net/ipv6/route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *      Linux INET6 implementation
   4  *      FIB front-end.
   5  *
   6  *      Authors:
   7  *      Pedro Roque             <[email protected]>
   8  */
   9
  10 /*      Changes:
  11  *
  12  *      YOSHIFUJI Hideaki @USAGI
  13  *              reworked default router selection.
  14  *              - respect outgoing interface
  15  *              - select from (probably) reachable routers (i.e.
  16  *              routers in REACHABLE, STALE, DELAY or PROBE states).
  17  *              - always select the same router if it is (probably)
  18  *              reachable.  otherwise, round-robin the list.
  19  *      Ville Nuorvala
  20  *              Fixed routing subtrees.
  21  */
  22
  23 #define pr_fmt(fmt) "IPv6: " fmt
  24
  25 #include <linux/capability.h>
  26 #include <linux/errno.h>
  27 #include <linux/export.h>
  28 #include <linux/types.h>
  29 #include <linux/times.h>
  30 #include <linux/socket.h>
  31 #include <linux/sockios.h>
  32 #include <linux/net.h>
  33 #include <linux/route.h>
  34 #include <linux/netdevice.h>
  35 #include <linux/in6.h>
  36 #include <linux/mroute6.h>
  37 #include <linux/init.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/proc_fs.h>
  40 #include <linux/seq_file.h>
  41 #include <linux/nsproxy.h>
  42 #include <linux/slab.h>
  43 #include <linux/jhash.h>
  44 #include <net/net_namespace.h>
  45 #include <net/snmp.h>
  46 #include <net/ipv6.h>
  47 #include <net/ip6_fib.h>
  48 #include <net/ip6_route.h>
  49 #include <net/ndisc.h>
  50 #include <net/addrconf.h>
  51 #include <net/tcp.h>
  52 #include <linux/rtnetlink.h>
  53 #include <net/dst.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/xfrm.h>
  56 #include <net/netevent.h>
  57 #include <net/netlink.h>
  58 #include <net/rtnh.h>
  59 #include <net/lwtunnel.h>
  60 #include <net/ip_tunnels.h>
  61 #include <net/l3mdev.h>
  62 #include <net/ip.h>
  63 #include <linux/uaccess.h>
  64
  65 #ifdef CONFIG_SYSCTL
  66 #include <linux/sysctl.h>
  67 #endif
  68
  69 static int ip6_rt_type_to_error(u8 fib6_type);
  70
  71 #define CREATE_TRACE_POINTS
  72 #include <trace/events/fib6.h>
  73 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  74 #undef CREATE_TRACE_POINTS
  75
  76 enum rt6_nud_state {
  77         RT6_NUD_FAIL_HARD = -3,
  78         RT6_NUD_FAIL_PROBE = -2,
  79         RT6_NUD_FAIL_DO_RR = -1,
  80         RT6_NUD_SUCCEED = 1
  81 };
  82
  83 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  84 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  85 static unsigned int      ip6_mtu(const struct dst_entry *dst);
  86 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  87 static void             ip6_dst_destroy(struct dst_entry *);
  88 static void             ip6_dst_ifdown(struct dst_entry *,
  89                                        struct net_device *dev, int how);
  90 static int               ip6_dst_gc(struct dst_ops *ops);
  91
  92 static int              ip6_pkt_discard(struct sk_buff *skb);
  93 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  94 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  95 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  96 static void             ip6_link_failure(struct sk_buff *skb);
  97 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
  98                                            struct sk_buff *skb, u32 mtu,
  99                                            bool confirm_neigh);
 100 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 101                                         struct sk_buff *skb);
 102 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 103                            int strict);
 104 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 105 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 106                          struct fib6_info *rt, struct dst_entry *dst,
 107                          struct in6_addr *dest, struct in6_addr *src,
 108                          int iif, int type, u32 portid, u32 seq,
 109                          unsigned int flags);
 110 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 111                                            const struct in6_addr *daddr,
 112                                            const struct in6_addr *saddr);
 113
 114 #ifdef CONFIG_IPV6_ROUTE_INFO
 115 static struct fib6_info *rt6_add_route_info(struct net *net,
 116                                            const struct in6_addr *prefix, int prefixlen,
 117                                            const struct in6_addr *gwaddr,
 118                                            struct net_device *dev,
 119                                            unsigned int pref);
 120 static struct fib6_info *rt6_get_route_info(struct net *net,
 121                                            const struct in6_addr *prefix, int prefixlen,
 122                                            const struct in6_addr *gwaddr,
 123                                            struct net_device *dev);
 124 #endif
 125
 126 struct uncached_list {
 127         spinlock_t              lock;
 128         struct list_head        head;
 129 };
 130
 131 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 132
 133 void rt6_uncached_list_add(struct rt6_info *rt)
 134 {
 135         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 136
 137         rt->rt6i_uncached_list = ul;
 138
 139         spin_lock_bh(&ul->lock);
 140         list_add_tail(&rt->rt6i_uncached, &ul->head);
 141         spin_unlock_bh(&ul->lock);
 142 }
 143
 144 void rt6_uncached_list_del(struct rt6_info *rt)
 145 {
 146         if (!list_empty(&rt->rt6i_uncached)) {
 147                 struct uncached_list *ul = rt->rt6i_uncached_list;
 148                 struct net *net = dev_net(rt->dst.dev);
 149
 150                 spin_lock_bh(&ul->lock);
 151                 list_del(&rt->rt6i_uncached);
 152                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 153                 spin_unlock_bh(&ul->lock);
 154         }
 155 }
 156
 157 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 158 {
 159         struct net_device *loopback_dev = net->loopback_dev;
 160         int cpu;
 161
 162         if (dev == loopback_dev)
 163                 return;
 164
 165         for_each_possible_cpu(cpu) {
 166                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 167                 struct rt6_info *rt;
 168
 169                 spin_lock_bh(&ul->lock);
 170                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 171                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 172                         struct net_device *rt_dev = rt->dst.dev;
 173
 174                         if (rt_idev->dev == dev) {
 175                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 176                                 in6_dev_put(rt_idev);
 177                         }
 178
 179                         if (rt_dev == dev) {
 180                                 rt->dst.dev = blackhole_netdev;
 181                                 dev_hold(rt->dst.dev);
 182                                 dev_put(rt_dev);
 183                         }
 184                 }
 185                 spin_unlock_bh(&ul->lock);
 186         }
 187 }
 188
 189 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 190                                              struct sk_buff *skb,
 191                                              const void *daddr)
 192 {
 193         if (!ipv6_addr_any(p))
 194                 return (const void *) p;
 195         else if (skb)
 196                 return &ipv6_hdr(skb)->daddr;
 197         return daddr;
 198 }
 199
 200 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 201                                    struct net_device *dev,
 202                                    struct sk_buff *skb,
 203                                    const void *daddr)
 204 {
 205         struct neighbour *n;
 206
 207         daddr = choose_neigh_daddr(gw, skb, daddr);
 208         n = __ipv6_neigh_lookup(dev, daddr);
 209         if (n)
 210                 return n;
 211
 212         n = neigh_create(&nd_tbl, daddr, dev);
 213         return IS_ERR(n) ? NULL : n;
 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(rt6_nexthop(rt, &in6addr_any),
 223                                 dst->dev, skb, daddr);
 224 }
 225
 226 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 227 {
 228         struct net_device *dev = dst->dev;
 229         struct rt6_info *rt = (struct rt6_info *)dst;
 230
 231         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 232         if (!daddr)
 233                 return;
 234         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 235                 return;
 236         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 237                 return;
 238         __ipv6_confirm_neigh(dev, daddr);
 239 }
 240
 241 static struct dst_ops ip6_dst_ops_template = {
 242         .family                 =       AF_INET6,
 243         .gc                     =       ip6_dst_gc,
 244         .gc_thresh              =       1024,
 245         .check                  =       ip6_dst_check,
 246         .default_advmss         =       ip6_default_advmss,
 247         .mtu                    =       ip6_mtu,
 248         .cow_metrics            =       dst_cow_metrics_generic,
 249         .destroy                =       ip6_dst_destroy,
 250         .ifdown                 =       ip6_dst_ifdown,
 251         .negative_advice        =       ip6_negative_advice,
 252         .link_failure           =       ip6_link_failure,
 253         .update_pmtu            =       ip6_rt_update_pmtu,
 254         .redirect               =       rt6_do_redirect,
 255         .local_out              =       __ip6_local_out,
 256         .neigh_lookup           =       ip6_dst_neigh_lookup,
 257         .confirm_neigh          =       ip6_confirm_neigh,
 258 };
 259
 260 static unsigned int ip6_blackhole_mtu(const struct dst_entry *dst)
 261 {
 262         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 263
 264         return mtu ? : dst->dev->mtu;
 265 }
 266
 267 static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, struct sock *sk,
 268                                          struct sk_buff *skb, u32 mtu,
 269                                          bool confirm_neigh)
 270 {
 271 }
 272
 273 static void ip6_rt_blackhole_redirect(struct dst_entry *dst, struct sock *sk,
 274                                       struct sk_buff *skb)
 275 {
 276 }
 277
 278 static struct dst_ops ip6_dst_blackhole_ops = {
 279         .family                 =       AF_INET6,
 280         .destroy                =       ip6_dst_destroy,
 281         .check                  =       ip6_dst_check,
 282         .mtu                    =       ip6_blackhole_mtu,
 283         .default_advmss         =       ip6_default_advmss,
 284         .update_pmtu            =       ip6_rt_blackhole_update_pmtu,
 285         .redirect               =       ip6_rt_blackhole_redirect,
 286         .cow_metrics            =       dst_cow_metrics_generic,
 287         .neigh_lookup           =       ip6_dst_neigh_lookup,
 288 };
 289
 290 static const u32 ip6_template_metrics[RTAX_MAX] = {
 291         [RTAX_HOPLIMIT - 1] = 0,
 292 };
 293
 294 static const struct fib6_info fib6_null_entry_template = {
 295         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 296         .fib6_protocol  = RTPROT_KERNEL,
 297         .fib6_metric    = ~(u32)0,
 298         .fib6_ref       = REFCOUNT_INIT(1),
 299         .fib6_type      = RTN_UNREACHABLE,
 300         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 301 };
 302
 303 static const struct rt6_info ip6_null_entry_template = {
 304         .dst = {
 305                 .__refcnt       = ATOMIC_INIT(1),
 306                 .__use          = 1,
 307                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 308                 .error          = -ENETUNREACH,
 309                 .input          = ip6_pkt_discard,
 310                 .output         = ip6_pkt_discard_out,
 311         },
 312         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 313 };
 314
 315 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 316
 317 static const struct rt6_info ip6_prohibit_entry_template = {
 318         .dst = {
 319                 .__refcnt       = ATOMIC_INIT(1),
 320                 .__use          = 1,
 321                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 322                 .error          = -EACCES,
 323                 .input          = ip6_pkt_prohibit,
 324                 .output         = ip6_pkt_prohibit_out,
 325         },
 326         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 327 };
 328
 329 static const struct rt6_info ip6_blk_hole_entry_template = {
 330         .dst = {
 331                 .__refcnt       = ATOMIC_INIT(1),
 332                 .__use          = 1,
 333                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 334                 .error          = -EINVAL,
 335                 .input          = dst_discard,
 336                 .output         = dst_discard_out,
 337         },
 338         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 339 };
 340
 341 #endif
 342
 343 static void rt6_info_init(struct rt6_info *rt)
 344 {
 345         struct dst_entry *dst = &rt->dst;
 346
 347         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 348         INIT_LIST_HEAD(&rt->rt6i_uncached);
 349 }
 350
 351 /* allocate dst with ip6_dst_ops */
 352 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 353                                int flags)
 354 {
 355         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 356                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 357
 358         if (rt) {
 359                 rt6_info_init(rt);
 360                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 361         }
 362
 363         return rt;
 364 }
 365 EXPORT_SYMBOL(ip6_dst_alloc);
 366
 367 static void ip6_dst_destroy(struct dst_entry *dst)
 368 {
 369         struct rt6_info *rt = (struct rt6_info *)dst;
 370         struct fib6_info *from;
 371         struct inet6_dev *idev;
 372
 373         ip_dst_metrics_put(dst);
 374         rt6_uncached_list_del(rt);
 375
 376         idev = rt->rt6i_idev;
 377         if (idev) {
 378                 rt->rt6i_idev = NULL;
 379                 in6_dev_put(idev);
 380         }
 381
 382         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 383         fib6_info_release(from);
 384 }
 385
 386 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 387                            int how)
 388 {
 389         struct rt6_info *rt = (struct rt6_info *)dst;
 390         struct inet6_dev *idev = rt->rt6i_idev;
 391         struct net_device *loopback_dev =
 392                 dev_net(dev)->loopback_dev;
 393
 394         if (idev && idev->dev != loopback_dev) {
 395                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 396                 if (loopback_idev) {
 397                         rt->rt6i_idev = loopback_idev;
 398                         in6_dev_put(idev);
 399                 }
 400         }
 401 }
 402
 403 static bool __rt6_check_expired(const struct rt6_info *rt)
 404 {
 405         if (rt->rt6i_flags & RTF_EXPIRES)
 406                 return time_after(jiffies, rt->dst.expires);
 407         else
 408                 return false;
 409 }
 410
 411 static bool rt6_check_expired(const struct rt6_info *rt)
 412 {
 413         struct fib6_info *from;
 414
 415         from = rcu_dereference(rt->from);
 416
 417         if (rt->rt6i_flags & RTF_EXPIRES) {
 418                 if (time_after(jiffies, rt->dst.expires))
 419                         return true;
 420         } else if (from) {
 421                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 422                         fib6_check_expired(from);
 423         }
 424         return false;
 425 }
 426
 427 void fib6_select_path(const struct net *net, struct fib6_result *res,
 428                       struct flowi6 *fl6, int oif, bool have_oif_match,
 429                       const struct sk_buff *skb, int strict)
 430 {
 431         struct fib6_info *sibling, *next_sibling;
 432         struct fib6_info *match = res->f6i;
 433
 434         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 435                 goto out;
 436
 437         if (match->nh && have_oif_match && res->nh)
 438                 return;
 439
 440         /* We might have already computed the hash for ICMPv6 errors. In such
 441          * case it will always be non-zero. Otherwise now is the time to do it.
 442          */
 443         if (!fl6->mp_hash &&
 444             (!match->nh || nexthop_is_multipath(match->nh)))
 445                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 446
 447         if (unlikely(match->nh)) {
 448                 nexthop_path_fib6_result(res, fl6->mp_hash);
 449                 return;
 450         }
 451
 452         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 453                 goto out;
 454
 455         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 456                                  fib6_siblings) {
 457                 const struct fib6_nh *nh = sibling->fib6_nh;
 458                 int nh_upper_bound;
 459
 460                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 461                 if (fl6->mp_hash > nh_upper_bound)
 462                         continue;
 463                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 464                         break;
 465                 match = sibling;
 466                 break;
 467         }
 468
 469 out:
 470         res->f6i = match;
 471         res->nh = match->fib6_nh;
 472 }
 473
 474 /*
 475  *      Route lookup. rcu_read_lock() should be held.
 476  */
 477
 478 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 479                                const struct in6_addr *saddr, int oif, int flags)
 480 {
 481         const struct net_device *dev;
 482
 483         if (nh->fib_nh_flags & RTNH_F_DEAD)
 484                 return false;
 485
 486         dev = nh->fib_nh_dev;
 487         if (oif) {
 488                 if (dev->ifindex == oif)
 489                         return true;
 490         } else {
 491                 if (ipv6_chk_addr(net, saddr, dev,
 492                                   flags & RT6_LOOKUP_F_IFACE))
 493                         return true;
 494         }
 495
 496         return false;
 497 }
 498
 499 struct fib6_nh_dm_arg {
 500         struct net              *net;
 501         const struct in6_addr   *saddr;
 502         int                     oif;
 503         int                     flags;
 504         struct fib6_nh          *nh;
 505 };
 506
 507 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 508 {
 509         struct fib6_nh_dm_arg *arg = _arg;
 510
 511         arg->nh = nh;
 512         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 513                                   arg->flags);
 514 }
 515
 516 /* returns fib6_nh from nexthop or NULL */
 517 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 518                                         struct fib6_result *res,
 519                                         const struct in6_addr *saddr,
 520                                         int oif, int flags)
 521 {
 522         struct fib6_nh_dm_arg arg = {
 523                 .net   = net,
 524                 .saddr = saddr,
 525                 .oif   = oif,
 526                 .flags = flags,
 527         };
 528
 529         if (nexthop_is_blackhole(nh))
 530                 return NULL;
 531
 532         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 533                 return arg.nh;
 534
 535         return NULL;
 536 }
 537
 538 static void rt6_device_match(struct net *net, struct fib6_result *res,
 539                              const struct in6_addr *saddr, int oif, int flags)
 540 {
 541         struct fib6_info *f6i = res->f6i;
 542         struct fib6_info *spf6i;
 543         struct fib6_nh *nh;
 544
 545         if (!oif && ipv6_addr_any(saddr)) {
 546                 if (unlikely(f6i->nh)) {
 547                         nh = nexthop_fib6_nh(f6i->nh);
 548                         if (nexthop_is_blackhole(f6i->nh))
 549                                 goto out_blackhole;
 550                 } else {
 551                         nh = f6i->fib6_nh;
 552                 }
 553                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 554                         goto out;
 555         }
 556
 557         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 558                 bool matched = false;
 559
 560                 if (unlikely(spf6i->nh)) {
 561                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 562                                               oif, flags);
 563                         if (nh)
 564                                 matched = true;
 565                 } else {
 566                         nh = spf6i->fib6_nh;
 567                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 568                                 matched = true;
 569                 }
 570                 if (matched) {
 571                         res->f6i = spf6i;
 572                         goto out;
 573                 }
 574         }
 575
 576         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 577                 res->f6i = net->ipv6.fib6_null_entry;
 578                 nh = res->f6i->fib6_nh;
 579                 goto out;
 580         }
 581
 582         if (unlikely(f6i->nh)) {
 583                 nh = nexthop_fib6_nh(f6i->nh);
 584                 if (nexthop_is_blackhole(f6i->nh))
 585                         goto out_blackhole;
 586         } else {
 587                 nh = f6i->fib6_nh;
 588         }
 589
 590         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 591                 res->f6i = net->ipv6.fib6_null_entry;
 592                 nh = res->f6i->fib6_nh;
 593         }
 594 out:
 595         res->nh = nh;
 596         res->fib6_type = res->f6i->fib6_type;
 597         res->fib6_flags = res->f6i->fib6_flags;
 598         return;
 599
 600 out_blackhole:
 601         res->fib6_flags |= RTF_REJECT;
 602         res->fib6_type = RTN_BLACKHOLE;
 603         res->nh = nh;
 604 }
 605
 606 #ifdef CONFIG_IPV6_ROUTER_PREF
 607 struct __rt6_probe_work {
 608         struct work_struct work;
 609         struct in6_addr target;
 610         struct net_device *dev;
 611 };
 612
 613 static void rt6_probe_deferred(struct work_struct *w)
 614 {
 615         struct in6_addr mcaddr;
 616         struct __rt6_probe_work *work =
 617                 container_of(w, struct __rt6_probe_work, work);
 618
 619         addrconf_addr_solict_mult(&work->target, &mcaddr);
 620         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 621         dev_put(work->dev);
 622         kfree(work);
 623 }
 624
 625 static void rt6_probe(struct fib6_nh *fib6_nh)
 626 {
 627         struct __rt6_probe_work *work = NULL;
 628         const struct in6_addr *nh_gw;
 629         unsigned long last_probe;
 630         struct neighbour *neigh;
 631         struct net_device *dev;
 632         struct inet6_dev *idev;
 633
 634         /*
 635          * Okay, this does not seem to be appropriate
 636          * for now, however, we need to check if it
 637          * is really so; aka Router Reachability Probing.
 638          *
 639          * Router Reachability Probe MUST be rate-limited
 640          * to no more than one per minute.
 641          */
 642         if (!fib6_nh->fib_nh_gw_family)
 643                 return;
 644
 645         nh_gw = &fib6_nh->fib_nh_gw6;
 646         dev = fib6_nh->fib_nh_dev;
 647         rcu_read_lock_bh();
 648         last_probe = READ_ONCE(fib6_nh->last_probe);
 649         idev = __in6_dev_get(dev);
 650         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 651         if (neigh) {
 652                 if (neigh->nud_state & NUD_VALID)
 653                         goto out;
 654
 655                 write_lock(&neigh->lock);
 656                 if (!(neigh->nud_state & NUD_VALID) &&
 657                     time_after(jiffies,
 658                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 659                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 660                         if (work)
 661                                 __neigh_set_probe_once(neigh);
 662                 }
 663                 write_unlock(&neigh->lock);
 664         } else if (time_after(jiffies, last_probe +
 665                                        idev->cnf.rtr_probe_interval)) {
 666                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 667         }
 668
 669         if (!work || cmpxchg(&fib6_nh->last_probe,
 670                              last_probe, jiffies) != last_probe) {
 671                 kfree(work);
 672         } else {
 673                 INIT_WORK(&work->work, rt6_probe_deferred);
 674                 work->target = *nh_gw;
 675                 dev_hold(dev);
 676                 work->dev = dev;
 677                 schedule_work(&work->work);
 678         }
 679
 680 out:
 681         rcu_read_unlock_bh();
 682 }
 683 #else
 684 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 685 {
 686 }
 687 #endif
 688
 689 /*
 690  * Default Router Selection (RFC 2461 6.3.6)
 691  */
 692 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 693 {
 694         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 695         struct neighbour *neigh;
 696
 697         rcu_read_lock_bh();
 698         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 699                                           &fib6_nh->fib_nh_gw6);
 700         if (neigh) {
 701                 read_lock(&neigh->lock);
 702                 if (neigh->nud_state & NUD_VALID)
 703                         ret = RT6_NUD_SUCCEED;
 704 #ifdef CONFIG_IPV6_ROUTER_PREF
 705                 else if (!(neigh->nud_state & NUD_FAILED))
 706                         ret = RT6_NUD_SUCCEED;
 707                 else
 708                         ret = RT6_NUD_FAIL_PROBE;
 709 #endif
 710                 read_unlock(&neigh->lock);
 711         } else {
 712                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 713                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 714         }
 715         rcu_read_unlock_bh();
 716
 717         return ret;
 718 }
 719
 720 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 721                            int strict)
 722 {
 723         int m = 0;
 724
 725         if (!oif || nh->fib_nh_dev->ifindex == oif)
 726                 m = 2;
 727
 728         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 729                 return RT6_NUD_FAIL_HARD;
 730 #ifdef CONFIG_IPV6_ROUTER_PREF
 731         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 732 #endif
 733         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 734             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 735                 int n = rt6_check_neigh(nh);
 736                 if (n < 0)
 737                         return n;
 738         }
 739         return m;
 740 }
 741
 742 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 743                        int oif, int strict, int *mpri, bool *do_rr)
 744 {
 745         bool match_do_rr = false;
 746         bool rc = false;
 747         int m;
 748
 749         if (nh->fib_nh_flags & RTNH_F_DEAD)
 750                 goto out;
 751
 752         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 753             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 754             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 755                 goto out;
 756
 757         m = rt6_score_route(nh, fib6_flags, oif, strict);
 758         if (m == RT6_NUD_FAIL_DO_RR) {
 759                 match_do_rr = true;
 760                 m = 0; /* lowest valid score */
 761         } else if (m == RT6_NUD_FAIL_HARD) {
 762                 goto out;
 763         }
 764
 765         if (strict & RT6_LOOKUP_F_REACHABLE)
 766                 rt6_probe(nh);
 767
 768         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 769         if (m > *mpri) {
 770                 *do_rr = match_do_rr;
 771                 *mpri = m;
 772                 rc = true;
 773         }
 774 out:
 775         return rc;
 776 }
 777
 778 struct fib6_nh_frl_arg {
 779         u32             flags;
 780         int             oif;
 781         int             strict;
 782         int             *mpri;
 783         bool            *do_rr;
 784         struct fib6_nh  *nh;
 785 };
 786
 787 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 788 {
 789         struct fib6_nh_frl_arg *arg = _arg;
 790
 791         arg->nh = nh;
 792         return find_match(nh, arg->flags, arg->oif, arg->strict,
 793                           arg->mpri, arg->do_rr);
 794 }
 795
 796 static void __find_rr_leaf(struct fib6_info *f6i_start,
 797                            struct fib6_info *nomatch, u32 metric,
 798                            struct fib6_result *res, struct fib6_info **cont,
 799                            int oif, int strict, bool *do_rr, int *mpri)
 800 {
 801         struct fib6_info *f6i;
 802
 803         for (f6i = f6i_start;
 804              f6i && f6i != nomatch;
 805              f6i = rcu_dereference(f6i->fib6_next)) {
 806                 bool matched = false;
 807                 struct fib6_nh *nh;
 808
 809                 if (cont && f6i->fib6_metric != metric) {
 810                         *cont = f6i;
 811                         return;
 812                 }
 813
 814                 if (fib6_check_expired(f6i))
 815                         continue;
 816
 817                 if (unlikely(f6i->nh)) {
 818                         struct fib6_nh_frl_arg arg = {
 819                                 .flags  = f6i->fib6_flags,
 820                                 .oif    = oif,
 821                                 .strict = strict,
 822                                 .mpri   = mpri,
 823                                 .do_rr  = do_rr
 824                         };
 825
 826                         if (nexthop_is_blackhole(f6i->nh)) {
 827                                 res->fib6_flags = RTF_REJECT;
 828                                 res->fib6_type = RTN_BLACKHOLE;
 829                                 res->f6i = f6i;
 830                                 res->nh = nexthop_fib6_nh(f6i->nh);
 831                                 return;
 832                         }
 833                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 834                                                      &arg)) {
 835                                 matched = true;
 836                                 nh = arg.nh;
 837                         }
 838                 } else {
 839                         nh = f6i->fib6_nh;
 840                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 841                                        mpri, do_rr))
 842                                 matched = true;
 843                 }
 844                 if (matched) {
 845                         res->f6i = f6i;
 846                         res->nh = nh;
 847                         res->fib6_flags = f6i->fib6_flags;
 848                         res->fib6_type = f6i->fib6_type;
 849                 }
 850         }
 851 }
 852
 853 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 854                          struct fib6_info *rr_head, int oif, int strict,
 855                          bool *do_rr, struct fib6_result *res)
 856 {
 857         u32 metric = rr_head->fib6_metric;
 858         struct fib6_info *cont = NULL;
 859         int mpri = -1;
 860
 861         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 862                        oif, strict, do_rr, &mpri);
 863
 864         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 865                        oif, strict, do_rr, &mpri);
 866
 867         if (res->f6i || !cont)
 868                 return;
 869
 870         __find_rr_leaf(cont, NULL, metric, res, NULL,
 871                        oif, strict, do_rr, &mpri);
 872 }
 873
 874 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 875                        struct fib6_result *res, int strict)
 876 {
 877         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 878         struct fib6_info *rt0;
 879         bool do_rr = false;
 880         int key_plen;
 881
 882         /* make sure this function or its helpers sets f6i */
 883         res->f6i = NULL;
 884
 885         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 886                 goto out;
 887
 888         rt0 = rcu_dereference(fn->rr_ptr);
 889         if (!rt0)
 890                 rt0 = leaf;
 891
 892         /* Double check to make sure fn is not an intermediate node
 893          * and fn->leaf does not points to its child's leaf
 894          * (This might happen if all routes under fn are deleted from
 895          * the tree and fib6_repair_tree() is called on the node.)
 896          */
 897         key_plen = rt0->fib6_dst.plen;
 898 #ifdef CONFIG_IPV6_SUBTREES
 899         if (rt0->fib6_src.plen)
 900                 key_plen = rt0->fib6_src.plen;
 901 #endif
 902         if (fn->fn_bit != key_plen)
 903                 goto out;
 904
 905         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 906         if (do_rr) {
 907                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 908
 909                 /* no entries matched; do round-robin */
 910                 if (!next || next->fib6_metric != rt0->fib6_metric)
 911                         next = leaf;
 912
 913                 if (next != rt0) {
 914                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 915                         /* make sure next is not being deleted from the tree */
 916                         if (next->fib6_node)
 917                                 rcu_assign_pointer(fn->rr_ptr, next);
 918                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 919                 }
 920         }
 921
 922 out:
 923         if (!res->f6i) {
 924                 res->f6i = net->ipv6.fib6_null_entry;
 925                 res->nh = res->f6i->fib6_nh;
 926                 res->fib6_flags = res->f6i->fib6_flags;
 927                 res->fib6_type = res->f6i->fib6_type;
 928         }
 929 }
 930
 931 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 932 {
 933         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 934                res->nh->fib_nh_gw_family;
 935 }
 936
 937 #ifdef CONFIG_IPV6_ROUTE_INFO
 938 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 939                   const struct in6_addr *gwaddr)
 940 {
 941         struct net *net = dev_net(dev);
 942         struct route_info *rinfo = (struct route_info *) opt;
 943         struct in6_addr prefix_buf, *prefix;
 944         unsigned int pref;
 945         unsigned long lifetime;
 946         struct fib6_info *rt;
 947
 948         if (len < sizeof(struct route_info)) {
 949                 return -EINVAL;
 950         }
 951
 952         /* Sanity check for prefix_len and length */
 953         if (rinfo->length > 3) {
 954                 return -EINVAL;
 955         } else if (rinfo->prefix_len > 128) {
 956                 return -EINVAL;
 957         } else if (rinfo->prefix_len > 64) {
 958                 if (rinfo->length < 2) {
 959                         return -EINVAL;
 960                 }
 961         } else if (rinfo->prefix_len > 0) {
 962                 if (rinfo->length < 1) {
 963                         return -EINVAL;
 964                 }
 965         }
 966
 967         pref = rinfo->route_pref;
 968         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 969                 return -EINVAL;
 970
 971         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 972
 973         if (rinfo->length == 3)
 974                 prefix = (struct in6_addr *)rinfo->prefix;
 975         else {
 976                 /* this function is safe */
 977                 ipv6_addr_prefix(&prefix_buf,
 978                                  (struct in6_addr *)rinfo->prefix,
 979                                  rinfo->prefix_len);
 980                 prefix = &prefix_buf;
 981         }
 982
 983         if (rinfo->prefix_len == 0)
 984                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 985         else
 986                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 987                                         gwaddr, dev);
 988
 989         if (rt && !lifetime) {
 990                 ip6_del_rt(net, rt, false);
 991                 rt = NULL;
 992         }
 993
 994         if (!rt && lifetime)
 995                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 996                                         dev, pref);
 997         else if (rt)
 998                 rt->fib6_flags = RTF_ROUTEINFO |
 999                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
1000
1001         if (rt) {
1002                 if (!addrconf_finite_timeout(lifetime))
1003                         fib6_clean_expires(rt);
1004                 else
1005                         fib6_set_expires(rt, jiffies + HZ * lifetime);
1006
1007                 fib6_info_release(rt);
1008         }
1009         return 0;
1010 }
1011 #endif
1012
1013 /*
1014  *      Misc support functions
1015  */
1016
1017 /* called with rcu_lock held */
1018 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1019 {
1020         struct net_device *dev = res->nh->fib_nh_dev;
1021
1022         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1023                 /* for copies of local routes, dst->dev needs to be the
1024                  * device if it is a master device, the master device if
1025                  * device is enslaved, and the loopback as the default
1026                  */
1027                 if (netif_is_l3_slave(dev) &&
1028                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1029                         dev = l3mdev_master_dev_rcu(dev);
1030                 else if (!netif_is_l3_master(dev))
1031                         dev = dev_net(dev)->loopback_dev;
1032                 /* last case is netif_is_l3_master(dev) is true in which
1033                  * case we want dev returned to be dev
1034                  */
1035         }
1036
1037         return dev;
1038 }
1039
1040 static const int fib6_prop[RTN_MAX + 1] = {
1041         [RTN_UNSPEC]    = 0,
1042         [RTN_UNICAST]   = 0,
1043         [RTN_LOCAL]     = 0,
1044         [RTN_BROADCAST] = 0,
1045         [RTN_ANYCAST]   = 0,
1046         [RTN_MULTICAST] = 0,
1047         [RTN_BLACKHOLE] = -EINVAL,
1048         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1049         [RTN_PROHIBIT]  = -EACCES,
1050         [RTN_THROW]     = -EAGAIN,
1051         [RTN_NAT]       = -EINVAL,
1052         [RTN_XRESOLVE]  = -EINVAL,
1053 };
1054
1055 static int ip6_rt_type_to_error(u8 fib6_type)
1056 {
1057         return fib6_prop[fib6_type];
1058 }
1059
1060 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1061 {
1062         unsigned short flags = 0;
1063
1064         if (rt->dst_nocount)
1065                 flags |= DST_NOCOUNT;
1066         if (rt->dst_nopolicy)
1067                 flags |= DST_NOPOLICY;
1068
1069         return flags;
1070 }
1071
1072 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1073 {
1074         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1075
1076         switch (fib6_type) {
1077         case RTN_BLACKHOLE:
1078                 rt->dst.output = dst_discard_out;
1079                 rt->dst.input = dst_discard;
1080                 break;
1081         case RTN_PROHIBIT:
1082                 rt->dst.output = ip6_pkt_prohibit_out;
1083                 rt->dst.input = ip6_pkt_prohibit;
1084                 break;
1085         case RTN_THROW:
1086         case RTN_UNREACHABLE:
1087         default:
1088                 rt->dst.output = ip6_pkt_discard_out;
1089                 rt->dst.input = ip6_pkt_discard;
1090                 break;
1091         }
1092 }
1093
1094 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1095 {
1096         struct fib6_info *f6i = res->f6i;
1097
1098         if (res->fib6_flags & RTF_REJECT) {
1099                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1100                 return;
1101         }
1102
1103         rt->dst.error = 0;
1104         rt->dst.output = ip6_output;
1105
1106         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1107                 rt->dst.input = ip6_input;
1108         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1109                 rt->dst.input = ip6_mc_input;
1110         } else {
1111                 rt->dst.input = ip6_forward;
1112         }
1113
1114         if (res->nh->fib_nh_lws) {
1115                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1116                 lwtunnel_set_redirect(&rt->dst);
1117         }
1118
1119         rt->dst.lastuse = jiffies;
1120 }
1121
1122 /* Caller must already hold reference to @from */
1123 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1124 {
1125         rt->rt6i_flags &= ~RTF_EXPIRES;
1126         rcu_assign_pointer(rt->from, from);
1127         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1128 }
1129
1130 /* Caller must already hold reference to f6i in result */
1131 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1132 {
1133         const struct fib6_nh *nh = res->nh;
1134         const struct net_device *dev = nh->fib_nh_dev;
1135         struct fib6_info *f6i = res->f6i;
1136
1137         ip6_rt_init_dst(rt, res);
1138
1139         rt->rt6i_dst = f6i->fib6_dst;
1140         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1141         rt->rt6i_flags = res->fib6_flags;
1142         if (nh->fib_nh_gw_family) {
1143                 rt->rt6i_gateway = nh->fib_nh_gw6;
1144                 rt->rt6i_flags |= RTF_GATEWAY;
1145         }
1146         rt6_set_from(rt, f6i);
1147 #ifdef CONFIG_IPV6_SUBTREES
1148         rt->rt6i_src = f6i->fib6_src;
1149 #endif
1150 }
1151
1152 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1153                                         struct in6_addr *saddr)
1154 {
1155         struct fib6_node *pn, *sn;
1156         while (1) {
1157                 if (fn->fn_flags & RTN_TL_ROOT)
1158                         return NULL;
1159                 pn = rcu_dereference(fn->parent);
1160                 sn = FIB6_SUBTREE(pn);
1161                 if (sn && sn != fn)
1162                         fn = fib6_node_lookup(sn, NULL, saddr);
1163                 else
1164                         fn = pn;
1165                 if (fn->fn_flags & RTN_RTINFO)
1166                         return fn;
1167         }
1168 }
1169
1170 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1171 {
1172         struct rt6_info *rt = *prt;
1173
1174         if (dst_hold_safe(&rt->dst))
1175                 return true;
1176         if (net) {
1177                 rt = net->ipv6.ip6_null_entry;
1178                 dst_hold(&rt->dst);
1179         } else {
1180                 rt = NULL;
1181         }
1182         *prt = rt;
1183         return false;
1184 }
1185
1186 /* called with rcu_lock held */
1187 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1188 {
1189         struct net_device *dev = res->nh->fib_nh_dev;
1190         struct fib6_info *f6i = res->f6i;
1191         unsigned short flags;
1192         struct rt6_info *nrt;
1193
1194         if (!fib6_info_hold_safe(f6i))
1195                 goto fallback;
1196
1197         flags = fib6_info_dst_flags(f6i);
1198         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1199         if (!nrt) {
1200                 fib6_info_release(f6i);
1201                 goto fallback;
1202         }
1203
1204         ip6_rt_copy_init(nrt, res);
1205         return nrt;
1206
1207 fallback:
1208         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1209         dst_hold(&nrt->dst);
1210         return nrt;
1211 }
1212
1213 static struct rt6_info *ip6_pol_route_lookup(struct net *net,
1214                                              struct fib6_table *table,
1215                                              struct flowi6 *fl6,
1216                                              const struct sk_buff *skb,
1217                                              int flags)
1218 {
1219         struct fib6_result res = {};
1220         struct fib6_node *fn;
1221         struct rt6_info *rt;
1222
1223         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1224                 flags &= ~RT6_LOOKUP_F_IFACE;
1225
1226         rcu_read_lock();
1227         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1228 restart:
1229         res.f6i = rcu_dereference(fn->leaf);
1230         if (!res.f6i)
1231                 res.f6i = net->ipv6.fib6_null_entry;
1232         else
1233                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1234                                  flags);
1235
1236         if (res.f6i == net->ipv6.fib6_null_entry) {
1237                 fn = fib6_backtrack(fn, &fl6->saddr);
1238                 if (fn)
1239                         goto restart;
1240
1241                 rt = net->ipv6.ip6_null_entry;
1242                 dst_hold(&rt->dst);
1243                 goto out;
1244         } else if (res.fib6_flags & RTF_REJECT) {
1245                 goto do_create;
1246         }
1247
1248         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1249                          fl6->flowi6_oif != 0, skb, flags);
1250
1251         /* Search through exception table */
1252         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1253         if (rt) {
1254                 if (ip6_hold_safe(net, &rt))
1255                         dst_use_noref(&rt->dst, jiffies);
1256         } else {
1257 do_create:
1258                 rt = ip6_create_rt_rcu(&res);
1259         }
1260
1261 out:
1262         trace_fib6_table_lookup(net, &res, table, fl6);
1263
1264         rcu_read_unlock();
1265
1266         return rt;
1267 }
1268
1269 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1270                                    const struct sk_buff *skb, int flags)
1271 {
1272         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1273 }
1274 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1275
1276 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1277                             const struct in6_addr *saddr, int oif,
1278                             const struct sk_buff *skb, int strict)
1279 {
1280         struct flowi6 fl6 = {
1281                 .flowi6_oif = oif,
1282                 .daddr = *daddr,
1283         };
1284         struct dst_entry *dst;
1285         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1286
1287         if (saddr) {
1288                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1289                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1290         }
1291
1292         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1293         if (dst->error == 0)
1294                 return (struct rt6_info *) dst;
1295
1296         dst_release(dst);
1297
1298         return NULL;
1299 }
1300 EXPORT_SYMBOL(rt6_lookup);
1301
1302 /* ip6_ins_rt is called with FREE table->tb6_lock.
1303  * It takes new route entry, the addition fails by any reason the
1304  * route is released.
1305  * Caller must hold dst before calling it.
1306  */
1307
1308 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1309                         struct netlink_ext_ack *extack)
1310 {
1311         int err;
1312         struct fib6_table *table;
1313
1314         table = rt->fib6_table;
1315         spin_lock_bh(&table->tb6_lock);
1316         err = fib6_add(&table->tb6_root, rt, info, extack);
1317         spin_unlock_bh(&table->tb6_lock);
1318
1319         return err;
1320 }
1321
1322 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1323 {
1324         struct nl_info info = { .nl_net = net, };
1325
1326         return __ip6_ins_rt(rt, &info, NULL);
1327 }
1328
1329 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1330                                            const struct in6_addr *daddr,
1331                                            const struct in6_addr *saddr)
1332 {
1333         struct fib6_info *f6i = res->f6i;
1334         struct net_device *dev;
1335         struct rt6_info *rt;
1336
1337         /*
1338          *      Clone the route.
1339          */
1340
1341         if (!fib6_info_hold_safe(f6i))
1342                 return NULL;
1343
1344         dev = ip6_rt_get_dev_rcu(res);
1345         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1346         if (!rt) {
1347                 fib6_info_release(f6i);
1348                 return NULL;
1349         }
1350
1351         ip6_rt_copy_init(rt, res);
1352         rt->rt6i_flags |= RTF_CACHE;
1353         rt->rt6i_dst.addr = *daddr;
1354         rt->rt6i_dst.plen = 128;
1355
1356         if (!rt6_is_gw_or_nonexthop(res)) {
1357                 if (f6i->fib6_dst.plen != 128 &&
1358                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1359                         rt->rt6i_flags |= RTF_ANYCAST;
1360 #ifdef CONFIG_IPV6_SUBTREES
1361                 if (rt->rt6i_src.plen && saddr) {
1362                         rt->rt6i_src.addr = *saddr;
1363                         rt->rt6i_src.plen = 128;
1364                 }
1365 #endif
1366         }
1367
1368         return rt;
1369 }
1370
1371 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1372 {
1373         struct fib6_info *f6i = res->f6i;
1374         unsigned short flags = fib6_info_dst_flags(f6i);
1375         struct net_device *dev;
1376         struct rt6_info *pcpu_rt;
1377
1378         if (!fib6_info_hold_safe(f6i))
1379                 return NULL;
1380
1381         rcu_read_lock();
1382         dev = ip6_rt_get_dev_rcu(res);
1383         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1384         rcu_read_unlock();
1385         if (!pcpu_rt) {
1386                 fib6_info_release(f6i);
1387                 return NULL;
1388         }
1389         ip6_rt_copy_init(pcpu_rt, res);
1390         pcpu_rt->rt6i_flags |= RTF_PCPU;
1391
1392         if (f6i->nh)
1393                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1394
1395         return pcpu_rt;
1396 }
1397
1398 static bool rt6_is_valid(const struct rt6_info *rt6)
1399 {
1400         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1401 }
1402
1403 /* It should be called with rcu_read_lock() acquired */
1404 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1405 {
1406         struct rt6_info *pcpu_rt;
1407
1408         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1409
1410         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1411                 struct rt6_info *prev, **p;
1412
1413                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1414                 prev = xchg(p, NULL);
1415                 if (prev) {
1416                         dst_dev_put(&prev->dst);
1417                         dst_release(&prev->dst);
1418                 }
1419
1420                 pcpu_rt = NULL;
1421         }
1422
1423         return pcpu_rt;
1424 }
1425
1426 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1427                                             const struct fib6_result *res)
1428 {
1429         struct rt6_info *pcpu_rt, *prev, **p;
1430
1431         pcpu_rt = ip6_rt_pcpu_alloc(res);
1432         if (!pcpu_rt)
1433                 return NULL;
1434
1435         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1436         prev = cmpxchg(p, NULL, pcpu_rt);
1437         BUG_ON(prev);
1438
1439         if (res->f6i->fib6_destroying) {
1440                 struct fib6_info *from;
1441
1442                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1443                 fib6_info_release(from);
1444         }
1445
1446         return pcpu_rt;
1447 }
1448
1449 /* exception hash table implementation
1450  */
1451 static DEFINE_SPINLOCK(rt6_exception_lock);
1452
1453 /* Remove rt6_ex from hash table and free the memory
1454  * Caller must hold rt6_exception_lock
1455  */
1456 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1457                                  struct rt6_exception *rt6_ex)
1458 {
1459         struct fib6_info *from;
1460         struct net *net;
1461
1462         if (!bucket || !rt6_ex)
1463                 return;
1464
1465         net = dev_net(rt6_ex->rt6i->dst.dev);
1466         net->ipv6.rt6_stats->fib_rt_cache--;
1467
1468         /* purge completely the exception to allow releasing the held resources:
1469          * some [sk] cache may keep the dst around for unlimited time
1470          */
1471         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1472         fib6_info_release(from);
1473         dst_dev_put(&rt6_ex->rt6i->dst);
1474
1475         hlist_del_rcu(&rt6_ex->hlist);
1476         dst_release(&rt6_ex->rt6i->dst);
1477         kfree_rcu(rt6_ex, rcu);
1478         WARN_ON_ONCE(!bucket->depth);
1479         bucket->depth--;
1480 }
1481
1482 /* Remove oldest rt6_ex in bucket and free the memory
1483  * Caller must hold rt6_exception_lock
1484  */
1485 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1486 {
1487         struct rt6_exception *rt6_ex, *oldest = NULL;
1488
1489         if (!bucket)
1490                 return;
1491
1492         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1493                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1494                         oldest = rt6_ex;
1495         }
1496         rt6_remove_exception(bucket, oldest);
1497 }
1498
1499 static u32 rt6_exception_hash(const struct in6_addr *dst,
1500                               const struct in6_addr *src)
1501 {
1502         static u32 seed __read_mostly;
1503         u32 val;
1504
1505         net_get_random_once(&seed, sizeof(seed));
1506         val = jhash2((const u32 *)dst, sizeof(*dst)/sizeof(u32), seed);
1507
1508 #ifdef CONFIG_IPV6_SUBTREES
1509         if (src)
1510                 val = jhash2((const u32 *)src, sizeof(*src)/sizeof(u32), val);
1511 #endif
1512         return hash_32(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1513 }
1514
1515 /* Helper function to find the cached rt in the hash table
1516  * and update bucket pointer to point to the bucket for this
1517  * (daddr, saddr) pair
1518  * Caller must hold rt6_exception_lock
1519  */
1520 static struct rt6_exception *
1521 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1522                               const struct in6_addr *daddr,
1523                               const struct in6_addr *saddr)
1524 {
1525         struct rt6_exception *rt6_ex;
1526         u32 hval;
1527
1528         if (!(*bucket) || !daddr)
1529                 return NULL;
1530
1531         hval = rt6_exception_hash(daddr, saddr);
1532         *bucket += hval;
1533
1534         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1535                 struct rt6_info *rt6 = rt6_ex->rt6i;
1536                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1537
1538 #ifdef CONFIG_IPV6_SUBTREES
1539                 if (matched && saddr)
1540                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1541 #endif
1542                 if (matched)
1543                         return rt6_ex;
1544         }
1545         return NULL;
1546 }
1547
1548 /* Helper function to find the cached rt in the hash table
1549  * and update bucket pointer to point to the bucket for this
1550  * (daddr, saddr) pair
1551  * Caller must hold rcu_read_lock()
1552  */
1553 static struct rt6_exception *
1554 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1555                          const struct in6_addr *daddr,
1556                          const struct in6_addr *saddr)
1557 {
1558         struct rt6_exception *rt6_ex;
1559         u32 hval;
1560
1561         WARN_ON_ONCE(!rcu_read_lock_held());
1562
1563         if (!(*bucket) || !daddr)
1564                 return NULL;
1565
1566         hval = rt6_exception_hash(daddr, saddr);
1567         *bucket += hval;
1568
1569         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1570                 struct rt6_info *rt6 = rt6_ex->rt6i;
1571                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1572
1573 #ifdef CONFIG_IPV6_SUBTREES
1574                 if (matched && saddr)
1575                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1576 #endif
1577                 if (matched)
1578                         return rt6_ex;
1579         }
1580         return NULL;
1581 }
1582
1583 static unsigned int fib6_mtu(const struct fib6_result *res)
1584 {
1585         const struct fib6_nh *nh = res->nh;
1586         unsigned int mtu;
1587
1588         if (res->f6i->fib6_pmtu) {
1589                 mtu = res->f6i->fib6_pmtu;
1590         } else {
1591                 struct net_device *dev = nh->fib_nh_dev;
1592                 struct inet6_dev *idev;
1593
1594                 rcu_read_lock();
1595                 idev = __in6_dev_get(dev);
1596                 mtu = idev->cnf.mtu6;
1597                 rcu_read_unlock();
1598         }
1599
1600         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1601
1602         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1603 }
1604
1605 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1606
1607 /* used when the flushed bit is not relevant, only access to the bucket
1608  * (ie., all bucket users except rt6_insert_exception);
1609  *
1610  * called under rcu lock; sometimes called with rt6_exception_lock held
1611  */
1612 static
1613 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1614                                                        spinlock_t *lock)
1615 {
1616         struct rt6_exception_bucket *bucket;
1617
1618         if (lock)
1619                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1620                                                    lockdep_is_held(lock));
1621         else
1622                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1623
1624         /* remove bucket flushed bit if set */
1625         if (bucket) {
1626                 unsigned long p = (unsigned long)bucket;
1627
1628                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1629                 bucket = (struct rt6_exception_bucket *)p;
1630         }
1631
1632         return bucket;
1633 }
1634
1635 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1636 {
1637         unsigned long p = (unsigned long)bucket;
1638
1639         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1640 }
1641
1642 /* called with rt6_exception_lock held */
1643 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1644                                               spinlock_t *lock)
1645 {
1646         struct rt6_exception_bucket *bucket;
1647         unsigned long p;
1648
1649         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1650                                            lockdep_is_held(lock));
1651
1652         p = (unsigned long)bucket;
1653         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1654         bucket = (struct rt6_exception_bucket *)p;
1655         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1656 }
1657
1658 static int rt6_insert_exception(struct rt6_info *nrt,
1659                                 const struct fib6_result *res)
1660 {
1661         struct net *net = dev_net(nrt->dst.dev);
1662         struct rt6_exception_bucket *bucket;
1663         struct fib6_info *f6i = res->f6i;
1664         struct in6_addr *src_key = NULL;
1665         struct rt6_exception *rt6_ex;
1666         struct fib6_nh *nh = res->nh;
1667         int err = 0;
1668
1669         spin_lock_bh(&rt6_exception_lock);
1670
1671         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1672                                           lockdep_is_held(&rt6_exception_lock));
1673         if (!bucket) {
1674                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1675                                  GFP_ATOMIC);
1676                 if (!bucket) {
1677                         err = -ENOMEM;
1678                         goto out;
1679                 }
1680                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1681         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1682                 err = -EINVAL;
1683                 goto out;
1684         }
1685
1686 #ifdef CONFIG_IPV6_SUBTREES
1687         /* fib6_src.plen != 0 indicates f6i is in subtree
1688          * and exception table is indexed by a hash of
1689          * both fib6_dst and fib6_src.
1690          * Otherwise, the exception table is indexed by
1691          * a hash of only fib6_dst.
1692          */
1693         if (f6i->fib6_src.plen)
1694                 src_key = &nrt->rt6i_src.addr;
1695 #endif
1696         /* rt6_mtu_change() might lower mtu on f6i.
1697          * Only insert this exception route if its mtu
1698          * is less than f6i's mtu value.
1699          */
1700         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1701                 err = -EINVAL;
1702                 goto out;
1703         }
1704
1705         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1706                                                src_key);
1707         if (rt6_ex)
1708                 rt6_remove_exception(bucket, rt6_ex);
1709
1710         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1711         if (!rt6_ex) {
1712                 err = -ENOMEM;
1713                 goto out;
1714         }
1715         rt6_ex->rt6i = nrt;
1716         rt6_ex->stamp = jiffies;
1717         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1718         bucket->depth++;
1719         net->ipv6.rt6_stats->fib_rt_cache++;
1720
1721         if (bucket->depth > FIB6_MAX_DEPTH)
1722                 rt6_exception_remove_oldest(bucket);
1723
1724 out:
1725         spin_unlock_bh(&rt6_exception_lock);
1726
1727         /* Update fn->fn_sernum to invalidate all cached dst */
1728         if (!err) {
1729                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1730                 fib6_update_sernum(net, f6i);
1731                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1732                 fib6_force_start_gc(net);
1733         }
1734
1735         return err;
1736 }
1737
1738 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1739 {
1740         struct rt6_exception_bucket *bucket;
1741         struct rt6_exception *rt6_ex;
1742         struct hlist_node *tmp;
1743         int i;
1744
1745         spin_lock_bh(&rt6_exception_lock);
1746
1747         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1748         if (!bucket)
1749                 goto out;
1750
1751         /* Prevent rt6_insert_exception() to recreate the bucket list */
1752         if (!from)
1753                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1754
1755         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1756                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1757                         if (!from ||
1758                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1759                                 rt6_remove_exception(bucket, rt6_ex);
1760                 }
1761                 WARN_ON_ONCE(!from && bucket->depth);
1762                 bucket++;
1763         }
1764 out:
1765         spin_unlock_bh(&rt6_exception_lock);
1766 }
1767
1768 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1769 {
1770         struct fib6_info *f6i = arg;
1771
1772         fib6_nh_flush_exceptions(nh, f6i);
1773
1774         return 0;
1775 }
1776
1777 void rt6_flush_exceptions(struct fib6_info *f6i)
1778 {
1779         if (f6i->nh)
1780                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1781                                          f6i);
1782         else
1783                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1784 }
1785
1786 /* Find cached rt in the hash table inside passed in rt
1787  * Caller has to hold rcu_read_lock()
1788  */
1789 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1790                                            const struct in6_addr *daddr,
1791                                            const struct in6_addr *saddr)
1792 {
1793         const struct in6_addr *src_key = NULL;
1794         struct rt6_exception_bucket *bucket;
1795         struct rt6_exception *rt6_ex;
1796         struct rt6_info *ret = NULL;
1797
1798 #ifdef CONFIG_IPV6_SUBTREES
1799         /* fib6i_src.plen != 0 indicates f6i is in subtree
1800          * and exception table is indexed by a hash of
1801          * both fib6_dst and fib6_src.
1802          * However, the src addr used to create the hash
1803          * might not be exactly the passed in saddr which
1804          * is a /128 addr from the flow.
1805          * So we need to use f6i->fib6_src to redo lookup
1806          * if the passed in saddr does not find anything.
1807          * (See the logic in ip6_rt_cache_alloc() on how
1808          * rt->rt6i_src is updated.)
1809          */
1810         if (res->f6i->fib6_src.plen)
1811                 src_key = saddr;
1812 find_ex:
1813 #endif
1814         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1815         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1816
1817         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1818                 ret = rt6_ex->rt6i;
1819
1820 #ifdef CONFIG_IPV6_SUBTREES
1821         /* Use fib6_src as src_key and redo lookup */
1822         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1823                 src_key = &res->f6i->fib6_src.addr;
1824                 goto find_ex;
1825         }
1826 #endif
1827
1828         return ret;
1829 }
1830
1831 /* Remove the passed in cached rt from the hash table that contains it */
1832 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1833                                     const struct rt6_info *rt)
1834 {
1835         const struct in6_addr *src_key = NULL;
1836         struct rt6_exception_bucket *bucket;
1837         struct rt6_exception *rt6_ex;
1838         int err;
1839
1840         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1841                 return -ENOENT;
1842
1843         spin_lock_bh(&rt6_exception_lock);
1844         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1845
1846 #ifdef CONFIG_IPV6_SUBTREES
1847         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1848          * and exception table is indexed by a hash of
1849          * both rt6i_dst and rt6i_src.
1850          * Otherwise, the exception table is indexed by
1851          * a hash of only rt6i_dst.
1852          */
1853         if (plen)
1854                 src_key = &rt->rt6i_src.addr;
1855 #endif
1856         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1857                                                &rt->rt6i_dst.addr,
1858                                                src_key);
1859         if (rt6_ex) {
1860                 rt6_remove_exception(bucket, rt6_ex);
1861                 err = 0;
1862         } else {
1863                 err = -ENOENT;
1864         }
1865
1866         spin_unlock_bh(&rt6_exception_lock);
1867         return err;
1868 }
1869
1870 struct fib6_nh_excptn_arg {
1871         struct rt6_info *rt;
1872         int             plen;
1873 };
1874
1875 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1876 {
1877         struct fib6_nh_excptn_arg *arg = _arg;
1878         int err;
1879
1880         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1881         if (err == 0)
1882                 return 1;
1883
1884         return 0;
1885 }
1886
1887 static int rt6_remove_exception_rt(struct rt6_info *rt)
1888 {
1889         struct fib6_info *from;
1890
1891         from = rcu_dereference(rt->from);
1892         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1893                 return -EINVAL;
1894
1895         if (from->nh) {
1896                 struct fib6_nh_excptn_arg arg = {
1897                         .rt = rt,
1898                         .plen = from->fib6_src.plen
1899                 };
1900                 int rc;
1901
1902                 /* rc = 1 means an entry was found */
1903                 rc = nexthop_for_each_fib6_nh(from->nh,
1904                                               rt6_nh_remove_exception_rt,
1905                                               &arg);
1906                 return rc ? 0 : -ENOENT;
1907         }
1908
1909         return fib6_nh_remove_exception(from->fib6_nh,
1910                                         from->fib6_src.plen, rt);
1911 }
1912
1913 /* Find rt6_ex which contains the passed in rt cache and
1914  * refresh its stamp
1915  */
1916 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1917                                      const struct rt6_info *rt)
1918 {
1919         const struct in6_addr *src_key = NULL;
1920         struct rt6_exception_bucket *bucket;
1921         struct rt6_exception *rt6_ex;
1922
1923         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1924 #ifdef CONFIG_IPV6_SUBTREES
1925         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1926          * and exception table is indexed by a hash of
1927          * both rt6i_dst and rt6i_src.
1928          * Otherwise, the exception table is indexed by
1929          * a hash of only rt6i_dst.
1930          */
1931         if (plen)
1932                 src_key = &rt->rt6i_src.addr;
1933 #endif
1934         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1935         if (rt6_ex)
1936                 rt6_ex->stamp = jiffies;
1937 }
1938
1939 struct fib6_nh_match_arg {
1940         const struct net_device *dev;
1941         const struct in6_addr   *gw;
1942         struct fib6_nh          *match;
1943 };
1944
1945 /* determine if fib6_nh has given device and gateway */
1946 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1947 {
1948         struct fib6_nh_match_arg *arg = _arg;
1949
1950         if (arg->dev != nh->fib_nh_dev ||
1951             (arg->gw && !nh->fib_nh_gw_family) ||
1952             (!arg->gw && nh->fib_nh_gw_family) ||
1953             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1954                 return 0;
1955
1956         arg->match = nh;
1957
1958         /* found a match, break the loop */
1959         return 1;
1960 }
1961
1962 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1963 {
1964         struct fib6_info *from;
1965         struct fib6_nh *fib6_nh;
1966
1967         rcu_read_lock();
1968
1969         from = rcu_dereference(rt->from);
1970         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1971                 goto unlock;
1972
1973         if (from->nh) {
1974                 struct fib6_nh_match_arg arg = {
1975                         .dev = rt->dst.dev,
1976                         .gw = &rt->rt6i_gateway,
1977                 };
1978
1979                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1980
1981                 if (!arg.match)
1982                         goto unlock;
1983                 fib6_nh = arg.match;
1984         } else {
1985                 fib6_nh = from->fib6_nh;
1986         }
1987         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1988 unlock:
1989         rcu_read_unlock();
1990 }
1991
1992 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1993                                          struct rt6_info *rt, int mtu)
1994 {
1995         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1996          * lowest MTU in the path: always allow updating the route PMTU to
1997          * reflect PMTU decreases.
1998          *
1999          * If the new MTU is higher, and the route PMTU is equal to the local
2000          * MTU, this means the old MTU is the lowest in the path, so allow
2001          * updating it: if other nodes now have lower MTUs, PMTU discovery will
2002          * handle this.
2003          */
2004
2005         if (dst_mtu(&rt->dst) >= mtu)
2006                 return true;
2007
2008         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2009                 return true;
2010
2011         return false;
2012 }
2013
2014 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2015                                        const struct fib6_nh *nh, int mtu)
2016 {
2017         struct rt6_exception_bucket *bucket;
2018         struct rt6_exception *rt6_ex;
2019         int i;
2020
2021         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2022         if (!bucket)
2023                 return;
2024
2025         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2026                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2027                         struct rt6_info *entry = rt6_ex->rt6i;
2028
2029                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2030                          * route), the metrics of its rt->from have already
2031                          * been updated.
2032                          */
2033                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2034                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2035                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2036                 }
2037                 bucket++;
2038         }
2039 }
2040
2041 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2042
2043 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2044                                             const struct in6_addr *gateway)
2045 {
2046         struct rt6_exception_bucket *bucket;
2047         struct rt6_exception *rt6_ex;
2048         struct hlist_node *tmp;
2049         int i;
2050
2051         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2052                 return;
2053
2054         spin_lock_bh(&rt6_exception_lock);
2055         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2056         if (bucket) {
2057                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2058                         hlist_for_each_entry_safe(rt6_ex, tmp,
2059                                                   &bucket->chain, hlist) {
2060                                 struct rt6_info *entry = rt6_ex->rt6i;
2061
2062                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2063                                     RTF_CACHE_GATEWAY &&
2064                                     ipv6_addr_equal(gateway,
2065                                                     &entry->rt6i_gateway)) {
2066                                         rt6_remove_exception(bucket, rt6_ex);
2067                                 }
2068                         }
2069                         bucket++;
2070                 }
2071         }
2072
2073         spin_unlock_bh(&rt6_exception_lock);
2074 }
2075
2076 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2077                                       struct rt6_exception *rt6_ex,
2078                                       struct fib6_gc_args *gc_args,
2079                                       unsigned long now)
2080 {
2081         struct rt6_info *rt = rt6_ex->rt6i;
2082
2083         /* we are pruning and obsoleting aged-out and non gateway exceptions
2084          * even if others have still references to them, so that on next
2085          * dst_check() such references can be dropped.
2086          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2087          * expired, independently from their aging, as per RFC 8201 section 4
2088          */
2089         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2090                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2091                         RT6_TRACE("aging clone %p\n", rt);
2092                         rt6_remove_exception(bucket, rt6_ex);
2093                         return;
2094                 }
2095         } else if (time_after(jiffies, rt->dst.expires)) {
2096                 RT6_TRACE("purging expired route %p\n", rt);
2097                 rt6_remove_exception(bucket, rt6_ex);
2098                 return;
2099         }
2100
2101         if (rt->rt6i_flags & RTF_GATEWAY) {
2102                 struct neighbour *neigh;
2103                 __u8 neigh_flags = 0;
2104
2105                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2106                 if (neigh)
2107                         neigh_flags = neigh->flags;
2108
2109                 if (!(neigh_flags & NTF_ROUTER)) {
2110                         RT6_TRACE("purging route %p via non-router but gateway\n",
2111                                   rt);
2112                         rt6_remove_exception(bucket, rt6_ex);
2113                         return;
2114                 }
2115         }
2116
2117         gc_args->more++;
2118 }
2119
2120 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2121                                    struct fib6_gc_args *gc_args,
2122                                    unsigned long now)
2123 {
2124         struct rt6_exception_bucket *bucket;
2125         struct rt6_exception *rt6_ex;
2126         struct hlist_node *tmp;
2127         int i;
2128
2129         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2130                 return;
2131
2132         rcu_read_lock_bh();
2133         spin_lock(&rt6_exception_lock);
2134         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2135         if (bucket) {
2136                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2137                         hlist_for_each_entry_safe(rt6_ex, tmp,
2138                                                   &bucket->chain, hlist) {
2139                                 rt6_age_examine_exception(bucket, rt6_ex,
2140                                                           gc_args, now);
2141                         }
2142                         bucket++;
2143                 }
2144         }
2145         spin_unlock(&rt6_exception_lock);
2146         rcu_read_unlock_bh();
2147 }
2148
2149 struct fib6_nh_age_excptn_arg {
2150         struct fib6_gc_args     *gc_args;
2151         unsigned long           now;
2152 };
2153
2154 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2155 {
2156         struct fib6_nh_age_excptn_arg *arg = _arg;
2157
2158         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2159         return 0;
2160 }
2161
2162 void rt6_age_exceptions(struct fib6_info *f6i,
2163                         struct fib6_gc_args *gc_args,
2164                         unsigned long now)
2165 {
2166         if (f6i->nh) {
2167                 struct fib6_nh_age_excptn_arg arg = {
2168                         .gc_args = gc_args,
2169                         .now = now
2170                 };
2171
2172                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2173                                          &arg);
2174         } else {
2175                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2176         }
2177 }
2178
2179 /* must be called with rcu lock held */
2180 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2181                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2182 {
2183         struct fib6_node *fn, *saved_fn;
2184
2185         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2186         saved_fn = fn;
2187
2188         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2189                 oif = 0;
2190
2191 redo_rt6_select:
2192         rt6_select(net, fn, oif, res, strict);
2193         if (res->f6i == net->ipv6.fib6_null_entry) {
2194                 fn = fib6_backtrack(fn, &fl6->saddr);
2195                 if (fn)
2196                         goto redo_rt6_select;
2197                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2198                         /* also consider unreachable route */
2199                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2200                         fn = saved_fn;
2201                         goto redo_rt6_select;
2202                 }
2203         }
2204
2205         trace_fib6_table_lookup(net, res, table, fl6);
2206
2207         return 0;
2208 }
2209
2210 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2211                                int oif, struct flowi6 *fl6,
2212                                const struct sk_buff *skb, int flags)
2213 {
2214         struct fib6_result res = {};
2215         struct rt6_info *rt = NULL;
2216         int strict = 0;
2217
2218         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2219                      !rcu_read_lock_held());
2220
2221         strict |= flags & RT6_LOOKUP_F_IFACE;
2222         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2223         if (net->ipv6.devconf_all->forwarding == 0)
2224                 strict |= RT6_LOOKUP_F_REACHABLE;
2225
2226         rcu_read_lock();
2227
2228         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2229         if (res.f6i == net->ipv6.fib6_null_entry)
2230                 goto out;
2231
2232         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2233
2234         /*Search through exception table */
2235         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2236         if (rt) {
2237                 goto out;
2238         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2239                             !res.nh->fib_nh_gw_family)) {
2240                 /* Create a RTF_CACHE clone which will not be
2241                  * owned by the fib6 tree.  It is for the special case where
2242                  * the daddr in the skb during the neighbor look-up is different
2243                  * from the fl6->daddr used to look-up route here.
2244                  */
2245                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2246
2247                 if (rt) {
2248                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2249                          * As rt6_uncached_list_add() does not consume refcnt,
2250                          * this refcnt is always returned to the caller even
2251                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2252                          */
2253                         rt6_uncached_list_add(rt);
2254                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2255                         rcu_read_unlock();
2256
2257                         return rt;
2258                 }
2259         } else {
2260                 /* Get a percpu copy */
2261                 local_bh_disable();
2262                 rt = rt6_get_pcpu_route(&res);
2263
2264                 if (!rt)
2265                         rt = rt6_make_pcpu_route(net, &res);
2266
2267                 local_bh_enable();
2268         }
2269 out:
2270         if (!rt)
2271                 rt = net->ipv6.ip6_null_entry;
2272         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2273                 ip6_hold_safe(net, &rt);
2274         rcu_read_unlock();
2275
2276         return rt;
2277 }
2278 EXPORT_SYMBOL_GPL(ip6_pol_route);
2279
2280 static struct rt6_info *ip6_pol_route_input(struct net *net,
2281                                             struct fib6_table *table,
2282                                             struct flowi6 *fl6,
2283                                             const struct sk_buff *skb,
2284                                             int flags)
2285 {
2286         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2287 }
2288
2289 struct dst_entry *ip6_route_input_lookup(struct net *net,
2290                                          struct net_device *dev,
2291                                          struct flowi6 *fl6,
2292                                          const struct sk_buff *skb,
2293                                          int flags)
2294 {
2295         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2296                 flags |= RT6_LOOKUP_F_IFACE;
2297
2298         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2299 }
2300 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2301
2302 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2303                                   struct flow_keys *keys,
2304                                   struct flow_keys *flkeys)
2305 {
2306         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2307         const struct ipv6hdr *key_iph = outer_iph;
2308         struct flow_keys *_flkeys = flkeys;
2309         const struct ipv6hdr *inner_iph;
2310         const struct icmp6hdr *icmph;
2311         struct ipv6hdr _inner_iph;
2312         struct icmp6hdr _icmph;
2313
2314         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2315                 goto out;
2316
2317         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2318                                    sizeof(_icmph), &_icmph);
2319         if (!icmph)
2320                 goto out;
2321
2322         if (!icmpv6_is_err(icmph->icmp6_type))
2323                 goto out;
2324
2325         inner_iph = skb_header_pointer(skb,
2326                                        skb_transport_offset(skb) + sizeof(*icmph),
2327                                        sizeof(_inner_iph), &_inner_iph);
2328         if (!inner_iph)
2329                 goto out;
2330
2331         key_iph = inner_iph;
2332         _flkeys = NULL;
2333 out:
2334         if (_flkeys) {
2335                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2336                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2337                 keys->tags.flow_label = _flkeys->tags.flow_label;
2338                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2339         } else {
2340                 keys->addrs.v6addrs.src = key_iph->saddr;
2341                 keys->addrs.v6addrs.dst = key_iph->daddr;
2342                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2343                 keys->basic.ip_proto = key_iph->nexthdr;
2344         }
2345 }
2346
2347 /* if skb is set it will be used and fl6 can be NULL */
2348 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2349                        const struct sk_buff *skb, struct flow_keys *flkeys)
2350 {
2351         struct flow_keys hash_keys;
2352         u32 mhash;
2353
2354         switch (ip6_multipath_hash_policy(net)) {
2355         case 0:
2356                 memset(&hash_keys, 0, sizeof(hash_keys));
2357                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2358                 if (skb) {
2359                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2360                 } else {
2361                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2362                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2363                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2364                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2365                 }
2366                 break;
2367         case 1:
2368                 if (skb) {
2369                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2370                         struct flow_keys keys;
2371
2372                         /* short-circuit if we already have L4 hash present */
2373                         if (skb->l4_hash)
2374                                 return skb_get_hash_raw(skb) >> 1;
2375
2376                         memset(&hash_keys, 0, sizeof(hash_keys));
2377
2378                         if (!flkeys) {
2379                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2380                                 flkeys = &keys;
2381                         }
2382                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2383                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2384                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2385                         hash_keys.ports.src = flkeys->ports.src;
2386                         hash_keys.ports.dst = flkeys->ports.dst;
2387                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2388                 } else {
2389                         memset(&hash_keys, 0, sizeof(hash_keys));
2390                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2391                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2392                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2393                         hash_keys.ports.src = fl6->fl6_sport;
2394                         hash_keys.ports.dst = fl6->fl6_dport;
2395                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2396                 }
2397                 break;
2398         case 2:
2399                 memset(&hash_keys, 0, sizeof(hash_keys));
2400                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2401                 if (skb) {
2402                         struct flow_keys keys;
2403
2404                         if (!flkeys) {
2405                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2406                                 flkeys = &keys;
2407                         }
2408
2409                         /* Inner can be v4 or v6 */
2410                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2411                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2412                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2413                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2414                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2415                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2416                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2417                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2418                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2419                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2420                         } else {
2421                                 /* Same as case 0 */
2422                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2423                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2424                         }
2425                 } else {
2426                         /* Same as case 0 */
2427                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2428                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2429                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2430                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2431                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2432                 }
2433                 break;
2434         }
2435         mhash = flow_hash_from_keys(&hash_keys);
2436
2437         return mhash >> 1;
2438 }
2439
2440 /* Called with rcu held */
2441 void ip6_route_input(struct sk_buff *skb)
2442 {
2443         const struct ipv6hdr *iph = ipv6_hdr(skb);
2444         struct net *net = dev_net(skb->dev);
2445         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2446         struct ip_tunnel_info *tun_info;
2447         struct flowi6 fl6 = {
2448                 .flowi6_iif = skb->dev->ifindex,
2449                 .daddr = iph->daddr,
2450                 .saddr = iph->saddr,
2451                 .flowlabel = ip6_flowinfo(iph),
2452                 .flowi6_mark = skb->mark,
2453                 .flowi6_proto = iph->nexthdr,
2454         };
2455         struct flow_keys *flkeys = NULL, _flkeys;
2456
2457         tun_info = skb_tunnel_info(skb);
2458         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2459                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2460
2461         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2462                 flkeys = &_flkeys;
2463
2464         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2465                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2466         skb_dst_drop(skb);
2467         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2468                                                       &fl6, skb, flags));
2469 }
2470
2471 static struct rt6_info *ip6_pol_route_output(struct net *net,
2472                                              struct fib6_table *table,
2473                                              struct flowi6 *fl6,
2474                                              const struct sk_buff *skb,
2475                                              int flags)
2476 {
2477         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2478 }
2479
2480 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2481                                                const struct sock *sk,
2482                                                struct flowi6 *fl6, int flags)
2483 {
2484         bool any_src;
2485
2486         if (ipv6_addr_type(&fl6->daddr) &
2487             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2488                 struct dst_entry *dst;
2489
2490                 /* This function does not take refcnt on the dst */
2491                 dst = l3mdev_link_scope_lookup(net, fl6);
2492                 if (dst)
2493                         return dst;
2494         }
2495
2496         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2497
2498         flags |= RT6_LOOKUP_F_DST_NOREF;
2499         any_src = ipv6_addr_any(&fl6->saddr);
2500         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2501             (fl6->flowi6_oif && any_src))
2502                 flags |= RT6_LOOKUP_F_IFACE;
2503
2504         if (!any_src)
2505                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2506         else if (sk)
2507                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2508
2509         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2510 }
2511 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2512
2513 struct dst_entry *ip6_route_output_flags(struct net *net,
2514                                          const struct sock *sk,
2515                                          struct flowi6 *fl6,
2516                                          int flags)
2517 {
2518         struct dst_entry *dst;
2519         struct rt6_info *rt6;
2520
2521         rcu_read_lock();
2522         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2523         rt6 = (struct rt6_info *)dst;
2524         /* For dst cached in uncached_list, refcnt is already taken. */
2525         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2526                 dst = &net->ipv6.ip6_null_entry->dst;
2527                 dst_hold(dst);
2528         }
2529         rcu_read_unlock();
2530
2531         return dst;
2532 }
2533 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2534
2535 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2536 {
2537         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2538         struct net_device *loopback_dev = net->loopback_dev;
2539         struct dst_entry *new = NULL;
2540
2541         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2542                        DST_OBSOLETE_DEAD, 0);
2543         if (rt) {
2544                 rt6_info_init(rt);
2545                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2546
2547                 new = &rt->dst;
2548                 new->__use = 1;
2549                 new->input = dst_discard;
2550                 new->output = dst_discard_out;
2551
2552                 dst_copy_metrics(new, &ort->dst);
2553
2554                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2555                 rt->rt6i_gateway = ort->rt6i_gateway;
2556                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2557
2558                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2559 #ifdef CONFIG_IPV6_SUBTREES
2560                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2561 #endif
2562         }
2563
2564         dst_release(dst_orig);
2565         return new ? new : ERR_PTR(-ENOMEM);
2566 }
2567
2568 /*
2569  *      Destination cache support functions
2570  */
2571
2572 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2573 {
2574         u32 rt_cookie = 0;
2575
2576         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2577                 return false;
2578
2579         if (fib6_check_expired(f6i))
2580                 return false;
2581
2582         return true;
2583 }
2584
2585 static struct dst_entry *rt6_check(struct rt6_info *rt,
2586                                    struct fib6_info *from,
2587                                    u32 cookie)
2588 {
2589         u32 rt_cookie = 0;
2590
2591         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2592             rt_cookie != cookie)
2593                 return NULL;
2594
2595         if (rt6_check_expired(rt))
2596                 return NULL;
2597
2598         return &rt->dst;
2599 }
2600
2601 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2602                                             struct fib6_info *from,
2603                                             u32 cookie)
2604 {
2605         if (!__rt6_check_expired(rt) &&
2606             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2607             fib6_check(from, cookie))
2608                 return &rt->dst;
2609         else
2610                 return NULL;
2611 }
2612
2613 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2614 {
2615         struct dst_entry *dst_ret;
2616         struct fib6_info *from;
2617         struct rt6_info *rt;
2618
2619         rt = container_of(dst, struct rt6_info, dst);
2620
2621         if (rt->sernum)
2622                 return rt6_is_valid(rt) ? dst : NULL;
2623
2624         rcu_read_lock();
2625
2626         /* All IPV6 dsts are created with ->obsolete set to the value
2627          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2628          * into this function always.
2629          */
2630
2631         from = rcu_dereference(rt->from);
2632
2633         if (from && (rt->rt6i_flags & RTF_PCPU ||
2634             unlikely(!list_empty(&rt->rt6i_uncached))))
2635                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2636         else
2637                 dst_ret = rt6_check(rt, from, cookie);
2638
2639         rcu_read_unlock();
2640
2641         return dst_ret;
2642 }
2643
2644 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2645 {
2646         struct rt6_info *rt = (struct rt6_info *) dst;
2647
2648         if (rt) {
2649                 if (rt->rt6i_flags & RTF_CACHE) {
2650                         rcu_read_lock();
2651                         if (rt6_check_expired(rt)) {
2652                                 rt6_remove_exception_rt(rt);
2653                                 dst = NULL;
2654                         }
2655                         rcu_read_unlock();
2656                 } else {
2657                         dst_release(dst);
2658                         dst = NULL;
2659                 }
2660         }
2661         return dst;
2662 }
2663
2664 static void ip6_link_failure(struct sk_buff *skb)
2665 {
2666         struct rt6_info *rt;
2667
2668         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2669
2670         rt = (struct rt6_info *) skb_dst(skb);
2671         if (rt) {
2672                 rcu_read_lock();
2673                 if (rt->rt6i_flags & RTF_CACHE) {
2674                         rt6_remove_exception_rt(rt);
2675                 } else {
2676                         struct fib6_info *from;
2677                         struct fib6_node *fn;
2678
2679                         from = rcu_dereference(rt->from);
2680                         if (from) {
2681                                 fn = rcu_dereference(from->fib6_node);
2682                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2683                                         fn->fn_sernum = -1;
2684                         }
2685                 }
2686                 rcu_read_unlock();
2687         }
2688 }
2689
2690 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2691 {
2692         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2693                 struct fib6_info *from;
2694
2695                 rcu_read_lock();
2696                 from = rcu_dereference(rt0->from);
2697                 if (from)
2698                         rt0->dst.expires = from->expires;
2699                 rcu_read_unlock();
2700         }
2701
2702         dst_set_expires(&rt0->dst, timeout);
2703         rt0->rt6i_flags |= RTF_EXPIRES;
2704 }
2705
2706 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2707 {
2708         struct net *net = dev_net(rt->dst.dev);
2709
2710         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2711         rt->rt6i_flags |= RTF_MODIFIED;
2712         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2713 }
2714
2715 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2716 {
2717         return !(rt->rt6i_flags & RTF_CACHE) &&
2718                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2719 }
2720
2721 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2722                                  const struct ipv6hdr *iph, u32 mtu,
2723                                  bool confirm_neigh)
2724 {
2725         const struct in6_addr *daddr, *saddr;
2726         struct rt6_info *rt6 = (struct rt6_info *)dst;
2727
2728         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2729          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2730          * [see also comment in rt6_mtu_change_route()]
2731          */
2732
2733         if (iph) {
2734                 daddr = &iph->daddr;
2735                 saddr = &iph->saddr;
2736         } else if (sk) {
2737                 daddr = &sk->sk_v6_daddr;
2738                 saddr = &inet6_sk(sk)->saddr;
2739         } else {
2740                 daddr = NULL;
2741                 saddr = NULL;
2742         }
2743
2744         if (confirm_neigh)
2745                 dst_confirm_neigh(dst, daddr);
2746
2747         mtu = max_t(u32, mtu, IPV6_MIN_MTU);
2748         if (mtu >= dst_mtu(dst))
2749                 return;
2750
2751         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2752                 rt6_do_update_pmtu(rt6, mtu);
2753                 /* update rt6_ex->stamp for cache */
2754                 if (rt6->rt6i_flags & RTF_CACHE)
2755                         rt6_update_exception_stamp_rt(rt6);
2756         } else if (daddr) {
2757                 struct fib6_result res = {};
2758                 struct rt6_info *nrt6;
2759
2760                 rcu_read_lock();
2761                 res.f6i = rcu_dereference(rt6->from);
2762                 if (!res.f6i)
2763                         goto out_unlock;
2764
2765                 res.fib6_flags = res.f6i->fib6_flags;
2766                 res.fib6_type = res.f6i->fib6_type;
2767
2768                 if (res.f6i->nh) {
2769                         struct fib6_nh_match_arg arg = {
2770                                 .dev = dst->dev,
2771                                 .gw = &rt6->rt6i_gateway,
2772                         };
2773
2774                         nexthop_for_each_fib6_nh(res.f6i->nh,
2775                                                  fib6_nh_find_match, &arg);
2776
2777                         /* fib6_info uses a nexthop that does not have fib6_nh
2778                          * using the dst->dev + gw. Should be impossible.
2779                          */
2780                         if (!arg.match)
2781                                 goto out_unlock;
2782
2783                         res.nh = arg.match;
2784                 } else {
2785                         res.nh = res.f6i->fib6_nh;
2786                 }
2787
2788                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2789                 if (nrt6) {
2790                         rt6_do_update_pmtu(nrt6, mtu);
2791                         if (rt6_insert_exception(nrt6, &res))
2792                                 dst_release_immediate(&nrt6->dst);
2793                 }
2794 out_unlock:
2795                 rcu_read_unlock();
2796         }
2797 }
2798
2799 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2800                                struct sk_buff *skb, u32 mtu,
2801                                bool confirm_neigh)
2802 {
2803         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2804                              confirm_neigh);
2805 }
2806
2807 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2808                      int oif, u32 mark, kuid_t uid)
2809 {
2810         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2811         struct dst_entry *dst;
2812         struct flowi6 fl6 = {
2813                 .flowi6_oif = oif,
2814                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2815                 .daddr = iph->daddr,
2816                 .saddr = iph->saddr,
2817                 .flowlabel = ip6_flowinfo(iph),
2818                 .flowi6_uid = uid,
2819         };
2820
2821         dst = ip6_route_output(net, NULL, &fl6);
2822         if (!dst->error)
2823                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2824         dst_release(dst);
2825 }
2826 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2827
2828 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2829 {
2830         int oif = sk->sk_bound_dev_if;
2831         struct dst_entry *dst;
2832
2833         if (!oif && skb->dev)
2834                 oif = l3mdev_master_ifindex(skb->dev);
2835
2836         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2837
2838         dst = __sk_dst_get(sk);
2839         if (!dst || !dst->obsolete ||
2840             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2841                 return;
2842
2843         bh_lock_sock(sk);
2844         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2845                 ip6_datagram_dst_update(sk, false);
2846         bh_unlock_sock(sk);
2847 }
2848 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2849
2850 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2851                            const struct flowi6 *fl6)
2852 {
2853 #ifdef CONFIG_IPV6_SUBTREES
2854         struct ipv6_pinfo *np = inet6_sk(sk);
2855 #endif
2856
2857         ip6_dst_store(sk, dst,
2858                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2859                       &sk->sk_v6_daddr : NULL,
2860 #ifdef CONFIG_IPV6_SUBTREES
2861                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2862                       &np->saddr :
2863 #endif
2864                       NULL);
2865 }
2866
2867 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2868                                   struct flowi6 *fl6,
2869                                   const struct in6_addr *gw,
2870                                   struct rt6_info **ret)
2871 {
2872         const struct fib6_nh *nh = res->nh;
2873
2874         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2875             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2876                 return false;
2877
2878         /* rt_cache's gateway might be different from its 'parent'
2879          * in the case of an ip redirect.
2880          * So we keep searching in the exception table if the gateway
2881          * is different.
2882          */
2883         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2884                 struct rt6_info *rt_cache;
2885
2886                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2887                 if (rt_cache &&
2888                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2889                         *ret = rt_cache;
2890                         return true;
2891                 }
2892                 return false;
2893         }
2894         return true;
2895 }
2896
2897 struct fib6_nh_rd_arg {
2898         struct fib6_result      *res;
2899         struct flowi6           *fl6;
2900         const struct in6_addr   *gw;
2901         struct rt6_info         **ret;
2902 };
2903
2904 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2905 {
2906         struct fib6_nh_rd_arg *arg = _arg;
2907
2908         arg->res->nh = nh;
2909         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2910 }
2911
2912 /* Handle redirects */
2913 struct ip6rd_flowi {
2914         struct flowi6 fl6;
2915         struct in6_addr gateway;
2916 };
2917
2918 static struct rt6_info *__ip6_route_redirect(struct net *net,
2919                                              struct fib6_table *table,
2920                                              struct flowi6 *fl6,
2921                                              const struct sk_buff *skb,
2922                                              int flags)
2923 {
2924         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2925         struct rt6_info *ret = NULL;
2926         struct fib6_result res = {};
2927         struct fib6_nh_rd_arg arg = {
2928                 .res = &res,
2929                 .fl6 = fl6,
2930                 .gw  = &rdfl->gateway,
2931                 .ret = &ret
2932         };
2933         struct fib6_info *rt;
2934         struct fib6_node *fn;
2935
2936         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2937          * this case we must match on the real ingress device, so reset it
2938          */
2939         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2940                 fl6->flowi6_oif = skb->dev->ifindex;
2941
2942         /* Get the "current" route for this destination and
2943          * check if the redirect has come from appropriate router.
2944          *
2945          * RFC 4861 specifies that redirects should only be
2946          * accepted if they come from the nexthop to the target.
2947          * Due to the way the routes are chosen, this notion
2948          * is a bit fuzzy and one might need to check all possible
2949          * routes.
2950          */
2951
2952         rcu_read_lock();
2953         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2954 restart:
2955         for_each_fib6_node_rt_rcu(fn) {
2956                 res.f6i = rt;
2957                 if (fib6_check_expired(rt))
2958                         continue;
2959                 if (rt->fib6_flags & RTF_REJECT)
2960                         break;
2961                 if (unlikely(rt->nh)) {
2962                         if (nexthop_is_blackhole(rt->nh))
2963                                 continue;
2964                         /* on match, res->nh is filled in and potentially ret */
2965                         if (nexthop_for_each_fib6_nh(rt->nh,
2966                                                      fib6_nh_redirect_match,
2967                                                      &arg))
2968                                 goto out;
2969                 } else {
2970                         res.nh = rt->fib6_nh;
2971                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2972                                                   &ret))
2973                                 goto out;
2974                 }
2975         }
2976
2977         if (!rt)
2978                 rt = net->ipv6.fib6_null_entry;
2979         else if (rt->fib6_flags & RTF_REJECT) {
2980                 ret = net->ipv6.ip6_null_entry;
2981                 goto out;
2982         }
2983
2984         if (rt == net->ipv6.fib6_null_entry) {
2985                 fn = fib6_backtrack(fn, &fl6->saddr);
2986                 if (fn)
2987                         goto restart;
2988         }
2989
2990         res.f6i = rt;
2991         res.nh = rt->fib6_nh;
2992 out:
2993         if (ret) {
2994                 ip6_hold_safe(net, &ret);
2995         } else {
2996                 res.fib6_flags = res.f6i->fib6_flags;
2997                 res.fib6_type = res.f6i->fib6_type;
2998                 ret = ip6_create_rt_rcu(&res);
2999         }
3000
3001         rcu_read_unlock();
3002
3003         trace_fib6_table_lookup(net, &res, table, fl6);
3004         return ret;
3005 };
3006
3007 static struct dst_entry *ip6_route_redirect(struct net *net,
3008                                             const struct flowi6 *fl6,
3009                                             const struct sk_buff *skb,
3010                                             const struct in6_addr *gateway)
3011 {
3012         int flags = RT6_LOOKUP_F_HAS_SADDR;
3013         struct ip6rd_flowi rdfl;
3014
3015         rdfl.fl6 = *fl6;
3016         rdfl.gateway = *gateway;
3017
3018         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3019                                 flags, __ip6_route_redirect);
3020 }
3021
3022 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3023                   kuid_t uid)
3024 {
3025         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3026         struct dst_entry *dst;
3027         struct flowi6 fl6 = {
3028                 .flowi6_iif = LOOPBACK_IFINDEX,
3029                 .flowi6_oif = oif,
3030                 .flowi6_mark = mark,
3031                 .daddr = iph->daddr,
3032                 .saddr = iph->saddr,
3033                 .flowlabel = ip6_flowinfo(iph),
3034                 .flowi6_uid = uid,
3035         };
3036
3037         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3038         rt6_do_redirect(dst, NULL, skb);
3039         dst_release(dst);
3040 }
3041 EXPORT_SYMBOL_GPL(ip6_redirect);
3042
3043 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3044 {
3045         const struct ipv6hdr *iph = ipv6_hdr(skb);
3046         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3047         struct dst_entry *dst;
3048         struct flowi6 fl6 = {
3049                 .flowi6_iif = LOOPBACK_IFINDEX,
3050                 .flowi6_oif = oif,
3051                 .daddr = msg->dest,
3052                 .saddr = iph->daddr,
3053                 .flowi6_uid = sock_net_uid(net, NULL),
3054         };
3055
3056         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3057         rt6_do_redirect(dst, NULL, skb);
3058         dst_release(dst);
3059 }
3060
3061 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3062 {
3063         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3064                      sk->sk_uid);
3065 }
3066 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3067
3068 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3069 {
3070         struct net_device *dev = dst->dev;
3071         unsigned int mtu = dst_mtu(dst);
3072         struct net *net = dev_net(dev);
3073
3074         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3075
3076         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3077                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3078
3079         /*
3080          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3081          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3082          * IPV6_MAXPLEN is also valid and means: "any MSS,
3083          * rely only on pmtu discovery"
3084          */
3085         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3086                 mtu = IPV6_MAXPLEN;
3087         return mtu;
3088 }
3089
3090 static unsigned int ip6_mtu(const struct dst_entry *dst)
3091 {
3092         struct inet6_dev *idev;
3093         unsigned int mtu;
3094
3095         mtu = dst_metric_raw(dst, RTAX_MTU);
3096         if (mtu)
3097                 goto out;
3098
3099         mtu = IPV6_MIN_MTU;
3100
3101         rcu_read_lock();
3102         idev = __in6_dev_get(dst->dev);
3103         if (idev)
3104                 mtu = idev->cnf.mtu6;
3105         rcu_read_unlock();
3106
3107 out:
3108         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3109
3110         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3111 }
3112
3113 /* MTU selection:
3114  * 1. mtu on route is locked - use it
3115  * 2. mtu from nexthop exception
3116  * 3. mtu from egress device
3117  *
3118  * based on ip6_dst_mtu_forward and exception logic of
3119  * rt6_find_cached_rt; called with rcu_read_lock
3120  */
3121 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3122                       const struct in6_addr *daddr,
3123                       const struct in6_addr *saddr)
3124 {
3125         const struct fib6_nh *nh = res->nh;
3126         struct fib6_info *f6i = res->f6i;
3127         struct inet6_dev *idev;
3128         struct rt6_info *rt;
3129         u32 mtu = 0;
3130
3131         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3132                 mtu = f6i->fib6_pmtu;
3133                 if (mtu)
3134                         goto out;
3135         }
3136
3137         rt = rt6_find_cached_rt(res, daddr, saddr);
3138         if (unlikely(rt)) {
3139                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3140         } else {
3141                 struct net_device *dev = nh->fib_nh_dev;
3142
3143                 mtu = IPV6_MIN_MTU;
3144                 idev = __in6_dev_get(dev);
3145                 if (idev && idev->cnf.mtu6 > mtu)
3146                         mtu = idev->cnf.mtu6;
3147         }
3148
3149         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3150 out:
3151         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3152 }
3153
3154 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3155                                   struct flowi6 *fl6)
3156 {
3157         struct dst_entry *dst;
3158         struct rt6_info *rt;
3159         struct inet6_dev *idev = in6_dev_get(dev);
3160         struct net *net = dev_net(dev);
3161
3162         if (unlikely(!idev))
3163                 return ERR_PTR(-ENODEV);
3164
3165         rt = ip6_dst_alloc(net, dev, 0);
3166         if (unlikely(!rt)) {
3167                 in6_dev_put(idev);
3168                 dst = ERR_PTR(-ENOMEM);
3169                 goto out;
3170         }
3171
3172         rt->dst.input = ip6_input;
3173         rt->dst.output  = ip6_output;
3174         rt->rt6i_gateway  = fl6->daddr;
3175         rt->rt6i_dst.addr = fl6->daddr;
3176         rt->rt6i_dst.plen = 128;
3177         rt->rt6i_idev     = idev;
3178         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3179
3180         /* Add this dst into uncached_list so that rt6_disable_ip() can
3181          * do proper release of the net_device
3182          */
3183         rt6_uncached_list_add(rt);
3184         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3185
3186         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3187
3188 out:
3189         return dst;
3190 }
3191
3192 static int ip6_dst_gc(struct dst_ops *ops)
3193 {
3194         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3195         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3196         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3197         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3198         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3199         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3200         int entries;
3201
3202         entries = dst_entries_get_fast(ops);
3203         if (entries > rt_max_size)
3204                 entries = dst_entries_get_slow(ops);
3205
3206         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3207             entries <= rt_max_size)
3208                 goto out;
3209
3210         net->ipv6.ip6_rt_gc_expire++;
3211         fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
3212         entries = dst_entries_get_slow(ops);
3213         if (entries < ops->gc_thresh)
3214                 net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
3215 out:
3216         net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
3217         return entries > rt_max_size;
3218 }
3219
3220 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3221                                const struct in6_addr *gw_addr, u32 tbid,
3222                                int flags, struct fib6_result *res)
3223 {
3224         struct flowi6 fl6 = {
3225                 .flowi6_oif = cfg->fc_ifindex,
3226                 .daddr = *gw_addr,
3227                 .saddr = cfg->fc_prefsrc,
3228         };
3229         struct fib6_table *table;
3230         int err;
3231
3232         table = fib6_get_table(net, tbid);
3233         if (!table)
3234                 return -EINVAL;
3235
3236         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3237                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3238
3239         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3240
3241         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3242         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3243                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3244                                  cfg->fc_ifindex != 0, NULL, flags);
3245
3246         return err;
3247 }
3248
3249 static int ip6_route_check_nh_onlink(struct net *net,
3250                                      struct fib6_config *cfg,
3251                                      const struct net_device *dev,
3252                                      struct netlink_ext_ack *extack)
3253 {
3254         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3255         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3256         struct fib6_result res = {};
3257         int err;
3258
3259         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3260         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3261             /* ignore match if it is the default route */
3262             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3263             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3264                 NL_SET_ERR_MSG(extack,
3265                                "Nexthop has invalid gateway or device mismatch");
3266                 err = -EINVAL;
3267         }
3268
3269         return err;
3270 }
3271
3272 static int ip6_route_check_nh(struct net *net,
3273                               struct fib6_config *cfg,
3274                               struct net_device **_dev,
3275                               struct inet6_dev **idev)
3276 {
3277         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3278         struct net_device *dev = _dev ? *_dev : NULL;
3279         int flags = RT6_LOOKUP_F_IFACE;
3280         struct fib6_result res = {};
3281         int err = -EHOSTUNREACH;
3282
3283         if (cfg->fc_table) {
3284                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3285                                           cfg->fc_table, flags, &res);
3286                 /* gw_addr can not require a gateway or resolve to a reject
3287                  * route. If a device is given, it must match the result.
3288                  */
3289                 if (err || res.fib6_flags & RTF_REJECT ||
3290                     res.nh->fib_nh_gw_family ||
3291                     (dev && dev != res.nh->fib_nh_dev))
3292                         err = -EHOSTUNREACH;
3293         }
3294
3295         if (err < 0) {
3296                 struct flowi6 fl6 = {
3297                         .flowi6_oif = cfg->fc_ifindex,
3298                         .daddr = *gw_addr,
3299                 };
3300
3301                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3302                 if (err || res.fib6_flags & RTF_REJECT ||
3303                     res.nh->fib_nh_gw_family)
3304                         err = -EHOSTUNREACH;
3305
3306                 if (err)
3307                         return err;
3308
3309                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3310                                  cfg->fc_ifindex != 0, NULL, flags);
3311         }
3312
3313         err = 0;
3314         if (dev) {
3315                 if (dev != res.nh->fib_nh_dev)
3316                         err = -EHOSTUNREACH;
3317         } else {
3318                 *_dev = dev = res.nh->fib_nh_dev;
3319                 dev_hold(dev);
3320                 *idev = in6_dev_get(dev);
3321         }
3322
3323         return err;
3324 }
3325
3326 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3327                            struct net_device **_dev, struct inet6_dev **idev,
3328                            struct netlink_ext_ack *extack)
3329 {
3330         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3331         int gwa_type = ipv6_addr_type(gw_addr);
3332         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3333         const struct net_device *dev = *_dev;
3334         bool need_addr_check = !dev;
3335         int err = -EINVAL;
3336
3337         /* if gw_addr is local we will fail to detect this in case
3338          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3339          * will return already-added prefix route via interface that
3340          * prefix route was assigned to, which might be non-loopback.
3341          */
3342         if (dev &&
3343             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3344                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3345                 goto out;
3346         }
3347
3348         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3349                 /* IPv6 strictly inhibits using not link-local
3350                  * addresses as nexthop address.
3351                  * Otherwise, router will not able to send redirects.
3352                  * It is very good, but in some (rare!) circumstances
3353                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3354                  * some exceptions. --ANK
3355                  * We allow IPv4-mapped nexthops to support RFC4798-type
3356                  * addressing
3357                  */
3358                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3359                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3360                         goto out;
3361                 }
3362
3363                 rcu_read_lock();
3364
3365                 if (cfg->fc_flags & RTNH_F_ONLINK)
3366                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3367                 else
3368                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3369
3370                 rcu_read_unlock();
3371
3372                 if (err)
3373                         goto out;
3374         }
3375
3376         /* reload in case device was changed */
3377         dev = *_dev;
3378
3379         err = -EINVAL;
3380         if (!dev) {
3381                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3382                 goto out;
3383         } else if (dev->flags & IFF_LOOPBACK) {
3384                 NL_SET_ERR_MSG(extack,
3385                                "Egress device can not be loopback device for this route");
3386                 goto out;
3387         }
3388
3389         /* if we did not check gw_addr above, do so now that the
3390          * egress device has been resolved.
3391          */
3392         if (need_addr_check &&
3393             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3394                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3395                 goto out;
3396         }
3397
3398         err = 0;
3399 out:
3400         return err;
3401 }
3402
3403 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3404 {
3405         if ((flags & RTF_REJECT) ||
3406             (dev && (dev->flags & IFF_LOOPBACK) &&
3407              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3408              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3409                 return true;
3410
3411         return false;
3412 }
3413
3414 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3415                  struct fib6_config *cfg, gfp_t gfp_flags,
3416                  struct netlink_ext_ack *extack)
3417 {
3418         struct net_device *dev = NULL;
3419         struct inet6_dev *idev = NULL;
3420         int addr_type;
3421         int err;
3422
3423         fib6_nh->fib_nh_family = AF_INET6;
3424 #ifdef CONFIG_IPV6_ROUTER_PREF
3425         fib6_nh->last_probe = jiffies;
3426 #endif
3427         if (cfg->fc_is_fdb) {
3428                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3429                 fib6_nh->fib_nh_gw_family = AF_INET6;
3430                 return 0;
3431         }
3432
3433         err = -ENODEV;
3434         if (cfg->fc_ifindex) {
3435                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3436                 if (!dev)
3437                         goto out;
3438                 idev = in6_dev_get(dev);
3439                 if (!idev)
3440                         goto out;
3441         }
3442
3443         if (cfg->fc_flags & RTNH_F_ONLINK) {
3444                 if (!dev) {
3445                         NL_SET_ERR_MSG(extack,
3446                                        "Nexthop device required for onlink");
3447                         goto out;
3448                 }
3449
3450                 if (!(dev->flags & IFF_UP)) {
3451                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3452                         err = -ENETDOWN;
3453                         goto out;
3454                 }
3455
3456                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3457         }
3458
3459         fib6_nh->fib_nh_weight = 1;
3460
3461         /* We cannot add true routes via loopback here,
3462          * they would result in kernel looping; promote them to reject routes
3463          */
3464         addr_type = ipv6_addr_type(&cfg->fc_dst);
3465         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3466                 /* hold loopback dev/idev if we haven't done so. */
3467                 if (dev != net->loopback_dev) {
3468                         if (dev) {
3469                                 dev_put(dev);
3470                                 in6_dev_put(idev);
3471                         }
3472                         dev = net->loopback_dev;
3473                         dev_hold(dev);
3474                         idev = in6_dev_get(dev);
3475                         if (!idev) {
3476                                 err = -ENODEV;
3477                                 goto out;
3478                         }
3479                 }
3480                 goto pcpu_alloc;
3481         }
3482
3483         if (cfg->fc_flags & RTF_GATEWAY) {
3484                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3485                 if (err)
3486                         goto out;
3487
3488                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3489                 fib6_nh->fib_nh_gw_family = AF_INET6;
3490         }
3491
3492         err = -ENODEV;
3493         if (!dev)
3494                 goto out;
3495
3496         if (idev->cnf.disable_ipv6) {
3497                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3498                 err = -EACCES;
3499                 goto out;
3500         }
3501
3502         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3503                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3504                 err = -ENETDOWN;
3505                 goto out;
3506         }
3507
3508         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3509             !netif_carrier_ok(dev))
3510                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3511
3512         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3513                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3514         if (err)
3515                 goto out;
3516
3517 pcpu_alloc:
3518         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3519         if (!fib6_nh->rt6i_pcpu) {
3520                 err = -ENOMEM;
3521                 goto out;
3522         }
3523
3524         fib6_nh->fib_nh_dev = dev;
3525         fib6_nh->fib_nh_oif = dev->ifindex;
3526         err = 0;
3527 out:
3528         if (idev)
3529                 in6_dev_put(idev);
3530
3531         if (err) {
3532                 lwtstate_put(fib6_nh->fib_nh_lws);
3533                 fib6_nh->fib_nh_lws = NULL;
3534                 if (dev)
3535                         dev_put(dev);
3536         }
3537
3538         return err;
3539 }
3540
3541 void fib6_nh_release(struct fib6_nh *fib6_nh)
3542 {
3543         struct rt6_exception_bucket *bucket;
3544
3545         rcu_read_lock();
3546
3547         fib6_nh_flush_exceptions(fib6_nh, NULL);
3548         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3549         if (bucket) {
3550                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3551                 kfree(bucket);
3552         }
3553
3554         rcu_read_unlock();
3555
3556         if (fib6_nh->rt6i_pcpu) {
3557                 int cpu;
3558
3559                 for_each_possible_cpu(cpu) {
3560                         struct rt6_info **ppcpu_rt;
3561                         struct rt6_info *pcpu_rt;
3562
3563                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3564                         pcpu_rt = *ppcpu_rt;
3565                         if (pcpu_rt) {
3566                                 dst_dev_put(&pcpu_rt->dst);
3567                                 dst_release(&pcpu_rt->dst);
3568                                 *ppcpu_rt = NULL;
3569                         }
3570                 }
3571
3572                 free_percpu(fib6_nh->rt6i_pcpu);
3573         }
3574
3575         fib_nh_common_release(&fib6_nh->nh_common);
3576 }
3577
3578 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3579                                               gfp_t gfp_flags,
3580                                               struct netlink_ext_ack *extack)
3581 {
3582         struct net *net = cfg->fc_nlinfo.nl_net;
3583         struct fib6_info *rt = NULL;
3584         struct nexthop *nh = NULL;
3585         struct fib6_table *table;
3586         struct fib6_nh *fib6_nh;
3587         int err = -EINVAL;
3588         int addr_type;
3589
3590         /* RTF_PCPU is an internal flag; can not be set by userspace */
3591         if (cfg->fc_flags & RTF_PCPU) {
3592                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3593                 goto out;
3594         }
3595
3596         /* RTF_CACHE is an internal flag; can not be set by userspace */
3597         if (cfg->fc_flags & RTF_CACHE) {
3598                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3599                 goto out;
3600         }
3601
3602         if (cfg->fc_type > RTN_MAX) {
3603                 NL_SET_ERR_MSG(extack, "Invalid route type");
3604                 goto out;
3605         }
3606
3607         if (cfg->fc_dst_len > 128) {
3608                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3609                 goto out;
3610         }
3611         if (cfg->fc_src_len > 128) {
3612                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3613                 goto out;
3614         }
3615 #ifndef CONFIG_IPV6_SUBTREES
3616         if (cfg->fc_src_len) {
3617                 NL_SET_ERR_MSG(extack,
3618                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3619                 goto out;
3620         }
3621 #endif
3622         if (cfg->fc_nh_id) {
3623                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3624                 if (!nh) {
3625                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3626                         goto out;
3627                 }
3628                 err = fib6_check_nexthop(nh, cfg, extack);
3629                 if (err)
3630                         goto out;
3631         }
3632
3633         err = -ENOBUFS;
3634         if (cfg->fc_nlinfo.nlh &&
3635             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3636                 table = fib6_get_table(net, cfg->fc_table);
3637                 if (!table) {
3638                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3639                         table = fib6_new_table(net, cfg->fc_table);
3640                 }
3641         } else {
3642                 table = fib6_new_table(net, cfg->fc_table);
3643         }
3644
3645         if (!table)
3646                 goto out;
3647
3648         err = -ENOMEM;
3649         rt = fib6_info_alloc(gfp_flags, !nh);
3650         if (!rt)
3651                 goto out;
3652
3653         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3654                                                extack);
3655         if (IS_ERR(rt->fib6_metrics)) {
3656                 err = PTR_ERR(rt->fib6_metrics);
3657                 /* Do not leave garbage there. */
3658                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3659                 goto out;
3660         }
3661
3662         if (cfg->fc_flags & RTF_ADDRCONF)
3663                 rt->dst_nocount = true;
3664
3665         if (cfg->fc_flags & RTF_EXPIRES)
3666                 fib6_set_expires(rt, jiffies +
3667                                 clock_t_to_jiffies(cfg->fc_expires));
3668         else
3669                 fib6_clean_expires(rt);
3670
3671         if (cfg->fc_protocol == RTPROT_UNSPEC)
3672                 cfg->fc_protocol = RTPROT_BOOT;
3673         rt->fib6_protocol = cfg->fc_protocol;
3674
3675         rt->fib6_table = table;
3676         rt->fib6_metric = cfg->fc_metric;
3677         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3678         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3679
3680         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3681         rt->fib6_dst.plen = cfg->fc_dst_len;
3682
3683 #ifdef CONFIG_IPV6_SUBTREES
3684         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3685         rt->fib6_src.plen = cfg->fc_src_len;
3686 #endif
3687         if (nh) {
3688                 if (!nexthop_get(nh)) {
3689                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3690                         goto out;
3691                 }
3692                 if (rt->fib6_src.plen) {
3693                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3694                         goto out;
3695                 }
3696                 rt->nh = nh;
3697                 fib6_nh = nexthop_fib6_nh(rt->nh);
3698         } else {
3699                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3700                 if (err)
3701                         goto out;
3702
3703                 fib6_nh = rt->fib6_nh;
3704
3705                 /* We cannot add true routes via loopback here, they would
3706                  * result in kernel looping; promote them to reject routes
3707                  */
3708                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3709                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3710                                    addr_type))
3711                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3712         }
3713
3714         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3715                 struct net_device *dev = fib6_nh->fib_nh_dev;
3716
3717                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3718                         NL_SET_ERR_MSG(extack, "Invalid source address");
3719                         err = -EINVAL;
3720                         goto out;
3721                 }
3722                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3723                 rt->fib6_prefsrc.plen = 128;
3724         } else
3725                 rt->fib6_prefsrc.plen = 0;
3726
3727         return rt;
3728 out:
3729         fib6_info_release(rt);
3730         return ERR_PTR(err);
3731 }
3732
3733 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3734                   struct netlink_ext_ack *extack)
3735 {
3736         struct fib6_info *rt;
3737         int err;
3738
3739         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3740         if (IS_ERR(rt))
3741                 return PTR_ERR(rt);
3742
3743         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3744         fib6_info_release(rt);
3745
3746         return err;
3747 }
3748
3749 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3750 {
3751         struct net *net = info->nl_net;
3752         struct fib6_table *table;
3753         int err;
3754
3755         if (rt == net->ipv6.fib6_null_entry) {
3756                 err = -ENOENT;
3757                 goto out;
3758         }
3759
3760         table = rt->fib6_table;
3761         spin_lock_bh(&table->tb6_lock);
3762         err = fib6_del(rt, info);
3763         spin_unlock_bh(&table->tb6_lock);
3764
3765 out:
3766         fib6_info_release(rt);
3767         return err;
3768 }
3769
3770 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3771 {
3772         struct nl_info info = {
3773                 .nl_net = net,
3774                 .skip_notify = skip_notify
3775         };
3776
3777         return __ip6_del_rt(rt, &info);
3778 }
3779
3780 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3781 {
3782         struct nl_info *info = &cfg->fc_nlinfo;
3783         struct net *net = info->nl_net;
3784         struct sk_buff *skb = NULL;
3785         struct fib6_table *table;
3786         int err = -ENOENT;
3787
3788         if (rt == net->ipv6.fib6_null_entry)
3789                 goto out_put;
3790         table = rt->fib6_table;
3791         spin_lock_bh(&table->tb6_lock);
3792
3793         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3794                 struct fib6_info *sibling, *next_sibling;
3795                 struct fib6_node *fn;
3796
3797                 /* prefer to send a single notification with all hops */
3798                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3799                 if (skb) {
3800                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3801
3802                         if (rt6_fill_node(net, skb, rt, NULL,
3803                                           NULL, NULL, 0, RTM_DELROUTE,
3804                                           info->portid, seq, 0) < 0) {
3805                                 kfree_skb(skb);
3806                                 skb = NULL;
3807                         } else
3808                                 info->skip_notify = 1;
3809                 }
3810
3811                 /* 'rt' points to the first sibling route. If it is not the
3812                  * leaf, then we do not need to send a notification. Otherwise,
3813                  * we need to check if the last sibling has a next route or not
3814                  * and emit a replace or delete notification, respectively.
3815                  */
3816                 info->skip_notify_kernel = 1;
3817                 fn = rcu_dereference_protected(rt->fib6_node,
3818                                             lockdep_is_held(&table->tb6_lock));
3819                 if (rcu_access_pointer(fn->leaf) == rt) {
3820                         struct fib6_info *last_sibling, *replace_rt;
3821
3822                         last_sibling = list_last_entry(&rt->fib6_siblings,
3823                                                        struct fib6_info,
3824                                                        fib6_siblings);
3825                         replace_rt = rcu_dereference_protected(
3826                                             last_sibling->fib6_next,
3827                                             lockdep_is_held(&table->tb6_lock));
3828                         if (replace_rt)
3829                                 call_fib6_entry_notifiers_replace(net,
3830                                                                   replace_rt);
3831                         else
3832                                 call_fib6_multipath_entry_notifiers(net,
3833                                                        FIB_EVENT_ENTRY_DEL,
3834                                                        rt, rt->fib6_nsiblings,
3835                                                        NULL);
3836                 }
3837                 list_for_each_entry_safe(sibling, next_sibling,
3838                                          &rt->fib6_siblings,
3839                                          fib6_siblings) {
3840                         err = fib6_del(sibling, info);
3841                         if (err)
3842                                 goto out_unlock;
3843                 }
3844         }
3845
3846         err = fib6_del(rt, info);
3847 out_unlock:
3848         spin_unlock_bh(&table->tb6_lock);
3849 out_put:
3850         fib6_info_release(rt);
3851
3852         if (skb) {
3853                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3854                             info->nlh, gfp_any());
3855         }
3856         return err;
3857 }
3858
3859 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3860 {
3861         int rc = -ESRCH;
3862
3863         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3864                 goto out;
3865
3866         if (cfg->fc_flags & RTF_GATEWAY &&
3867             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3868                 goto out;
3869
3870         rc = rt6_remove_exception_rt(rt);
3871 out:
3872         return rc;
3873 }
3874
3875 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3876                              struct fib6_nh *nh)
3877 {
3878         struct fib6_result res = {
3879                 .f6i = rt,
3880                 .nh = nh,
3881         };
3882         struct rt6_info *rt_cache;
3883
3884         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3885         if (rt_cache)
3886                 return __ip6_del_cached_rt(rt_cache, cfg);
3887
3888         return 0;
3889 }
3890
3891 struct fib6_nh_del_cached_rt_arg {
3892         struct fib6_config *cfg;
3893         struct fib6_info *f6i;
3894 };
3895
3896 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3897 {
3898         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3899         int rc;
3900
3901         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3902         return rc != -ESRCH ? rc : 0;
3903 }
3904
3905 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3906 {
3907         struct fib6_nh_del_cached_rt_arg arg = {
3908                 .cfg = cfg,
3909                 .f6i = f6i
3910         };
3911
3912         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3913 }
3914
3915 static int ip6_route_del(struct fib6_config *cfg,
3916                          struct netlink_ext_ack *extack)
3917 {
3918         struct fib6_table *table;
3919         struct fib6_info *rt;
3920         struct fib6_node *fn;
3921         int err = -ESRCH;
3922
3923         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3924         if (!table) {
3925                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3926                 return err;
3927         }
3928
3929         rcu_read_lock();
3930
3931         fn = fib6_locate(&table->tb6_root,
3932                          &cfg->fc_dst, cfg->fc_dst_len,
3933                          &cfg->fc_src, cfg->fc_src_len,
3934                          !(cfg->fc_flags & RTF_CACHE));
3935
3936         if (fn) {
3937                 for_each_fib6_node_rt_rcu(fn) {
3938                         struct fib6_nh *nh;
3939
3940                         if (rt->nh && cfg->fc_nh_id &&
3941                             rt->nh->id != cfg->fc_nh_id)
3942                                 continue;
3943
3944                         if (cfg->fc_flags & RTF_CACHE) {
3945                                 int rc = 0;
3946
3947                                 if (rt->nh) {
3948                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3949                                 } else if (cfg->fc_nh_id) {
3950                                         continue;
3951                                 } else {
3952                                         nh = rt->fib6_nh;
3953                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3954                                 }
3955                                 if (rc != -ESRCH) {
3956                                         rcu_read_unlock();
3957                                         return rc;
3958                                 }
3959                                 continue;
3960                         }
3961
3962                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3963                                 continue;
3964                         if (cfg->fc_protocol &&
3965                             cfg->fc_protocol != rt->fib6_protocol)
3966                                 continue;
3967
3968                         if (rt->nh) {
3969                                 if (!fib6_info_hold_safe(rt))
3970                                         continue;
3971                                 rcu_read_unlock();
3972
3973                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3974                         }
3975                         if (cfg->fc_nh_id)
3976                                 continue;
3977
3978                         nh = rt->fib6_nh;
3979                         if (cfg->fc_ifindex &&
3980                             (!nh->fib_nh_dev ||
3981                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
3982                                 continue;
3983                         if (cfg->fc_flags & RTF_GATEWAY &&
3984                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
3985                                 continue;
3986                         if (!fib6_info_hold_safe(rt))
3987                                 continue;
3988                         rcu_read_unlock();
3989
3990                         /* if gateway was specified only delete the one hop */
3991                         if (cfg->fc_flags & RTF_GATEWAY)
3992                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3993
3994                         return __ip6_del_rt_siblings(rt, cfg);
3995                 }
3996         }
3997         rcu_read_unlock();
3998
3999         return err;
4000 }
4001
4002 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4003 {
4004         struct netevent_redirect netevent;
4005         struct rt6_info *rt, *nrt = NULL;
4006         struct fib6_result res = {};
4007         struct ndisc_options ndopts;
4008         struct inet6_dev *in6_dev;
4009         struct neighbour *neigh;
4010         struct rd_msg *msg;
4011         int optlen, on_link;
4012         u8 *lladdr;
4013
4014         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4015         optlen -= sizeof(*msg);
4016
4017         if (optlen < 0) {
4018                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4019                 return;
4020         }
4021
4022         msg = (struct rd_msg *)icmp6_hdr(skb);
4023
4024         if (ipv6_addr_is_multicast(&msg->dest)) {
4025                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4026                 return;
4027         }
4028
4029         on_link = 0;
4030         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4031                 on_link = 1;
4032         } else if (ipv6_addr_type(&msg->target) !=
4033                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4034                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4035                 return;
4036         }
4037
4038         in6_dev = __in6_dev_get(skb->dev);
4039         if (!in6_dev)
4040                 return;
4041         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4042                 return;
4043
4044         /* RFC2461 8.1:
4045          *      The IP source address of the Redirect MUST be the same as the current
4046          *      first-hop router for the specified ICMP Destination Address.
4047          */
4048
4049         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4050                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4051                 return;
4052         }
4053
4054         lladdr = NULL;
4055         if (ndopts.nd_opts_tgt_lladdr) {
4056                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4057                                              skb->dev);
4058                 if (!lladdr) {
4059                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4060                         return;
4061                 }
4062         }
4063
4064         rt = (struct rt6_info *) dst;
4065         if (rt->rt6i_flags & RTF_REJECT) {
4066                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4067                 return;
4068         }
4069
4070         /* Redirect received -> path was valid.
4071          * Look, redirects are sent only in response to data packets,
4072          * so that this nexthop apparently is reachable. --ANK
4073          */
4074         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4075
4076         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4077         if (!neigh)
4078                 return;
4079
4080         /*
4081          *      We have finally decided to accept it.
4082          */
4083
4084         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4085                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4086                      NEIGH_UPDATE_F_OVERRIDE|
4087                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4088                                      NEIGH_UPDATE_F_ISROUTER)),
4089                      NDISC_REDIRECT, &ndopts);
4090
4091         rcu_read_lock();
4092         res.f6i = rcu_dereference(rt->from);
4093         if (!res.f6i)
4094                 goto out;
4095
4096         if (res.f6i->nh) {
4097                 struct fib6_nh_match_arg arg = {
4098                         .dev = dst->dev,
4099                         .gw = &rt->rt6i_gateway,
4100                 };
4101
4102                 nexthop_for_each_fib6_nh(res.f6i->nh,
4103                                          fib6_nh_find_match, &arg);
4104
4105                 /* fib6_info uses a nexthop that does not have fib6_nh
4106                  * using the dst->dev. Should be impossible
4107                  */
4108                 if (!arg.match)
4109                         goto out;
4110                 res.nh = arg.match;
4111         } else {
4112                 res.nh = res.f6i->fib6_nh;
4113         }
4114
4115         res.fib6_flags = res.f6i->fib6_flags;
4116         res.fib6_type = res.f6i->fib6_type;
4117         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4118         if (!nrt)
4119                 goto out;
4120
4121         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4122         if (on_link)
4123                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4124
4125         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4126
4127         /* rt6_insert_exception() will take care of duplicated exceptions */
4128         if (rt6_insert_exception(nrt, &res)) {
4129                 dst_release_immediate(&nrt->dst);
4130                 goto out;
4131         }
4132
4133         netevent.old = &rt->dst;
4134         netevent.new = &nrt->dst;
4135         netevent.daddr = &msg->dest;
4136         netevent.neigh = neigh;
4137         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4138
4139 out:
4140         rcu_read_unlock();
4141         neigh_release(neigh);
4142 }
4143
4144 #ifdef CONFIG_IPV6_ROUTE_INFO
4145 static struct fib6_info *rt6_get_route_info(struct net *net,
4146                                            const struct in6_addr *prefix, int prefixlen,
4147                                            const struct in6_addr *gwaddr,
4148                                            struct net_device *dev)
4149 {
4150         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4151         int ifindex = dev->ifindex;
4152         struct fib6_node *fn;
4153         struct fib6_info *rt = NULL;
4154         struct fib6_table *table;
4155
4156         table = fib6_get_table(net, tb_id);
4157         if (!table)
4158                 return NULL;
4159
4160         rcu_read_lock();
4161         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4162         if (!fn)
4163                 goto out;
4164
4165         for_each_fib6_node_rt_rcu(fn) {
4166                 /* these routes do not use nexthops */
4167                 if (rt->nh)
4168                         continue;
4169                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4170                         continue;
4171                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4172                     !rt->fib6_nh->fib_nh_gw_family)
4173                         continue;
4174                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4175                         continue;
4176                 if (!fib6_info_hold_safe(rt))
4177                         continue;
4178                 break;
4179         }
4180 out:
4181         rcu_read_unlock();
4182         return rt;
4183 }
4184
4185 static struct fib6_info *rt6_add_route_info(struct net *net,
4186                                            const struct in6_addr *prefix, int prefixlen,
4187                                            const struct in6_addr *gwaddr,
4188                                            struct net_device *dev,
4189                                            unsigned int pref)
4190 {
4191         struct fib6_config cfg = {
4192                 .fc_metric      = IP6_RT_PRIO_USER,
4193                 .fc_ifindex     = dev->ifindex,
4194                 .fc_dst_len     = prefixlen,
4195                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4196                                   RTF_UP | RTF_PREF(pref),
4197                 .fc_protocol = RTPROT_RA,
4198                 .fc_type = RTN_UNICAST,
4199                 .fc_nlinfo.portid = 0,
4200                 .fc_nlinfo.nlh = NULL,
4201                 .fc_nlinfo.nl_net = net,
4202         };
4203
4204         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO,
4205         cfg.fc_dst = *prefix;
4206         cfg.fc_gateway = *gwaddr;
4207
4208         /* We should treat it as a default route if prefix length is 0. */
4209         if (!prefixlen)
4210                 cfg.fc_flags |= RTF_DEFAULT;
4211
4212         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4213
4214         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4215 }
4216 #endif
4217
4218 struct fib6_info *rt6_get_dflt_router(struct net *net,
4219                                      const struct in6_addr *addr,
4220                                      struct net_device *dev)
4221 {
4222         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4223         struct fib6_info *rt;
4224         struct fib6_table *table;
4225
4226         table = fib6_get_table(net, tb_id);
4227         if (!table)
4228                 return NULL;
4229
4230         rcu_read_lock();
4231         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4232                 struct fib6_nh *nh;
4233
4234                 /* RA routes do not use nexthops */
4235                 if (rt->nh)
4236                         continue;
4237
4238                 nh = rt->fib6_nh;
4239                 if (dev == nh->fib_nh_dev &&
4240                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4241                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4242                         break;
4243         }
4244         if (rt && !fib6_info_hold_safe(rt))
4245                 rt = NULL;
4246         rcu_read_unlock();
4247         return rt;
4248 }
4249
4250 struct fib6_info *rt6_add_dflt_router(struct net *net,
4251                                      const struct in6_addr *gwaddr,
4252                                      struct net_device *dev,
4253                                      unsigned int pref)
4254 {
4255         struct fib6_config cfg = {
4256                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4257                 .fc_metric      = IP6_RT_PRIO_USER,
4258                 .fc_ifindex     = dev->ifindex,
4259                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4260                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4261                 .fc_protocol = RTPROT_RA,
4262                 .fc_type = RTN_UNICAST,
4263                 .fc_nlinfo.portid = 0,
4264                 .fc_nlinfo.nlh = NULL,
4265                 .fc_nlinfo.nl_net = net,
4266         };
4267
4268         cfg.fc_gateway = *gwaddr;
4269
4270         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4271                 struct fib6_table *table;
4272
4273                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4274                 if (table)
4275                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4276         }
4277
4278         return rt6_get_dflt_router(net, gwaddr, dev);
4279 }
4280
4281 static void __rt6_purge_dflt_routers(struct net *net,
4282                                      struct fib6_table *table)
4283 {
4284         struct fib6_info *rt;
4285
4286 restart:
4287         rcu_read_lock();
4288         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4289                 struct net_device *dev = fib6_info_nh_dev(rt);
4290                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4291
4292                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4293                     (!idev || idev->cnf.accept_ra != 2) &&
4294                     fib6_info_hold_safe(rt)) {
4295                         rcu_read_unlock();
4296                         ip6_del_rt(net, rt, false);
4297                         goto restart;
4298                 }
4299         }
4300         rcu_read_unlock();
4301
4302         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4303 }
4304
4305 void rt6_purge_dflt_routers(struct net *net)
4306 {
4307         struct fib6_table *table;
4308         struct hlist_head *head;
4309         unsigned int h;
4310
4311         rcu_read_lock();
4312
4313         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4314                 head = &net->ipv6.fib_table_hash[h];
4315                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4316                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4317                                 __rt6_purge_dflt_routers(net, table);
4318                 }
4319         }
4320
4321         rcu_read_unlock();
4322 }
4323
4324 static void rtmsg_to_fib6_config(struct net *net,
4325                                  struct in6_rtmsg *rtmsg,
4326                                  struct fib6_config *cfg)
4327 {
4328         *cfg = (struct fib6_config){
4329                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4330                          : RT6_TABLE_MAIN,
4331                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4332                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4333                 .fc_expires = rtmsg->rtmsg_info,
4334                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4335                 .fc_src_len = rtmsg->rtmsg_src_len,
4336                 .fc_flags = rtmsg->rtmsg_flags,
4337                 .fc_type = rtmsg->rtmsg_type,
4338
4339                 .fc_nlinfo.nl_net = net,
4340
4341                 .fc_dst = rtmsg->rtmsg_dst,
4342                 .fc_src = rtmsg->rtmsg_src,
4343                 .fc_gateway = rtmsg->rtmsg_gateway,
4344         };
4345 }
4346
4347 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4348 {
4349         struct fib6_config cfg;
4350         int err;
4351
4352         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4353                 return -EINVAL;
4354         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4355                 return -EPERM;
4356
4357         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4358
4359         rtnl_lock();
4360         switch (cmd) {
4361         case SIOCADDRT:
4362                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4363                 break;
4364         case SIOCDELRT:
4365                 err = ip6_route_del(&cfg, NULL);
4366                 break;
4367         }
4368         rtnl_unlock();
4369         return err;
4370 }
4371
4372 /*
4373  *      Drop the packet on the floor
4374  */
4375
4376 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4377 {
4378         struct dst_entry *dst = skb_dst(skb);
4379         struct net *net = dev_net(dst->dev);
4380         struct inet6_dev *idev;
4381         int type;
4382
4383         if (netif_is_l3_master(skb->dev) &&
4384             dst->dev == net->loopback_dev)
4385                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4386         else
4387                 idev = ip6_dst_idev(dst);
4388
4389         switch (ipstats_mib_noroutes) {
4390         case IPSTATS_MIB_INNOROUTES:
4391                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4392                 if (type == IPV6_ADDR_ANY) {
4393                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4394                         break;
4395                 }
4396                 fallthrough;
4397         case IPSTATS_MIB_OUTNOROUTES:
4398                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4399                 break;
4400         }
4401
4402         /* Start over by dropping the dst for l3mdev case */
4403         if (netif_is_l3_master(skb->dev))
4404                 skb_dst_drop(skb);
4405
4406         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4407         kfree_skb(skb);
4408         return 0;
4409 }
4410
4411 static int ip6_pkt_discard(struct sk_buff *skb)
4412 {
4413         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4414 }
4415
4416 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4417 {
4418         skb->dev = skb_dst(skb)->dev;
4419         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4420 }
4421
4422 static int ip6_pkt_prohibit(struct sk_buff *skb)
4423 {
4424         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4425 }
4426
4427 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4428 {
4429         skb->dev = skb_dst(skb)->dev;
4430         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4431 }
4432
4433 /*
4434  *      Allocate a dst for local (unicast / anycast) address.
4435  */
4436
4437 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4438                                      struct inet6_dev *idev,
4439                                      const struct in6_addr *addr,
4440                                      bool anycast, gfp_t gfp_flags)
4441 {
4442         struct fib6_config cfg = {
4443                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4444                 .fc_ifindex = idev->dev->ifindex,
4445                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4446                 .fc_dst = *addr,
4447                 .fc_dst_len = 128,
4448                 .fc_protocol = RTPROT_KERNEL,
4449                 .fc_nlinfo.nl_net = net,
4450                 .fc_ignore_dev_down = true,
4451         };
4452         struct fib6_info *f6i;
4453
4454         if (anycast) {
4455                 cfg.fc_type = RTN_ANYCAST;
4456                 cfg.fc_flags |= RTF_ANYCAST;
4457         } else {
4458                 cfg.fc_type = RTN_LOCAL;
4459                 cfg.fc_flags |= RTF_LOCAL;
4460         }
4461
4462         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4463         if (!IS_ERR(f6i))
4464                 f6i->dst_nocount = true;
4465         return f6i;
4466 }
4467
4468 /* remove deleted ip from prefsrc entries */
4469 struct arg_dev_net_ip {
4470         struct net_device *dev;
4471         struct net *net;
4472         struct in6_addr *addr;
4473 };
4474
4475 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4476 {
4477         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4478         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4479         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4480
4481         if (!rt->nh &&
4482             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4483             rt != net->ipv6.fib6_null_entry &&
4484             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4485                 spin_lock_bh(&rt6_exception_lock);
4486                 /* remove prefsrc entry */
4487                 rt->fib6_prefsrc.plen = 0;
4488                 spin_unlock_bh(&rt6_exception_lock);
4489         }
4490         return 0;
4491 }
4492
4493 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4494 {
4495         struct net *net = dev_net(ifp->idev->dev);
4496         struct arg_dev_net_ip adni = {
4497                 .dev = ifp->idev->dev,
4498                 .net = net,
4499                 .addr = &ifp->addr,
4500         };
4501         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4502 }
4503
4504 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4505
4506 /* Remove routers and update dst entries when gateway turn into host. */
4507 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4508 {
4509         struct in6_addr *gateway = (struct in6_addr *)arg;
4510         struct fib6_nh *nh;
4511
4512         /* RA routes do not use nexthops */
4513         if (rt->nh)
4514                 return 0;
4515
4516         nh = rt->fib6_nh;
4517         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4518             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4519                 return -1;
4520
4521         /* Further clean up cached routes in exception table.
4522          * This is needed because cached route may have a different
4523          * gateway than its 'parent' in the case of an ip redirect.
4524          */
4525         fib6_nh_exceptions_clean_tohost(nh, gateway);
4526
4527         return 0;
4528 }
4529
4530 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4531 {
4532         fib6_clean_all(net, fib6_clean_tohost, gateway);
4533 }
4534
4535 struct arg_netdev_event {
4536         const struct net_device *dev;
4537         union {
4538                 unsigned char nh_flags;
4539                 unsigned long event;
4540         };
4541 };
4542
4543 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4544 {
4545         struct fib6_info *iter;
4546         struct fib6_node *fn;
4547
4548         fn = rcu_dereference_protected(rt->fib6_node,
4549                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4550         iter = rcu_dereference_protected(fn->leaf,
4551                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4552         while (iter) {
4553                 if (iter->fib6_metric == rt->fib6_metric &&
4554                     rt6_qualify_for_ecmp(iter))
4555                         return iter;
4556                 iter = rcu_dereference_protected(iter->fib6_next,
4557                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4558         }
4559
4560         return NULL;
4561 }
4562
4563 /* only called for fib entries with builtin fib6_nh */
4564 static bool rt6_is_dead(const struct fib6_info *rt)
4565 {
4566         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4567             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4568              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4569                 return true;
4570
4571         return false;
4572 }
4573
4574 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4575 {
4576         struct fib6_info *iter;
4577         int total = 0;
4578
4579         if (!rt6_is_dead(rt))
4580                 total += rt->fib6_nh->fib_nh_weight;
4581
4582         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4583                 if (!rt6_is_dead(iter))
4584                         total += iter->fib6_nh->fib_nh_weight;
4585         }
4586
4587         return total;
4588 }
4589
4590 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4591 {
4592         int upper_bound = -1;
4593
4594         if (!rt6_is_dead(rt)) {
4595                 *weight += rt->fib6_nh->fib_nh_weight;
4596                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4597                                                     total) - 1;
4598         }
4599         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4600 }
4601
4602 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4603 {
4604         struct fib6_info *iter;
4605         int weight = 0;
4606
4607         rt6_upper_bound_set(rt, &weight, total);
4608
4609         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4610                 rt6_upper_bound_set(iter, &weight, total);
4611 }
4612
4613 void rt6_multipath_rebalance(struct fib6_info *rt)
4614 {
4615         struct fib6_info *first;
4616         int total;
4617
4618         /* In case the entire multipath route was marked for flushing,
4619          * then there is no need to rebalance upon the removal of every
4620          * sibling route.
4621          */
4622         if (!rt->fib6_nsiblings || rt->should_flush)
4623                 return;
4624
4625         /* During lookup routes are evaluated in order, so we need to
4626          * make sure upper bounds are assigned from the first sibling
4627          * onwards.
4628          */
4629         first = rt6_multipath_first_sibling(rt);
4630         if (WARN_ON_ONCE(!first))
4631                 return;
4632
4633         total = rt6_multipath_total_weight(first);
4634         rt6_multipath_upper_bound_set(first, total);
4635 }
4636
4637 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4638 {
4639         const struct arg_netdev_event *arg = p_arg;
4640         struct net *net = dev_net(arg->dev);
4641
4642         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4643             rt->fib6_nh->fib_nh_dev == arg->dev) {
4644                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4645                 fib6_update_sernum_upto_root(net, rt);
4646                 rt6_multipath_rebalance(rt);
4647         }
4648
4649         return 0;
4650 }
4651
4652 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4653 {
4654         struct arg_netdev_event arg = {
4655                 .dev = dev,
4656                 {
4657                         .nh_flags = nh_flags,
4658                 },
4659         };
4660
4661         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4662                 arg.nh_flags |= RTNH_F_LINKDOWN;
4663
4664         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4665 }
4666
4667 /* only called for fib entries with inline fib6_nh */
4668 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4669                                    const struct net_device *dev)
4670 {
4671         struct fib6_info *iter;
4672
4673         if (rt->fib6_nh->fib_nh_dev == dev)
4674                 return true;
4675         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4676                 if (iter->fib6_nh->fib_nh_dev == dev)
4677                         return true;
4678
4679         return false;
4680 }
4681
4682 static void rt6_multipath_flush(struct fib6_info *rt)
4683 {
4684         struct fib6_info *iter;
4685
4686         rt->should_flush = 1;
4687         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4688                 iter->should_flush = 1;
4689 }
4690
4691 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4692                                              const struct net_device *down_dev)
4693 {
4694         struct fib6_info *iter;
4695         unsigned int dead = 0;
4696
4697         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4698             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4699                 dead++;
4700         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4701                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4702                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4703                         dead++;
4704
4705         return dead;
4706 }
4707
4708 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4709                                        const struct net_device *dev,
4710                                        unsigned char nh_flags)
4711 {
4712         struct fib6_info *iter;
4713
4714         if (rt->fib6_nh->fib_nh_dev == dev)
4715                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4716         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4717                 if (iter->fib6_nh->fib_nh_dev == dev)
4718                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4719 }
4720
4721 /* called with write lock held for table with rt */
4722 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4723 {
4724         const struct arg_netdev_event *arg = p_arg;
4725         const struct net_device *dev = arg->dev;
4726         struct net *net = dev_net(dev);
4727
4728         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4729                 return 0;
4730
4731         switch (arg->event) {
4732         case NETDEV_UNREGISTER:
4733                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4734         case NETDEV_DOWN:
4735                 if (rt->should_flush)
4736                         return -1;
4737                 if (!rt->fib6_nsiblings)
4738                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4739                 if (rt6_multipath_uses_dev(rt, dev)) {
4740                         unsigned int count;
4741
4742                         count = rt6_multipath_dead_count(rt, dev);
4743                         if (rt->fib6_nsiblings + 1 == count) {
4744                                 rt6_multipath_flush(rt);
4745                                 return -1;
4746                         }
4747                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4748                                                    RTNH_F_LINKDOWN);
4749                         fib6_update_sernum(net, rt);
4750                         rt6_multipath_rebalance(rt);
4751                 }
4752                 return -2;
4753         case NETDEV_CHANGE:
4754                 if (rt->fib6_nh->fib_nh_dev != dev ||
4755                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4756                         break;
4757                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4758                 rt6_multipath_rebalance(rt);
4759                 break;
4760         }
4761
4762         return 0;
4763 }
4764
4765 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4766 {
4767         struct arg_netdev_event arg = {
4768                 .dev = dev,
4769                 {
4770                         .event = event,
4771                 },
4772         };
4773         struct net *net = dev_net(dev);
4774
4775         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4776                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4777         else
4778                 fib6_clean_all(net, fib6_ifdown, &arg);
4779 }
4780
4781 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4782 {
4783         rt6_sync_down_dev(dev, event);
4784         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4785         neigh_ifdown(&nd_tbl, dev);
4786 }
4787
4788 struct rt6_mtu_change_arg {
4789         struct net_device *dev;
4790         unsigned int mtu;
4791         struct fib6_info *f6i;
4792 };
4793
4794 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4795 {
4796         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4797         struct fib6_info *f6i = arg->f6i;
4798
4799         /* For administrative MTU increase, there is no way to discover
4800          * IPv6 PMTU increase, so PMTU increase should be updated here.
4801          * Since RFC 1981 doesn't include administrative MTU increase
4802          * update PMTU increase is a MUST. (i.e. jumbo frame)
4803          */
4804         if (nh->fib_nh_dev == arg->dev) {
4805                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4806                 u32 mtu = f6i->fib6_pmtu;
4807
4808                 if (mtu >= arg->mtu ||
4809                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4810                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4811
4812                 spin_lock_bh(&rt6_exception_lock);
4813                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4814                 spin_unlock_bh(&rt6_exception_lock);
4815         }
4816
4817         return 0;
4818 }
4819
4820 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4821 {
4822         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4823         struct inet6_dev *idev;
4824
4825         /* In IPv6 pmtu discovery is not optional,
4826            so that RTAX_MTU lock cannot disable it.
4827            We still use this lock to block changes
4828            caused by addrconf/ndisc.
4829         */
4830
4831         idev = __in6_dev_get(arg->dev);
4832         if (!idev)
4833                 return 0;
4834
4835         if (fib6_metric_locked(f6i, RTAX_MTU))
4836                 return 0;
4837
4838         arg->f6i = f6i;
4839         if (f6i->nh) {
4840                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4841                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4842                                                 arg);
4843         }
4844
4845         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4846 }
4847
4848 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4849 {
4850         struct rt6_mtu_change_arg arg = {
4851                 .dev = dev,
4852                 .mtu = mtu,
4853         };
4854
4855         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4856 }
4857
4858 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4859         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4860         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4861         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4862         [RTA_OIF]               = { .type = NLA_U32 },
4863         [RTA_IIF]               = { .type = NLA_U32 },
4864         [RTA_PRIORITY]          = { .type = NLA_U32 },
4865         [RTA_METRICS]           = { .type = NLA_NESTED },
4866         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4867         [RTA_PREF]              = { .type = NLA_U8 },
4868         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4869         [RTA_ENCAP]             = { .type = NLA_NESTED },
4870         [RTA_EXPIRES]           = { .type = NLA_U32 },
4871         [RTA_UID]               = { .type = NLA_U32 },
4872         [RTA_MARK]              = { .type = NLA_U32 },
4873         [RTA_TABLE]             = { .type = NLA_U32 },
4874         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4875         [RTA_SPORT]             = { .type = NLA_U16 },
4876         [RTA_DPORT]             = { .type = NLA_U16 },
4877         [RTA_NH_ID]             = { .type = NLA_U32 },
4878 };
4879
4880 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4881                               struct fib6_config *cfg,
4882                               struct netlink_ext_ack *extack)
4883 {
4884         struct rtmsg *rtm;
4885         struct nlattr *tb[RTA_MAX+1];
4886         unsigned int pref;
4887         int err;
4888
4889         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4890                                      rtm_ipv6_policy, extack);
4891         if (err < 0)
4892                 goto errout;
4893
4894         err = -EINVAL;
4895         rtm = nlmsg_data(nlh);
4896
4897         *cfg = (struct fib6_config){
4898                 .fc_table = rtm->rtm_table,
4899                 .fc_dst_len = rtm->rtm_dst_len,
4900                 .fc_src_len = rtm->rtm_src_len,
4901                 .fc_flags = RTF_UP,
4902                 .fc_protocol = rtm->rtm_protocol,
4903                 .fc_type = rtm->rtm_type,
4904
4905                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4906                 .fc_nlinfo.nlh = nlh,
4907                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4908         };
4909
4910         if (rtm->rtm_type == RTN_UNREACHABLE ||
4911             rtm->rtm_type == RTN_BLACKHOLE ||
4912             rtm->rtm_type == RTN_PROHIBIT ||
4913             rtm->rtm_type == RTN_THROW)
4914                 cfg->fc_flags |= RTF_REJECT;
4915
4916         if (rtm->rtm_type == RTN_LOCAL)
4917                 cfg->fc_flags |= RTF_LOCAL;
4918
4919         if (rtm->rtm_flags & RTM_F_CLONED)
4920                 cfg->fc_flags |= RTF_CACHE;
4921
4922         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4923
4924         if (tb[RTA_NH_ID]) {
4925                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4926                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4927                         NL_SET_ERR_MSG(extack,
4928                                        "Nexthop specification and nexthop id are mutually exclusive");
4929                         goto errout;
4930                 }
4931                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4932         }
4933
4934         if (tb[RTA_GATEWAY]) {
4935                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4936                 cfg->fc_flags |= RTF_GATEWAY;
4937         }
4938         if (tb[RTA_VIA]) {
4939                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4940                 goto errout;
4941         }
4942
4943         if (tb[RTA_DST]) {
4944                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4945
4946                 if (nla_len(tb[RTA_DST]) < plen)
4947                         goto errout;
4948
4949                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4950         }
4951
4952         if (tb[RTA_SRC]) {
4953                 int plen = (rtm->rtm_src_len + 7) >> 3;
4954
4955                 if (nla_len(tb[RTA_SRC]) < plen)
4956                         goto errout;
4957
4958                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4959         }
4960
4961         if (tb[RTA_PREFSRC])
4962                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4963
4964         if (tb[RTA_OIF])
4965                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4966
4967         if (tb[RTA_PRIORITY])
4968                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4969
4970         if (tb[RTA_METRICS]) {
4971                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4972                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4973         }
4974
4975         if (tb[RTA_TABLE])
4976                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
4977
4978         if (tb[RTA_MULTIPATH]) {
4979                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
4980                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
4981
4982                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
4983                                                      cfg->fc_mp_len, extack);
4984                 if (err < 0)
4985                         goto errout;
4986         }
4987
4988         if (tb[RTA_PREF]) {
4989                 pref = nla_get_u8(tb[RTA_PREF]);
4990                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
4991                     pref != ICMPV6_ROUTER_PREF_HIGH)
4992                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
4993                 cfg->fc_flags |= RTF_PREF(pref);
4994         }
4995
4996         if (tb[RTA_ENCAP])
4997                 cfg->fc_encap = tb[RTA_ENCAP];
4998
4999         if (tb[RTA_ENCAP_TYPE]) {
5000                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5001
5002                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5003                 if (err < 0)
5004                         goto errout;
5005         }
5006
5007         if (tb[RTA_EXPIRES]) {
5008                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5009
5010                 if (addrconf_finite_timeout(timeout)) {
5011                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5012                         cfg->fc_flags |= RTF_EXPIRES;
5013                 }
5014         }
5015
5016         err = 0;
5017 errout:
5018         return err;
5019 }
5020
5021 struct rt6_nh {
5022         struct fib6_info *fib6_info;
5023         struct fib6_config r_cfg;
5024         struct list_head next;
5025 };
5026
5027 static int ip6_route_info_append(struct net *net,
5028                                  struct list_head *rt6_nh_list,
5029                                  struct fib6_info *rt,
5030                                  struct fib6_config *r_cfg)
5031 {
5032         struct rt6_nh *nh;
5033         int err = -EEXIST;
5034
5035         list_for_each_entry(nh, rt6_nh_list, next) {
5036                 /* check if fib6_info already exists */
5037                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5038                         return err;
5039         }
5040
5041         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5042         if (!nh)
5043                 return -ENOMEM;
5044         nh->fib6_info = rt;
5045         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5046         list_add_tail(&nh->next, rt6_nh_list);
5047
5048         return 0;
5049 }
5050
5051 static void ip6_route_mpath_notify(struct fib6_info *rt,
5052                                    struct fib6_info *rt_last,
5053                                    struct nl_info *info,
5054                                    __u16 nlflags)
5055 {
5056         /* if this is an APPEND route, then rt points to the first route
5057          * inserted and rt_last points to last route inserted. Userspace
5058          * wants a consistent dump of the route which starts at the first
5059          * nexthop. Since sibling routes are always added at the end of
5060          * the list, find the first sibling of the last route appended
5061          */
5062         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5063                 rt = list_first_entry(&rt_last->fib6_siblings,
5064                                       struct fib6_info,
5065                                       fib6_siblings);
5066         }
5067
5068         if (rt)
5069                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5070 }
5071
5072 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5073 {
5074         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5075         bool should_notify = false;
5076         struct fib6_info *leaf;
5077         struct fib6_node *fn;
5078
5079         rcu_read_lock();
5080         fn = rcu_dereference(rt->fib6_node);
5081         if (!fn)
5082                 goto out;
5083
5084         leaf = rcu_dereference(fn->leaf);
5085         if (!leaf)
5086                 goto out;
5087
5088         if (rt == leaf ||
5089             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5090              rt6_qualify_for_ecmp(leaf)))
5091                 should_notify = true;
5092 out:
5093         rcu_read_unlock();
5094
5095         return should_notify;
5096 }
5097
5098 static int ip6_route_multipath_add(struct fib6_config *cfg,
5099                                    struct netlink_ext_ack *extack)
5100 {
5101         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5102         struct nl_info *info = &cfg->fc_nlinfo;
5103         struct fib6_config r_cfg;
5104         struct rtnexthop *rtnh;
5105         struct fib6_info *rt;
5106         struct rt6_nh *err_nh;
5107         struct rt6_nh *nh, *nh_safe;
5108         __u16 nlflags;
5109         int remaining;
5110         int attrlen;
5111         int err = 1;
5112         int nhn = 0;
5113         int replace = (cfg->fc_nlinfo.nlh &&
5114                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5115         LIST_HEAD(rt6_nh_list);
5116
5117         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5118         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5119                 nlflags |= NLM_F_APPEND;
5120
5121         remaining = cfg->fc_mp_len;
5122         rtnh = (struct rtnexthop *)cfg->fc_mp;
5123
5124         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5125          * fib6_info structs per nexthop
5126          */
5127         while (rtnh_ok(rtnh, remaining)) {
5128                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5129                 if (rtnh->rtnh_ifindex)
5130                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5131
5132                 attrlen = rtnh_attrlen(rtnh);
5133                 if (attrlen > 0) {
5134                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5135
5136                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5137                         if (nla) {
5138                                 r_cfg.fc_gateway = nla_get_in6_addr(nla);
5139                                 r_cfg.fc_flags |= RTF_GATEWAY;
5140                         }
5141                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5142                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5143                         if (nla)
5144                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5145                 }
5146
5147                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5148                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5149                 if (IS_ERR(rt)) {
5150                         err = PTR_ERR(rt);
5151                         rt = NULL;
5152                         goto cleanup;
5153                 }
5154                 if (!rt6_qualify_for_ecmp(rt)) {
5155                         err = -EINVAL;
5156                         NL_SET_ERR_MSG(extack,
5157                                        "Device only routes can not be added for IPv6 using the multipath API.");
5158                         fib6_info_release(rt);
5159                         goto cleanup;
5160                 }
5161
5162                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5163
5164                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5165                                             rt, &r_cfg);
5166                 if (err) {
5167                         fib6_info_release(rt);
5168                         goto cleanup;
5169                 }
5170
5171                 rtnh = rtnh_next(rtnh, &remaining);
5172         }
5173
5174         if (list_empty(&rt6_nh_list)) {
5175                 NL_SET_ERR_MSG(extack,
5176                                "Invalid nexthop configuration - no valid nexthops");
5177                 return -EINVAL;
5178         }
5179
5180         /* for add and replace send one notification with all nexthops.
5181          * Skip the notification in fib6_add_rt2node and send one with
5182          * the full route when done
5183          */
5184         info->skip_notify = 1;
5185
5186         /* For add and replace, send one notification with all nexthops. For
5187          * append, send one notification with all appended nexthops.
5188          */
5189         info->skip_notify_kernel = 1;
5190
5191         err_nh = NULL;
5192         list_for_each_entry(nh, &rt6_nh_list, next) {
5193                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5194                 fib6_info_release(nh->fib6_info);
5195
5196                 if (!err) {
5197                         /* save reference to last route successfully inserted */
5198                         rt_last = nh->fib6_info;
5199
5200                         /* save reference to first route for notification */
5201                         if (!rt_notif)
5202                                 rt_notif = nh->fib6_info;
5203                 }
5204
5205                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5206                 nh->fib6_info = NULL;
5207                 if (err) {
5208                         if (replace && nhn)
5209                                 NL_SET_ERR_MSG_MOD(extack,
5210                                                    "multipath route replace failed (check consistency of installed routes)");
5211                         err_nh = nh;
5212                         goto add_errout;
5213                 }
5214
5215                 /* Because each route is added like a single route we remove
5216                  * these flags after the first nexthop: if there is a collision,
5217                  * we have already failed to add the first nexthop:
5218                  * fib6_add_rt2node() has rejected it; when replacing, old
5219                  * nexthops have been replaced by first new, the rest should
5220                  * be added to it.
5221                  */
5222                 cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5223                                                      NLM_F_REPLACE);
5224                 cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5225                 nhn++;
5226         }
5227
5228         /* An in-kernel notification should only be sent in case the new
5229          * multipath route is added as the first route in the node, or if
5230          * it was appended to it. We pass 'rt_notif' since it is the first
5231          * sibling and might allow us to skip some checks in the replace case.
5232          */
5233         if (ip6_route_mpath_should_notify(rt_notif)) {
5234                 enum fib_event_type fib_event;
5235
5236                 if (rt_notif->fib6_nsiblings != nhn - 1)
5237                         fib_event = FIB_EVENT_ENTRY_APPEND;
5238                 else
5239                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5240
5241                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5242                                                           fib_event, rt_notif,
5243                                                           nhn - 1, extack);
5244                 if (err) {
5245                         /* Delete all the siblings that were just added */
5246                         err_nh = NULL;
5247                         goto add_errout;
5248                 }
5249         }
5250
5251         /* success ... tell user about new route */
5252         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5253         goto cleanup;
5254
5255 add_errout:
5256         /* send notification for routes that were added so that
5257          * the delete notifications sent by ip6_route_del are
5258          * coherent
5259          */
5260         if (rt_notif)
5261                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5262
5263         /* Delete routes that were already added */
5264         list_for_each_entry(nh, &rt6_nh_list, next) {
5265                 if (err_nh == nh)
5266                         break;
5267                 ip6_route_del(&nh->r_cfg, extack);
5268         }
5269
5270 cleanup:
5271         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5272                 if (nh->fib6_info)
5273                         fib6_info_release(nh->fib6_info);
5274                 list_del(&nh->next);
5275                 kfree(nh);
5276         }
5277
5278         return err;
5279 }
5280
5281 static int ip6_route_multipath_del(struct fib6_config *cfg,
5282                                    struct netlink_ext_ack *extack)
5283 {
5284         struct fib6_config r_cfg;
5285         struct rtnexthop *rtnh;
5286         int remaining;
5287         int attrlen;
5288         int err = 1, last_err = 0;
5289
5290         remaining = cfg->fc_mp_len;
5291         rtnh = (struct rtnexthop *)cfg->fc_mp;
5292
5293         /* Parse a Multipath Entry */
5294         while (rtnh_ok(rtnh, remaining)) {
5295                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5296                 if (rtnh->rtnh_ifindex)
5297                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5298
5299                 attrlen = rtnh_attrlen(rtnh);
5300                 if (attrlen > 0) {
5301                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5302
5303                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5304                         if (nla) {
5305                                 nla_memcpy(&r_cfg.fc_gateway, nla, 16);
5306                                 r_cfg.fc_flags |= RTF_GATEWAY;
5307                         }
5308                 }
5309                 err = ip6_route_del(&r_cfg, extack);
5310                 if (err)
5311                         last_err = err;
5312
5313                 rtnh = rtnh_next(rtnh, &remaining);
5314         }
5315
5316         return last_err;
5317 }
5318
5319 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5320                               struct netlink_ext_ack *extack)
5321 {
5322         struct fib6_config cfg;
5323         int err;
5324
5325         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5326         if (err < 0)
5327                 return err;
5328
5329         if (cfg.fc_nh_id &&
5330             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5331                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5332                 return -EINVAL;
5333         }
5334
5335         if (cfg.fc_mp)
5336                 return ip6_route_multipath_del(&cfg, extack);
5337         else {
5338                 cfg.fc_delete_all_nh = 1;
5339                 return ip6_route_del(&cfg, extack);
5340         }
5341 }
5342
5343 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5344                               struct netlink_ext_ack *extack)
5345 {
5346         struct fib6_config cfg;
5347         int err;
5348
5349         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5350         if (err < 0)
5351                 return err;
5352
5353         if (cfg.fc_metric == 0)
5354                 cfg.fc_metric = IP6_RT_PRIO_USER;
5355
5356         if (cfg.fc_mp)
5357                 return ip6_route_multipath_add(&cfg, extack);
5358         else
5359                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5360 }
5361
5362 /* add the overhead of this fib6_nh to nexthop_len */
5363 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5364 {
5365         int *nexthop_len = arg;
5366
5367         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5368                      + NLA_ALIGN(sizeof(struct rtnexthop))
5369                      + nla_total_size(16); /* RTA_GATEWAY */
5370
5371         if (nh->fib_nh_lws) {
5372                 /* RTA_ENCAP_TYPE */
5373                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5374                 /* RTA_ENCAP */
5375                 *nexthop_len += nla_total_size(2);
5376         }
5377
5378         return 0;
5379 }
5380
5381 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5382 {
5383         int nexthop_len;
5384
5385         if (f6i->nh) {
5386                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5387                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5388                                          &nexthop_len);
5389         } else {
5390                 struct fib6_nh *nh = f6i->fib6_nh;
5391
5392                 nexthop_len = 0;
5393                 if (f6i->fib6_nsiblings) {
5394                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5395                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5396                                     + nla_total_size(16) /* RTA_GATEWAY */
5397                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5398
5399                         nexthop_len *= f6i->fib6_nsiblings;
5400                 }
5401                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5402         }
5403
5404         return NLMSG_ALIGN(sizeof(struct rtmsg))
5405                + nla_total_size(16) /* RTA_SRC */
5406                + nla_total_size(16) /* RTA_DST */
5407                + nla_total_size(16) /* RTA_GATEWAY */
5408                + nla_total_size(16) /* RTA_PREFSRC */
5409                + nla_total_size(4) /* RTA_TABLE */
5410                + nla_total_size(4) /* RTA_IIF */
5411                + nla_total_size(4) /* RTA_OIF */
5412                + nla_total_size(4) /* RTA_PRIORITY */
5413                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5414                + nla_total_size(sizeof(struct rta_cacheinfo))
5415                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5416                + nla_total_size(1) /* RTA_PREF */
5417                + nexthop_len;
5418 }
5419
5420 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5421                                  unsigned char *flags)
5422 {
5423         if (nexthop_is_multipath(nh)) {
5424                 struct nlattr *mp;
5425
5426                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5427                 if (!mp)
5428                         goto nla_put_failure;
5429
5430                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5431                         goto nla_put_failure;
5432
5433                 nla_nest_end(skb, mp);
5434         } else {
5435                 struct fib6_nh *fib6_nh;
5436
5437                 fib6_nh = nexthop_fib6_nh(nh);
5438                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5439                                      flags, false) < 0)
5440                         goto nla_put_failure;
5441         }
5442
5443         return 0;
5444
5445 nla_put_failure:
5446         return -EMSGSIZE;
5447 }
5448
5449 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5450                          struct fib6_info *rt, struct dst_entry *dst,
5451                          struct in6_addr *dest, struct in6_addr *src,
5452                          int iif, int type, u32 portid, u32 seq,
5453                          unsigned int flags)
5454 {
5455         struct rt6_info *rt6 = (struct rt6_info *)dst;
5456         struct rt6key *rt6_dst, *rt6_src;
5457         u32 *pmetrics, table, rt6_flags;
5458         unsigned char nh_flags = 0;
5459         struct nlmsghdr *nlh;
5460         struct rtmsg *rtm;
5461         long expires = 0;
5462
5463         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5464         if (!nlh)
5465                 return -EMSGSIZE;
5466
5467         if (rt6) {
5468                 rt6_dst = &rt6->rt6i_dst;
5469                 rt6_src = &rt6->rt6i_src;
5470                 rt6_flags = rt6->rt6i_flags;
5471         } else {
5472                 rt6_dst = &rt->fib6_dst;
5473                 rt6_src = &rt->fib6_src;
5474                 rt6_flags = rt->fib6_flags;
5475         }
5476
5477         rtm = nlmsg_data(nlh);
5478         rtm->rtm_family = AF_INET6;
5479         rtm->rtm_dst_len = rt6_dst->plen;
5480         rtm->rtm_src_len = rt6_src->plen;
5481         rtm->rtm_tos = 0;
5482         if (rt->fib6_table)
5483                 table = rt->fib6_table->tb6_id;
5484         else
5485                 table = RT6_TABLE_UNSPEC;
5486         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5487         if (nla_put_u32(skb, RTA_TABLE, table))
5488                 goto nla_put_failure;
5489
5490         rtm->rtm_type = rt->fib6_type;
5491         rtm->rtm_flags = 0;
5492         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5493         rtm->rtm_protocol = rt->fib6_protocol;
5494
5495         if (rt6_flags & RTF_CACHE)
5496                 rtm->rtm_flags |= RTM_F_CLONED;
5497
5498         if (dest) {
5499                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5500                         goto nla_put_failure;
5501                 rtm->rtm_dst_len = 128;
5502         } else if (rtm->rtm_dst_len)
5503                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5504                         goto nla_put_failure;
5505 #ifdef CONFIG_IPV6_SUBTREES
5506         if (src) {
5507                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5508                         goto nla_put_failure;
5509                 rtm->rtm_src_len = 128;
5510         } else if (rtm->rtm_src_len &&
5511                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5512                 goto nla_put_failure;
5513 #endif
5514         if (iif) {
5515 #ifdef CONFIG_IPV6_MROUTE
5516                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5517                         int err = ip6mr_get_route(net, skb, rtm, portid);
5518
5519                         if (err == 0)
5520                                 return 0;
5521                         if (err < 0)
5522                                 goto nla_put_failure;
5523                 } else
5524 #endif
5525                         if (nla_put_u32(skb, RTA_IIF, iif))
5526                                 goto nla_put_failure;
5527         } else if (dest) {
5528                 struct in6_addr saddr_buf;
5529                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5530                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5531                         goto nla_put_failure;
5532         }
5533
5534         if (rt->fib6_prefsrc.plen) {
5535                 struct in6_addr saddr_buf;
5536                 saddr_buf = rt->fib6_prefsrc.addr;
5537                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5538                         goto nla_put_failure;
5539         }
5540
5541         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5542         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5543                 goto nla_put_failure;
5544
5545         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5546                 goto nla_put_failure;
5547
5548         /* For multipath routes, walk the siblings list and add
5549          * each as a nexthop within RTA_MULTIPATH.
5550          */
5551         if (rt6) {
5552                 if (rt6_flags & RTF_GATEWAY &&
5553                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5554                         goto nla_put_failure;
5555
5556                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5557                         goto nla_put_failure;
5558         } else if (rt->fib6_nsiblings) {
5559                 struct fib6_info *sibling, *next_sibling;
5560                 struct nlattr *mp;
5561
5562                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5563                 if (!mp)
5564                         goto nla_put_failure;
5565
5566                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5567                                     rt->fib6_nh->fib_nh_weight, AF_INET6) < 0)
5568                         goto nla_put_failure;
5569
5570                 list_for_each_entry_safe(sibling, next_sibling,
5571                                          &rt->fib6_siblings, fib6_siblings) {
5572                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5573                                             sibling->fib6_nh->fib_nh_weight,
5574                                             AF_INET6) < 0)
5575                                 goto nla_put_failure;
5576                 }
5577
5578                 nla_nest_end(skb, mp);
5579         } else if (rt->nh) {
5580                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5581                         goto nla_put_failure;
5582
5583                 if (nexthop_is_blackhole(rt->nh))
5584                         rtm->rtm_type = RTN_BLACKHOLE;
5585
5586                 if (net->ipv4.sysctl_nexthop_compat_mode &&
5587                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5588                         goto nla_put_failure;
5589
5590                 rtm->rtm_flags |= nh_flags;
5591         } else {
5592                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5593                                      &nh_flags, false) < 0)
5594                         goto nla_put_failure;
5595
5596                 rtm->rtm_flags |= nh_flags;
5597         }
5598
5599         if (rt6_flags & RTF_EXPIRES) {
5600                 expires = dst ? dst->expires : rt->expires;
5601                 expires -= jiffies;
5602         }
5603
5604         if (!dst) {
5605                 if (rt->offload)
5606                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5607                 if (rt->trap)
5608                         rtm->rtm_flags |= RTM_F_TRAP;
5609         }
5610
5611         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5612                 goto nla_put_failure;
5613
5614         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5615                 goto nla_put_failure;
5616
5617
5618         nlmsg_end(skb, nlh);
5619         return 0;
5620
5621 nla_put_failure:
5622         nlmsg_cancel(skb, nlh);
5623         return -EMSGSIZE;
5624 }
5625
5626 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5627 {
5628         const struct net_device *dev = arg;
5629
5630         if (nh->fib_nh_dev == dev)
5631                 return 1;
5632
5633         return 0;
5634 }
5635
5636 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5637                                const struct net_device *dev)
5638 {
5639         if (f6i->nh) {
5640                 struct net_device *_dev = (struct net_device *)dev;
5641
5642                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5643                                                   fib6_info_nh_uses_dev,
5644                                                   _dev);
5645         }
5646
5647         if (f6i->fib6_nh->fib_nh_dev == dev)
5648                 return true;
5649
5650         if (f6i->fib6_nsiblings) {
5651                 struct fib6_info *sibling, *next_sibling;
5652
5653                 list_for_each_entry_safe(sibling, next_sibling,
5654                                          &f6i->fib6_siblings, fib6_siblings) {
5655                         if (sibling->fib6_nh->fib_nh_dev == dev)
5656                                 return true;
5657                 }
5658         }
5659
5660         return false;
5661 }
5662
5663 struct fib6_nh_exception_dump_walker {
5664         struct rt6_rtnl_dump_arg *dump;
5665         struct fib6_info *rt;
5666         unsigned int flags;
5667         unsigned int skip;
5668         unsigned int count;
5669 };
5670
5671 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5672 {
5673         struct fib6_nh_exception_dump_walker *w = arg;
5674         struct rt6_rtnl_dump_arg *dump = w->dump;
5675         struct rt6_exception_bucket *bucket;
5676         struct rt6_exception *rt6_ex;
5677         int i, err;
5678
5679         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5680         if (!bucket)
5681                 return 0;
5682
5683         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5684                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5685                         if (w->skip) {
5686                                 w->skip--;
5687                                 continue;
5688                         }
5689
5690                         /* Expiration of entries doesn't bump sernum, insertion
5691                          * does. Removal is triggered by insertion, so we can
5692                          * rely on the fact that if entries change between two
5693                          * partial dumps, this node is scanned again completely,
5694                          * see rt6_insert_exception() and fib6_dump_table().
5695                          *
5696                          * Count expired entries we go through as handled
5697                          * entries that we'll skip next time, in case of partial
5698                          * node dump. Otherwise, if entries expire meanwhile,
5699                          * we'll skip the wrong amount.
5700                          */
5701                         if (rt6_check_expired(rt6_ex->rt6i)) {
5702                                 w->count++;
5703                                 continue;
5704                         }
5705
5706                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5707                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5708                                             RTM_NEWROUTE,
5709                                             NETLINK_CB(dump->cb->skb).portid,
5710                                             dump->cb->nlh->nlmsg_seq, w->flags);
5711                         if (err)
5712                                 return err;
5713
5714                         w->count++;
5715                 }
5716                 bucket++;
5717         }
5718
5719         return 0;
5720 }
5721
5722 /* Return -1 if done with node, number of handled routes on partial dump */
5723 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5724 {
5725         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5726         struct fib_dump_filter *filter = &arg->filter;
5727         unsigned int flags = NLM_F_MULTI;
5728         struct net *net = arg->net;
5729         int count = 0;
5730
5731         if (rt == net->ipv6.fib6_null_entry)
5732                 return -1;
5733
5734         if ((filter->flags & RTM_F_PREFIX) &&
5735             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5736                 /* success since this is not a prefix route */
5737                 return -1;
5738         }
5739         if (filter->filter_set &&
5740             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5741              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5742              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5743                 return -1;
5744         }
5745
5746         if (filter->filter_set ||
5747             !filter->dump_routes || !filter->dump_exceptions) {
5748                 flags |= NLM_F_DUMP_FILTERED;
5749         }
5750
5751         if (filter->dump_routes) {
5752                 if (skip) {
5753                         skip--;
5754                 } else {
5755                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5756                                           0, RTM_NEWROUTE,
5757                                           NETLINK_CB(arg->cb->skb).portid,
5758                                           arg->cb->nlh->nlmsg_seq, flags)) {
5759                                 return 0;
5760                         }
5761                         count++;
5762                 }
5763         }
5764
5765         if (filter->dump_exceptions) {
5766                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5767                                                            .rt = rt,
5768                                                            .flags = flags,
5769                                                            .skip = skip,
5770                                                            .count = 0 };
5771                 int err;
5772
5773                 rcu_read_lock();
5774                 if (rt->nh) {
5775                         err = nexthop_for_each_fib6_nh(rt->nh,
5776                                                        rt6_nh_dump_exceptions,
5777                                                        &w);
5778                 } else {
5779                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5780                 }
5781                 rcu_read_unlock();
5782
5783                 if (err)
5784                         return count += w.count;
5785         }
5786
5787         return -1;
5788 }
5789
5790 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5791                                         const struct nlmsghdr *nlh,
5792                                         struct nlattr **tb,
5793                                         struct netlink_ext_ack *extack)
5794 {
5795         struct rtmsg *rtm;
5796         int i, err;
5797
5798         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5799                 NL_SET_ERR_MSG_MOD(extack,
5800                                    "Invalid header for get route request");
5801                 return -EINVAL;
5802         }
5803
5804         if (!netlink_strict_get_check(skb))
5805                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5806                                               rtm_ipv6_policy, extack);
5807
5808         rtm = nlmsg_data(nlh);
5809         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5810             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5811             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5812             rtm->rtm_type) {
5813                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5814                 return -EINVAL;
5815         }
5816         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5817                 NL_SET_ERR_MSG_MOD(extack,
5818                                    "Invalid flags for get route request");
5819                 return -EINVAL;
5820         }
5821
5822         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5823                                             rtm_ipv6_policy, extack);
5824         if (err)
5825                 return err;
5826
5827         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5828             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5829                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5830                 return -EINVAL;
5831         }
5832
5833         for (i = 0; i <= RTA_MAX; i++) {
5834                 if (!tb[i])
5835                         continue;
5836
5837                 switch (i) {
5838                 case RTA_SRC:
5839                 case RTA_DST:
5840                 case RTA_IIF:
5841                 case RTA_OIF:
5842                 case RTA_MARK:
5843                 case RTA_UID:
5844                 case RTA_SPORT:
5845                 case RTA_DPORT:
5846                 case RTA_IP_PROTO:
5847                         break;
5848                 default:
5849                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5850                         return -EINVAL;
5851                 }
5852         }
5853
5854         return 0;
5855 }
5856
5857 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5858                               struct netlink_ext_ack *extack)
5859 {
5860         struct net *net = sock_net(in_skb->sk);
5861         struct nlattr *tb[RTA_MAX+1];
5862         int err, iif = 0, oif = 0;
5863         struct fib6_info *from;
5864         struct dst_entry *dst;
5865         struct rt6_info *rt;
5866         struct sk_buff *skb;
5867         struct rtmsg *rtm;
5868         struct flowi6 fl6 = {};
5869         bool fibmatch;
5870
5871         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5872         if (err < 0)
5873                 goto errout;
5874
5875         err = -EINVAL;
5876         rtm = nlmsg_data(nlh);
5877         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5878         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5879
5880         if (tb[RTA_SRC]) {
5881                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5882                         goto errout;
5883
5884                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5885         }
5886
5887         if (tb[RTA_DST]) {
5888                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5889                         goto errout;
5890
5891                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5892         }
5893
5894         if (tb[RTA_IIF])
5895                 iif = nla_get_u32(tb[RTA_IIF]);
5896
5897         if (tb[RTA_OIF])
5898                 oif = nla_get_u32(tb[RTA_OIF]);
5899
5900         if (tb[RTA_MARK])
5901                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5902
5903         if (tb[RTA_UID])
5904                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5905                                            nla_get_u32(tb[RTA_UID]));
5906         else
5907                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5908
5909         if (tb[RTA_SPORT])
5910                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5911
5912         if (tb[RTA_DPORT])
5913                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5914
5915         if (tb[RTA_IP_PROTO]) {
5916                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5917                                                   &fl6.flowi6_proto, AF_INET6,
5918                                                   extack);
5919                 if (err)
5920                         goto errout;
5921         }
5922
5923         if (iif) {
5924                 struct net_device *dev;
5925                 int flags = 0;
5926
5927                 rcu_read_lock();
5928
5929                 dev = dev_get_by_index_rcu(net, iif);
5930                 if (!dev) {
5931                         rcu_read_unlock();
5932                         err = -ENODEV;
5933                         goto errout;
5934                 }
5935
5936                 fl6.flowi6_iif = iif;
5937
5938                 if (!ipv6_addr_any(&fl6.saddr))
5939                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5940
5941                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
5942
5943                 rcu_read_unlock();
5944         } else {
5945                 fl6.flowi6_oif = oif;
5946
5947                 dst = ip6_route_output(net, NULL, &fl6);
5948         }
5949
5950
5951         rt = container_of(dst, struct rt6_info, dst);
5952         if (rt->dst.error) {
5953                 err = rt->dst.error;
5954                 ip6_rt_put(rt);
5955                 goto errout;
5956         }
5957
5958         if (rt == net->ipv6.ip6_null_entry) {
5959                 err = rt->dst.error;
5960                 ip6_rt_put(rt);
5961                 goto errout;
5962         }
5963
5964         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
5965         if (!skb) {
5966                 ip6_rt_put(rt);
5967                 err = -ENOBUFS;
5968                 goto errout;
5969         }
5970
5971         skb_dst_set(skb, &rt->dst);
5972
5973         rcu_read_lock();
5974         from = rcu_dereference(rt->from);
5975         if (from) {
5976                 if (fibmatch)
5977                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
5978                                             iif, RTM_NEWROUTE,
5979                                             NETLINK_CB(in_skb).portid,
5980                                             nlh->nlmsg_seq, 0);
5981                 else
5982                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
5983                                             &fl6.saddr, iif, RTM_NEWROUTE,
5984                                             NETLINK_CB(in_skb).portid,
5985                                             nlh->nlmsg_seq, 0);
5986         } else {
5987                 err = -ENETUNREACH;
5988         }
5989         rcu_read_unlock();
5990
5991         if (err < 0) {
5992                 kfree_skb(skb);
5993                 goto errout;
5994         }
5995
5996         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
5997 errout:
5998         return err;
5999 }
6000
6001 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6002                      unsigned int nlm_flags)
6003 {
6004         struct sk_buff *skb;
6005         struct net *net = info->nl_net;
6006         u32 seq;
6007         int err;
6008
6009         err = -ENOBUFS;
6010         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6011
6012         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6013         if (!skb)
6014                 goto errout;
6015
6016         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6017                             event, info->portid, seq, nlm_flags);
6018         if (err < 0) {
6019                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6020                 WARN_ON(err == -EMSGSIZE);
6021                 kfree_skb(skb);
6022                 goto errout;
6023         }
6024         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6025                     info->nlh, gfp_any());
6026         return;
6027 errout:
6028         if (err < 0)
6029                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6030 }
6031
6032 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6033                     struct nl_info *info)
6034 {
6035         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6036         struct sk_buff *skb;
6037         int err = -ENOBUFS;
6038
6039         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
6040          * is implemented and supported for nexthop objects
6041          */
6042         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
6043
6044         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6045         if (!skb)
6046                 goto errout;
6047
6048         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6049                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6050         if (err < 0) {
6051                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6052                 WARN_ON(err == -EMSGSIZE);
6053                 kfree_skb(skb);
6054                 goto errout;
6055         }
6056         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6057                     info->nlh, gfp_any());
6058         return;
6059 errout:
6060         if (err < 0)
6061                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6062 }
6063
6064 static int ip6_route_dev_notify(struct notifier_block *this,
6065                                 unsigned long event, void *ptr)
6066 {
6067         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6068         struct net *net = dev_net(dev);
6069
6070         if (!(dev->flags & IFF_LOOPBACK))
6071                 return NOTIFY_OK;
6072
6073         if (event == NETDEV_REGISTER) {
6074                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6075                 net->ipv6.ip6_null_entry->dst.dev = dev;
6076                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6077 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6078                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6079                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6080                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6081                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6082 #endif
6083          } else if (event == NETDEV_UNREGISTER &&
6084                     dev->reg_state != NETREG_UNREGISTERED) {
6085                 /* NETDEV_UNREGISTER could be fired for multiple times by
6086                  * netdev_wait_allrefs(). Make sure we only call this once.
6087                  */
6088                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6089 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6090                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6091                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6092 #endif
6093         }
6094
6095         return NOTIFY_OK;
6096 }
6097
6098 /*
6099  *      /proc
6100  */
6101
6102 #ifdef CONFIG_PROC_FS
6103 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6104 {
6105         struct net *net = (struct net *)seq->private;
6106         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6107                    net->ipv6.rt6_stats->fib_nodes,
6108                    net->ipv6.rt6_stats->fib_route_nodes,
6109                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6110                    net->ipv6.rt6_stats->fib_rt_entries,
6111                    net->ipv6.rt6_stats->fib_rt_cache,
6112                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6113                    net->ipv6.rt6_stats->fib_discarded_routes);
6114
6115         return 0;
6116 }
6117 #endif  /* CONFIG_PROC_FS */
6118
6119 #ifdef CONFIG_SYSCTL
6120
6121 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6122                               void *buffer, size_t *lenp, loff_t *ppos)
6123 {
6124         struct net *net;
6125         int delay;
6126         int ret;
6127         if (!write)
6128                 return -EINVAL;
6129
6130         net = (struct net *)ctl->extra1;
6131         delay = net->ipv6.sysctl.flush_delay;
6132         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6133         if (ret)
6134                 return ret;
6135
6136         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6137         return 0;
6138 }
6139
6140 static struct ctl_table ipv6_route_table_template[] = {
6141         {
6142                 .procname       =       "flush",
6143                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6144                 .maxlen         =       sizeof(int),
6145                 .mode           =       0200,
6146                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6147         },
6148         {
6149                 .procname       =       "gc_thresh",
6150                 .data           =       &ip6_dst_ops_template.gc_thresh,
6151                 .maxlen         =       sizeof(int),
6152                 .mode           =       0644,
6153                 .proc_handler   =       proc_dointvec,
6154         },
6155         {
6156                 .procname       =       "max_size",
6157                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6158                 .maxlen         =       sizeof(int),
6159                 .mode           =       0644,
6160                 .proc_handler   =       proc_dointvec,
6161         },
6162         {
6163                 .procname       =       "gc_min_interval",
6164                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6165                 .maxlen         =       sizeof(int),
6166                 .mode           =       0644,
6167                 .proc_handler   =       proc_dointvec_jiffies,
6168         },
6169         {
6170                 .procname       =       "gc_timeout",
6171                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6172                 .maxlen         =       sizeof(int),
6173                 .mode           =       0644,
6174                 .proc_handler   =       proc_dointvec_jiffies,
6175         },
6176         {
6177                 .procname       =       "gc_interval",
6178                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6179                 .maxlen         =       sizeof(int),
6180                 .mode           =       0644,
6181                 .proc_handler   =       proc_dointvec_jiffies,
6182         },
6183         {
6184                 .procname       =       "gc_elasticity",
6185                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6186                 .maxlen         =       sizeof(int),
6187                 .mode           =       0644,
6188                 .proc_handler   =       proc_dointvec,
6189         },
6190         {
6191                 .procname       =       "mtu_expires",
6192                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6193                 .maxlen         =       sizeof(int),
6194                 .mode           =       0644,
6195                 .proc_handler   =       proc_dointvec_jiffies,
6196         },
6197         {
6198                 .procname       =       "min_adv_mss",
6199                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6200                 .maxlen         =       sizeof(int),
6201                 .mode           =       0644,
6202                 .proc_handler   =       proc_dointvec,
6203         },
6204         {
6205                 .procname       =       "gc_min_interval_ms",
6206                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6207                 .maxlen         =       sizeof(int),
6208                 .mode           =       0644,
6209                 .proc_handler   =       proc_dointvec_ms_jiffies,
6210         },
6211         {
6212                 .procname       =       "skip_notify_on_dev_down",
6213                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6214                 .maxlen         =       sizeof(int),
6215                 .mode           =       0644,
6216                 .proc_handler   =       proc_dointvec_minmax,
6217                 .extra1         =       SYSCTL_ZERO,
6218                 .extra2         =       SYSCTL_ONE,
6219         },
6220         { }
6221 };
6222
6223 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6224 {
6225         struct ctl_table *table;
6226
6227         table = kmemdup(ipv6_route_table_template,
6228                         sizeof(ipv6_route_table_template),
6229                         GFP_KERNEL);
6230
6231         if (table) {
6232                 table[0].data = &net->ipv6.sysctl.flush_delay;
6233                 table[0].extra1 = net;
6234                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6235                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6236                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6237                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6238                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6239                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6240                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6241                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6242                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6243                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6244
6245                 /* Don't export sysctls to unprivileged users */
6246                 if (net->user_ns != &init_user_ns)
6247                         table[0].procname = NULL;
6248         }
6249
6250         return table;
6251 }
6252 #endif
6253
6254 static int __net_init ip6_route_net_init(struct net *net)
6255 {
6256         int ret = -ENOMEM;
6257
6258         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6259                sizeof(net->ipv6.ip6_dst_ops));
6260
6261         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6262                 goto out_ip6_dst_ops;
6263
6264         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6265         if (!net->ipv6.fib6_null_entry)
6266                 goto out_ip6_dst_entries;
6267         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6268                sizeof(*net->ipv6.fib6_null_entry));
6269
6270         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6271                                            sizeof(*net->ipv6.ip6_null_entry),
6272                                            GFP_KERNEL);
6273         if (!net->ipv6.ip6_null_entry)
6274                 goto out_fib6_null_entry;
6275         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6276         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6277                          ip6_template_metrics, true);
6278         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6279
6280 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6281         net->ipv6.fib6_has_custom_rules = false;
6282         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6283                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6284                                                GFP_KERNEL);
6285         if (!net->ipv6.ip6_prohibit_entry)
6286                 goto out_ip6_null_entry;
6287         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6288         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6289                          ip6_template_metrics, true);
6290         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6291
6292         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6293                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6294                                                GFP_KERNEL);
6295         if (!net->ipv6.ip6_blk_hole_entry)
6296                 goto out_ip6_prohibit_entry;
6297         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6298         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6299                          ip6_template_metrics, true);
6300         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6301 #ifdef CONFIG_IPV6_SUBTREES
6302         net->ipv6.fib6_routes_require_src = 0;
6303 #endif
6304 #endif
6305
6306         net->ipv6.sysctl.flush_delay = 0;
6307         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6308         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6309         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6310         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6311         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6312         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6313         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6314         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6315
6316         net->ipv6.ip6_rt_gc_expire = 30*HZ;
6317
6318         ret = 0;
6319 out:
6320         return ret;
6321
6322 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6323 out_ip6_prohibit_entry:
6324         kfree(net->ipv6.ip6_prohibit_entry);
6325 out_ip6_null_entry:
6326         kfree(net->ipv6.ip6_null_entry);
6327 #endif
6328 out_fib6_null_entry:
6329         kfree(net->ipv6.fib6_null_entry);
6330 out_ip6_dst_entries:
6331         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6332 out_ip6_dst_ops:
6333         goto out;
6334 }
6335
6336 static void __net_exit ip6_route_net_exit(struct net *net)
6337 {
6338         kfree(net->ipv6.fib6_null_entry);
6339         kfree(net->ipv6.ip6_null_entry);
6340 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6341         kfree(net->ipv6.ip6_prohibit_entry);
6342         kfree(net->ipv6.ip6_blk_hole_entry);
6343 #endif
6344         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6345 }
6346
6347 static int __net_init ip6_route_net_init_late(struct net *net)
6348 {
6349 #ifdef CONFIG_PROC_FS
6350         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6351                         sizeof(struct ipv6_route_iter));
6352         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6353                         rt6_stats_seq_show, NULL);
6354 #endif
6355         return 0;
6356 }
6357
6358 static void __net_exit ip6_route_net_exit_late(struct net *net)
6359 {
6360 #ifdef CONFIG_PROC_FS
6361         remove_proc_entry("ipv6_route", net->proc_net);
6362         remove_proc_entry("rt6_stats", net->proc_net);
6363 #endif
6364 }
6365
6366 static struct pernet_operations ip6_route_net_ops = {
6367         .init = ip6_route_net_init,
6368         .exit = ip6_route_net_exit,
6369 };
6370
6371 static int __net_init ipv6_inetpeer_init(struct net *net)
6372 {
6373         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6374
6375         if (!bp)
6376                 return -ENOMEM;
6377         inet_peer_base_init(bp);
6378         net->ipv6.peers = bp;
6379         return 0;
6380 }
6381
6382 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6383 {
6384         struct inet_peer_base *bp = net->ipv6.peers;
6385
6386         net->ipv6.peers = NULL;
6387         inetpeer_invalidate_tree(bp);
6388         kfree(bp);
6389 }
6390
6391 static struct pernet_operations ipv6_inetpeer_ops = {
6392         .init   =       ipv6_inetpeer_init,
6393         .exit   =       ipv6_inetpeer_exit,
6394 };
6395
6396 static struct pernet_operations ip6_route_net_late_ops = {
6397         .init = ip6_route_net_init_late,
6398         .exit = ip6_route_net_exit_late,
6399 };
6400
6401 static struct notifier_block ip6_route_dev_notifier = {
6402         .notifier_call = ip6_route_dev_notify,
6403         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6404 };
6405
6406 void __init ip6_route_init_special_entries(void)
6407 {
6408         /* Registering of the loopback is done before this portion of code,
6409          * the loopback reference in rt6_info will not be taken, do it
6410          * manually for init_net */
6411         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6412         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6413         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6414   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6415         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6416         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6417         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6418         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6419   #endif
6420 }
6421
6422 #if IS_BUILTIN(CONFIG_IPV6)
6423 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6424 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6425
6426 static const struct bpf_iter_reg ipv6_route_reg_info = {
6427         .target                 = "ipv6_route",
6428         .seq_ops                = &ipv6_route_seq_ops,
6429         .init_seq_private       = bpf_iter_init_seq_net,
6430         .fini_seq_private       = bpf_iter_fini_seq_net,
6431         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6432         .ctx_arg_info_size      = 1,
6433         .ctx_arg_info           = {
6434                 { offsetof(struct bpf_iter__ipv6_route, rt),
6435                   PTR_TO_BTF_ID_OR_NULL },
6436         },
6437 };
6438
6439 static int __init bpf_iter_register(void)
6440 {
6441         return bpf_iter_reg_target(&ipv6_route_reg_info);
6442 }
6443
6444 static void bpf_iter_unregister(void)
6445 {
6446         bpf_iter_unreg_target(&ipv6_route_reg_info);
6447 }
6448 #endif
6449 #endif
6450
6451 int __init ip6_route_init(void)
6452 {
6453         int ret;
6454         int cpu;
6455
6456         ret = -ENOMEM;
6457         ip6_dst_ops_template.kmem_cachep =
6458                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6459                                   SLAB_HWCACHE_ALIGN, NULL);
6460         if (!ip6_dst_ops_template.kmem_cachep)
6461                 goto out;
6462
6463         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6464         if (ret)
6465                 goto out_kmem_cache;
6466
6467         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6468         if (ret)
6469                 goto out_dst_entries;
6470
6471         ret = register_pernet_subsys(&ip6_route_net_ops);
6472         if (ret)
6473                 goto out_register_inetpeer;
6474
6475         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6476
6477         ret = fib6_init();
6478         if (ret)
6479                 goto out_register_subsys;
6480
6481         ret = xfrm6_init();
6482         if (ret)
6483                 goto out_fib6_init;
6484
6485         ret = fib6_rules_init();
6486         if (ret)
6487                 goto xfrm6_init;
6488
6489         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6490         if (ret)
6491                 goto fib6_rules_init;
6492
6493         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6494                                    inet6_rtm_newroute, NULL, 0);
6495         if (ret < 0)
6496                 goto out_register_late_subsys;
6497
6498         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6499                                    inet6_rtm_delroute, NULL, 0);
6500         if (ret < 0)
6501                 goto out_register_late_subsys;
6502
6503         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6504                                    inet6_rtm_getroute, NULL,
6505                                    RTNL_FLAG_DOIT_UNLOCKED);
6506         if (ret < 0)
6507                 goto out_register_late_subsys;
6508
6509         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6510         if (ret)
6511                 goto out_register_late_subsys;
6512
6513 #if IS_BUILTIN(CONFIG_IPV6)
6514 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6515         ret = bpf_iter_register();
6516         if (ret)
6517                 goto out_register_late_subsys;
6518 #endif
6519 #endif
6520
6521         for_each_possible_cpu(cpu) {
6522                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6523
6524                 INIT_LIST_HEAD(&ul->head);
6525                 spin_lock_init(&ul->lock);
6526         }
6527
6528 out:
6529         return ret;
6530
6531 out_register_late_subsys:
6532         rtnl_unregister_all(PF_INET6);
6533         unregister_pernet_subsys(&ip6_route_net_late_ops);
6534 fib6_rules_init:
6535         fib6_rules_cleanup();
6536 xfrm6_init:
6537         xfrm6_fini();
6538 out_fib6_init:
6539         fib6_gc_cleanup();
6540 out_register_subsys:
6541         unregister_pernet_subsys(&ip6_route_net_ops);
6542 out_register_inetpeer:
6543         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6544 out_dst_entries:
6545         dst_entries_destroy(&ip6_dst_blackhole_ops);
6546 out_kmem_cache:
6547         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6548         goto out;
6549 }
6550
6551 void ip6_route_cleanup(void)
6552 {
6553 #if IS_BUILTIN(CONFIG_IPV6)
6554 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6555         bpf_iter_unregister();
6556 #endif
6557 #endif
6558         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6559         unregister_pernet_subsys(&ip6_route_net_late_ops);
6560         fib6_rules_cleanup();
6561         xfrm6_fini();
6562         fib6_gc_cleanup();
6563         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6564         unregister_pernet_subsys(&ip6_route_net_ops);
6565         dst_entries_destroy(&ip6_dst_blackhole_ops);
6566         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6567 }