_type)
 907 {
 908         return fib6_prop[fib6_type];
 909 }
 910
 911 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
 912 {
 913         unsigned short flags = 0;
 914
 915         if (rt->dst_nocount)
 916                 flags |= DST_NOCOUNT;
 917         if (rt->dst_nopolicy)
 918                 flags |= DST_NOPOLICY;
 919         if (rt->dst_host)
 920                 flags |= DST_HOST;
 921
 922         return flags;
 923 }
 924
 925 static void ip6_rt_init_dst_reject(struct rt6_info *rt, struct fib6_info *ort)
 926 {
 927         rt->dst.error = ip6_rt_type_to_error(ort->fib6_type);
 928
 929         switch (ort->fib6_type) {
 930         case RTN_BLACKHOLE:
 931                 rt->dst.output = dst_discard_out;
 932                 rt->dst.input = dst_discard;
 933                 break;
 934         case RTN_PROHIBIT:
 935                 rt->dst.output = ip6_pkt_prohibit_out;
 936                 rt->dst.input = ip6_pkt_prohibit;
 937                 break;
 938         case RTN_THROW:
 939         case RTN_UNREACHABLE:
 940         default:
 941                 rt->dst.output = ip6_pkt_discard_out;
 942                 rt->dst.input = ip6_pkt_discard;
 943                 break;
 944         }
 945 }
 946
 947 static void ip6_rt_init_dst(struct rt6_info *rt, struct fib6_info *ort)
 948 {
 949         if (ort->fib6_flags & RTF_REJECT) {
 950                 ip6_rt_init_dst_reject(rt, ort);
 951                 return;
 952         }
 953
 954         rt->dst.error = 0;
 955         rt->dst.output = ip6_output;
 956
 957         if (ort->fib6_type == RTN_LOCAL || ort->fib6_type == RTN_ANYCAST) {
 958                 rt->dst.input = ip6_input;
 959         } else if (ipv6_addr_type(&ort->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
 960                 rt->dst.input = ip6_mc_input;
 961         } else {
 962                 rt->dst.input = ip6_forward;
 963         }
 964
 965         if (ort->fib6_nh.nh_lwtstate) {
 966                 rt->dst.lwtstate = lwtstate_get(ort->fib6_nh.nh_lwtstate);
 967                 lwtunnel_set_redirect(&rt->dst);
 968         }
 969
 970         rt->dst.lastuse = jiffies;
 971 }
 972
 973 /* Caller must already hold reference to @from */
 974 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
 975 {
 976         rt->rt6i_flags &= ~RTF_EXPIRES;
 977         rcu_assign_pointer(rt->from, from);
 978         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
 979 }
 980
 981 /* Caller must already hold reference to @ort */
 982 static void ip6_rt_copy_init(struct rt6_info *rt, struct fib6_info *ort)
 983 {
 984         struct net_device *dev = fib6_info_nh_dev(ort);
 985
 986         ip6_rt_init_dst(rt, ort);
 987
 988         rt->rt6i_dst = ort->fib6_dst;
 989         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
 990         rt->rt6i_gateway = ort->fib6_nh.nh_gw;
 991         rt->rt6i_flags = ort->fib6_flags;
 992         rt6_set_from(rt, ort);
 993 #ifdef CONFIG_IPV6_SUBTREES
 994         rt->rt6i_src = ort->fib6_src;
 995 #endif
 996 }
 997
 998 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
 999                                         struct in6_addr *saddr)
1000 {
1001         struct fib6_node *pn, *sn;
1002         while (1) {
1003                 if (fn->fn_flags & RTN_TL_ROOT)
1004                         return NULL;
1005                 pn = rcu_dereference(fn->parent);
1006                 sn = FIB6_SUBTREE(pn);
1007                 if (sn && sn != fn)
1008                         fn = fib6_node_lookup(sn, NULL, saddr);
1009                 else
1010                         fn = pn;
1011                 if (fn->fn_flags & RTN_RTINFO)
1012                         return fn;
1013         }
1014 }
1015
1016 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt,
1017                           bool null_fallback)
1018 {
1019         struct rt6_info *rt = *prt;
1020
1021         if (dst_hold_safe(&rt->dst))
1022                 return true;
1023         if (null_fallback) {
1024                 rt = net->ipv6.ip6_null_entry;
1025                 dst_hold(&rt->dst);
1026         } else {
1027                 rt = NULL;
1028         }
1029         *prt = rt;
1030         return false;
1031 }
1032
1033 /* called with rcu_lock held */
1034 static struct rt6_info *ip6_create_rt_rcu(struct fib6_info *rt)
1035 {
1036         unsigned short flags = fib6_info_dst_flags(rt);
1037         struct net_device *dev = rt->fib6_nh.nh_dev;
1038         struct rt6_info *nrt;
1039
1040         if (!fib6_info_hold_safe(rt))
1041                 return NULL;
1042
1043         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1044         if (nrt)
1045                 ip6_rt_copy_init(nrt, rt);
1046         else
1047                 fib6_info_release(rt);
1048
1049         return nrt;
1050 }
1051
1052 static struct rt6_info *ip6_pol_route_lookup(struct net *net,
1053                                              struct fib6_table *table,
1054                                              struct flowi6 *fl6,
1055                                              const struct sk_buff *skb,
1056                                              int flags)
1057 {
1058         struct fib6_info *f6i;
1059         struct fib6_node *fn;
1060         struct rt6_info *rt;
1061
1062         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1063                 flags &= ~RT6_LOOKUP_F_IFACE;
1064
1065         rcu_read_lock();
1066         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1067 restart:
1068         f6i = rcu_dereference(fn->leaf);
1069         if (!f6i) {
1070                 f6i = net->ipv6.fib6_null_entry;
1071         } else {
1072                 f6i = rt6_device_match(net, f6i, &fl6->saddr,
1073                                       fl6->flowi6_oif, flags);
1074                 if (f6i->fib6_nsiblings && fl6->flowi6_oif == 0)
1075                         f6i = fib6_multipath_select(net, f6i, fl6,
1076                                                     fl6->flowi6_oif, skb,
1077                                                     flags);
1078         }
1079         if (f6i == net->ipv6.fib6_null_entry) {
1080                 fn = fib6_backtrack(fn, &fl6->saddr);
1081                 if (fn)
1082                         goto restart;
1083         }
1084
1085         trace_fib6_table_lookup(net, f6i, table, fl6);
1086
1087         /* Search through exception table */
1088         rt = rt6_find_cached_rt(f6i, &fl6->daddr, &fl6->saddr);
1089         if (rt) {
1090                 if (ip6_hold_safe(net, &rt, true))
1091                         dst_use_noref(&rt->dst, jiffies);
1092         } else if (f6i == net->ipv6.fib6_null_entry) {
1093                 rt = net->ipv6.ip6_null_entry;
1094                 dst_hold(&rt->dst);
1095         } else {
1096                 rt = ip6_create_rt_rcu(f6i);
1097                 if (!rt) {
1098                         rt = net->ipv6.ip6_null_entry;
1099                         dst_hold(&rt->dst);
1100                 }
1101         }
1102
1103         rcu_read_unlock();
1104
1105         return rt;
1106 }
1107
1108 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1109                                    const struct sk_buff *skb, int flags)
1110 {
1111         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1112 }
1113 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1114
1115 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1116                             const struct in6_addr *saddr, int oif,
1117                             const struct sk_buff *skb, int strict)
1118 {
1119         struct flowi6 fl6 = {
1120                 .flowi6_oif = oif,
1121                 .daddr = *daddr,
1122         };
1123         struct dst_entry *dst;
1124         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1125
1126         if (saddr) {
1127                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1128                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1129         }
1130
1131         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1132         if (dst->error == 0)
1133                 return (struct rt6_info *) dst;
1134
1135         dst_release(dst);
1136
1137         return NULL;
1138 }
1139 EXPORT_SYMBOL(rt6_lookup);
1140
1141 /* ip6_ins_rt is called with FREE table->tb6_lock.
1142  * It takes new route entry, the addition fails by any reason the
1143  * route is released.
1144  * Caller must hold dst before calling it.
1145  */
1146
1147 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1148                         struct netlink_ext_ack *extack)
1149 {
1150         int err;
1151         struct fib6_table *table;
1152
1153         table = rt->fib6_table;
1154         spin_lock_bh(&table->tb6_lock);
1155         err = fib6_add(&table->tb6_root, rt, info, extack);
1156         spin_unlock_bh(&table->tb6_lock);
1157
1158         return err;
1159 }
1160
1161 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1162 {
1163         struct nl_info info = { .nl_net = net, };
1164
1165         return __ip6_ins_rt(rt, &info, NULL);
1166 }
1167
1168 static struct rt6_info *ip6_rt_cache_alloc(struct fib6_info *ort,
1169                                            const struct in6_addr *daddr,
1170                                            const struct in6_addr *saddr)
1171 {
1172         struct net_device *dev;
1173         struct rt6_info *rt;
1174
1175         /*
1176          *      Clone the route.
1177          */
1178
1179         if (!fib6_info_hold_safe(ort))
1180                 return NULL;
1181
1182         dev = ip6_rt_get_dev_rcu(ort);
1183         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1184         if (!rt) {
1185                 fib6_info_release(ort);
1186                 return NULL;
1187         }
1188
1189         ip6_rt_copy_init(rt, ort);
1190         rt->rt6i_flags |= RTF_CACHE;
1191         rt->dst.flags |= DST_HOST;
1192         rt->rt6i_dst.addr = *daddr;
1193         rt->rt6i_dst.plen = 128;
1194
1195         if (!rt6_is_gw_or_nonexthop(ort)) {
1196                 if (ort->fib6_dst.plen != 128 &&
1197                     ipv6_addr_equal(&ort->fib6_dst.addr, daddr))
1198                         rt->rt6i_flags |= RTF_ANYCAST;
1199 #ifdef CONFIG_IPV6_SUBTREES
1200                 if (rt->rt6i_src.plen && saddr) {
1201                         rt->rt6i_src.addr = *saddr;
1202                         rt->rt6i_src.plen = 128;
1203                 }
1204 #endif
1205         }
1206
1207         return rt;
1208 }
1209
1210 static struct rt6_info *ip6_rt_pcpu_alloc(struct fib6_info *rt)
1211 {
1212         unsigned short flags = fib6_info_dst_flags(rt);
1213         struct net_device *dev;
1214         struct rt6_info *pcpu_rt;
1215
1216         if (!fib6_info_hold_safe(rt))
1217                 return NULL;
1218
1219         rcu_read_lock();
1220         dev = ip6_rt_get_dev_rcu(rt);
1221         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags);
1222         rcu_read_unlock();
1223         if (!pcpu_rt) {
1224                 fib6_info_release(rt);
1225                 return NULL;
1226         }
1227         ip6_rt_copy_init(pcpu_rt, rt);
1228         pcpu_rt->rt6i_flags |= RTF_PCPU;
1229         return pcpu_rt;
1230 }
1231
1232 /* It should be called with rcu_read_lock() acquired */
1233 static struct rt6_info *rt6_get_pcpu_route(struct fib6_info *rt)
1234 {
1235         struct rt6_info *pcpu_rt, **p;
1236
1237         p = this_cpu_ptr(rt->rt6i_pcpu);
1238         pcpu_rt = *p;
1239
1240         if (pcpu_rt)
1241                 ip6_hold_safe(NULL, &pcpu_rt, false);
1242
1243         return pcpu_rt;
1244 }
1245
1246 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1247                                             struct fib6_info *rt)
1248 {
1249         struct rt6_info *pcpu_rt, *prev, **p;
1250
1251         pcpu_rt = ip6_rt_pcpu_alloc(rt);
1252         if (!pcpu_rt) {
1253                 dst_hold(&net->ipv6.ip6_null_entry->dst);
1254                 return net->ipv6.ip6_null_entry;
1255         }
1256
1257         dst_hold(&pcpu_rt->dst);
1258         p = this_cpu_ptr(rt->rt6i_pcpu);
1259         prev = cmpxchg(p, NULL, pcpu_rt);
1260         BUG_ON(prev);
1261
1262         return pcpu_rt;
1263 }
1264
1265 /* exception hash table implementation
1266  */
1267 static DEFINE_SPINLOCK(rt6_exception_lock);
1268
1269 /* Remove rt6_ex from hash table and free the memory
1270  * Caller must hold rt6_exception_lock
1271  */
1272 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1273                                  struct rt6_exception *rt6_ex)
1274 {
1275         struct net *net;
1276
1277         if (!bucket || !rt6_ex)
1278                 return;
1279
1280         net = dev_net(rt6_ex->rt6i->dst.dev);
1281         hlist_del_rcu(&rt6_ex->hlist);
1282         dst_release(&rt6_ex->rt6i->dst);
1283         kfree_rcu(rt6_ex, rcu);
1284         WARN_ON_ONCE(!bucket->depth);
1285         bucket->depth--;
1286         net->ipv6.rt6_stats->fib_rt_cache--;
1287 }
1288
1289 /* Remove oldest rt6_ex in bucket and free the memory
1290  * Caller must hold rt6_exception_lock
1291  */
1292 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1293 {
1294         struct rt6_exception *rt6_ex, *oldest = NULL;
1295
1296         if (!bucket)
1297                 return;
1298
1299         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1300                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1301                         oldest = rt6_ex;
1302         }
1303         rt6_remove_exception(bucket, oldest);
1304 }
1305
1306 static u32 rt6_exception_hash(const struct in6_addr *dst,
1307                               const struct in6_addr *src)
1308 {
1309         static u32 seed __read_mostly;
1310         u32 val;
1311
1312         net_get_random_once(&seed, sizeof(seed));
1313         val = jhash(dst, sizeof(*dst), seed);
1314
1315 #ifdef CONFIG_IPV6_SUBTREES
1316         if (src)
1317                 val = jhash(src, sizeof(*src), val);
1318 #endif
1319         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1320 }
1321
1322 /* Helper function to find the cached rt in the hash table
1323  * and update bucket pointer to point to the bucket for this
1324  * (daddr, saddr) pair
1325  * Caller must hold rt6_exception_lock
1326  */
1327 static struct rt6_exception *
1328 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1329                               const struct in6_addr *daddr,
1330                               const struct in6_addr *saddr)
1331 {
1332         struct rt6_exception *rt6_ex;
1333         u32 hval;
1334
1335         if (!(*bucket) || !daddr)
1336                 return NULL;
1337
1338         hval = rt6_exception_hash(daddr, saddr);
1339         *bucket += hval;
1340
1341         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1342                 struct rt6_info *rt6 = rt6_ex->rt6i;
1343                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1344
1345 #ifdef CONFIG_IPV6_SUBTREES
1346                 if (matched && saddr)
1347                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1348 #endif
1349                 if (matched)
1350                         return rt6_ex;
1351         }
1352         return NULL;
1353 }
1354
1355 /* Helper function to find the cached rt in the hash table
1356  * and update bucket pointer to point to the bucket for this
1357  * (daddr, saddr) pair
1358  * Caller must hold rcu_read_lock()
1359  */
1360 static struct rt6_exception *
1361 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1362                          const struct in6_addr *daddr,
1363                          const struct in6_addr *saddr)
1364 {
1365         struct rt6_exception *rt6_ex;
1366         u32 hval;
1367
1368         WARN_ON_ONCE(!rcu_read_lock_held());
1369
1370         if (!(*bucket) || !daddr)
1371                 return NULL;
1372
1373         hval = rt6_exception_hash(daddr, saddr);
1374         *bucket += hval;
1375
1376         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1377                 struct rt6_info *rt6 = rt6_ex->rt6i;
1378                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1379
1380 #ifdef CONFIG_IPV6_SUBTREES
1381                 if (matched && saddr)
1382                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1383 #endif
1384                 if (matched)
1385                         return rt6_ex;
1386         }
1387         return NULL;
1388 }
1389
1390 static unsigned int fib6_mtu(const struct fib6_info *rt)
1391 {
1392         unsigned int mtu;
1393
1394         if (rt->fib6_pmtu) {
1395                 mtu = rt->fib6_pmtu;
1396         } else {
1397                 struct net_device *dev = fib6_info_nh_dev(rt);
1398                 struct inet6_dev *idev;
1399
1400                 rcu_read_lock();
1401                 idev = __in6_dev_get(dev);
1402                 mtu = idev->cnf.mtu6;
1403                 rcu_read_unlock();
1404         }
1405
1406         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1407
1408         return mtu - lwtunnel_headroom(rt->fib6_nh.nh_lwtstate, mtu);
1409 }
1410
1411 static int rt6_insert_exception(struct rt6_info *nrt,
1412                                 struct fib6_info *ort)
1413 {
1414         struct net *net = dev_net(nrt->dst.dev);
1415         struct rt6_exception_bucket *bucket;
1416         struct in6_addr *src_key = NULL;
1417         struct rt6_exception *rt6_ex;
1418         int err = 0;
1419
1420         spin_lock_bh(&rt6_exception_lock);
1421
1422         if (ort->exception_bucket_flushed) {
1423                 err = -EINVAL;
1424                 goto out;
1425         }
1426
1427         bucket = rcu_dereference_protected(ort->rt6i_exception_bucket,
1428                                         lockdep_is_held(&rt6_exception_lock));
1429         if (!bucket) {
1430                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1431                                  GFP_ATOMIC);
1432                 if (!bucket) {
1433                         err = -ENOMEM;
1434                         goto out;
1435                 }
1436                 rcu_assign_pointer(ort->rt6i_exception_bucket, bucket);
1437         }
1438
1439 #ifdef CONFIG_IPV6_SUBTREES
1440         /* rt6i_src.plen != 0 indicates ort is in subtree
1441          * and exception table is indexed by a hash of
1442          * both rt6i_dst and rt6i_src.
1443          * Otherwise, the exception table is indexed by
1444          * a hash of only rt6i_dst.
1445          */
1446         if (ort->fib6_src.plen)
1447                 src_key = &nrt->rt6i_src.addr;
1448 #endif
1449         /* rt6_mtu_change() might lower mtu on ort.
1450          * Only insert this exception route if its mtu
1451          * is less than ort's mtu value.
1452          */
1453         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(ort)) {
1454                 err = -EINVAL;
1455                 goto out;
1456         }
1457
1458         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1459                                                src_key);
1460         if (rt6_ex)
1461                 rt6_remove_exception(bucket, rt6_ex);
1462
1463         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1464         if (!rt6_ex) {
1465                 err = -ENOMEM;
1466                 goto out;
1467         }
1468         rt6_ex->rt6i = nrt;
1469         rt6_ex->stamp = jiffies;
1470         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1471         bucket->depth++;
1472         net->ipv6.rt6_stats->fib_rt_cache++;
1473
1474         if (bucket->depth > FIB6_MAX_DEPTH)
1475                 rt6_exception_remove_oldest(bucket);
1476
1477 out:
1478         spin_unlock_bh(&rt6_exception_lock);
1479
1480         /* Update fn->fn_sernum to invalidate all cached dst */
1481         if (!err) {
1482                 spin_lock_bh(&ort->fib6_table->tb6_lock);
1483                 fib6_update_sernum(net, ort);
1484                 spin_unlock_bh(&ort->fib6_table->tb6_lock);
1485                 fib6_force_start_gc(net);
1486         }
1487
1488         return err;
1489 }
1490
1491 void rt6_flush_exceptions(struct fib6_info *rt)
1492 {
1493         struct rt6_exception_bucket *bucket;
1494         struct rt6_exception *rt6_ex;
1495         struct hlist_node *tmp;
1496         int i;
1497
1498         spin_lock_bh(&rt6_exception_lock);
1499         /* Prevent rt6_insert_exception() to recreate the bucket list */
1500         rt->exception_bucket_flushed = 1;
1501
1502         bucket = rcu_dereference_protected(rt->rt6i_exception_bucket,
1503                                     lockdep_is_held(&rt6_exception_lock));
1504         if (!bucket)
1505                 goto out;
1506
1507         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1508                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist)
1509                         rt6_remove_exception(bucket, rt6_ex);
1510                 WARN_ON_ONCE(bucket->depth);
1511                 bucket++;
1512         }
1513
1514 out:
1515         spin_unlock_bh(&rt6_exception_lock);
1516 }
1517
1518 /* Find cached rt in the hash table inside passed in rt
1519  * Caller has to hold rcu_read_lock()
1520  */
1521 static struct rt6_info *rt6_find_cached_rt(struct fib6_info *rt,
1522                                            struct in6_addr *daddr,
1523                                            struct in6_addr *saddr)
1524 {
1525         struct rt6_exception_bucket *bucket;
1526         struct in6_addr *src_key = NULL;
1527         struct rt6_exception *rt6_ex;
1528         struct rt6_info *res = NULL;
1529
1530         bucket = rcu_dereference(rt->rt6i_exception_bucket);
1531
1532 #ifdef CONFIG_IPV6_SUBTREES
1533         /* rt6i_src.plen != 0 indicates rt is in subtree
1534          * and exception table is indexed by a hash of
1535          * both rt6i_dst and rt6i_src.
1536          * Otherwise, the exception table is indexed by
1537          * a hash of only rt6i_dst.
1538          */
1539         if (rt->fib6_src.plen)
1540                 src_key = saddr;
1541 #endif
1542         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1543
1544         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1545                 res = rt6_ex->rt6i;
1546
1547         return res;
1548 }
1549
1550 /* Remove the passed in cached rt from the hash table that contains it */
1551 static int rt6_remove_exception_rt(struct rt6_info *rt)
1552 {
1553         struct rt6_exception_bucket *bucket;
1554         struct in6_addr *src_key = NULL;
1555         struct rt6_exception *rt6_ex;
1556         struct fib6_info *from;
1557         int err;
1558
1559         from = rcu_dereference(rt->from);
1560         if (!from ||
1561             !(rt->rt6i_flags & RTF_CACHE))
1562                 return -EINVAL;
1563
1564         if (!rcu_access_pointer(from->rt6i_exception_bucket))
1565                 return -ENOENT;
1566
1567         spin_lock_bh(&rt6_exception_lock);
1568         bucket = rcu_dereference_protected(from->rt6i_exception_bucket,
1569                                     lockdep_is_held(&rt6_exception_lock));
1570 #ifdef CONFIG_IPV6_SUBTREES
1571         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1572          * and exception table is indexed by a hash of
1573          * both rt6i_dst and rt6i_src.
1574          * Otherwise, the exception table is indexed by
1575          * a hash of only rt6i_dst.
1576          */
1577         if (from->fib6_src.plen)
1578                 src_key = &rt->rt6i_src.addr;
1579 #endif
1580         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1581                                                &rt->rt6i_dst.addr,
1582                                                src_key);
1583         if (rt6_ex) {
1584                 rt6_remove_exception(bucket, rt6_ex);
1585                 err = 0;
1586         } else {
1587                 err = -ENOENT;
1588         }
1589
1590         spin_unlock_bh(&rt6_exception_lock);
1591         return err;
1592 }
1593
1594 /* Find rt6_ex which contains the passed in rt cache and
1595  * refresh its stamp
1596  */
1597 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1598 {
1599         struct rt6_exception_bucket *bucket;
1600         struct fib6_info *from = rt->from;
1601         struct in6_addr *src_key = NULL;
1602         struct rt6_exception *rt6_ex;
1603
1604         if (!from ||
1605             !(rt->rt6i_flags & RTF_CACHE))
1606                 return;
1607
1608         rcu_read_lock();
1609         bucket = rcu_dereference(from->rt6i_exception_bucket);
1610
1611 #ifdef CONFIG_IPV6_SUBTREES
1612         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1613          * and exception table is indexed by a hash of
1614          * both rt6i_dst and rt6i_src.
1615          * Otherwise, the exception table is indexed by
1616          * a hash of only rt6i_dst.
1617          */
1618         if (from->fib6_src.plen)
1619                 src_key = &rt->rt6i_src.addr;
1620 #endif
1621         rt6_ex = __rt6_find_exception_rcu(&bucket,
1622                                           &rt->rt6i_dst.addr,
1623                                           src_key);
1624         if (rt6_ex)
1625                 rt6_ex->stamp = jiffies;
1626
1627         rcu_read_unlock();
1628 }
1629
1630 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1631                                          struct rt6_info *rt, int mtu)
1632 {
1633         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1634          * lowest MTU in the path: always allow updating the route PMTU to
1635          * reflect PMTU decreases.
1636          *
1637          * If the new MTU is higher, and the route PMTU is equal to the local
1638          * MTU, this means the old MTU is the lowest in the path, so allow
1639          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1640          * handle this.
1641          */
1642
1643         if (dst_mtu(&rt->dst) >= mtu)
1644                 return true;
1645
1646         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
1647                 return true;
1648
1649         return false;
1650 }
1651
1652 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
1653                                        struct fib6_info *rt, int mtu)
1654 {
1655         struct rt6_exception_bucket *bucket;
1656         struct rt6_exception *rt6_ex;
1657         int i;
1658
1659         bucket = rcu_dereference_protected(rt->rt6i_exception_bucket,
1660                                         lockdep_is_held(&rt6_exception_lock));
1661
1662         if (!bucket)
1663                 return;
1664
1665         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1666                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1667                         struct rt6_info *entry = rt6_ex->rt6i;
1668
1669                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
1670                          * route), the metrics of its rt->from have already
1671                          * been updated.
1672                          */
1673                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
1674                             rt6_mtu_change_route_allowed(idev, entry, mtu))
1675                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
1676                 }
1677                 bucket++;
1678         }
1679 }
1680
1681 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
1682
1683 static void rt6_exceptions_clean_tohost(struct fib6_info *rt,
1684                                         struct in6_addr *gateway)
1685 {
1686         struct rt6_exception_bucket *bucket;
1687         struct rt6_exception *rt6_ex;
1688         struct hlist_node *tmp;
1689         int i;
1690
1691         if (!rcu_access_pointer(rt->rt6i_exception_bucket))
1692                 return;
1693
1694         spin_lock_bh(&rt6_exception_lock);
1695         bucket = rcu_dereference_protected(rt->rt6i_exception_bucket,
1696                                      lockdep_is_held(&rt6_exception_lock));
1697
1698         if (bucket) {
1699                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1700                         hlist_for_each_entry_safe(rt6_ex, tmp,
1701                                                   &bucket->chain, hlist) {
1702                                 struct rt6_info *entry = rt6_ex->rt6i;
1703
1704                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
1705                                     RTF_CACHE_GATEWAY &&
1706                                     ipv6_addr_equal(gateway,
1707                                                     &entry->rt6i_gateway)) {
1708                                         rt6_remove_exception(bucket, rt6_ex);
1709                                 }
1710                         }
1711                         bucket++;
1712                 }
1713         }
1714
1715         spin_unlock_bh(&rt6_exception_lock);
1716 }
1717
1718 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
1719                                       struct rt6_exception *rt6_ex,
1720                                       struct fib6_gc_args *gc_args,
1721                                       unsigned long now)
1722 {
1723         struct rt6_info *rt = rt6_ex->rt6i;
1724
1725         /* we are pruning and obsoleting aged-out and non gateway exceptions
1726          * even if others have still references to them, so that on next
1727          * dst_check() such references can be dropped.
1728          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
1729          * expired, independently from their aging, as per RFC 8201 section 4
1730          */
1731         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
1732                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
1733                         RT6_TRACE("aging clone %p\n", rt);
1734                         rt6_remove_exception(bucket, rt6_ex);
1735                         return;
1736                 }
1737         } else if (time_after(jiffies, rt->dst.expires)) {
1738                 RT6_TRACE("purging expired route %p\n", rt);
1739                 rt6_remove_exception(bucket, rt6_ex);
1740                 return;
1741         }
1742
1743         if (rt->rt6i_flags & RTF_GATEWAY) {
1744                 struct neighbour *neigh;
1745                 __u8 neigh_flags = 0;
1746
1747                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
1748                 if (neigh)
1749                         neigh_flags = neigh->flags;
1750
1751                 if (!(neigh_flags & NTF_ROUTER)) {
1752                         RT6_TRACE("purging route %p via non-router but gateway\n",
1753                                   rt);
1754                         rt6_remove_exception(bucket, rt6_ex);
1755                         return;
1756                 }
1757         }
1758
1759         gc_args->more++;
1760 }
1761
1762 void rt6_age_exceptions(struct fib6_info *rt,
1763                         struct fib6_gc_args *gc_args,
1764                         unsigned long now)
1765 {
1766         struct rt6_exception_bucket *bucket;
1767         struct rt6_exception *rt6_ex;
1768         struct hlist_node *tmp;
1769         int i;
1770
1771         if (!rcu_access_pointer(rt->rt6i_exception_bucket))
1772                 return;
1773
1774         rcu_read_lock_bh();
1775         spin_lock(&rt6_exception_lock);
1776         bucket = rcu_dereference_protected(rt->rt6i_exception_bucket,
1777                                     lockdep_is_held(&rt6_exception_lock));
1778
1779         if (bucket) {
1780                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1781                         hlist_for_each_entry_safe(rt6_ex, tmp,
1782                                                   &bucket->chain, hlist) {
1783                                 rt6_age_examine_exception(bucket, rt6_ex,
1784                                                           gc_args, now);
1785                         }
1786                         bucket++;
1787                 }
1788         }
1789         spin_unlock(&rt6_exception_lock);
1790         rcu_read_unlock_bh();
1791 }
1792
1793 /* must be called with rcu lock held */
1794 struct fib6_info *fib6_table_lookup(struct net *net, struct fib6_table *table,
1795                                     int oif, struct flowi6 *fl6, int strict)
1796 {
1797         struct fib6_node *fn, *saved_fn;
1798         struct fib6_info *f6i;
1799
1800         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1801         saved_fn = fn;
1802
1803         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1804                 oif = 0;
1805
1806 redo_rt6_select:
1807         f6i = rt6_select(net, fn, oif, strict);
1808         if (f6i == net->ipv6.fib6_null_entry) {
1809                 fn = fib6_backtrack(fn, &fl6->saddr);
1810                 if (fn)
1811                         goto redo_rt6_select;
1812                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
1813                         /* also consider unreachable route */
1814                         strict &= ~RT6_LOOKUP_F_REACHABLE;
1815                         fn = saved_fn;
1816                         goto redo_rt6_select;
1817                 }
1818         }
1819
1820         trace_fib6_table_lookup(net, f6i, table, fl6);
1821
1822         return f6i;
1823 }
1824
1825 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
1826                                int oif, struct flowi6 *fl6,
1827                                const struct sk_buff *skb, int flags)
1828 {
1829         struct fib6_info *f6i;
1830         struct rt6_info *rt;
1831         int strict = 0;
1832
1833         strict |= flags & RT6_LOOKUP_F_IFACE;
1834         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
1835         if (net->ipv6.devconf_all->forwarding == 0)
1836                 strict |= RT6_LOOKUP_F_REACHABLE;
1837
1838         rcu_read_lock();
1839
1840         f6i = fib6_table_lookup(net, table, oif, fl6, strict);
1841         if (f6i->fib6_nsiblings)
1842                 f6i = fib6_multipath_select(net, f6i, fl6, oif, skb, strict);
1843
1844         if (f6i == net->ipv6.fib6_null_entry) {
1845                 rt = net->ipv6.ip6_null_entry;
1846                 rcu_read_unlock();
1847                 dst_hold(&rt->dst);
1848                 return rt;
1849         }
1850
1851         /*Search through exception table */
1852         rt = rt6_find_cached_rt(f6i, &fl6->daddr, &fl6->saddr);
1853         if (rt) {
1854                 if (ip6_hold_safe(net, &rt, true))
1855                         dst_use_noref(&rt->dst, jiffies);
1856
1857                 rcu_read_unlock();
1858                 return rt;
1859         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
1860                             !(f6i->fib6_flags & RTF_GATEWAY))) {
1861                 /* Create a RTF_CACHE clone which will not be
1862                  * owned by the fib6 tree.  It is for the special case where
1863                  * the daddr in the skb during the neighbor look-up is different
1864                  * from the fl6->daddr used to look-up route here.
1865                  */
1866                 struct rt6_info *uncached_rt;
1867
1868                 uncached_rt = ip6_rt_cache_alloc(f6i, &fl6->daddr, NULL);
1869
1870                 rcu_read_unlock();
1871
1872                 if (uncached_rt) {
1873                         /* Uncached_rt's refcnt is taken during ip6_rt_cache_alloc()
1874                          * No need for another dst_hold()
1875                          */
1876                         rt6_uncached_list_add(uncached_rt);
1877                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
1878                 } else {
1879                         uncached_rt = net->ipv6.ip6_null_entry;
1880                         dst_hold(&uncached_rt->dst);
1881                 }
1882
1883                 return uncached_rt;
1884         } else {
1885                 /* Get a percpu copy */
1886
1887                 struct rt6_info *pcpu_rt;
1888
1889                 local_bh_disable();
1890                 pcpu_rt = rt6_get_pcpu_route(f6i);
1891
1892                 if (!pcpu_rt)
1893                         pcpu_rt = rt6_make_pcpu_route(net, f6i);
1894
1895                 local_bh_enable();
1896                 rcu_read_unlock();
1897
1898                 return pcpu_rt;
1899         }
1900 }
1901 EXPORT_SYMBOL_GPL(ip6_pol_route);
1902
1903 static struct rt6_info *ip6_pol_route_input(struct net *net,
1904                                             struct fib6_table *table,
1905                                             struct flowi6 *fl6,
1906                                             const struct sk_buff *skb,
1907                                             int flags)
1908 {
1909         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
1910 }
1911
1912 struct dst_entry *ip6_route_input_lookup(struct net *net,
1913                                          struct net_device *dev,
1914                                          struct flowi6 *fl6,
1915                                          const struct sk_buff *skb,
1916                                          int flags)
1917 {
1918         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
1919                 flags |= RT6_LOOKUP_F_IFACE;
1920
1921         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
1922 }
1923 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
1924
1925 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
1926                                   struct flow_keys *keys,
1927                                   struct flow_keys *flkeys)
1928 {
1929         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
1930         const struct ipv6hdr *key_iph = outer_iph;
1931         struct flow_keys *_flkeys = flkeys;
1932         const struct ipv6hdr *inner_iph;
1933         const struct icmp6hdr *icmph;
1934         struct ipv6hdr _inner_iph;
1935         struct icmp6hdr _icmph;
1936
1937         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
1938                 goto out;
1939
1940         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
1941                                    sizeof(_icmph), &_icmph);
1942         if (!icmph)
1943                 goto out;
1944
1945         if (icmph->icmp6_type != ICMPV6_DEST_UNREACH &&
1946             icmph->icmp6_type != ICMPV6_PKT_TOOBIG &&
1947             icmph->icmp6_type != ICMPV6_TIME_EXCEED &&
1948             icmph->icmp6_type != ICMPV6_PARAMPROB)
1949                 goto out;
1950
1951         inner_iph = skb_header_pointer(skb,
1952                                        skb_transport_offset(skb) + sizeof(*icmph),
1953                                        sizeof(_inner_iph), &_inner_iph);
1954         if (!inner_iph)
1955                 goto out;
1956
1957         key_iph = inner_iph;
1958         _flkeys = NULL;
1959 out:
1960         if (_flkeys) {
1961                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
1962                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
1963                 keys->tags.flow_label = _flkeys->tags.flow_label;
1964                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
1965         } else {
1966                 keys->addrs.v6addrs.src = key_iph->saddr;
1967                 keys->addrs.v6addrs.dst = key_iph->daddr;
1968                 keys->tags.flow_label = ip6_flowlabel(key_iph);
1969                 keys->basic.ip_proto = key_iph->nexthdr;
1970         }
1971 }
1972
1973 /* if skb is set it will be used and fl6 can be NULL */
1974 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
1975                        const struct sk_buff *skb, struct flow_keys *flkeys)
1976 {
1977         struct flow_keys hash_keys;
1978         u32 mhash;
1979
1980         switch (ip6_multipath_hash_policy(net)) {
1981         case 0:
1982                 memset(&hash_keys, 0, sizeof(hash_keys));
1983                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
1984                 if (skb) {
1985                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
1986                 } else {
1987                         hash_keys.addrs.v6addrs.src = fl6->saddr;
1988                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
1989                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
1990                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
1991                 }
1992                 break;
1993         case 1:
1994                 if (skb) {
1995                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
1996                         struct flow_keys keys;
1997
1998                         /* short-circuit if we already have L4 hash present */
1999                         if (skb->l4_hash)
2000                                 return skb_get_hash_raw(skb) >> 1;
2001
2002                         memset(&hash_keys, 0, sizeof(hash_keys));
2003
2004                         if (!flkeys) {
2005                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2006                                 flkeys = &keys;
2007                         }
2008                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2009                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2010                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2011                         hash_keys.ports.src = flkeys->ports.src;
2012                         hash_keys.ports.dst = flkeys->ports.dst;
2013                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2014                 } else {
2015                         memset(&hash_keys, 0, sizeof(hash_keys));
2016                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2017                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2018                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2019                         hash_keys.ports.src = fl6->fl6_sport;
2020                         hash_keys.ports.dst = fl6->fl6_dport;
2021                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2022                 }
2023                 break;
2024         }
2025         mhash = flow_hash_from_keys(&hash_keys);
2026
2027         return mhash >> 1;
2028 }
2029
2030 void ip6_route_input(struct sk_buff *skb)
2031 {
2032         const struct ipv6hdr *iph = ipv6_hdr(skb);
2033         struct net *net = dev_net(skb->dev);
2034         int flags = RT6_LOOKUP_F_HAS_SADDR;
2035         struct ip_tunnel_info *tun_info;
2036         struct flowi6 fl6 = {
2037                 .flowi6_iif = skb->dev->ifindex,
2038                 .daddr = iph->daddr,
2039                 .saddr = iph->saddr,
2040                 .flowlabel = ip6_flowinfo(iph),
2041                 .flowi6_mark = skb->mark,
2042                 .flowi6_proto = iph->nexthdr,
2043         };
2044         struct flow_keys *flkeys = NULL, _flkeys;
2045
2046         tun_info = skb_tunnel_info(skb);
2047         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2048                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2049
2050         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2051                 flkeys = &_flkeys;
2052
2053         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2054                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2055         skb_dst_drop(skb);
2056         skb_dst_set(skb,
2057                     ip6_route_input_lookup(net, skb->dev, &fl6, skb, flags));
2058 }
2059
2060 static struct rt6_info *ip6_pol_route_output(struct net *net,
2061                                              struct fib6_table *table,
2062                                              struct flowi6 *fl6,
2063                                              const struct sk_buff *skb,
2064                                              int flags)
2065 {
2066         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2067 }
2068
2069 struct dst_entry *ip6_route_output_flags(struct net *net, const struct sock *sk,
2070                                          struct flowi6 *fl6, int flags)
2071 {
2072         bool any_src;
2073
2074         if (ipv6_addr_type(&fl6->daddr) &
2075             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2076                 struct dst_entry *dst;
2077
2078                 dst = l3mdev_link_scope_lookup(net, fl6);
2079                 if (dst)
2080                         return dst;
2081         }
2082
2083         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2084
2085         any_src = ipv6_addr_any(&fl6->saddr);
2086         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2087             (fl6->flowi6_oif && any_src))
2088                 flags |= RT6_LOOKUP_F_IFACE;
2089
2090         if (!any_src)
2091                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2092         else if (sk)
2093                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2094
2095         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2096 }
2097 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2098
2099 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2100 {
2101         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2102         struct net_device *loopback_dev = net->loopback_dev;
2103         struct dst_entry *new = NULL;
2104
2105         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2106                        DST_OBSOLETE_DEAD, 0);
2107         if (rt) {
2108                 rt6_info_init(rt);
2109                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2110
2111                 new = &rt->dst;
2112                 new->__use = 1;
2113                 new->input = dst_discard;
2114                 new->output = dst_discard_out;
2115
2116                 dst_copy_metrics(new, &ort->dst);
2117
2118                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2119                 rt->rt6i_gateway = ort->rt6i_gateway;
2120                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2121
2122                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2123 #ifdef CONFIG_IPV6_SUBTREES
2124                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2125 #endif
2126         }
2127
2128         dst_release(dst_orig);
2129         return new ? new : ERR_PTR(-ENOMEM);
2130 }
2131
2132 /*
2133  *      Destination cache support functions
2134  */
2135
2136 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2137 {
2138         u32 rt_cookie = 0;
2139
2140         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2141                 return false;
2142
2143         if (fib6_check_expired(f6i))
2144                 return false;
2145
2146         return true;
2147 }
2148
2149 static struct dst_entry *rt6_check(struct rt6_info *rt,
2150                                    struct fib6_info *from,
2151                                    u32 cookie)
2152 {
2153         u32 rt_cookie = 0;
2154
2155         if ((from && !fib6_get_cookie_safe(from, &rt_cookie)) ||
2156             rt_cookie != cookie)
2157                 return NULL;
2158
2159         if (rt6_check_expired(rt))
2160                 return NULL;
2161
2162         return &rt->dst;
2163 }
2164
2165 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2166                                             struct fib6_info *from,
2167                                             u32 cookie)
2168 {
2169         if (!__rt6_check_expired(rt) &&
2170             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2171             fib6_check(from, cookie))
2172                 return &rt->dst;
2173         else
2174                 return NULL;
2175 }
2176
2177 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2178 {
2179         struct dst_entry *dst_ret;
2180         struct fib6_info *from;
2181         struct rt6_info *rt;
2182
2183         rt = container_of(dst, struct rt6_info, dst);
2184
2185         rcu_read_lock();
2186
2187         /* All IPV6 dsts are created with ->obsolete set to the value
2188          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2189          * into this function always.
2190          */
2191
2192         from = rcu_dereference(rt->from);
2193
2194         if (from && (rt->rt6i_flags & RTF_PCPU ||
2195             unlikely(!list_empty(&rt->rt6i_uncached))))
2196                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2197         else
2198                 dst_ret = rt6_check(rt, from, cookie);
2199
2200         rcu_read_unlock();
2201
2202         return dst_ret;
2203 }
2204
2205 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2206 {
2207         struct rt6_info *rt = (struct rt6_info *) dst;
2208
2209         if (rt) {
2210                 if (rt->rt6i_flags & RTF_CACHE) {
2211                         rcu_read_lock();
2212                         if (rt6_check_expired(rt)) {
2213                                 rt6_remove_exception_rt(rt);
2214                                 dst = NULL;
2215                         }
2216                         rcu_read_unlock();
2217                 } else {
2218                         dst_release(dst);
2219                         dst = NULL;
2220                 }
2221         }
2222         return dst;
2223 }
2224
2225 static void ip6_link_failure(struct sk_buff *skb)
2226 {
2227         struct rt6_info *rt;
2228
2229         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2230
2231         rt = (struct rt6_info *) skb_dst(skb);
2232         if (rt) {
2233                 rcu_read_lock();
2234                 if (rt->rt6i_flags & RTF_CACHE) {
2235                         rt6_remove_exception_rt(rt);
2236                 } else {
2237                         struct fib6_info *from;
2238                         struct fib6_node *fn;
2239
2240                         from = rcu_dereference(rt->from);
2241                         if (from) {
2242                                 fn = rcu_dereference(from->fib6_node);
2243                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2244                                         fn->fn_sernum = -1;
2245                         }
2246                 }
2247                 rcu_read_unlock();
2248         }
2249 }
2250
2251 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2252 {
2253         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2254                 struct fib6_info *from;
2255
2256                 rcu_read_lock();
2257                 from = rcu_dereference(rt0->from);
2258                 if (from)
2259                         rt0->dst.expires = from->expires;
2260                 rcu_read_unlock();
2261         }
2262
2263         dst_set_expires(&rt0->dst, timeout);
2264         rt0->rt6i_flags |= RTF_EXPIRES;
2265 }
2266
2267 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2268 {
2269         struct net *net = dev_net(rt->dst.dev);
2270
2271         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2272         rt->rt6i_flags |= RTF_MODIFIED;
2273         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2274 }
2275
2276 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2277 {
2278         bool from_set;
2279
2280         rcu_read_lock();
2281         from_set = !!rcu_dereference(rt->from);
2282         rcu_read_unlock();
2283
2284         return !(rt->rt6i_flags & RTF_CACHE) &&
2285                 (rt->rt6i_flags & RTF_PCPU || from_set);
2286 }
2287
2288 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2289                                  const struct ipv6hdr *iph, u32 mtu)
2290 {
2291         const struct in6_addr *daddr, *saddr;
2292         struct rt6_info *rt6 = (struct rt6_info *)dst;
2293
2294         if (dst_metric_locked(dst, RTAX_MTU))
2295                 return;
2296
2297         if (iph) {
2298                 daddr = &iph->daddr;
2299                 saddr = &iph->saddr;
2300         } else if (sk) {
2301                 daddr = &sk->sk_v6_daddr;
2302                 saddr = &inet6_sk(sk)->saddr;
2303         } else {
2304                 daddr = NULL;
2305                 saddr = NULL;
2306         }
2307         dst_confirm_neigh(dst, daddr);
2308         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2309         if (mtu >= dst_mtu(dst))
2310                 return;
2311
2312         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2313                 rt6_do_update_pmtu(rt6, mtu);
2314                 /* update rt6_ex->stamp for cache */
2315                 if (rt6->rt6i_flags & RTF_CACHE)
2316                         rt6_update_exception_stamp_rt(rt6);
2317         } else if (daddr) {
2318                 struct fib6_info *from;
2319                 struct rt6_info *nrt6;
2320
2321                 rcu_read_lock();
2322                 from = rcu_dereference(rt6->from);
2323                 nrt6 = ip6_rt_cache_alloc(from, daddr, saddr);
2324                 if (nrt6) {
2325                         rt6_do_update_pmtu(nrt6, mtu);
2326                         if (rt6_insert_exception(nrt6, from))
2327                                 dst_release_immediate(&nrt6->dst);
2328                 }
2329                 rcu_read_unlock();
2330         }
2331 }
2332
2333 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2334                                struct sk_buff *skb, u32 mtu)
2335 {
2336         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu);
2337 }
2338
2339 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2340                      int oif, u32 mark, kuid_t uid)
2341 {
2342         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2343         struct dst_entry *dst;
2344         struct flowi6 fl6 = {
2345                 .flowi6_oif = oif,
2346                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2347                 .daddr = iph->daddr,
2348                 .saddr = iph->saddr,
2349                 .flowlabel = ip6_flowinfo(iph),
2350                 .flowi6_uid = uid,
2351         };
2352
2353         dst = ip6_route_output(net, NULL, &fl6);
2354         if (!dst->error)
2355                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu));
2356         dst_release(dst);
2357 }
2358 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2359
2360 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2361 {
2362         int oif = sk->sk_bound_dev_if;
2363         struct dst_entry *dst;
2364
2365         if (!oif && skb->dev)
2366                 oif = l3mdev_master_ifindex(skb->dev);
2367
2368         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2369
2370         dst = __sk_dst_get(sk);
2371         if (!dst || !dst->obsolete ||
2372             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2373                 return;
2374
2375         bh_lock_sock(sk);
2376         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2377                 ip6_datagram_dst_update(sk, false);
2378         bh_unlock_sock(sk);
2379 }
2380 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2381
2382 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2383                            const struct flowi6 *fl6)
2384 {
2385 #ifdef CONFIG_IPV6_SUBTREES
2386         struct ipv6_pinfo *np = inet6_sk(sk);
2387 #endif
2388
2389         ip6_dst_store(sk, dst,
2390                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2391                       &sk->sk_v6_daddr : NULL,
2392 #ifdef CONFIG_IPV6_SUBTREES
2393                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2394                       &np->saddr :
2395 #endif
2396                       NULL);
2397 }
2398
2399 /* Handle redirects */
2400 struct ip6rd_flowi {
2401         struct flowi6 fl6;
2402         struct in6_addr gateway;
2403 };
2404
2405 static struct rt6_info *__ip6_route_redirect(struct net *net,
2406                                              struct fib6_table *table,
2407                                              struct flowi6 *fl6,
2408                                              const struct sk_buff *skb,
2409                                              int flags)
2410 {
2411         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2412         struct rt6_info *ret = NULL, *rt_cache;
2413         struct fib6_info *rt;
2414         struct fib6_node *fn;
2415
2416         /* Get the "current" route for this destination and
2417          * check if the redirect has come from appropriate router.
2418          *
2419          * RFC 4861 specifies that redirects should only be
2420          * accepted if they come from the nexthop to the target.
2421          * Due to the way the routes are chosen, this notion
2422          * is a bit fuzzy and one might need to check all possible
2423          * routes.
2424          */
2425
2426         rcu_read_lock();
2427         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2428 restart:
2429         for_each_fib6_node_rt_rcu(fn) {
2430                 if (rt->fib6_nh.nh_flags & RTNH_F_DEAD)
2431                         continue;
2432                 if (fib6_check_expired(rt))
2433                         continue;
2434                 if (rt->fib6_flags & RTF_REJECT)
2435                         break;
2436                 if (!(rt->fib6_flags & RTF_GATEWAY))
2437                         continue;
2438                 if (fl6->flowi6_oif != rt->fib6_nh.nh_dev->ifindex)
2439                         continue;
2440                 /* rt_cache's gateway might be different from its 'parent'
2441                  * in the case of an ip redirect.
2442                  * So we keep searching in the exception table if the gateway
2443                  * is different.
2444                  */
2445                 if (!ipv6_addr_equal(&rdfl->gateway, &rt->fib6_nh.nh_gw)) {
2446                         rt_cache = rt6_find_cached_rt(rt,
2447                                                       &fl6->daddr,
2448                                                       &fl6->saddr);
2449                         if (rt_cache &&
2450                             ipv6_addr_equal(&rdfl->gateway,
2451                                             &rt_cache->rt6i_gateway)) {
2452                                 ret = rt_cache;
2453                                 break;
2454                         }
2455                         continue;
2456                 }
2457                 break;
2458         }
2459
2460         if (!rt)
2461                 rt = net->ipv6.fib6_null_entry;
2462         else if (rt->fib6_flags & RTF_REJECT) {
2463                 ret = net->ipv6.ip6_null_entry;
2464                 goto out;
2465         }
2466
2467         if (rt == net->ipv6.fib6_null_entry) {
2468                 fn = fib6_backtrack(fn, &fl6->saddr);
2469                 if (fn)
2470                         goto restart;
2471         }
2472
2473 out:
2474         if (ret)
2475                 ip6_hold_safe(net, &ret, true);
2476         else
2477                 ret = ip6_create_rt_rcu(rt);
2478
2479         rcu_read_unlock();
2480
2481         trace_fib6_table_lookup(net, rt, table, fl6);
2482         return ret;
2483 };
2484
2485 static struct dst_entry *ip6_route_redirect(struct net *net,
2486                                             const struct flowi6 *fl6,
2487                                             const struct sk_buff *skb,
2488                                             const struct in6_addr *gateway)
2489 {
2490         int flags = RT6_LOOKUP_F_HAS_SADDR;
2491         struct ip6rd_flowi rdfl;
2492
2493         rdfl.fl6 = *fl6;
2494         rdfl.gateway = *gateway;
2495
2496         return fib6_rule_lookup(net, &rdfl.fl6, skb,
2497                                 flags, __ip6_route_redirect);
2498 }
2499
2500 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
2501                   kuid_t uid)
2502 {
2503         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2504         struct dst_entry *dst;
2505         struct flowi6 fl6 = {
2506                 .flowi6_iif = LOOPBACK_IFINDEX,
2507                 .flowi6_oif = oif,
2508                 .flowi6_mark = mark,
2509                 .daddr = iph->daddr,
2510                 .saddr = iph->saddr,
2511                 .flowlabel = ip6_flowinfo(iph),
2512                 .flowi6_uid = uid,
2513         };
2514
2515         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
2516         rt6_do_redirect(dst, NULL, skb);
2517         dst_release(dst);
2518 }
2519 EXPORT_SYMBOL_GPL(ip6_redirect);
2520
2521 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
2522 {
2523         const struct ipv6hdr *iph = ipv6_hdr(skb);
2524         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
2525         struct dst_entry *dst;
2526         struct flowi6 fl6 = {
2527                 .flowi6_iif = LOOPBACK_IFINDEX,
2528                 .flowi6_oif = oif,
2529                 .daddr = msg->dest,
2530                 .saddr = iph->daddr,
2531                 .flowi6_uid = sock_net_uid(net, NULL),
2532         };
2533
2534         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
2535         rt6_do_redirect(dst, NULL, skb);
2536         dst_release(dst);
2537 }
2538
2539 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
2540 {
2541         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
2542                      sk->sk_uid);
2543 }
2544 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
2545
2546 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
2547 {
2548         struct net_device *dev = dst->dev;
2549         unsigned int mtu = dst_mtu(dst);
2550         struct net *net = dev_net(dev);
2551
2552         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
2553
2554         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
2555                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
2556
2557         /*
2558          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
2559          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
2560          * IPV6_MAXPLEN is also valid and means: "any MSS,
2561          * rely only on pmtu discovery"
2562          */
2563         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
2564                 mtu = IPV6_MAXPLEN;
2565         return mtu;
2566 }
2567
2568 static unsigned int ip6_mtu(const struct dst_entry *dst)
2569 {
2570         struct inet6_dev *idev;
2571         unsigned int mtu;
2572
2573         mtu = dst_metric_raw(dst, RTAX_MTU);
2574         if (mtu)
2575                 goto out;
2576
2577         mtu = IPV6_MIN_MTU;
2578
2579         rcu_read_lock();
2580         idev = __in6_dev_get(dst->dev);
2581         if (idev)
2582                 mtu = idev->cnf.mtu6;
2583         rcu_read_unlock();
2584
2585 out:
2586         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
2587
2588         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
2589 }
2590
2591 /* MTU selection:
2592  * 1. mtu on route is locked - use it
2593  * 2. mtu from nexthop exception
2594  * 3. mtu from egress device
2595  *
2596  * based on ip6_dst_mtu_forward and exception logic of
2597  * rt6_find_cached_rt; called with rcu_read_lock
2598  */
2599 u32 ip6_mtu_from_fib6(struct fib6_info *f6i, struct in6_addr *daddr,
2600                       struct in6_addr *saddr)
2601 {
2602         struct rt6_exception_bucket *bucket;
2603         struct rt6_exception *rt6_ex;
2604         struct in6_addr *src_key;
2605         struct inet6_dev *idev;
2606         u32 mtu = 0;
2607
2608         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
2609                 mtu = f6i->fib6_pmtu;
2610                 if (mtu)
2611                         goto out;
2612         }
2613
2614         src_key = NULL;
2615 #ifdef CONFIG_IPV6_SUBTREES
2616         if (f6i->fib6_src.plen)
2617                 src_key = saddr;
2618 #endif
2619
2620         bucket = rcu_dereference(f6i->rt6i_exception_bucket);
2621         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
2622         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
2623                 mtu = dst_metric_raw(&rt6_ex->rt6i->dst, RTAX_MTU);
2624
2625         if (likely(!mtu)) {
2626                 struct net_device *dev = fib6_info_nh_dev(f6i);
2627
2628                 mtu = IPV6_MIN_MTU;
2629                 idev = __in6_dev_get(dev);
2630                 if (idev && idev->cnf.mtu6 > mtu)
2631                         mtu = idev->cnf.mtu6;
2632         }
2633
2634         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
2635 out:
2636         return mtu - lwtunnel_headroom(fib6_info_nh_lwt(f6i), mtu);
2637 }
2638
2639 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
2640                                   struct flowi6 *fl6)
2641 {
2642         struct dst_entry *dst;
2643         struct rt6_info *rt;
2644         struct inet6_dev *idev = in6_dev_get(dev);
2645         struct net *net = dev_net(dev);
2646
2647         if (unlikely(!idev))
2648                 return ERR_PTR(-ENODEV);
2649
2650         rt = ip6_dst_alloc(net, dev, 0);
2651         if (unlikely(!rt)) {
2652                 in6_dev_put(idev);
2653                 dst = ERR_PTR(-ENOMEM);
2654                 goto out;
2655         }
2656
2657         rt->dst.flags |= DST_HOST;
2658         rt->dst.input = ip6_input;
2659         rt->dst.output  = ip6_output;
2660         rt->rt6i_gateway  = fl6->daddr;
2661         rt->rt6i_dst.addr = fl6->daddr;
2662         rt->rt6i_dst.plen = 128;
2663         rt->rt6i_idev     = idev;
2664         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
2665
2666         /* Add this dst into uncached_list so that rt6_disable_ip() can
2667          * do proper release of the net_device
2668          */
2669         rt6_uncached_list_add(rt);
2670         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2671
2672         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
2673
2674 out:
2675         return dst;
2676 }
2677
2678 static int ip6_dst_gc(struct dst_ops *ops)
2679 {
2680         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
2681         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
2682         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
2683         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
2684         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
2685         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
2686         int entries;
2687
2688         entries = dst_entries_get_fast(ops);
2689         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
2690             entries <= rt_max_size)
2691                 goto out;
2692
2693         net->ipv6.ip6_rt_gc_expire++;
2694         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
2695         entries = dst_entries_get_slow(ops);
2696         if (entries < ops->gc_thresh)
2697                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
2698 out:
2699         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
2700         return entries > rt_max_size;
2701 }
2702
2703 static struct rt6_info *ip6_nh_lookup_table(struct net *net,
2704                                             struct fib6_config *cfg,
2705                                             const struct in6_addr *gw_addr,
2706                                             u32 tbid, int flags)
2707 {
2708         struct flowi6 fl6 = {
2709                 .flowi6_oif = cfg->fc_ifindex,
2710                 .daddr = *gw_addr,
2711                 .saddr = cfg->fc_prefsrc,
2712         };
2713         struct fib6_table *table;
2714         struct rt6_info *rt;
2715
2716         table = fib6_get_table(net, tbid);
2717         if (!table)
2718                 return NULL;
2719
2720         if (!ipv6_addr_any(&cfg->fc_prefsrc))
2721                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2722
2723         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
2724         rt = ip6_pol_route(net, table, cfg->fc_ifindex, &fl6, NULL, flags);
2725
2726         /* if table lookup failed, fall back to full lookup */
2727         if (rt == net->ipv6.ip6_null_entry) {
2728                 ip6_rt_put(rt);
2729                 rt = NULL;
2730         }
2731
2732         return rt;
2733 }
2734
2735 static int ip6_route_check_nh_onlink(struct net *net,
2736                                      struct fib6_config *cfg,
2737                                      const struct net_device *dev,
2738                                      struct netlink_ext_ack *extack)
2739 {
2740         u32 tbid = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
2741         const struct in6_addr *gw_addr = &cfg->fc_gateway;
2742         u32 flags = RTF_LOCAL | RTF_ANYCAST | RTF_REJECT;
2743         struct rt6_info *grt;
2744         int err;
2745
2746         err = 0;
2747         grt = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0);
2748         if (grt) {
2749                 if (!grt->dst.error &&
2750                     /* ignore match if it is the default route */
2751                     grt->from && !ipv6_addr_any(&grt->from->fib6_dst.addr) &&
2752                     (grt->rt6i_flags & flags || dev != grt->dst.dev)) {
2753                         NL_SET_ERR_MSG(extack,
2754                                        "Nexthop has invalid gateway or device mismatch");
2755                         err = -EINVAL;
2756                 }
2757
2758                 ip6_rt_put(grt);
2759         }
2760
2761         return err;
2762 }
2763
2764 static int ip6_route_check_nh(struct net *net,
2765                               struct fib6_config *cfg,
2766                               struct net_device **_dev,
2767                               struct inet6_dev **idev)
2768 {
2769         const struct in6_addr *gw_addr = &cfg->fc_gateway;
2770         struct net_device *dev = _dev ? *_dev : NULL;
2771         struct rt6_info *grt = NULL;
2772         int err = -EHOSTUNREACH;
2773
2774         if (cfg->fc_table) {
2775                 int flags = RT6_LOOKUP_F_IFACE;
2776
2777                 grt = ip6_nh_lookup_table(net, cfg, gw_addr,
2778                                           cfg->fc_table, flags);
2779                 if (grt) {
2780                         if (grt->rt6i_flags & RTF_GATEWAY ||
2781                             (dev && dev != grt->dst.dev)) {
2782                                 ip6_rt_put(grt);
2783                                 grt = NULL;
2784                         }
2785                 }
2786         }
2787
2788         if (!grt)
2789                 grt = rt6_lookup(net, gw_addr, NULL, cfg->fc_ifindex, NULL, 1);
2790
2791         if (!grt)
2792                 goto out;
2793
2794         if (dev) {
2795                 if (dev != grt->dst.dev) {
2796                         ip6_rt_put(grt);
2797                         goto out;
2798                 }
2799         } else {
2800                 *_dev = dev = grt->dst.dev;
2801                 *idev = grt->rt6i_idev;
2802                 dev_hold(dev);
2803                 in6_dev_hold(grt->rt6i_idev);
2804         }
2805
2806         if (!(grt->rt6i_flags & RTF_GATEWAY))
2807                 err = 0;
2808
2809         ip6_rt_put(grt);
2810
2811 out:
2812         return err;
2813 }
2814
2815 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
2816                            struct net_device **_dev, struct inet6_dev **idev,
2817                            struct netlink_ext_ack *extack)
2818 {
2819         const struct in6_addr *gw_addr = &cfg->fc_gateway;
2820         int gwa_type = ipv6_addr_type(gw_addr);
2821         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
2822         const struct net_device *dev = *_dev;
2823         bool need_addr_check = !dev;
2824         int err = -EINVAL;
2825
2826         /* if gw_addr is local we will fail to detect this in case
2827          * address is still TENTATIVE (DAD in progress). rt6_lookup()
2828          * will return already-added prefix route via interface that
2829          * prefix route was assigned to, which might be non-loopback.
2830          */
2831         if (dev &&
2832             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
2833                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
2834                 goto out;
2835         }
2836
2837         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
2838                 /* IPv6 strictly inhibits using not link-local
2839                  * addresses as nexthop address.
2840                  * Otherwise, router will not able to send redirects.
2841                  * It is very good, but in some (rare!) circumstances
2842                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
2843                  * some exceptions. --ANK
2844                  * We allow IPv4-mapped nexthops to support RFC4798-type
2845                  * addressing
2846                  */
2847                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
2848                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
2849                         goto out;
2850                 }
2851
2852                 if (cfg->fc_flags & RTNH_F_ONLINK)
2853                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
2854                 else
2855                         err = ip6_route_check_nh(net, cfg, _dev, idev);
2856
2857                 if (err)
2858                         goto out;
2859         }
2860
2861         /* reload in case device was changed */
2862         dev = *_dev;
2863
2864         err = -EINVAL;
2865         if (!dev) {
2866                 NL_SET_ERR_MSG(extack, "Egress device not specified");
2867                 goto out;
2868         } else if (dev->flags & IFF_LOOPBACK) {
2869                 NL_SET_ERR_MSG(extack,
2870                                "Egress device can not be loopback device for this route");
2871                 goto out;
2872         }
2873
2874         /* if we did not check gw_addr above, do so now that the
2875          * egress device has been resolved.
2876          */
2877         if (need_addr_check &&
2878             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
2879                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
2880                 goto out;
2881         }
2882
2883         err = 0;
2884 out:
2885         return err;
2886 }
2887
2888 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
2889                                               gfp_t gfp_flags,
2890                                               struct netlink_ext_ack *extack)
2891 {
2892         struct net *net = cfg->fc_nlinfo.nl_net;
2893         struct fib6_info *rt = NULL;
2894         struct net_device *dev = NULL;
2895         struct inet6_dev *idev = NULL;
2896         struct fib6_table *table;
2897         int addr_type;
2898         int err = -EINVAL;
2899
2900         /* RTF_PCPU is an internal flag; can not be set by userspace */
2901         if (cfg->fc_flags & RTF_PCPU) {
2902                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
2903                 goto out;
2904         }
2905
2906         /* RTF_CACHE is an internal flag; can not be set by userspace */
2907         if (cfg->fc_flags & RTF_CACHE) {
2908                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
2909                 goto out;
2910         }
2911
2912         if (cfg->fc_type > RTN_MAX) {
2913                 NL_SET_ERR_MSG(extack, "Invalid route type");
2914                 goto out;
2915         }
2916
2917         if (cfg->fc_dst_len > 128) {
2918                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
2919                 goto out;
2920         }
2921         if (cfg->fc_src_len > 128) {
2922                 NL_SET_ERR_MSG(extack, "Invalid source address length");
2923                 goto out;
2924         }
2925 #ifndef CONFIG_IPV6_SUBTREES
2926         if (cfg->fc_src_len) {
2927                 NL_SET_ERR_MSG(extack,
2928                                "Specifying source address requires IPV6_SUBTREES to be enabled");
2929                 goto out;
2930         }
2931 #endif
2932         if (cfg->fc_ifindex) {
2933                 err = -ENODEV;
2934                 dev = dev_get_by_index(net, cfg->fc_ifindex);
2935                 if (!dev)
2936                         goto out;
2937                 idev = in6_dev_get(dev);
2938                 if (!idev)
2939                         goto out;
2940         }
2941
2942         if (cfg->fc_metric == 0)
2943                 cfg->fc_metric = IP6_RT_PRIO_USER;
2944
2945         if (cfg->fc_flags & RTNH_F_ONLINK) {
2946                 if (!dev) {
2947                         NL_SET_ERR_MSG(extack,
2948                                        "Nexthop device required for onlink");
2949                         err = -ENODEV;
2950                         goto out;
2951                 }
2952
2953                 if (!(dev->flags & IFF_UP)) {
2954                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
2955                         err = -ENETDOWN;
2956                         goto out;
2957                 }
2958         }
2959
2960         err = -ENOBUFS;
2961         if (cfg->fc_nlinfo.nlh &&
2962             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
2963                 table = fib6_get_table(net, cfg->fc_table);
2964                 if (!table) {
2965                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
2966                         table = fib6_new_table(net, cfg->fc_table);
2967                 }
2968         } else {
2969                 table = fib6_new_table(net, cfg->fc_table);
2970         }
2971
2972         if (!table)
2973                 goto out;
2974
2975         err = -ENOMEM;
2976         rt = fib6_info_alloc(gfp_flags);
2977         if (!rt)
2978                 goto out;
2979
2980         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len);
2981         if (IS_ERR(rt->fib6_metrics)) {
2982                 err = PTR_ERR(rt->fib6_metrics);
2983                 /* Do not leave garbage there. */
2984                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
2985                 goto out;
2986         }
2987
2988         if (cfg->fc_flags & RTF_ADDRCONF)
2989                 rt->dst_nocount = true;
2990
2991         if (cfg->fc_flags & RTF_EXPIRES)
2992                 fib6_set_expires(rt, jiffies +
2993                                 clock_t_to_jiffies(cfg->fc_expires));
2994         else
2995                 fib6_clean_expires(rt);
2996
2997         if (cfg->fc_protocol == RTPROT_UNSPEC)
2998                 cfg->fc_protocol = RTPROT_BOOT;
2999         rt->fib6_protocol = cfg->fc_protocol;
3000
3001         addr_type = ipv6_addr_type(&cfg->fc_dst);
3002
3003         if (cfg->fc_encap) {
3004                 struct lwtunnel_state *lwtstate;
3005
3006                 err = lwtunnel_build_state(cfg->fc_encap_type,
3007                                            cfg->fc_encap, AF_INET6, cfg,
3008                                            &lwtstate, extack);
3009                 if (err)
3010                         goto out;
3011                 rt->fib6_nh.nh_lwtstate = lwtstate_get(lwtstate);
3012         }
3013
3014         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3015         rt->fib6_dst.plen = cfg->fc_dst_len;
3016         if (rt->fib6_dst.plen == 128)
3017                 rt->dst_host = true;
3018
3019 #ifdef CONFIG_IPV6_SUBTREES
3020         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3021         rt->fib6_src.plen = cfg->fc_src_len;
3022 #endif
3023
3024         rt->fib6_metric = cfg->fc_metric;
3025         rt->fib6_nh.nh_weight = 1;
3026
3027         rt->fib6_type = cfg->fc_type;
3028
3029         /* We cannot add true routes via loopback here,
3030            they would result in kernel looping; promote them to reject routes
3031          */
3032         if ((cfg->fc_flags & RTF_REJECT) ||
3033             (dev && (dev->flags & IFF_LOOPBACK) &&
3034              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3035              !(cfg->fc_flags & RTF_LOCAL))) {
3036                 /* hold loopback dev/idev if we haven't done so. */
3037                 if (dev != net->loopback_dev) {
3038                         if (dev) {
3039                                 dev_put(dev);
3040                                 in6_dev_put(idev);
3041                         }
3042                         dev = net->loopback_dev;
3043                         dev_hold(dev);
3044                         idev = in6_dev_get(dev);
3045                         if (!idev) {
3046                                 err = -ENODEV;
3047                                 goto out;
3048                         }
3049                 }
3050                 rt->fib6_flags = RTF_REJECT|RTF_NONEXTHOP;
3051                 goto install_route;
3052         }
3053
3054         if (cfg->fc_flags & RTF_GATEWAY) {
3055                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3056                 if (err)
3057                         goto out;
3058
3059                 rt->fib6_nh.nh_gw = cfg->fc_gateway;
3060         }
3061
3062         err = -ENODEV;
3063         if (!dev)
3064                 goto out;
3065
3066         if (idev->cnf.disable_ipv6) {
3067                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3068                 err = -EACCES;
3069                 goto out;
3070         }
3071
3072         if (!(dev->flags & IFF_UP)) {
3073                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3074                 err = -ENETDOWN;
3075                 goto out;
3076         }
3077
3078         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3079                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3080                         NL_SET_ERR_MSG(extack, "Invalid source address");
3081                         err = -EINVAL;
3082                         goto out;
3083                 }
3084                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3085                 rt->fib6_prefsrc.plen = 128;
3086         } else
3087                 rt->fib6_prefsrc.plen = 0;
3088
3089         rt->fib6_flags = cfg->fc_flags;
3090
3091 install_route:
3092         if (!(rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3093             !netif_carrier_ok(dev))
3094                 rt->fib6_nh.nh_flags |= RTNH_F_LINKDOWN;
3095         rt->fib6_nh.nh_flags |= (cfg->fc_flags & RTNH_F_ONLINK);
3096         rt->fib6_nh.nh_dev = dev;
3097         rt->fib6_table = table;
3098
3099         if (idev)
3100                 in6_dev_put(idev);
3101
3102         return rt;
3103 out:
3104         if (dev)
3105                 dev_put(dev);
3106         if (idev)
3107                 in6_dev_put(idev);
3108
3109         fib6_info_release(rt);
3110         return ERR_PTR(err);
3111 }
3112
3113 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3114                   struct netlink_ext_ack *extack)
3115 {
3116         struct fib6_info *rt;
3117         int err;
3118
3119         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3120         if (IS_ERR(rt))
3121                 return PTR_ERR(rt);
3122
3123         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3124         fib6_info_release(rt);
3125
3126         return err;
3127 }
3128
3129 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3130 {
3131         struct net *net = info->nl_net;
3132         struct fib6_table *table;
3133         int err;
3134
3135         if (rt == net->ipv6.fib6_null_entry) {
3136                 err = -ENOENT;
3137                 goto out;
3138         }
3139
3140         table = rt->fib6_table;
3141         spin_lock_bh(&table->tb6_lock);
3142         err = fib6_del(rt, info);
3143         spin_unlock_bh(&table->tb6_lock);
3144
3145 out:
3146         fib6_info_release(rt);
3147         return err;
3148 }
3149
3150 int ip6_del_rt(struct net *net, struct fib6_info *rt)
3151 {
3152         struct nl_info info = { .nl_net = net };
3153
3154         return __ip6_del_rt(rt, &info);
3155 }
3156
3157 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3158 {
3159         struct nl_info *info = &cfg->fc_nlinfo;
3160         struct net *net = info->nl_net;
3161         struct sk_buff *skb = NULL;
3162         struct fib6_table *table;
3163         int err = -ENOENT;
3164
3165         if (rt == net->ipv6.fib6_null_entry)
3166                 goto out_put;
3167         table = rt->fib6_table;
3168         spin_lock_bh(&table->tb6_lock);
3169
3170         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3171                 struct fib6_info *sibling, *next_sibling;
3172
3173                 /* prefer to send a single notification with all hops */
3174                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3175                 if (skb) {
3176                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3177
3178                         if (rt6_fill_node(net, skb, rt, NULL,
3179                                           NULL, NULL, 0, RTM_DELROUTE,
3180                                           info->portid, seq, 0) < 0) {
3181                                 kfree_skb(skb);
3182                                 skb = NULL;
3183                         } else
3184                                 info->skip_notify = 1;
3185                 }
3186
3187                 list_for_each_entry_safe(sibling, next_sibling,
3188                                          &rt->fib6_siblings,
3189                                          fib6_siblings) {
3190                         err = fib6_del(sibling, info);
3191                         if (err)
3192                                 goto out_unlock;
3193                 }
3194         }
3195
3196         err = fib6_del(rt, info);
3197 out_unlock:
3198         spin_unlock_bh(&table->tb6_lock);
3199 out_put:
3200         fib6_info_release(rt);
3201
3202         if (skb) {
3203                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3204                             info->nlh, gfp_any());
3205         }
3206         return err;
3207 }
3208
3209 static int ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3210 {
3211         int rc = -ESRCH;
3212
3213         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3214                 goto out;
3215
3216         if (cfg->fc_flags & RTF_GATEWAY &&
3217             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3218                 goto out;
3219
3220         rc = rt6_remove_exception_rt(rt);
3221 out:
3222         return rc;
3223 }
3224
3225 static int ip6_route_del(struct fib6_config *cfg,
3226                          struct netlink_ext_ack *extack)
3227 {
3228         struct rt6_info *rt_cache;
3229         struct fib6_table *table;
3230         struct fib6_info *rt;
3231         struct fib6_node *fn;
3232         int err = -ESRCH;
3233
3234         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3235         if (!table) {
3236                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3237                 return err;
3238         }
3239
3240         rcu_read_lock();
3241
3242         fn = fib6_locate(&table->tb6_root,
3243                          &cfg->fc_dst, cfg->fc_dst_len,
3244                          &cfg->fc_src, cfg->fc_src_len,
3245                          !(cfg->fc_flags & RTF_CACHE));
3246
3247         if (fn) {
3248                 for_each_fib6_node_rt_rcu(fn) {
3249                         if (cfg->fc_flags & RTF_CACHE) {
3250                                 int rc;
3251
3252                                 rt_cache = rt6_find_cached_rt(rt, &cfg->fc_dst,
3253                                                               &cfg->fc_src);
3254                                 if (rt_cache) {
3255                                         rc = ip6_del_cached_rt(rt_cache, cfg);
3256                                         if (rc != -ESRCH) {
3257                                                 rcu_read_unlock();
3258                                                 return rc;
3259                                         }
3260                                 }
3261                                 continue;
3262                         }
3263                         if (cfg->fc_ifindex &&
3264                             (!rt->fib6_nh.nh_dev ||
3265                              rt->fib6_nh.nh_dev->ifindex != cfg->fc_ifindex))
3266                                 continue;
3267                         if (cfg->fc_flags & RTF_GATEWAY &&
3268                             !ipv6_addr_equal(&cfg->fc_gateway, &rt->fib6_nh.nh_gw))
3269                                 continue;
3270                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3271                                 continue;
3272                         if (cfg->fc_protocol && cfg->fc_protocol != rt->fib6_protocol)
3273                                 continue;
3274                         if (!fib6_info_hold_safe(rt))
3275                                 continue;
3276                         rcu_read_unlock();
3277
3278                         /* if gateway was specified only delete the one hop */
3279                         if (cfg->fc_flags & RTF_GATEWAY)
3280                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3281
3282                         return __ip6_del_rt_siblings(rt, cfg);
3283                 }
3284         }
3285         rcu_read_unlock();
3286
3287         return err;
3288 }
3289
3290 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
3291 {
3292         struct netevent_redirect netevent;
3293         struct rt6_info *rt, *nrt = NULL;
3294         struct ndisc_options ndopts;
3295         struct inet6_dev *in6_dev;
3296         struct neighbour *neigh;
3297         struct fib6_info *from;
3298         struct rd_msg *msg;
3299         int optlen, on_link;
3300         u8 *lladdr;
3301
3302         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
3303         optlen -= sizeof(*msg);
3304
3305         if (optlen < 0) {
3306                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
3307                 return;
3308         }
3309
3310         msg = (struct rd_msg *)icmp6_hdr(skb);
3311
3312         if (ipv6_addr_is_multicast(&msg->dest)) {
3313                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
3314                 return;
3315         }
3316
3317         on_link = 0;
3318         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
3319                 on_link = 1;
3320         } else if (ipv6_addr_type(&msg->target) !=
3321                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
3322                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
3323                 return;
3324         }
3325
3326         in6_dev = __in6_dev_get(skb->dev);
3327         if (!in6_dev)
3328                 return;
3329         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
3330                 return;
3331
3332         /* RFC2461 8.1:
3333          *      The IP source address of the Redirect MUST be the same as the current
3334          *      first-hop router for the specified ICMP Destination Address.
3335          */
3336
3337         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
3338                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
3339                 return;
3340         }
3341
3342         lladdr = NULL;
3343         if (ndopts.nd_opts_tgt_lladdr) {
3344                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
3345                                              skb->dev);
3346                 if (!lladdr) {
3347                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
3348                         return;
3349                 }
3350         }
3351
3352         rt = (struct rt6_info *) dst;
3353         if (rt->rt6i_flags & RTF_REJECT) {
3354                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
3355                 return;
3356         }
3357
3358         /* Redirect received -> path was valid.
3359          * Look, redirects are sent only in response to data packets,
3360          * so that this nexthop apparently is reachable. --ANK
3361          */
3362         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
3363
3364         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
3365         if (!neigh)
3366                 return;
3367
3368         /*
3369          *      We have finally decided to accept it.
3370          */
3371
3372         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
3373                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
3374                      NEIGH_UPDATE_F_OVERRIDE|
3375                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
3376                                      NEIGH_UPDATE_F_ISROUTER)),
3377                      NDISC_REDIRECT, &ndopts);
3378
3379         rcu_read_lock();
3380         from = rcu_dereference(rt->from);
3381         /* This fib6_info_hold() is safe here because we hold reference to rt
3382          * and rt already holds reference to fib6_info.
3383          */
3384         fib6_info_hold(from);
3385         rcu_read_unlock();
3386
3387         nrt = ip6_rt_cache_alloc(from, &msg->dest, NULL);
3388         if (!nrt)
3389                 goto out;
3390
3391         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
3392         if (on_link)
3393                 nrt->rt6i_flags &= ~RTF_GATEWAY;
3394
3395         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
3396
3397         /* No need to remove rt from the exception table if rt is
3398          * a cached route because rt6_insert_exception() will
3399          * takes care of it
3400          */
3401         if (rt6_insert_exception(nrt, from)) {
3402                 dst_release_immediate(&nrt->dst);
3403                 goto out;
3404         }
3405
3406         netevent.old = &rt->dst;
3407         netevent.new = &nrt->dst;
3408         netevent.daddr = &msg->dest;
3409         netevent.neigh = neigh;
3410         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
3411
3412 out:
3413         fib6_info_release(from);
3414         neigh_release(neigh);
3415 }
3416
3417 #ifdef CONFIG_IPV6_ROUTE_INFO
3418 static struct fib6_info *rt6_get_route_info(struct net *net,
3419                                            const struct in6_addr *prefix, int prefixlen,
3420                                            const struct in6_addr *gwaddr,
3421                                            struct net_device *dev)
3422 {
3423         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
3424         int ifindex = dev->ifindex;
3425         struct fib6_node *fn;
3426         struct fib6_info *rt = NULL;
3427         struct fib6_table *table;
3428
3429         table = fib6_get_table(net, tb_id);
3430         if (!table)
3431                 return NULL;
3432
3433         rcu_read_lock();
3434         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
3435         if (!fn)
3436                 goto out;
3437
3438         for_each_fib6_node_rt_rcu(fn) {
3439                 if (rt->fib6_nh.nh_dev->ifindex != ifindex)
3440                         continue;
3441                 if ((rt->fib6_flags & (RTF_ROUTEINFO|RTF_GATEWAY)) != (RTF_ROUTEINFO|RTF_GATEWAY))
3442                         continue;
3443                 if (!ipv6_addr_equal(&rt->fib6_nh.nh_gw, gwaddr))
3444                         continue;
3445                 if (!fib6_info_hold_safe(rt))
3446                         continue;
3447                 break;
3448         }
3449 out:
3450         rcu_read_unlock();
3451         return rt;
3452 }
3453
3454 static struct fib6_info *rt6_add_route_info(struct net *net,
3455                                            const struct in6_addr *prefix, int prefixlen,
3456                                            const struct in6_addr *gwaddr,
3457                                            struct net_device *dev,
3458                                            unsigned int pref)
3459 {
3460         struct fib6_config cfg = {
3461                 .fc_metric      = IP6_RT_PRIO_USER,
3462                 .fc_ifindex     = dev->ifindex,
3463                 .fc_dst_len     = prefixlen,
3464                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
3465                                   RTF_UP | RTF_PREF(pref),
3466                 .fc_protocol = RTPROT_RA,
3467                 .fc_type = RTN_UNICAST,
3468                 .fc_nlinfo.portid = 0,
3469                 .fc_nlinfo.nlh = NULL,
3470                 .fc_nlinfo.nl_net = net,
3471         };
3472
3473         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
3474         cfg.fc_dst = *prefix;
3475         cfg.fc_gateway = *gwaddr;
3476
3477         /* We should treat it as a default route if prefix length is 0. */
3478         if (!prefixlen)
3479                 cfg.fc_flags |= RTF_DEFAULT;
3480
3481         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
3482
3483         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
3484 }
3485 #endif
3486
3487 struct fib6_info *rt6_get_dflt_router(struct net *net,
3488                                      const struct in6_addr *addr,
3489                                      struct net_device *dev)
3490 {
3491         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
3492         struct fib6_info *rt;
3493         struct fib6_table *table;
3494
3495         table = fib6_get_table(net, tb_id);
3496         if (!table)
3497                 return NULL;
3498
3499         rcu_read_lock();
3500         for_each_fib6_node_rt_rcu(&table->tb6_root) {
3501                 if (dev == rt->fib6_nh.nh_dev &&
3502                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
3503                     ipv6_addr_equal(&rt->fib6_nh.nh_gw, addr))
3504                         break;
3505         }
3506         if (rt && !fib6_info_hold_safe(rt))
3507                 rt = NULL;
3508         rcu_read_unlock();
3509         return rt;
3510 }
3511
3512 struct fib6_info *rt6_add_dflt_router(struct net *net,
3513                                      const struct in6_addr *gwaddr,
3514                                      struct net_device *dev,
3515                                      unsigned int pref)
3516 {
3517         struct fib6_config cfg = {
3518                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
3519                 .fc_metric      = IP6_RT_PRIO_USER,
3520                 .fc_ifindex     = dev->ifindex,
3521                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
3522                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
3523                 .fc_protocol = RTPROT_RA,
3524                 .fc_type = RTN_UNICAST,
3525                 .fc_nlinfo.portid = 0,
3526                 .fc_nlinfo.nlh = NULL,
3527                 .fc_nlinfo.nl_net = net,
3528         };
3529
3530         cfg.fc_gateway = *gwaddr;
3531
3532         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
3533                 struct fib6_table *table;
3534
3535                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
3536                 if (table)
3537                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
3538         }
3539
3540         return rt6_get_dflt_router(net, gwaddr, dev);
3541 }
3542
3543 static void __rt6_purge_dflt_routers(struct net *net,
3544                                      struct fib6_table *table)
3545 {
3546         struct fib6_info *rt;
3547
3548 restart:
3549         rcu_read_lock();
3550         for_each_fib6_node_rt_rcu(&table->tb6_root) {
3551                 struct net_device *dev = fib6_info_nh_dev(rt);
3552                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
3553
3554                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
3555                     (!idev || idev->cnf.accept_ra != 2) &&
3556                     fib6_info_hold_safe(rt)) {
3557                         rcu_read_unlock();
3558                         ip6_del_rt(net, rt);
3559                         goto restart;
3560                 }
3561         }
3562         rcu_read_unlock();
3563
3564         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
3565 }
3566
3567 void rt6_purge_dflt_routers(struct net *net)
3568 {
3569         struct fib6_table *table;
3570         struct hlist_head *head;
3571         unsigned int h;
3572
3573         rcu_read_lock();
3574
3575         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
3576                 head = &net->ipv6.fib_table_hash[h];
3577                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
3578                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
3579                                 __rt6_purge_dflt_routers(net, table);
3580                 }
3581         }
3582
3583         rcu_read_unlock();
3584 }
3585
3586 static void rtmsg_to_fib6_config(struct net *net,
3587                                  struct in6_rtmsg *rtmsg,
3588                                  struct fib6_config *cfg)
3589 {
3590         *cfg = (struct fib6_config){
3591                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
3592                          : RT6_TABLE_MAIN,
3593                 .fc_ifindex = rtmsg->rtmsg_ifindex,
3594                 .fc_metric = rtmsg->rtmsg_metric,
3595                 .fc_expires = rtmsg->rtmsg_info,
3596                 .fc_dst_len = rtmsg->rtmsg_dst_len,
3597                 .fc_src_len = rtmsg->rtmsg_src_len,
3598                 .fc_flags = rtmsg->rtmsg_flags,
3599                 .fc_type = rtmsg->rtmsg_type,
3600
3601                 .fc_nlinfo.nl_net = net,
3602
3603                 .fc_dst = rtmsg->rtmsg_dst,
3604                 .fc_src = rtmsg->rtmsg_src,
3605                 .fc_gateway = rtmsg->rtmsg_gateway,
3606         };
3607 }
3608
3609 int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3610 {
3611         struct fib6_config cfg;
3612         struct in6_rtmsg rtmsg;
3613         int err;
3614
3615         switch (cmd) {
3616         case SIOCADDRT:         /* Add a route */
3617         case SIOCDELRT:         /* Delete a route */
3618                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3619                         return -EPERM;
3620                 err = copy_from_user(&rtmsg, arg,
3621                                      sizeof(struct in6_rtmsg));
3622                 if (err)
3623                         return -EFAULT;
3624
3625                 rtmsg_to_fib6_config(net, &rtmsg, &cfg);
3626
3627                 rtnl_lock();
3628                 switch (cmd) {
3629                 case SIOCADDRT:
3630                         err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
3631                         break;
3632                 case SIOCDELRT:
3633                         err = ip6_route_del(&cfg, NULL);
3634                         break;
3635                 default:
3636                         err = -EINVAL;
3637                 }
3638                 rtnl_unlock();
3639
3640                 return err;
3641         }
3642
3643         return -EINVAL;
3644 }
3645
3646 /*
3647  *      Drop the packet on the floor
3648  */
3649
3650 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
3651 {
3652         int type;
3653         struct dst_entry *dst = skb_dst(skb);
3654         switch (ipstats_mib_noroutes) {
3655         case IPSTATS_MIB_INNOROUTES:
3656                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
3657                 if (type == IPV6_ADDR_ANY) {
3658                         IP6_INC_STATS(dev_net(dst->dev),
3659                                       __in6_dev_get_safely(skb->dev),
3660                                       IPSTATS_MIB_INADDRERRORS);
3661                         break;
3662                 }
3663                 /* FALLTHROUGH */
3664         case IPSTATS_MIB_OUTNOROUTES:
3665                 IP6_INC_STATS(dev_net(dst->dev), ip6_dst_idev(dst),
3666                               ipstats_mib_noroutes);
3667                 break;
3668         }
3669         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
3670         kfree_skb(skb);
3671         return 0;
3672 }
3673
3674 static int ip6_pkt_discard(struct sk_buff *skb)
3675 {
3676         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
3677 }
3678
3679 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
3680 {
3681         skb->dev = skb_dst(skb)->dev;
3682         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
3683 }
3684
3685 static int ip6_pkt_prohibit(struct sk_buff *skb)
3686 {
3687         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
3688 }
3689
3690 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
3691 {
3692         skb->dev = skb_dst(skb)->dev;
3693         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
3694 }
3695
3696 /*
3697  *      Allocate a dst for local (unicast / anycast) address.
3698  */
3699
3700 struct fib6_info *addrconf_f6i_alloc(struct net *net,
3701                                      struct inet6_dev *idev,
3702                                      const struct in6_addr *addr,
3703                                      bool anycast, gfp_t gfp_flags)
3704 {
3705         u32 tb_id;
3706         struct net_device *dev = idev->dev;
3707         struct fib6_info *f6i;
3708
3709         f6i = fib6_info_alloc(gfp_flags);
3710         if (!f6i)
3711                 return ERR_PTR(-ENOMEM);
3712
3713         f6i->fib6_metrics = ip_fib_metrics_init(net, NULL, 0);
3714         f6i->dst_nocount = true;
3715         f6i->dst_host = true;
3716         f6i->fib6_protocol = RTPROT_KERNEL;
3717         f6i->fib6_flags = RTF_UP | RTF_NONEXTHOP;
3718         if (anycast) {
3719                 f6i->fib6_type = RTN_ANYCAST;
3720                 f6i->fib6_flags |= RTF_ANYCAST;
3721         } else {
3722                 f6i->fib6_type = RTN_LOCAL;
3723                 f6i->fib6_flags |= RTF_LOCAL;
3724         }
3725
3726         f6i->fib6_nh.nh_gw = *addr;
3727         dev_hold(dev);
3728         f6i->fib6_nh.nh_dev = dev;
3729         f6i->fib6_dst.addr = *addr;
3730         f6i->fib6_dst.plen = 128;
3731         tb_id = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL;
3732         f6i->fib6_table = fib6_get_table(net, tb_id);
3733
3734         return f6i;
3735 }
3736
3737 /* remove deleted ip from prefsrc entries */
3738 struct arg_dev_net_ip {
3739         struct net_device *dev;
3740         struct net *net;
3741         struct in6_addr *addr;
3742 };
3743
3744 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
3745 {
3746         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
3747         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
3748         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
3749
3750         if (((void *)rt->fib6_nh.nh_dev == dev || !dev) &&
3751             rt != net->ipv6.fib6_null_entry &&
3752             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
3753                 spin_lock_bh(&rt6_exception_lock);
3754                 /* remove prefsrc entry */
3755                 rt->fib6_prefsrc.plen = 0;
3756                 spin_unlock_bh(&rt6_exception_lock);
3757         }
3758         return 0;
3759 }
3760
3761 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
3762 {
3763         struct net *net = dev_net(ifp->idev->dev);
3764         struct arg_dev_net_ip adni = {
3765                 .dev = ifp->idev->dev,
3766                 .net = net,
3767                 .addr = &ifp->addr,
3768         };
3769         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
3770 }
3771
3772 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT | RTF_GATEWAY)
3773
3774 /* Remove routers and update dst entries when gateway turn into host. */
3775 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
3776 {
3777         struct in6_addr *gateway = (struct in6_addr *)arg;
3778
3779         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
3780             ipv6_addr_equal(gateway, &rt->fib6_nh.nh_gw)) {
3781                 return -1;
3782         }
3783
3784         /* Further clean up cached routes in exception table.
3785          * This is needed because cached route may have a different
3786          * gateway than its 'parent' in the case of an ip redirect.
3787          */
3788         rt6_exceptions_clean_tohost(rt, gateway);
3789
3790         return 0;
3791 }
3792
3793 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
3794 {
3795         fib6_clean_all(net, fib6_clean_tohost, gateway);
3796 }
3797
3798 struct arg_netdev_event {
3799         const struct net_device *dev;
3800         union {
3801                 unsigned int nh_flags;
3802                 unsigned long event;
3803         };
3804 };
3805
3806 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
3807 {
3808         struct fib6_info *iter;
3809         struct fib6_node *fn;
3810
3811         fn = rcu_dereference_protected(rt->fib6_node,
3812                         lockdep_is_held(&rt->fib6_table->tb6_lock));
3813         iter = rcu_dereference_protected(fn->leaf,
3814                         lockdep_is_held(&rt->fib6_table->tb6_lock));
3815         while (iter) {
3816                 if (iter->fib6_metric == rt->fib6_metric &&
3817                     rt6_qualify_for_ecmp(iter))
3818                         return iter;
3819                 iter = rcu_dereference_protected(iter->fib6_next,
3820                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
3821         }
3822
3823         return NULL;
3824 }
3825
3826 static bool rt6_is_dead(const struct fib6_info *rt)
3827 {
3828         if (rt->fib6_nh.nh_flags & RTNH_F_DEAD ||
3829             (rt->fib6_nh.nh_flags & RTNH_F_LINKDOWN &&
3830              fib6_ignore_linkdown(rt)))
3831                 return true;
3832
3833         return false;
3834 }
3835
3836 static int rt6_multipath_total_weight(const struct fib6_info *rt)
3837 {
3838         struct fib6_info *iter;
3839         int total = 0;
3840
3841         if (!rt6_is_dead(rt))
3842                 total += rt->fib6_nh.nh_weight;
3843
3844         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
3845                 if (!rt6_is_dead(iter))
3846                         total += iter->fib6_nh.nh_weight;
3847         }
3848
3849         return total;
3850 }
3851
3852 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
3853 {
3854         int upper_bound = -1;
3855
3856         if (!rt6_is_dead(rt)) {
3857                 *weight += rt->fib6_nh.nh_weight;
3858                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
3859                                                     total) - 1;
3860         }
3861         atomic_set(&rt->fib6_nh.nh_upper_bound, upper_bound);
3862 }
3863
3864 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
3865 {
3866         struct fib6_info *iter;
3867         int weight = 0;
3868
3869         rt6_upper_bound_set(rt, &weight, total);
3870
3871         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
3872                 rt6_upper_bound_set(iter, &weight, total);
3873 }
3874
3875 void rt6_multipath_rebalance(struct fib6_info *rt)
3876 {
3877         struct fib6_info *first;
3878         int total;
3879
3880         /* In case the entire multipath route was marked for flushing,
3881          * then there is no need to rebalance upon the removal of every
3882          * sibling route.
3883          */
3884         if (!rt->fib6_nsiblings || rt->should_flush)
3885                 return;
3886
3887         /* During lookup routes are evaluated in order, so we need to
3888          * make sure upper bounds are assigned from the first sibling
3889          * onwards.
3890          */
3891         first = rt6_multipath_first_sibling(rt);
3892         if (WARN_ON_ONCE(!first))
3893                 return;
3894
3895         total = rt6_multipath_total_weight(first);
3896         rt6_multipath_upper_bound_set(first, total);
3897 }
3898
3899 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
3900 {
3901         const struct arg_netdev_event *arg = p_arg;
3902         struct net *net = dev_net(arg->dev);
3903
3904         if (rt != net->ipv6.fib6_null_entry && rt->fib6_nh.nh_dev == arg->dev) {
3905                 rt->fib6_nh.nh_flags &= ~arg->nh_flags;
3906                 fib6_update_sernum_upto_root(net, rt);
3907                 rt6_multipath_rebalance(rt);
3908         }
3909
3910         return 0;
3911 }
3912
3913 void rt6_sync_up(struct net_device *dev, unsigned int nh_flags)
3914 {
3915         struct arg_netdev_event arg = {
3916                 .dev = dev,
3917                 {
3918                         .nh_flags = nh_flags,
3919                 },
3920         };
3921
3922         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
3923                 arg.nh_flags |= RTNH_F_LINKDOWN;
3924
3925         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
3926 }
3927
3928 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
3929                                    const struct net_device *dev)
3930 {
3931         struct fib6_info *iter;
3932
3933         if (rt->fib6_nh.nh_dev == dev)
3934                 return true;
3935         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
3936                 if (iter->fib6_nh.nh_dev == dev)
3937                         return true;
3938
3939         return false;
3940 }
3941
3942 static void rt6_multipath_flush(struct fib6_info *rt)
3943 {
3944         struct fib6_info *iter;
3945
3946         rt->should_flush = 1;
3947         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
3948                 iter->should_flush = 1;
3949 }
3950
3951 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
3952                                              const struct net_device *down_dev)
3953 {
3954         struct fib6_info *iter;
3955         unsigned int dead = 0;
3956
3957         if (rt->fib6_nh.nh_dev == down_dev ||
3958             rt->fib6_nh.nh_flags & RTNH_F_DEAD)
3959                 dead++;
3960         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
3961                 if (iter->fib6_nh.nh_dev == down_dev ||
3962                     iter->fib6_nh.nh_flags & RTNH_F_DEAD)
3963                         dead++;
3964
3965         return dead;
3966 }
3967
3968 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
3969                                        const struct net_device *dev,
3970                                        unsigned int nh_flags)
3971 {
3972         struct fib6_info *iter;
3973
3974         if (rt->fib6_nh.nh_dev == dev)
3975                 rt->fib6_nh.nh_flags |= nh_flags;
3976         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
3977                 if (iter->fib6_nh.nh_dev == dev)
3978                         iter->fib6_nh.nh_flags |= nh_flags;
3979 }
3980
3981 /* called with write lock held for table with rt */
3982 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
3983 {
3984         const struct arg_netdev_event *arg = p_arg;
3985         const struct net_device *dev = arg->dev;
3986         struct net *net = dev_net(dev);
3987
3988         if (rt == net->ipv6.fib6_null_entry)
3989                 return 0;
3990
3991         switch (arg->event) {
3992         case NETDEV_UNREGISTER:
3993                 return rt->fib6_nh.nh_dev == dev ? -1 : 0;
3994         case NETDEV_DOWN:
3995                 if (rt->should_flush)
3996                         return -1;
3997                 if (!rt->fib6_nsiblings)
3998                         return rt->fib6_nh.nh_dev == dev ? -1 : 0;
3999                 if (rt6_multipath_uses_dev(rt, dev)) {
4000                         unsigned int count;
4001
4002                         count = rt6_multipath_dead_count(rt, dev);
4003                         if (rt->fib6_nsiblings + 1 == count) {
4004                                 rt6_multipath_flush(rt);
4005                                 return -1;
4006                         }
4007                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4008                                                    RTNH_F_LINKDOWN);
4009                         fib6_update_sernum(net, rt);
4010                         rt6_multipath_rebalance(rt);
4011                 }
4012                 return -2;
4013         case NETDEV_CHANGE:
4014                 if (rt->fib6_nh.nh_dev != dev ||
4015                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4016                         break;
4017                 rt->fib6_nh.nh_flags |= RTNH_F_LINKDOWN;
4018                 rt6_multipath_rebalance(rt);
4019                 break;
4020         }
4021
4022         return 0;
4023 }
4024
4025 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4026 {
4027         struct arg_netdev_event arg = {
4028                 .dev = dev,
4029                 {
4030                         .event = event,
4031                 },
4032         };
4033         struct net *net = dev_net(dev);
4034
4035         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4036                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4037         else
4038                 fib6_clean_all(net, fib6_ifdown, &arg);
4039 }
4040
4041 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4042 {
4043         rt6_sync_down_dev(dev, event);
4044         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4045         neigh_ifdown(&nd_tbl, dev);
4046 }
4047
4048 struct rt6_mtu_change_arg {
4049         struct net_device *dev;
4050         unsigned int mtu;
4051 };
4052
4053 static int rt6_mtu_change_route(struct fib6_info *rt, void *p_arg)
4054 {
4055         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4056         struct inet6_dev *idev;
4057
4058         /* In IPv6 pmtu discovery is not optional,
4059            so that RTAX_MTU lock cannot disable it.
4060            We still use this lock to block changes
4061            caused by addrconf/ndisc.
4062         */
4063
4064         idev = __in6_dev_get(arg->dev);
4065         if (!idev)
4066                 return 0;
4067
4068         /* For administrative MTU increase, there is no way to discover
4069            IPv6 PMTU increase, so PMTU increase should be updated here.
4070            Since RFC 1981 doesn't include administrative MTU increase
4071            update PMTU increase is a MUST. (i.e. jumbo frame)
4072          */
4073         if (rt->fib6_nh.nh_dev == arg->dev &&
4074             !fib6_metric_locked(rt, RTAX_MTU)) {
4075                 u32 mtu = rt->fib6_pmtu;
4076
4077                 if (mtu >= arg->mtu ||
4078                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4079                         fib6_metric_set(rt, RTAX_MTU, arg->mtu);
4080
4081                 spin_lock_bh(&rt6_exception_lock);
4082                 rt6_exceptions_update_pmtu(idev, rt, arg->mtu);
4083                 spin_unlock_bh(&rt6_exception_lock);
4084         }
4085         return 0;
4086 }
4087
4088 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4089 {
4090         struct rt6_mtu_change_arg arg = {
4091                 .dev = dev,
4092                 .mtu = mtu,
4093         };
4094
4095         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4096 }
4097
4098 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4099         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4100         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4101         [RTA_OIF]               = { .type = NLA_U32 },
4102         [RTA_IIF]               = { .type = NLA_U32 },
4103         [RTA_PRIORITY]          = { .type = NLA_U32 },
4104         [RTA_METRICS]           = { .type = NLA_NESTED },
4105         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4106         [RTA_PREF]              = { .type = NLA_U8 },
4107         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4108         [RTA_ENCAP]             = { .type = NLA_NESTED },
4109         [RTA_EXPIRES]           = { .type = NLA_U32 },
4110         [RTA_UID]               = { .type = NLA_U32 },
4111         [RTA_MARK]              = { .type = NLA_U32 },
4112         [RTA_TABLE]             = { .type = NLA_U32 },
4113         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4114         [RTA_SPORT]             = { .type = NLA_U16 },
4115         [RTA_DPORT]             = { .type = NLA_U16 },
4116 };
4117
4118 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4119                               struct fib6_config *cfg,
4120                               struct netlink_ext_ack *extack)
4121 {
4122         struct rtmsg *rtm;
4123         struct nlattr *tb[RTA_MAX+1];
4124         unsigned int pref;
4125         int err;
4126
4127         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv6_policy,
4128                           extack);
4129         if (err < 0)
4130                 goto errout;
4131
4132         err = -EINVAL;
4133         rtm = nlmsg_data(nlh);
4134
4135         *cfg = (struct fib6_config){
4136                 .fc_table = rtm->rtm_table,
4137                 .fc_dst_len = rtm->rtm_dst_len,
4138                 .fc_src_len = rtm->rtm_src_len,
4139                 .fc_flags = RTF_UP,
4140                 .fc_protocol = rtm->rtm_protocol,
4141                 .fc_type = rtm->rtm_type,
4142
4143                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4144                 .fc_nlinfo.nlh = nlh,
4145                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4146         };
4147
4148         if (rtm->rtm_type == RTN_UNREACHABLE ||
4149             rtm->rtm_type == RTN_BLACKHOLE ||
4150             rtm->rtm_type == RTN_PROHIBIT ||
4151             rtm->rtm_type == RTN_THROW)
4152                 cfg->fc_flags |= RTF_REJECT;
4153
4154         if (rtm->rtm_type == RTN_LOCAL)
4155                 cfg->fc_flags |= RTF_LOCAL;
4156
4157         if (rtm->rtm_flags & RTM_F_CLONED)
4158                 cfg->fc_flags |= RTF_CACHE;
4159
4160         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4161
4162         if (tb[RTA_GATEWAY]) {
4163                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4164                 cfg->fc_flags |= RTF_GATEWAY;
4165         }
4166
4167         if (tb[RTA_DST]) {
4168                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4169
4170                 if (nla_len(tb[RTA_DST]) < plen)
4171                         goto errout;
4172
4173                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4174         }
4175
4176         if (tb[RTA_SRC]) {
4177                 int plen = (rtm->rtm_src_len + 7) >> 3;
4178
4179                 if (nla_len(tb[RTA_SRC]) < plen)
4180                         goto errout;
4181
4182                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4183         }
4184
4185         if (tb[RTA_PREFSRC])
4186                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4187
4188         if (tb[RTA_OIF])
4189                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4190
4191         if (tb[RTA_PRIORITY])
4192                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4193
4194         if (tb[RTA_METRICS]) {
4195                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4196                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4197         }
4198
4199         if (tb[RTA_TABLE])
4200                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4201
4202         if (tb[RTA_MULTIPATH]) {
4203                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4204                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4205
4206                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4207                                                      cfg->fc_mp_len, extack);
4208                 if (err < 0)
4209                         goto errout;
4210         }
4211
4212         if (tb[RTA_PREF]) {
4213                 pref = nla_get_u8(tb[RTA_PREF]);
4214                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4215                     pref != ICMPV6_ROUTER_PREF_HIGH)
4216                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4217                 cfg->fc_flags |= RTF_PREF(pref);
4218         }
4219
4220         if (tb[RTA_ENCAP])
4221                 cfg->fc_encap = tb[RTA_ENCAP];
4222
4223         if (tb[RTA_ENCAP_TYPE]) {
4224                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
4225
4226                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
4227                 if (err < 0)
4228                         goto errout;
4229         }
4230
4231         if (tb[RTA_EXPIRES]) {
4232                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
4233
4234                 if (addrconf_finite_timeout(timeout)) {
4235                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
4236                         cfg->fc_flags |= RTF_EXPIRES;
4237                 }
4238         }
4239
4240         err = 0;
4241 errout:
4242         return err;
4243 }
4244
4245 struct rt6_nh {
4246         struct fib6_info *fib6_info;
4247         struct fib6_config r_cfg;
4248         struct list_head next;
4249 };
4250
4251 static void ip6_print_replace_route_err(struct list_head *rt6_nh_list)
4252 {
4253         struct rt6_nh *nh;
4254
4255         list_for_each_entry(nh, rt6_nh_list, next) {
4256                 pr_warn("IPV6: multipath route replace failed (check consistency of installed routes): %pI6c nexthop %pI6c ifi %d\n",
4257                         &nh->r_cfg.fc_dst, &nh->r_cfg.fc_gateway,
4258                         nh->r_cfg.fc_ifindex);
4259         }
4260 }
4261
4262 static int ip6_route_info_append(struct net *net,
4263                                  struct list_head *rt6_nh_list,
4264                                  struct fib6_info *rt,
4265                                  struct fib6_config *r_cfg)
4266 {
4267         struct rt6_nh *nh;
4268         int err = -EEXIST;
4269
4270         list_for_each_entry(nh, rt6_nh_list, next) {
4271                 /* check if fib6_info already exists */
4272                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
4273                         return err;
4274         }
4275
4276         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
4277         if (!nh)
4278                 return -ENOMEM;
4279         nh->fib6_info = rt;
4280         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
4281         list_add_tail(&nh->next, rt6_nh_list);
4282
4283         return 0;
4284 }
4285
4286 static void ip6_route_mpath_notify(struct fib6_info *rt,
4287                                    struct fib6_info *rt_last,
4288                                    struct nl_info *info,
4289                                    __u16 nlflags)
4290 {
4291         /* if this is an APPEND route, then rt points to the first route
4292          * inserted and rt_last points to last route inserted. Userspace
4293          * wants a consistent dump of the route which starts at the first
4294          * nexthop. Since sibling routes are always added at the end of
4295          * the list, find the first sibling of the last route appended
4296          */
4297         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
4298                 rt = list_first_entry(&rt_last->fib6_siblings,
4299                                       struct fib6_info,
4300                                       fib6_siblings);
4301         }
4302
4303         if (rt)
4304                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
4305 }
4306
4307 static int ip6_route_multipath_add(struct fib6_config *cfg,
4308                                    struct netlink_ext_ack *extack)
4309 {
4310         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
4311         struct nl_info *info = &cfg->fc_nlinfo;
4312         struct fib6_config r_cfg;
4313         struct rtnexthop *rtnh;
4314         struct fib6_info *rt;
4315         struct rt6_nh *err_nh;
4316         struct rt6_nh *nh, *nh_safe;
4317         __u16 nlflags;
4318         int remaining;
4319         int attrlen;
4320         int err = 1;
4321         int nhn = 0;
4322         int replace = (cfg->fc_nlinfo.nlh &&
4323                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
4324         LIST_HEAD(rt6_nh_list);
4325
4326         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
4327         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
4328                 nlflags |= NLM_F_APPEND;
4329
4330         remaining = cfg->fc_mp_len;
4331         rtnh = (struct rtnexthop *)cfg->fc_mp;
4332
4333         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
4334          * fib6_info structs per nexthop
4335          */
4336         while (rtnh_ok(rtnh, remaining)) {
4337                 memcpy(&r_cfg, cfg, sizeof(*cfg));
4338                 if (rtnh->rtnh_ifindex)
4339                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
4340
4341                 attrlen = rtnh_attrlen(rtnh);
4342                 if (attrlen > 0) {
4343                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
4344
4345                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
4346                         if (nla) {
4347                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
4348                                 r_cfg.fc_flags |= RTF_GATEWAY;
4349                         }
4350                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
4351                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
4352                         if (nla)
4353                                 r_cfg.fc_encap_type = nla_get_u16(nla);
4354                 }
4355
4356                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
4357                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
4358                 if (IS_ERR(rt)) {
4359                         err = PTR_ERR(rt);
4360                         rt = NULL;
4361                         goto cleanup;
4362                 }
4363                 if (!rt6_qualify_for_ecmp(rt)) {
4364                         err = -EINVAL;
4365                         NL_SET_ERR_MSG(extack,
4366                                        "Device only routes can not be added for IPv6 using the multipath API.");
4367                         fib6_info_release(rt);
4368                         goto cleanup;
4369                 }
4370
4371                 rt->fib6_nh.nh_weight = rtnh->rtnh_hops + 1;
4372
4373                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
4374                                             rt, &r_cfg);
4375                 if (err) {
4376                         fib6_info_release(rt);
4377                         goto cleanup;
4378                 }
4379
4380                 rtnh = rtnh_next(rtnh, &remaining);
4381         }
4382
4383         /* for add and replace send one notification with all nexthops.
4384          * Skip the notification in fib6_add_rt2node and send one with
4385          * the full route when done
4386          */
4387         info->skip_notify = 1;
4388
4389         err_nh = NULL;
4390         list_for_each_entry(nh, &rt6_nh_list, next) {
4391                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
4392                 fib6_info_release(nh->fib6_info);
4393
4394                 if (!err) {
4395                         /* save reference to last route successfully inserted */
4396                         rt_last = nh->fib6_info;
4397
4398                         /* save reference to first route for notification */
4399                         if (!rt_notif)
4400                                 rt_notif = nh->fib6_info;
4401                 }
4402
4403                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
4404                 nh->fib6_info = NULL;
4405                 if (err) {
4406                         if (replace && nhn)
4407                                 ip6_print_replace_route_err(&rt6_nh_list);
4408                         err_nh = nh;
4409                         goto add_errout;
4410                 }
4411
4412                 /* Because each route is added like a single route we remove
4413                  * these flags after the first nexthop: if there is a collision,
4414                  * we have already failed to add the first nexthop:
4415                  * fib6_add_rt2node() has rejected it; when replacing, old
4416                  * nexthops have been replaced by first new, the rest should
4417                  * be added to it.
4418                  */
4419                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
4420                                                      NLM_F_REPLACE);
4421                 nhn++;
4422         }
4423
4424         /* success ... tell user about new route */
4425         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
4426         goto cleanup;
4427
4428 add_errout:
4429         /* send notification for routes that were added so that
4430          * the delete notifications sent by ip6_route_del are
4431          * coherent
4432          */
4433         if (rt_notif)
4434                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
4435
4436         /* Delete routes that were already added */
4437         list_for_each_entry(nh, &rt6_nh_list, next) {
4438                 if (err_nh == nh)
4439                         break;
4440                 ip6_route_del(&nh->r_cfg, extack);
4441         }
4442
4443 cleanup:
4444         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
4445                 if (nh->fib6_info)
4446                         fib6_info_release(nh->fib6_info);
4447                 list_del(&nh->next);
4448                 kfree(nh);
4449         }
4450
4451         return err;
4452 }
4453
4454 static int ip6_route_multipath_del(struct fib6_config *cfg,
4455                                    struct netlink_ext_ack *extack)
4456 {
4457         struct fib6_config r_cfg;
4458         struct rtnexthop *rtnh;
4459         int remaining;
4460         int attrlen;
4461         int err = 1, last_err = 0;
4462
4463         remaining = cfg->fc_mp_len;
4464         rtnh = (struct rtnexthop *)cfg->fc_mp;
4465
4466         /* Parse a Multipath Entry */
4467         while (rtnh_ok(rtnh, remaining)) {
4468                 memcpy(&r_cfg, cfg, sizeof(*cfg));
4469                 if (rtnh->rtnh_ifindex)
4470                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
4471
4472                 attrlen = rtnh_attrlen(rtnh);
4473                 if (attrlen > 0) {
4474                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
4475
4476                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
4477                         if (nla) {
4478                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
4479                                 r_cfg.fc_flags |= RTF_GATEWAY;
4480                         }
4481                 }
4482                 err = ip6_route_del(&r_cfg, extack);
4483                 if (err)
4484                         last_err = err;
4485
4486                 rtnh = rtnh_next(rtnh, &remaining);
4487         }
4488
4489         return last_err;
4490 }
4491
4492 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
4493                               struct netlink_ext_ack *extack)
4494 {
4495         struct fib6_config cfg;
4496         int err;
4497
4498         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
4499         if (err < 0)
4500                 return err;
4501
4502         if (cfg.fc_mp)
4503                 return ip6_route_multipath_del(&cfg, extack);
4504         else {
4505                 cfg.fc_delete_all_nh = 1;
4506                 return ip6_route_del(&cfg, extack);
4507         }
4508 }
4509
4510 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
4511                               struct netlink_ext_ack *extack)
4512 {
4513         struct fib6_config cfg;
4514         int err;
4515
4516         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
4517         if (err < 0)
4518                 return err;
4519
4520         if (cfg.fc_mp)
4521                 return ip6_route_multipath_add(&cfg, extack);
4522         else
4523                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
4524 }
4525
4526 static size_t rt6_nlmsg_size(struct fib6_info *rt)
4527 {
4528         int nexthop_len = 0;
4529
4530         if (rt->fib6_nsiblings) {
4531                 nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
4532                             + NLA_ALIGN(sizeof(struct rtnexthop))
4533                             + nla_total_size(16) /* RTA_GATEWAY */
4534                             + lwtunnel_get_encap_size(rt->fib6_nh.nh_lwtstate);
4535
4536                 nexthop_len *= rt->fib6_nsiblings;
4537         }
4538
4539         return NLMSG_ALIGN(sizeof(struct rtmsg))
4540                + nla_total_size(16) /* RTA_SRC */
4541                + nla_total_size(16) /* RTA_DST */
4542                + nla_total_size(16) /* RTA_GATEWAY */
4543                + nla_total_size(16) /* RTA_PREFSRC */
4544                + nla_total_size(4) /* RTA_TABLE */
4545                + nla_total_size(4) /* RTA_IIF */
4546                + nla_total_size(4) /* RTA_OIF */
4547                + nla_total_size(4) /* RTA_PRIORITY */
4548                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
4549                + nla_total_size(sizeof(struct rta_cacheinfo))
4550                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
4551                + nla_total_size(1) /* RTA_PREF */
4552                + lwtunnel_get_encap_size(rt->fib6_nh.nh_lwtstate)
4553                + nexthop_len;
4554 }
4555
4556 static int rt6_nexthop_info(struct sk_buff *skb, struct fib6_info *rt,
4557                             unsigned int *flags, bool skip_oif)
4558 {
4559         if (rt->fib6_nh.nh_flags & RTNH_F_DEAD)
4560                 *flags |= RTNH_F_DEAD;
4561
4562         if (rt->fib6_nh.nh_flags & RTNH_F_LINKDOWN) {
4563                 *flags |= RTNH_F_LINKDOWN;
4564
4565                 rcu_read_lock();
4566                 if (fib6_ignore_linkdown(rt))
4567                         *flags |= RTNH_F_DEAD;
4568                 rcu_read_unlock();
4569         }
4570
4571         if (rt->fib6_flags & RTF_GATEWAY) {
4572                 if (nla_put_in6_addr(skb, RTA_GATEWAY, &rt->fib6_nh.nh_gw) < 0)
4573                         goto nla_put_failure;
4574         }
4575
4576         *flags |= (rt->fib6_nh.nh_flags & RTNH_F_ONLINK);
4577         if (rt->fib6_nh.nh_flags & RTNH_F_OFFLOAD)
4578                 *flags |= RTNH_F_OFFLOAD;
4579
4580         /* not needed for multipath encoding b/c it has a rtnexthop struct */
4581         if (!skip_oif && rt->fib6_nh.nh_dev &&
4582             nla_put_u32(skb, RTA_OIF, rt->fib6_nh.nh_dev->ifindex))
4583                 goto nla_put_failure;
4584
4585         if (rt->fib6_nh.nh_lwtstate &&
4586             lwtunnel_fill_encap(skb, rt->fib6_nh.nh_lwtstate) < 0)
4587                 goto nla_put_failure;
4588
4589         return 0;
4590
4591 nla_put_failure:
4592         return -EMSGSIZE;
4593 }
4594
4595 /* add multipath next hop */
4596 static int rt6_add_nexthop(struct sk_buff *skb, struct fib6_info *rt)
4597 {
4598         const struct net_device *dev = rt->fib6_nh.nh_dev;
4599         struct rtnexthop *rtnh;
4600         unsigned int flags = 0;
4601
4602         rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
4603         if (!rtnh)
4604                 goto nla_put_failure;
4605
4606         rtnh->rtnh_hops = rt->fib6_nh.nh_weight - 1;
4607         rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
4608
4609         if (rt6_nexthop_info(skb, rt, &flags, true) < 0)
4610                 goto nla_put_failure;
4611
4612         rtnh->rtnh_flags = flags;
4613
4614         /* length of rtnetlink header + attributes */
4615         rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
4616
4617         return 0;
4618
4619 nla_put_failure:
4620         return -EMSGSIZE;
4621 }
4622
4623 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
4624                          struct fib6_info *rt, struct dst_entry *dst,
4625                          struct in6_addr *dest, struct in6_addr *src,
4626                          int iif, int type, u32 portid, u32 seq,
4627                          unsigned int flags)
4628 {
4629         struct rt6_info *rt6 = (struct rt6_info *)dst;
4630         struct rt6key *rt6_dst, *rt6_src;
4631         u32 *pmetrics, table, rt6_flags;
4632         struct nlmsghdr *nlh;
4633         struct rtmsg *rtm;
4634         long expires = 0;
4635
4636         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
4637         if (!nlh)
4638                 return -EMSGSIZE;
4639
4640         if (rt6) {
4641                 rt6_dst = &rt6->rt6i_dst;
4642                 rt6_src = &rt6->rt6i_src;
4643                 rt6_flags = rt6->rt6i_flags;
4644         } else {
4645                 rt6_dst = &rt->fib6_dst;
4646                 rt6_src = &rt->fib6_src;
4647                 rt6_flags = rt->fib6_flags;
4648         }
4649
4650         rtm = nlmsg_data(nlh);
4651         rtm->rtm_family = AF_INET6;
4652         rtm->rtm_dst_len = rt6_dst->plen;
4653         rtm->rtm_src_len = rt6_src->plen;
4654         rtm->rtm_tos = 0;
4655         if (rt->fib6_table)
4656                 table = rt->fib6_table->tb6_id;
4657         else
4658                 table = RT6_TABLE_UNSPEC;
4659         rtm->rtm_table = table;
4660         if (nla_put_u32(skb, RTA_TABLE, table))
4661                 goto nla_put_failure;
4662
4663         rtm->rtm_type = rt->fib6_type;
4664         rtm->rtm_flags = 0;
4665         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
4666         rtm->rtm_protocol = rt->fib6_protocol;
4667
4668         if (rt6_flags & RTF_CACHE)
4669                 rtm->rtm_flags |= RTM_F_CLONED;
4670
4671         if (dest) {
4672                 if (nla_put_in6_addr(skb, RTA_DST, dest))
4673                         goto nla_put_failure;
4674                 rtm->rtm_dst_len = 128;
4675         } else if (rtm->rtm_dst_len)
4676                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
4677                         goto nla_put_failure;
4678 #ifdef CONFIG_IPV6_SUBTREES
4679         if (src) {
4680                 if (nla_put_in6_addr(skb, RTA_SRC, src))
4681                         goto nla_put_failure;
4682                 rtm->rtm_src_len = 128;
4683         } else if (rtm->rtm_src_len &&
4684                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
4685                 goto nla_put_failure;
4686 #endif
4687         if (iif) {
4688 #ifdef CONFIG_IPV6_MROUTE
4689                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
4690                         int err = ip6mr_get_route(net, skb, rtm, portid);
4691
4692                         if (err == 0)
4693                                 return 0;
4694                         if (err < 0)
4695                                 goto nla_put_failure;
4696                 } else
4697 #endif
4698                         if (nla_put_u32(skb, RTA_IIF, iif))
4699                                 goto nla_put_failure;
4700         } else if (dest) {
4701                 struct in6_addr saddr_buf;
4702                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
4703                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
4704                         goto nla_put_failure;
4705         }
4706
4707         if (rt->fib6_prefsrc.plen) {
4708                 struct in6_addr saddr_buf;
4709                 saddr_buf = rt->fib6_prefsrc.addr;
4710                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
4711                         goto nla_put_failure;
4712         }
4713
4714         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
4715         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
4716                 goto nla_put_failure;
4717
4718         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
4719                 goto nla_put_failure;
4720
4721         /* For multipath routes, walk the siblings list and add
4722          * each as a nexthop within RTA_MULTIPATH.
4723          */
4724         if (rt6) {
4725                 if (rt6_flags & RTF_GATEWAY &&
4726                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
4727                         goto nla_put_failure;
4728
4729                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
4730                         goto nla_put_failure;
4731         } else if (rt->fib6_nsiblings) {
4732                 struct fib6_info *sibling, *next_sibling;
4733                 struct nlattr *mp;
4734
4735                 mp = nla_nest_start(skb, RTA_MULTIPATH);
4736                 if (!mp)
4737                         goto nla_put_failure;
4738
4739                 if (rt6_add_nexthop(skb, rt) < 0)
4740                         goto nla_put_failure;
4741
4742                 list_for_each_entry_safe(sibling, next_sibling,
4743                                          &rt->fib6_siblings, fib6_siblings) {
4744                         if (rt6_add_nexthop(skb, sibling) < 0)
4745                                 goto nla_put_failure;
4746                 }
4747
4748                 nla_nest_end(skb, mp);
4749         } else {
4750                 if (rt6_nexthop_info(skb, rt, &rtm->rtm_flags, false) < 0)
4751                         goto nla_put_failure;
4752         }
4753
4754         if (rt6_flags & RTF_EXPIRES) {
4755                 expires = dst ? dst->expires : rt->expires;
4756                 expires -= jiffies;
4757         }
4758
4759         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
4760                 goto nla_put_failure;
4761
4762         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
4763                 goto nla_put_failure;
4764
4765
4766         nlmsg_end(skb, nlh);
4767         return 0;
4768
4769 nla_put_failure:
4770         nlmsg_cancel(skb, nlh);
4771         return -EMSGSIZE;
4772 }
4773
4774 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
4775                                const struct net_device *dev)
4776 {
4777         if (f6i->fib6_nh.nh_dev == dev)
4778                 return true;
4779
4780         if (f6i->fib6_nsiblings) {
4781                 struct fib6_info *sibling, *next_sibling;
4782
4783                 list_for_each_entry_safe(sibling, next_sibling,
4784                                          &f6i->fib6_siblings, fib6_siblings) {
4785                         if (sibling->fib6_nh.nh_dev == dev)
4786                                 return true;
4787                 }
4788         }
4789
4790         return false;
4791 }
4792
4793 int rt6_dump_route(struct fib6_info *rt, void *p_arg)
4794 {
4795         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
4796         struct fib_dump_filter *filter = &arg->filter;
4797         unsigned int flags = NLM_F_MULTI;
4798         struct net *net = arg->net;
4799
4800         if (rt == net->ipv6.fib6_null_entry)
4801                 return 0;
4802
4803         if ((filter->flags & RTM_F_PREFIX) &&
4804             !(rt->fib6_flags & RTF_PREFIX_RT)) {
4805                 /* success since this is not a prefix route */
4806                 return 1;
4807         }
4808         if (filter->filter_set) {
4809                 if ((filter->rt_type && rt->fib6_type != filter->rt_type) ||
4810                     (filter->dev && !fib6_info_uses_dev(rt, filter->dev)) ||
4811                     (filter->protocol && rt->fib6_protocol != filter->protocol)) {
4812                         return 1;
4813                 }
4814                 flags |= NLM_F_DUMP_FILTERED;
4815         }
4816
4817         return rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL, 0,
4818                              RTM_NEWROUTE, NETLINK_CB(arg->cb->skb).portid,
4819                              arg->cb->nlh->nlmsg_seq, flags);
4820 }
4821
4822 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
4823                               struct netlink_ext_ack *extack)
4824 {
4825         struct net *net = sock_net(in_skb->sk);
4826         struct nlattr *tb[RTA_MAX+1];
4827         int err, iif = 0, oif = 0;
4828         struct fib6_info *from;
4829         struct dst_entry *dst;
4830         struct rt6_info *rt;
4831         struct sk_buff *skb;
4832         struct rtmsg *rtm;
4833         struct flowi6 fl6 = {};
4834         bool fibmatch;
4835
4836         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv6_policy,
4837                           extack);
4838         if (err < 0)
4839                 goto errout;
4840
4841         err = -EINVAL;
4842         rtm = nlmsg_data(nlh);
4843         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
4844         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
4845
4846         if (tb[RTA_SRC]) {
4847                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
4848                         goto errout;
4849
4850                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
4851         }
4852
4853         if (tb[RTA_DST]) {
4854                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
4855                         goto errout;
4856
4857                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
4858         }
4859
4860         if (tb[RTA_IIF])
4861                 iif = nla_get_u32(tb[RTA_IIF]);
4862
4863         if (tb[RTA_OIF])
4864                 oif = nla_get_u32(tb[RTA_OIF]);
4865
4866         if (tb[RTA_MARK])
4867                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
4868
4869         if (tb[RTA_UID])
4870                 fl6.flowi6_uid = make_kuid(current_user_ns(),
4871                                            nla_get_u32(tb[RTA_UID]));
4872         else
4873                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
4874
4875         if (tb[RTA_SPORT])
4876                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
4877
4878         if (tb[RTA_DPORT])
4879                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
4880
4881         if (tb[RTA_IP_PROTO]) {
4882                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
4883                                                   &fl6.flowi6_proto, extack);
4884                 if (err)
4885                         goto errout;
4886         }
4887
4888         if (iif) {
4889                 struct net_device *dev;
4890                 int flags = 0;
4891
4892                 rcu_read_lock();
4893
4894                 dev = dev_get_by_index_rcu(net, iif);
4895                 if (!dev) {
4896                         rcu_read_unlock();
4897                         err = -ENODEV;
4898                         goto errout;
4899                 }
4900
4901                 fl6.flowi6_iif = iif;
4902
4903                 if (!ipv6_addr_any(&fl6.saddr))
4904                         flags |= RT6_LOOKUP_F_HAS_SADDR;
4905
4906                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
4907
4908                 rcu_read_unlock();
4909         } else {
4910                 fl6.flowi6_oif = oif;
4911
4912                 dst = ip6_route_output(net, NULL, &fl6);
4913         }
4914
4915
4916         rt = container_of(dst, struct rt6_info, dst);
4917         if (rt->dst.error) {
4918                 err = rt->dst.error;
4919                 ip6_rt_put(rt);
4920                 goto errout;
4921         }
4922
4923         if (rt == net->ipv6.ip6_null_entry) {
4924                 err = rt->dst.error;
4925                 ip6_rt_put(rt);
4926                 goto errout;
4927         }
4928
4929         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
4930         if (!skb) {
4931                 ip6_rt_put(rt);
4932                 err = -ENOBUFS;
4933                 goto errout;
4934         }
4935
4936         skb_dst_set(skb, &rt->dst);
4937
4938         rcu_read_lock();
4939         from = rcu_dereference(rt->from);
4940
4941         if (fibmatch)
4942                 err = rt6_fill_node(net, skb, from, NULL, NULL, NULL, iif,
4943                                     RTM_NEWROUTE, NETLINK_CB(in_skb).portid,
4944                                     nlh->nlmsg_seq, 0);
4945         else
4946                 err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
4947                                     &fl6.saddr, iif, RTM_NEWROUTE,
4948                                     NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
4949                                     0);
4950         rcu_read_unlock();
4951
4952         if (err < 0) {
4953                 kfree_skb(skb);
4954                 goto errout;
4955         }
4956
4957         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
4958 errout:
4959         return err;
4960 }
4961
4962 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
4963                      unsigned int nlm_flags)
4964 {
4965         struct sk_buff *skb;
4966         struct net *net = info->nl_net;
4967         u32 seq;
4968         int err;
4969
4970         err = -ENOBUFS;
4971         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
4972
4973         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
4974         if (!skb)
4975                 goto errout;
4976
4977         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
4978                             event, info->portid, seq, nlm_flags);
4979         if (err < 0) {
4980                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
4981                 WARN_ON(err == -EMSGSIZE);
4982                 kfree_skb(skb);
4983                 goto errout;
4984         }
4985         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
4986                     info->nlh, gfp_any());
4987         return;
4988 errout:
4989         if (err < 0)
4990                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
4991 }
4992
4993 static int ip6_route_dev_notify(struct notifier_block *this,
4994                                 unsigned long event, void *ptr)
4995 {
4996         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4997         struct net *net = dev_net(dev);
4998
4999         if (!(dev->flags & IFF_LOOPBACK))
5000                 return NOTIFY_OK;
5001
5002         if (event == NETDEV_REGISTER) {
5003                 net->ipv6.fib6_null_entry->fib6_nh.nh_dev = dev;
5004                 net->ipv6.ip6_null_entry->dst.dev = dev;
5005                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
5006 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5007                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
5008                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
5009                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
5010                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
5011 #endif
5012          } else if (event == NETDEV_UNREGISTER &&
5013                     dev->reg_state != NETREG_UNREGISTERED) {
5014                 /* NETDEV_UNREGISTER could be fired for multiple times by
5015                  * netdev_wait_allrefs(). Make sure we only call this once.
5016                  */
5017                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
5018 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5019                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
5020                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
5021 #endif
5022         }
5023
5024         return NOTIFY_OK;
5025 }
5026
5027 /*
5028  *      /proc
5029  */
5030
5031 #ifdef CONFIG_PROC_FS
5032 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
5033 {
5034         struct net *net = (struct net *)seq->private;
5035         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
5036                    net->ipv6.rt6_stats->fib_nodes,
5037                    net->ipv6.rt6_stats->fib_route_nodes,
5038                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
5039                    net->ipv6.rt6_stats->fib_rt_entries,
5040                    net->ipv6.rt6_stats->fib_rt_cache,
5041                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
5042                    net->ipv6.rt6_stats->fib_discarded_routes);
5043
5044         return 0;
5045 }
5046 #endif  /* CONFIG_PROC_FS */
5047
5048 #ifdef CONFIG_SYSCTL
5049
5050 static
5051 int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
5052                               void __user *buffer, size_t *lenp, loff_t *ppos)
5053 {
5054         struct net *net;
5055         int delay;
5056         if (!write)
5057                 return -EINVAL;
5058
5059         net = (struct net *)ctl->extra1;
5060         delay = net->ipv6.sysctl.flush_delay;
5061         proc_dointvec(ctl, write, buffer, lenp, ppos);
5062         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
5063         return 0;
5064 }
5065
5066 static int zero;
5067 static int one = 1;
5068
5069 static struct ctl_table ipv6_route_table_template[] = {
5070         {
5071                 .procname       =       "flush",
5072                 .data           =       &init_net.ipv6.sysctl.flush_delay,
5073                 .maxlen         =       sizeof(int),
5074                 .mode           =       0200,
5075                 .proc_handler   =       ipv6_sysctl_rtcache_flush
5076         },
5077         {
5078                 .procname       =       "gc_thresh",
5079                 .data           =       &ip6_dst_ops_template.gc_thresh,
5080                 .maxlen         =       sizeof(int),
5081                 .mode           =       0644,
5082                 .proc_handler   =       proc_dointvec,
5083         },
5084         {
5085                 .procname       =       "max_size",
5086                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
5087                 .maxlen         =       sizeof(int),
5088                 .mode           =       0644,
5089                 .proc_handler   =       proc_dointvec,
5090         },
5091         {
5092                 .procname       =       "gc_min_interval",
5093                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
5094                 .maxlen         =       sizeof(int),
5095                 .mode           =       0644,
5096                 .proc_handler   =       proc_dointvec_jiffies,
5097         },
5098         {
5099                 .procname       =       "gc_timeout",
5100                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
5101                 .maxlen         =       sizeof(int),
5102                 .mode           =       0644,
5103                 .proc_handler   =       proc_dointvec_jiffies,
5104         },
5105         {
5106                 .procname       =       "gc_interval",
5107                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
5108                 .maxlen         =       sizeof(int),
5109                 .mode           =       0644,
5110                 .proc_handler   =       proc_dointvec_jiffies,
5111         },
5112         {
5113                 .procname       =       "gc_elasticity",
5114                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
5115                 .maxlen         =       sizeof(int),
5116                 .mode           =       0644,
5117                 .proc_handler   =       proc_dointvec,
5118         },
5119         {
5120                 .procname       =       "mtu_expires",
5121                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
5122                 .maxlen         =       sizeof(int),
5123                 .mode           =       0644,
5124                 .proc_handler   =       proc_dointvec_jiffies,
5125         },
5126         {
5127                 .procname       =       "min_adv_mss",
5128                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
5129                 .maxlen         =       sizeof(int),
5130                 .mode           =       0644,
5131                 .proc_handler   =       proc_dointvec,
5132         },
5133         {
5134                 .procname       =       "gc_min_interval_ms",
5135                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
5136                 .maxlen         =       sizeof(int),
5137                 .mode           =       0644,
5138                 .proc_handler   =       proc_dointvec_ms_jiffies,
5139         },
5140         {
5141                 .procname       =       "skip_notify_on_dev_down",
5142                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
5143                 .maxlen         =       sizeof(int),
5144                 .mode           =       0644,
5145                 .proc_handler   =       proc_dointvec,
5146                 .extra1         =       &zero,
5147                 .extra2         =       &one,
5148         },
5149         { }
5150 };
5151
5152 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
5153 {
5154         struct ctl_table *table;
5155
5156         table = kmemdup(ipv6_route_table_template,
5157                         sizeof(ipv6_route_table_template),
5158                         GFP_KERNEL);
5159
5160         if (table) {
5161                 table[0].data = &net->ipv6.sysctl.flush_delay;
5162                 table[0].extra1 = net;
5163                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
5164                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
5165                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
5166                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
5167                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
5168                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
5169                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
5170                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
5171                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
5172                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
5173
5174                 /* Don't export sysctls to unprivileged users */
5175                 if (net->user_ns != &init_user_ns)
5176                         table[0].procname = NULL;
5177         }
5178
5179         return table;
5180 }
5181 #endif
5182
5183 static int __net_init ip6_route_net_init(struct net *net)
5184 {
5185         int ret = -ENOMEM;
5186
5187         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
5188                sizeof(net->ipv6.ip6_dst_ops));
5189
5190         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
5191                 goto out_ip6_dst_ops;
5192
5193         net->ipv6.fib6_null_entry = kmemdup(&fib6_null_entry_template,
5194                                             sizeof(*net->ipv6.fib6_null_entry),
5195                                             GFP_KERNEL);
5196         if (!net->ipv6.fib6_null_entry)
5197                 goto out_ip6_dst_entries;
5198
5199         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
5200                                            sizeof(*net->ipv6.ip6_null_entry),
5201                                            GFP_KERNEL);
5202         if (!net->ipv6.ip6_null_entry)
5203                 goto out_fib6_null_entry;
5204         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5205         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
5206                          ip6_template_metrics, true);
5207
5208 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5209         net->ipv6.fib6_has_custom_rules = false;
5210         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
5211                                                sizeof(*net->ipv6.ip6_prohibit_entry),
5212                                                GFP_KERNEL);
5213         if (!net->ipv6.ip6_prohibit_entry)
5214                 goto out_ip6_null_entry;
5215         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5216         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
5217                          ip6_template_metrics, true);
5218
5219         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
5220                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
5221                                                GFP_KERNEL);
5222         if (!net->ipv6.ip6_blk_hole_entry)
5223                 goto out_ip6_prohibit_entry;
5224         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
5225         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
5226                          ip6_template_metrics, true);
5227 #endif
5228
5229         net->ipv6.sysctl.flush_delay = 0;
5230         net->ipv6.sysctl.ip6_rt_max_size = 4096;
5231         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
5232         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
5233         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
5234         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
5235         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
5236         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
5237         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
5238
5239         net->ipv6.ip6_rt_gc_expire = 30*HZ;
5240
5241         ret = 0;
5242 out:
5243         return ret;
5244
5245 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5246 out_ip6_prohibit_entry:
5247         kfree(net->ipv6.ip6_prohibit_entry);
5248 out_ip6_null_entry:
5249         kfree(net->ipv6.ip6_null_entry);
5250 #endif
5251 out_fib6_null_entry:
5252         kfree(net->ipv6.fib6_null_entry);
5253 out_ip6_dst_entries:
5254         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
5255 out_ip6_dst_ops:
5256         goto out;
5257 }
5258
5259 static void __net_exit ip6_route_net_exit(struct net *net)
5260 {
5261         kfree(net->ipv6.fib6_null_entry);
5262         kfree(net->ipv6.ip6_null_entry);
5263 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5264         kfree(net->ipv6.ip6_prohibit_entry);
5265         kfree(net->ipv6.ip6_blk_hole_entry);
5266 #endif
5267         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
5268 }
5269
5270 static int __net_init ip6_route_net_init_late(struct net *net)
5271 {
5272 #ifdef CONFIG_PROC_FS
5273         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
5274                         sizeof(struct ipv6_route_iter));
5275         proc_create_net_single("rt6_stats", 0444, net->proc_net,
5276                         rt6_stats_seq_show, NULL);
5277 #endif
5278         return 0;
5279 }
5280
5281 static void __net_exit ip6_route_net_exit_late(struct net *net)
5282 {
5283 #ifdef CONFIG_PROC_FS
5284         remove_proc_entry("ipv6_route", net->proc_net);
5285         remove_proc_entry("rt6_stats", net->proc_net);
5286 #endif
5287 }
5288
5289 static struct pernet_operations ip6_route_net_ops = {
5290         .init = ip6_route_net_init,
5291         .exit = ip6_route_net_exit,
5292 };
5293
5294 static int __net_init ipv6_inetpeer_init(struct net *net)
5295 {
5296         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
5297
5298         if (!bp)
5299                 return -ENOMEM;
5300         inet_peer_base_init(bp);
5301         net->ipv6.peers = bp;
5302         return 0;
5303 }
5304
5305 static void __net_exit ipv6_inetpeer_exit(struct net *net)
5306 {
5307         struct inet_peer_base *bp = net->ipv6.peers;
5308
5309         net->ipv6.peers = NULL;
5310         inetpeer_invalidate_tree(bp);
5311         kfree(bp);
5312 }
5313
5314 static struct pernet_operations ipv6_inetpeer_ops = {
5315         .init   =       ipv6_inetpeer_init,
5316         .exit   =       ipv6_inetpeer_exit,
5317 };
5318
5319 static struct pernet_operations ip6_route_net_late_ops = {
5320         .init = ip6_route_net_init_late,
5321         .exit = ip6_route_net_exit_late,
5322 };
5323
5324 static struct notifier_block ip6_route_dev_notifier = {
5325         .notifier_call = ip6_route_dev_notify,
5326         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
5327 };
5328
5329 void __init ip6_route_init_special_entries(void)
5330 {
5331         /* Registering of the loopback is done before this portion of code,
5332          * the loopback reference in rt6_info will not be taken, do it
5333          * manually for init_net */
5334         init_net.ipv6.fib6_null_entry->fib6_nh.nh_dev = init_net.loopback_dev;
5335         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
5336         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5337   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
5338         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
5339         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5340         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
5341         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
5342   #endif
5343 }
5344
5345 int __init ip6_route_init(void)
5346 {
5347         int ret;
5348         int cpu;
5349
5350         ret = -ENOMEM;
5351         ip6_dst_ops_template.kmem_cachep =
5352                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
5353                                   SLAB_HWCACHE_ALIGN, NULL);
5354         if (!ip6_dst_ops_template.kmem_cachep)
5355                 goto out;
5356
5357         ret = dst_entries_init(&ip6_dst_blackhole_ops);
5358         if (ret)
5359                 goto out_kmem_cache;
5360
5361         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
5362         if (ret)
5363                 goto out_dst_entries;
5364
5365         ret = register_pernet_subsys(&ip6_route_net_ops);
5366         if (ret)
5367                 goto out_register_inetpeer;
5368
5369         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
5370
5371         ret = fib6_init();
5372         if (ret)
5373                 goto out_register_subsys;
5374
5375         ret = xfrm6_init();
5376         if (ret)
5377                 goto out_fib6_init;
5378
5379         ret = fib6_rules_init();
5380         if (ret)
5381                 goto xfrm6_init;
5382
5383         ret = register_pernet_subsys(&ip6_route_net_late_ops);
5384         if (ret)
5385                 goto fib6_rules_init;
5386
5387         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
5388                                    inet6_rtm_newroute, NULL, 0);
5389         if (ret < 0)
5390                 goto out_register_late_subsys;
5391
5392         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
5393                                    inet6_rtm_delroute, NULL, 0);
5394         if (ret < 0)
5395                 goto out_register_late_subsys;
5396
5397         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
5398                                    inet6_rtm_getroute, NULL,
5399                                    RTNL_FLAG_DOIT_UNLOCKED);
5400         if (ret < 0)
5401                 goto out_register_late_subsys;
5402
5403         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
5404         if (ret)
5405                 goto out_register_late_subsys;
5406
5407         for_each_possible_cpu(cpu) {
5408                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
5409
5410                 INIT_LIST_HEAD(&ul->head);
5411                 spin_lock_init(&ul->lock);
5412         }
5413
5414 out:
5415         return ret;
5416
5417 out_register_late_subsys:
5418         rtnl_unregister_all(PF_INET6);
5419         unregister_pernet_subsys(&ip6_route_net_late_ops);
5420 fib6_rules_init:
5421         fib6_rules_cleanup();
5422 xfrm6_init:
5423         xfrm6_fini();
5424 out_fib6_init:
5425         fib6_gc_cleanup();
5426 out_register_subsys:
5427         unregister_pernet_subsys(&ip6_route_net_ops);
5428 out_register_inetpeer:
5429         unregister_pernet_subsys(&ipv6_inetpeer_ops);
5430 out_dst_entries:
5431         dst_entries_destroy(&ip6_dst_blackhole_ops);
5432 out_kmem_cache:
5433         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
5434         goto out;
5435 }
5436
5437 void ip6_route_cleanup(void)
5438 {
5439         unregister_netdevice_notifier(&ip6_route_dev_notifier);
5440         unregister_pernet_subsys(&ip6_route_net_late_ops);
5441         fib6_rules_cleanup();
5442         xfrm6_fini();
5443         fib6_gc_cleanup();
5444         unregister_pernet_subsys(&ipv6_inetpeer_ops);
5445         unregister_pernet_subsys(&ip6_route_net_ops);
5446         dst_entries_destroy(&ip6_dst_blackhole_ops);
5447         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
5448 }