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             <roque@di.fc.ul.pt>
   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 <linux/siphash.h>
  45 #include <net/net_namespace.h>
  46 #include <net/snmp.h>
  47 #include <net/ipv6.h>
  48 #include <net/ip6_fib.h>
  49 #include <net/ip6_route.h>
  50 #include <net/ndisc.h>
  51 #include <net/addrconf.h>
  52 #include <net/tcp.h>
  53 #include <linux/rtnetlink.h>
  54 #include <net/dst.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/xfrm.h>
  57 #include <net/netevent.h>
  58 #include <net/netlink.h>
  59 #include <net/rtnh.h>
  60 #include <net/lwtunnel.h>
  61 #include <net/ip_tunnels.h>
  62 #include <net/l3mdev.h>
  63 #include <net/ip.h>
  64 #include <linux/uaccess.h>
  65 #include <linux/btf_ids.h>
  66
  67 #ifdef CONFIG_SYSCTL
  68 #include <linux/sysctl.h>
  69 #endif
  70
  71 static int ip6_rt_type_to_error(u8 fib6_type);
  72
  73 #define CREATE_TRACE_POINTS
  74 #include <trace/events/fib6.h>
  75 EXPORT_TRACEPOINT_SYMBOL_GPL(fib6_table_lookup);
  76 #undef CREATE_TRACE_POINTS
  77
  78 enum rt6_nud_state {
  79         RT6_NUD_FAIL_HARD = -3,
  80         RT6_NUD_FAIL_PROBE = -2,
  81         RT6_NUD_FAIL_DO_RR = -1,
  82         RT6_NUD_SUCCEED = 1
  83 };
  84
  85 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie);
  86 static unsigned int      ip6_default_advmss(const struct dst_entry *dst);
  87 static unsigned int      ip6_mtu(const struct dst_entry *dst);
  88 static struct dst_entry *ip6_negative_advice(struct dst_entry *);
  89 static void             ip6_dst_destroy(struct dst_entry *);
  90 static void             ip6_dst_ifdown(struct dst_entry *,
  91                                        struct net_device *dev, int how);
  92 static int               ip6_dst_gc(struct dst_ops *ops);
  93
  94 static int              ip6_pkt_discard(struct sk_buff *skb);
  95 static int              ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  96 static int              ip6_pkt_prohibit(struct sk_buff *skb);
  97 static int              ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb);
  98 static void             ip6_link_failure(struct sk_buff *skb);
  99 static void             ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
 100                                            struct sk_buff *skb, u32 mtu,
 101                                            bool confirm_neigh);
 102 static void             rt6_do_redirect(struct dst_entry *dst, struct sock *sk,
 103                                         struct sk_buff *skb);
 104 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 105                            int strict);
 106 static size_t rt6_nlmsg_size(struct fib6_info *f6i);
 107 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
 108                          struct fib6_info *rt, struct dst_entry *dst,
 109                          struct in6_addr *dest, struct in6_addr *src,
 110                          int iif, int type, u32 portid, u32 seq,
 111                          unsigned int flags);
 112 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
 113                                            const struct in6_addr *daddr,
 114                                            const struct in6_addr *saddr);
 115
 116 #ifdef CONFIG_IPV6_ROUTE_INFO
 117 static struct fib6_info *rt6_add_route_info(struct net *net,
 118                                            const struct in6_addr *prefix, int prefixlen,
 119                                            const struct in6_addr *gwaddr,
 120                                            struct net_device *dev,
 121                                            unsigned int pref);
 122 static struct fib6_info *rt6_get_route_info(struct net *net,
 123                                            const struct in6_addr *prefix, int prefixlen,
 124                                            const struct in6_addr *gwaddr,
 125                                            struct net_device *dev);
 126 #endif
 127
 128 struct uncached_list {
 129         spinlock_t              lock;
 130         struct list_head        head;
 131 };
 132
 133 static DEFINE_PER_CPU_ALIGNED(struct uncached_list, rt6_uncached_list);
 134
 135 void rt6_uncached_list_add(struct rt6_info *rt)
 136 {
 137         struct uncached_list *ul = raw_cpu_ptr(&rt6_uncached_list);
 138
 139         rt->rt6i_uncached_list = ul;
 140
 141         spin_lock_bh(&ul->lock);
 142         list_add_tail(&rt->rt6i_uncached, &ul->head);
 143         spin_unlock_bh(&ul->lock);
 144 }
 145
 146 void rt6_uncached_list_del(struct rt6_info *rt)
 147 {
 148         if (!list_empty(&rt->rt6i_uncached)) {
 149                 struct uncached_list *ul = rt->rt6i_uncached_list;
 150                 struct net *net = dev_net(rt->dst.dev);
 151
 152                 spin_lock_bh(&ul->lock);
 153                 list_del(&rt->rt6i_uncached);
 154                 atomic_dec(&net->ipv6.rt6_stats->fib_rt_uncache);
 155                 spin_unlock_bh(&ul->lock);
 156         }
 157 }
 158
 159 static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 160 {
 161         struct net_device *loopback_dev = net->loopback_dev;
 162         int cpu;
 163
 164         if (dev == loopback_dev)
 165                 return;
 166
 167         for_each_possible_cpu(cpu) {
 168                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
 169                 struct rt6_info *rt;
 170
 171                 spin_lock_bh(&ul->lock);
 172                 list_for_each_entry(rt, &ul->head, rt6i_uncached) {
 173                         struct inet6_dev *rt_idev = rt->rt6i_idev;
 174                         struct net_device *rt_dev = rt->dst.dev;
 175
 176                         if (rt_idev->dev == dev) {
 177                                 rt->rt6i_idev = in6_dev_get(loopback_dev);
 178                                 in6_dev_put(rt_idev);
 179                         }
 180
 181                         if (rt_dev == dev) {
 182                                 rt->dst.dev = blackhole_netdev;
 183                                 dev_hold(rt->dst.dev);
 184                                 dev_put(rt_dev);
 185                         }
 186                 }
 187                 spin_unlock_bh(&ul->lock);
 188         }
 189 }
 190
 191 static inline const void *choose_neigh_daddr(const struct in6_addr *p,
 192                                              struct sk_buff *skb,
 193                                              const void *daddr)
 194 {
 195         if (!ipv6_addr_any(p))
 196                 return (const void *) p;
 197         else if (skb)
 198                 return &ipv6_hdr(skb)->daddr;
 199         return daddr;
 200 }
 201
 202 struct neighbour *ip6_neigh_lookup(const struct in6_addr *gw,
 203                                    struct net_device *dev,
 204                                    struct sk_buff *skb,
 205                                    const void *daddr)
 206 {
 207         struct neighbour *n;
 208
 209         daddr = choose_neigh_daddr(gw, skb, daddr);
 210         n = __ipv6_neigh_lookup(dev, daddr);
 211         if (n)
 212                 return n;
 213
 214         n = neigh_create(&nd_tbl, daddr, dev);
 215         return IS_ERR(n) ? NULL : n;
 216 }
 217
 218 static struct neighbour *ip6_dst_neigh_lookup(const struct dst_entry *dst,
 219                                               struct sk_buff *skb,
 220                                               const void *daddr)
 221 {
 222         const struct rt6_info *rt = container_of(dst, struct rt6_info, dst);
 223
 224         return ip6_neigh_lookup(rt6_nexthop(rt, &in6addr_any),
 225                                 dst->dev, skb, daddr);
 226 }
 227
 228 static void ip6_confirm_neigh(const struct dst_entry *dst, const void *daddr)
 229 {
 230         struct net_device *dev = dst->dev;
 231         struct rt6_info *rt = (struct rt6_info *)dst;
 232
 233         daddr = choose_neigh_daddr(rt6_nexthop(rt, &in6addr_any), NULL, daddr);
 234         if (!daddr)
 235                 return;
 236         if (dev->flags & (IFF_NOARP | IFF_LOOPBACK))
 237                 return;
 238         if (ipv6_addr_is_multicast((const struct in6_addr *)daddr))
 239                 return;
 240         __ipv6_confirm_neigh(dev, daddr);
 241 }
 242
 243 static struct dst_ops ip6_dst_ops_template = {
 244         .family                 =       AF_INET6,
 245         .gc                     =       ip6_dst_gc,
 246         .gc_thresh              =       1024,
 247         .check                  =       ip6_dst_check,
 248         .default_advmss         =       ip6_default_advmss,
 249         .mtu                    =       ip6_mtu,
 250         .cow_metrics            =       dst_cow_metrics_generic,
 251         .destroy                =       ip6_dst_destroy,
 252         .ifdown                 =       ip6_dst_ifdown,
 253         .negative_advice        =       ip6_negative_advice,
 254         .link_failure           =       ip6_link_failure,
 255         .update_pmtu            =       ip6_rt_update_pmtu,
 256         .redirect               =       rt6_do_redirect,
 257         .local_out              =       __ip6_local_out,
 258         .neigh_lookup           =       ip6_dst_neigh_lookup,
 259         .confirm_neigh          =       ip6_confirm_neigh,
 260 };
 261
 262 static struct dst_ops ip6_dst_blackhole_ops = {
 263         .family                 = AF_INET6,
 264         .default_advmss         = ip6_default_advmss,
 265         .neigh_lookup           = ip6_dst_neigh_lookup,
 266         .check                  = ip6_dst_check,
 267         .destroy                = ip6_dst_destroy,
 268         .cow_metrics            = dst_cow_metrics_generic,
 269         .update_pmtu            = dst_blackhole_update_pmtu,
 270         .redirect               = dst_blackhole_redirect,
 271         .mtu                    = dst_blackhole_mtu,
 272 };
 273
 274 static const u32 ip6_template_metrics[RTAX_MAX] = {
 275         [RTAX_HOPLIMIT - 1] = 0,
 276 };
 277
 278 static const struct fib6_info fib6_null_entry_template = {
 279         .fib6_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 280         .fib6_protocol  = RTPROT_KERNEL,
 281         .fib6_metric    = ~(u32)0,
 282         .fib6_ref       = REFCOUNT_INIT(1),
 283         .fib6_type      = RTN_UNREACHABLE,
 284         .fib6_metrics   = (struct dst_metrics *)&dst_default_metrics,
 285 };
 286
 287 static const struct rt6_info ip6_null_entry_template = {
 288         .dst = {
 289                 .__refcnt       = ATOMIC_INIT(1),
 290                 .__use          = 1,
 291                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 292                 .error          = -ENETUNREACH,
 293                 .input          = ip6_pkt_discard,
 294                 .output         = ip6_pkt_discard_out,
 295         },
 296         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 297 };
 298
 299 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
 300
 301 static const struct rt6_info ip6_prohibit_entry_template = {
 302         .dst = {
 303                 .__refcnt       = ATOMIC_INIT(1),
 304                 .__use          = 1,
 305                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 306                 .error          = -EACCES,
 307                 .input          = ip6_pkt_prohibit,
 308                 .output         = ip6_pkt_prohibit_out,
 309         },
 310         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 311 };
 312
 313 static const struct rt6_info ip6_blk_hole_entry_template = {
 314         .dst = {
 315                 .__refcnt       = ATOMIC_INIT(1),
 316                 .__use          = 1,
 317                 .obsolete       = DST_OBSOLETE_FORCE_CHK,
 318                 .error          = -EINVAL,
 319                 .input          = dst_discard,
 320                 .output         = dst_discard_out,
 321         },
 322         .rt6i_flags     = (RTF_REJECT | RTF_NONEXTHOP),
 323 };
 324
 325 #endif
 326
 327 static void rt6_info_init(struct rt6_info *rt)
 328 {
 329         struct dst_entry *dst = &rt->dst;
 330
 331         memset(dst + 1, 0, sizeof(*rt) - sizeof(*dst));
 332         INIT_LIST_HEAD(&rt->rt6i_uncached);
 333 }
 334
 335 /* allocate dst with ip6_dst_ops */
 336 struct rt6_info *ip6_dst_alloc(struct net *net, struct net_device *dev,
 337                                int flags)
 338 {
 339         struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 340                                         1, DST_OBSOLETE_FORCE_CHK, flags);
 341
 342         if (rt) {
 343                 rt6_info_init(rt);
 344                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
 345         }
 346
 347         return rt;
 348 }
 349 EXPORT_SYMBOL(ip6_dst_alloc);
 350
 351 static void ip6_dst_destroy(struct dst_entry *dst)
 352 {
 353         struct rt6_info *rt = (struct rt6_info *)dst;
 354         struct fib6_info *from;
 355         struct inet6_dev *idev;
 356
 357         ip_dst_metrics_put(dst);
 358         rt6_uncached_list_del(rt);
 359
 360         idev = rt->rt6i_idev;
 361         if (idev) {
 362                 rt->rt6i_idev = NULL;
 363                 in6_dev_put(idev);
 364         }
 365
 366         from = xchg((__force struct fib6_info **)&rt->from, NULL);
 367         fib6_info_release(from);
 368 }
 369
 370 static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev,
 371                            int how)
 372 {
 373         struct rt6_info *rt = (struct rt6_info *)dst;
 374         struct inet6_dev *idev = rt->rt6i_idev;
 375         struct net_device *loopback_dev =
 376                 dev_net(dev)->loopback_dev;
 377
 378         if (idev && idev->dev != loopback_dev) {
 379                 struct inet6_dev *loopback_idev = in6_dev_get(loopback_dev);
 380                 if (loopback_idev) {
 381                         rt->rt6i_idev = loopback_idev;
 382                         in6_dev_put(idev);
 383                 }
 384         }
 385 }
 386
 387 static bool __rt6_check_expired(const struct rt6_info *rt)
 388 {
 389         if (rt->rt6i_flags & RTF_EXPIRES)
 390                 return time_after(jiffies, rt->dst.expires);
 391         else
 392                 return false;
 393 }
 394
 395 static bool rt6_check_expired(const struct rt6_info *rt)
 396 {
 397         struct fib6_info *from;
 398
 399         from = rcu_dereference(rt->from);
 400
 401         if (rt->rt6i_flags & RTF_EXPIRES) {
 402                 if (time_after(jiffies, rt->dst.expires))
 403                         return true;
 404         } else if (from) {
 405                 return rt->dst.obsolete != DST_OBSOLETE_FORCE_CHK ||
 406                         fib6_check_expired(from);
 407         }
 408         return false;
 409 }
 410
 411 void fib6_select_path(const struct net *net, struct fib6_result *res,
 412                       struct flowi6 *fl6, int oif, bool have_oif_match,
 413                       const struct sk_buff *skb, int strict)
 414 {
 415         struct fib6_info *sibling, *next_sibling;
 416         struct fib6_info *match = res->f6i;
 417
 418         if (!match->nh && (!match->fib6_nsiblings || have_oif_match))
 419                 goto out;
 420
 421         if (match->nh && have_oif_match && res->nh)
 422                 return;
 423
 424         /* We might have already computed the hash for ICMPv6 errors. In such
 425          * case it will always be non-zero. Otherwise now is the time to do it.
 426          */
 427         if (!fl6->mp_hash &&
 428             (!match->nh || nexthop_is_multipath(match->nh)))
 429                 fl6->mp_hash = rt6_multipath_hash(net, fl6, skb, NULL);
 430
 431         if (unlikely(match->nh)) {
 432                 nexthop_path_fib6_result(res, fl6->mp_hash);
 433                 return;
 434         }
 435
 436         if (fl6->mp_hash <= atomic_read(&match->fib6_nh->fib_nh_upper_bound))
 437                 goto out;
 438
 439         list_for_each_entry_safe(sibling, next_sibling, &match->fib6_siblings,
 440                                  fib6_siblings) {
 441                 const struct fib6_nh *nh = sibling->fib6_nh;
 442                 int nh_upper_bound;
 443
 444                 nh_upper_bound = atomic_read(&nh->fib_nh_upper_bound);
 445                 if (fl6->mp_hash > nh_upper_bound)
 446                         continue;
 447                 if (rt6_score_route(nh, sibling->fib6_flags, oif, strict) < 0)
 448                         break;
 449                 match = sibling;
 450                 break;
 451         }
 452
 453 out:
 454         res->f6i = match;
 455         res->nh = match->fib6_nh;
 456 }
 457
 458 /*
 459  *      Route lookup. rcu_read_lock() should be held.
 460  */
 461
 462 static bool __rt6_device_match(struct net *net, const struct fib6_nh *nh,
 463                                const struct in6_addr *saddr, int oif, int flags)
 464 {
 465         const struct net_device *dev;
 466
 467         if (nh->fib_nh_flags & RTNH_F_DEAD)
 468                 return false;
 469
 470         dev = nh->fib_nh_dev;
 471         if (oif) {
 472                 if (dev->ifindex == oif)
 473                         return true;
 474         } else {
 475                 if (ipv6_chk_addr(net, saddr, dev,
 476                                   flags & RT6_LOOKUP_F_IFACE))
 477                         return true;
 478         }
 479
 480         return false;
 481 }
 482
 483 struct fib6_nh_dm_arg {
 484         struct net              *net;
 485         const struct in6_addr   *saddr;
 486         int                     oif;
 487         int                     flags;
 488         struct fib6_nh          *nh;
 489 };
 490
 491 static int __rt6_nh_dev_match(struct fib6_nh *nh, void *_arg)
 492 {
 493         struct fib6_nh_dm_arg *arg = _arg;
 494
 495         arg->nh = nh;
 496         return __rt6_device_match(arg->net, nh, arg->saddr, arg->oif,
 497                                   arg->flags);
 498 }
 499
 500 /* returns fib6_nh from nexthop or NULL */
 501 static struct fib6_nh *rt6_nh_dev_match(struct net *net, struct nexthop *nh,
 502                                         struct fib6_result *res,
 503                                         const struct in6_addr *saddr,
 504                                         int oif, int flags)
 505 {
 506         struct fib6_nh_dm_arg arg = {
 507                 .net   = net,
 508                 .saddr = saddr,
 509                 .oif   = oif,
 510                 .flags = flags,
 511         };
 512
 513         if (nexthop_is_blackhole(nh))
 514                 return NULL;
 515
 516         if (nexthop_for_each_fib6_nh(nh, __rt6_nh_dev_match, &arg))
 517                 return arg.nh;
 518
 519         return NULL;
 520 }
 521
 522 static void rt6_device_match(struct net *net, struct fib6_result *res,
 523                              const struct in6_addr *saddr, int oif, int flags)
 524 {
 525         struct fib6_info *f6i = res->f6i;
 526         struct fib6_info *spf6i;
 527         struct fib6_nh *nh;
 528
 529         if (!oif && ipv6_addr_any(saddr)) {
 530                 if (unlikely(f6i->nh)) {
 531                         nh = nexthop_fib6_nh(f6i->nh);
 532                         if (nexthop_is_blackhole(f6i->nh))
 533                                 goto out_blackhole;
 534                 } else {
 535                         nh = f6i->fib6_nh;
 536                 }
 537                 if (!(nh->fib_nh_flags & RTNH_F_DEAD))
 538                         goto out;
 539         }
 540
 541         for (spf6i = f6i; spf6i; spf6i = rcu_dereference(spf6i->fib6_next)) {
 542                 bool matched = false;
 543
 544                 if (unlikely(spf6i->nh)) {
 545                         nh = rt6_nh_dev_match(net, spf6i->nh, res, saddr,
 546                                               oif, flags);
 547                         if (nh)
 548                                 matched = true;
 549                 } else {
 550                         nh = spf6i->fib6_nh;
 551                         if (__rt6_device_match(net, nh, saddr, oif, flags))
 552                                 matched = true;
 553                 }
 554                 if (matched) {
 555                         res->f6i = spf6i;
 556                         goto out;
 557                 }
 558         }
 559
 560         if (oif && flags & RT6_LOOKUP_F_IFACE) {
 561                 res->f6i = net->ipv6.fib6_null_entry;
 562                 nh = res->f6i->fib6_nh;
 563                 goto out;
 564         }
 565
 566         if (unlikely(f6i->nh)) {
 567                 nh = nexthop_fib6_nh(f6i->nh);
 568                 if (nexthop_is_blackhole(f6i->nh))
 569                         goto out_blackhole;
 570         } else {
 571                 nh = f6i->fib6_nh;
 572         }
 573
 574         if (nh->fib_nh_flags & RTNH_F_DEAD) {
 575                 res->f6i = net->ipv6.fib6_null_entry;
 576                 nh = res->f6i->fib6_nh;
 577         }
 578 out:
 579         res->nh = nh;
 580         res->fib6_type = res->f6i->fib6_type;
 581         res->fib6_flags = res->f6i->fib6_flags;
 582         return;
 583
 584 out_blackhole:
 585         res->fib6_flags |= RTF_REJECT;
 586         res->fib6_type = RTN_BLACKHOLE;
 587         res->nh = nh;
 588 }
 589
 590 #ifdef CONFIG_IPV6_ROUTER_PREF
 591 struct __rt6_probe_work {
 592         struct work_struct work;
 593         struct in6_addr target;
 594         struct net_device *dev;
 595 };
 596
 597 static void rt6_probe_deferred(struct work_struct *w)
 598 {
 599         struct in6_addr mcaddr;
 600         struct __rt6_probe_work *work =
 601                 container_of(w, struct __rt6_probe_work, work);
 602
 603         addrconf_addr_solict_mult(&work->target, &mcaddr);
 604         ndisc_send_ns(work->dev, &work->target, &mcaddr, NULL, 0);
 605         dev_put(work->dev);
 606         kfree(work);
 607 }
 608
 609 static void rt6_probe(struct fib6_nh *fib6_nh)
 610 {
 611         struct __rt6_probe_work *work = NULL;
 612         const struct in6_addr *nh_gw;
 613         unsigned long last_probe;
 614         struct neighbour *neigh;
 615         struct net_device *dev;
 616         struct inet6_dev *idev;
 617
 618         /*
 619          * Okay, this does not seem to be appropriate
 620          * for now, however, we need to check if it
 621          * is really so; aka Router Reachability Probing.
 622          *
 623          * Router Reachability Probe MUST be rate-limited
 624          * to no more than one per minute.
 625          */
 626         if (!fib6_nh->fib_nh_gw_family)
 627                 return;
 628
 629         nh_gw = &fib6_nh->fib_nh_gw6;
 630         dev = fib6_nh->fib_nh_dev;
 631         rcu_read_lock_bh();
 632         last_probe = READ_ONCE(fib6_nh->last_probe);
 633         idev = __in6_dev_get(dev);
 634         neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 635         if (neigh) {
 636                 if (neigh->nud_state & NUD_VALID)
 637                         goto out;
 638
 639                 write_lock(&neigh->lock);
 640                 if (!(neigh->nud_state & NUD_VALID) &&
 641                     time_after(jiffies,
 642                                neigh->updated + idev->cnf.rtr_probe_interval)) {
 643                         work = kmalloc(sizeof(*work), GFP_ATOMIC);
 644                         if (work)
 645                                 __neigh_set_probe_once(neigh);
 646                 }
 647                 write_unlock(&neigh->lock);
 648         } else if (time_after(jiffies, last_probe +
 649                                        idev->cnf.rtr_probe_interval)) {
 650                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
 651         }
 652
 653         if (!work || cmpxchg(&fib6_nh->last_probe,
 654                              last_probe, jiffies) != last_probe) {
 655                 kfree(work);
 656         } else {
 657                 INIT_WORK(&work->work, rt6_probe_deferred);
 658                 work->target = *nh_gw;
 659                 dev_hold(dev);
 660                 work->dev = dev;
 661                 schedule_work(&work->work);
 662         }
 663
 664 out:
 665         rcu_read_unlock_bh();
 666 }
 667 #else
 668 static inline void rt6_probe(struct fib6_nh *fib6_nh)
 669 {
 670 }
 671 #endif
 672
 673 /*
 674  * Default Router Selection (RFC 2461 6.3.6)
 675  */
 676 static enum rt6_nud_state rt6_check_neigh(const struct fib6_nh *fib6_nh)
 677 {
 678         enum rt6_nud_state ret = RT6_NUD_FAIL_HARD;
 679         struct neighbour *neigh;
 680
 681         rcu_read_lock_bh();
 682         neigh = __ipv6_neigh_lookup_noref(fib6_nh->fib_nh_dev,
 683                                           &fib6_nh->fib_nh_gw6);
 684         if (neigh) {
 685                 read_lock(&neigh->lock);
 686                 if (neigh->nud_state & NUD_VALID)
 687                         ret = RT6_NUD_SUCCEED;
 688 #ifdef CONFIG_IPV6_ROUTER_PREF
 689                 else if (!(neigh->nud_state & NUD_FAILED))
 690                         ret = RT6_NUD_SUCCEED;
 691                 else
 692                         ret = RT6_NUD_FAIL_PROBE;
 693 #endif
 694                 read_unlock(&neigh->lock);
 695         } else {
 696                 ret = IS_ENABLED(CONFIG_IPV6_ROUTER_PREF) ?
 697                       RT6_NUD_SUCCEED : RT6_NUD_FAIL_DO_RR;
 698         }
 699         rcu_read_unlock_bh();
 700
 701         return ret;
 702 }
 703
 704 static int rt6_score_route(const struct fib6_nh *nh, u32 fib6_flags, int oif,
 705                            int strict)
 706 {
 707         int m = 0;
 708
 709         if (!oif || nh->fib_nh_dev->ifindex == oif)
 710                 m = 2;
 711
 712         if (!m && (strict & RT6_LOOKUP_F_IFACE))
 713                 return RT6_NUD_FAIL_HARD;
 714 #ifdef CONFIG_IPV6_ROUTER_PREF
 715         m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(fib6_flags)) << 2;
 716 #endif
 717         if ((strict & RT6_LOOKUP_F_REACHABLE) &&
 718             !(fib6_flags & RTF_NONEXTHOP) && nh->fib_nh_gw_family) {
 719                 int n = rt6_check_neigh(nh);
 720                 if (n < 0)
 721                         return n;
 722         }
 723         return m;
 724 }
 725
 726 static bool find_match(struct fib6_nh *nh, u32 fib6_flags,
 727                        int oif, int strict, int *mpri, bool *do_rr)
 728 {
 729         bool match_do_rr = false;
 730         bool rc = false;
 731         int m;
 732
 733         if (nh->fib_nh_flags & RTNH_F_DEAD)
 734                 goto out;
 735
 736         if (ip6_ignore_linkdown(nh->fib_nh_dev) &&
 737             nh->fib_nh_flags & RTNH_F_LINKDOWN &&
 738             !(strict & RT6_LOOKUP_F_IGNORE_LINKSTATE))
 739                 goto out;
 740
 741         m = rt6_score_route(nh, fib6_flags, oif, strict);
 742         if (m == RT6_NUD_FAIL_DO_RR) {
 743                 match_do_rr = true;
 744                 m = 0; /* lowest valid score */
 745         } else if (m == RT6_NUD_FAIL_HARD) {
 746                 goto out;
 747         }
 748
 749         if (strict & RT6_LOOKUP_F_REACHABLE)
 750                 rt6_probe(nh);
 751
 752         /* note that m can be RT6_NUD_FAIL_PROBE at this point */
 753         if (m > *mpri) {
 754                 *do_rr = match_do_rr;
 755                 *mpri = m;
 756                 rc = true;
 757         }
 758 out:
 759         return rc;
 760 }
 761
 762 struct fib6_nh_frl_arg {
 763         u32             flags;
 764         int             oif;
 765         int             strict;
 766         int             *mpri;
 767         bool            *do_rr;
 768         struct fib6_nh  *nh;
 769 };
 770
 771 static int rt6_nh_find_match(struct fib6_nh *nh, void *_arg)
 772 {
 773         struct fib6_nh_frl_arg *arg = _arg;
 774
 775         arg->nh = nh;
 776         return find_match(nh, arg->flags, arg->oif, arg->strict,
 777                           arg->mpri, arg->do_rr);
 778 }
 779
 780 static void __find_rr_leaf(struct fib6_info *f6i_start,
 781                            struct fib6_info *nomatch, u32 metric,
 782                            struct fib6_result *res, struct fib6_info **cont,
 783                            int oif, int strict, bool *do_rr, int *mpri)
 784 {
 785         struct fib6_info *f6i;
 786
 787         for (f6i = f6i_start;
 788              f6i && f6i != nomatch;
 789              f6i = rcu_dereference(f6i->fib6_next)) {
 790                 bool matched = false;
 791                 struct fib6_nh *nh;
 792
 793                 if (cont && f6i->fib6_metric != metric) {
 794                         *cont = f6i;
 795                         return;
 796                 }
 797
 798                 if (fib6_check_expired(f6i))
 799                         continue;
 800
 801                 if (unlikely(f6i->nh)) {
 802                         struct fib6_nh_frl_arg arg = {
 803                                 .flags  = f6i->fib6_flags,
 804                                 .oif    = oif,
 805                                 .strict = strict,
 806                                 .mpri   = mpri,
 807                                 .do_rr  = do_rr
 808                         };
 809
 810                         if (nexthop_is_blackhole(f6i->nh)) {
 811                                 res->fib6_flags = RTF_REJECT;
 812                                 res->fib6_type = RTN_BLACKHOLE;
 813                                 res->f6i = f6i;
 814                                 res->nh = nexthop_fib6_nh(f6i->nh);
 815                                 return;
 816                         }
 817                         if (nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_find_match,
 818                                                      &arg)) {
 819                                 matched = true;
 820                                 nh = arg.nh;
 821                         }
 822                 } else {
 823                         nh = f6i->fib6_nh;
 824                         if (find_match(nh, f6i->fib6_flags, oif, strict,
 825                                        mpri, do_rr))
 826                                 matched = true;
 827                 }
 828                 if (matched) {
 829                         res->f6i = f6i;
 830                         res->nh = nh;
 831                         res->fib6_flags = f6i->fib6_flags;
 832                         res->fib6_type = f6i->fib6_type;
 833                 }
 834         }
 835 }
 836
 837 static void find_rr_leaf(struct fib6_node *fn, struct fib6_info *leaf,
 838                          struct fib6_info *rr_head, int oif, int strict,
 839                          bool *do_rr, struct fib6_result *res)
 840 {
 841         u32 metric = rr_head->fib6_metric;
 842         struct fib6_info *cont = NULL;
 843         int mpri = -1;
 844
 845         __find_rr_leaf(rr_head, NULL, metric, res, &cont,
 846                        oif, strict, do_rr, &mpri);
 847
 848         __find_rr_leaf(leaf, rr_head, metric, res, &cont,
 849                        oif, strict, do_rr, &mpri);
 850
 851         if (res->f6i || !cont)
 852                 return;
 853
 854         __find_rr_leaf(cont, NULL, metric, res, NULL,
 855                        oif, strict, do_rr, &mpri);
 856 }
 857
 858 static void rt6_select(struct net *net, struct fib6_node *fn, int oif,
 859                        struct fib6_result *res, int strict)
 860 {
 861         struct fib6_info *leaf = rcu_dereference(fn->leaf);
 862         struct fib6_info *rt0;
 863         bool do_rr = false;
 864         int key_plen;
 865
 866         /* make sure this function or its helpers sets f6i */
 867         res->f6i = NULL;
 868
 869         if (!leaf || leaf == net->ipv6.fib6_null_entry)
 870                 goto out;
 871
 872         rt0 = rcu_dereference(fn->rr_ptr);
 873         if (!rt0)
 874                 rt0 = leaf;
 875
 876         /* Double check to make sure fn is not an intermediate node
 877          * and fn->leaf does not points to its child's leaf
 878          * (This might happen if all routes under fn are deleted from
 879          * the tree and fib6_repair_tree() is called on the node.)
 880          */
 881         key_plen = rt0->fib6_dst.plen;
 882 #ifdef CONFIG_IPV6_SUBTREES
 883         if (rt0->fib6_src.plen)
 884                 key_plen = rt0->fib6_src.plen;
 885 #endif
 886         if (fn->fn_bit != key_plen)
 887                 goto out;
 888
 889         find_rr_leaf(fn, leaf, rt0, oif, strict, &do_rr, res);
 890         if (do_rr) {
 891                 struct fib6_info *next = rcu_dereference(rt0->fib6_next);
 892
 893                 /* no entries matched; do round-robin */
 894                 if (!next || next->fib6_metric != rt0->fib6_metric)
 895                         next = leaf;
 896
 897                 if (next != rt0) {
 898                         spin_lock_bh(&leaf->fib6_table->tb6_lock);
 899                         /* make sure next is not being deleted from the tree */
 900                         if (next->fib6_node)
 901                                 rcu_assign_pointer(fn->rr_ptr, next);
 902                         spin_unlock_bh(&leaf->fib6_table->tb6_lock);
 903                 }
 904         }
 905
 906 out:
 907         if (!res->f6i) {
 908                 res->f6i = net->ipv6.fib6_null_entry;
 909                 res->nh = res->f6i->fib6_nh;
 910                 res->fib6_flags = res->f6i->fib6_flags;
 911                 res->fib6_type = res->f6i->fib6_type;
 912         }
 913 }
 914
 915 static bool rt6_is_gw_or_nonexthop(const struct fib6_result *res)
 916 {
 917         return (res->f6i->fib6_flags & RTF_NONEXTHOP) ||
 918                res->nh->fib_nh_gw_family;
 919 }
 920
 921 #ifdef CONFIG_IPV6_ROUTE_INFO
 922 int rt6_route_rcv(struct net_device *dev, u8 *opt, int len,
 923                   const struct in6_addr *gwaddr)
 924 {
 925         struct net *net = dev_net(dev);
 926         struct route_info *rinfo = (struct route_info *) opt;
 927         struct in6_addr prefix_buf, *prefix;
 928         unsigned int pref;
 929         unsigned long lifetime;
 930         struct fib6_info *rt;
 931
 932         if (len < sizeof(struct route_info)) {
 933                 return -EINVAL;
 934         }
 935
 936         /* Sanity check for prefix_len and length */
 937         if (rinfo->length > 3) {
 938                 return -EINVAL;
 939         } else if (rinfo->prefix_len > 128) {
 940                 return -EINVAL;
 941         } else if (rinfo->prefix_len > 64) {
 942                 if (rinfo->length < 2) {
 943                         return -EINVAL;
 944                 }
 945         } else if (rinfo->prefix_len > 0) {
 946                 if (rinfo->length < 1) {
 947                         return -EINVAL;
 948                 }
 949         }
 950
 951         pref = rinfo->route_pref;
 952         if (pref == ICMPV6_ROUTER_PREF_INVALID)
 953                 return -EINVAL;
 954
 955         lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ);
 956
 957         if (rinfo->length == 3)
 958                 prefix = (struct in6_addr *)rinfo->prefix;
 959         else {
 960                 /* this function is safe */
 961                 ipv6_addr_prefix(&prefix_buf,
 962                                  (struct in6_addr *)rinfo->prefix,
 963                                  rinfo->prefix_len);
 964                 prefix = &prefix_buf;
 965         }
 966
 967         if (rinfo->prefix_len == 0)
 968                 rt = rt6_get_dflt_router(net, gwaddr, dev);
 969         else
 970                 rt = rt6_get_route_info(net, prefix, rinfo->prefix_len,
 971                                         gwaddr, dev);
 972
 973         if (rt && !lifetime) {
 974                 ip6_del_rt(net, rt, false);
 975                 rt = NULL;
 976         }
 977
 978         if (!rt && lifetime)
 979                 rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr,
 980                                         dev, pref);
 981         else if (rt)
 982                 rt->fib6_flags = RTF_ROUTEINFO |
 983                                  (rt->fib6_flags & ~RTF_PREF_MASK) | RTF_PREF(pref);
 984
 985         if (rt) {
 986                 if (!addrconf_finite_timeout(lifetime))
 987                         fib6_clean_expires(rt);
 988                 else
 989                         fib6_set_expires(rt, jiffies + HZ * lifetime);
 990
 991                 fib6_info_release(rt);
 992         }
 993         return 0;
 994 }
 995 #endif
 996
 997 /*
 998  *      Misc support functions
 999  */
1000
1001 /* called with rcu_lock held */
1002 static struct net_device *ip6_rt_get_dev_rcu(const struct fib6_result *res)
1003 {
1004         struct net_device *dev = res->nh->fib_nh_dev;
1005
1006         if (res->fib6_flags & (RTF_LOCAL | RTF_ANYCAST)) {
1007                 /* for copies of local routes, dst->dev needs to be the
1008                  * device if it is a master device, the master device if
1009                  * device is enslaved, and the loopback as the default
1010                  */
1011                 if (netif_is_l3_slave(dev) &&
1012                     !rt6_need_strict(&res->f6i->fib6_dst.addr))
1013                         dev = l3mdev_master_dev_rcu(dev);
1014                 else if (!netif_is_l3_master(dev))
1015                         dev = dev_net(dev)->loopback_dev;
1016                 /* last case is netif_is_l3_master(dev) is true in which
1017                  * case we want dev returned to be dev
1018                  */
1019         }
1020
1021         return dev;
1022 }
1023
1024 static const int fib6_prop[RTN_MAX + 1] = {
1025         [RTN_UNSPEC]    = 0,
1026         [RTN_UNICAST]   = 0,
1027         [RTN_LOCAL]     = 0,
1028         [RTN_BROADCAST] = 0,
1029         [RTN_ANYCAST]   = 0,
1030         [RTN_MULTICAST] = 0,
1031         [RTN_BLACKHOLE] = -EINVAL,
1032         [RTN_UNREACHABLE] = -EHOSTUNREACH,
1033         [RTN_PROHIBIT]  = -EACCES,
1034         [RTN_THROW]     = -EAGAIN,
1035         [RTN_NAT]       = -EINVAL,
1036         [RTN_XRESOLVE]  = -EINVAL,
1037 };
1038
1039 static int ip6_rt_type_to_error(u8 fib6_type)
1040 {
1041         return fib6_prop[fib6_type];
1042 }
1043
1044 static unsigned short fib6_info_dst_flags(struct fib6_info *rt)
1045 {
1046         unsigned short flags = 0;
1047
1048         if (rt->dst_nocount)
1049                 flags |= DST_NOCOUNT;
1050         if (rt->dst_nopolicy)
1051                 flags |= DST_NOPOLICY;
1052
1053         return flags;
1054 }
1055
1056 static void ip6_rt_init_dst_reject(struct rt6_info *rt, u8 fib6_type)
1057 {
1058         rt->dst.error = ip6_rt_type_to_error(fib6_type);
1059
1060         switch (fib6_type) {
1061         case RTN_BLACKHOLE:
1062                 rt->dst.output = dst_discard_out;
1063                 rt->dst.input = dst_discard;
1064                 break;
1065         case RTN_PROHIBIT:
1066                 rt->dst.output = ip6_pkt_prohibit_out;
1067                 rt->dst.input = ip6_pkt_prohibit;
1068                 break;
1069         case RTN_THROW:
1070         case RTN_UNREACHABLE:
1071         default:
1072                 rt->dst.output = ip6_pkt_discard_out;
1073                 rt->dst.input = ip6_pkt_discard;
1074                 break;
1075         }
1076 }
1077
1078 static void ip6_rt_init_dst(struct rt6_info *rt, const struct fib6_result *res)
1079 {
1080         struct fib6_info *f6i = res->f6i;
1081
1082         if (res->fib6_flags & RTF_REJECT) {
1083                 ip6_rt_init_dst_reject(rt, res->fib6_type);
1084                 return;
1085         }
1086
1087         rt->dst.error = 0;
1088         rt->dst.output = ip6_output;
1089
1090         if (res->fib6_type == RTN_LOCAL || res->fib6_type == RTN_ANYCAST) {
1091                 rt->dst.input = ip6_input;
1092         } else if (ipv6_addr_type(&f6i->fib6_dst.addr) & IPV6_ADDR_MULTICAST) {
1093                 rt->dst.input = ip6_mc_input;
1094         } else {
1095                 rt->dst.input = ip6_forward;
1096         }
1097
1098         if (res->nh->fib_nh_lws) {
1099                 rt->dst.lwtstate = lwtstate_get(res->nh->fib_nh_lws);
1100                 lwtunnel_set_redirect(&rt->dst);
1101         }
1102
1103         rt->dst.lastuse = jiffies;
1104 }
1105
1106 /* Caller must already hold reference to @from */
1107 static void rt6_set_from(struct rt6_info *rt, struct fib6_info *from)
1108 {
1109         rt->rt6i_flags &= ~RTF_EXPIRES;
1110         rcu_assign_pointer(rt->from, from);
1111         ip_dst_init_metrics(&rt->dst, from->fib6_metrics);
1112 }
1113
1114 /* Caller must already hold reference to f6i in result */
1115 static void ip6_rt_copy_init(struct rt6_info *rt, const struct fib6_result *res)
1116 {
1117         const struct fib6_nh *nh = res->nh;
1118         const struct net_device *dev = nh->fib_nh_dev;
1119         struct fib6_info *f6i = res->f6i;
1120
1121         ip6_rt_init_dst(rt, res);
1122
1123         rt->rt6i_dst = f6i->fib6_dst;
1124         rt->rt6i_idev = dev ? in6_dev_get(dev) : NULL;
1125         rt->rt6i_flags = res->fib6_flags;
1126         if (nh->fib_nh_gw_family) {
1127                 rt->rt6i_gateway = nh->fib_nh_gw6;
1128                 rt->rt6i_flags |= RTF_GATEWAY;
1129         }
1130         rt6_set_from(rt, f6i);
1131 #ifdef CONFIG_IPV6_SUBTREES
1132         rt->rt6i_src = f6i->fib6_src;
1133 #endif
1134 }
1135
1136 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
1137                                         struct in6_addr *saddr)
1138 {
1139         struct fib6_node *pn, *sn;
1140         while (1) {
1141                 if (fn->fn_flags & RTN_TL_ROOT)
1142                         return NULL;
1143                 pn = rcu_dereference(fn->parent);
1144                 sn = FIB6_SUBTREE(pn);
1145                 if (sn && sn != fn)
1146                         fn = fib6_node_lookup(sn, NULL, saddr);
1147                 else
1148                         fn = pn;
1149                 if (fn->fn_flags & RTN_RTINFO)
1150                         return fn;
1151         }
1152 }
1153
1154 static bool ip6_hold_safe(struct net *net, struct rt6_info **prt)
1155 {
1156         struct rt6_info *rt = *prt;
1157
1158         if (dst_hold_safe(&rt->dst))
1159                 return true;
1160         if (net) {
1161                 rt = net->ipv6.ip6_null_entry;
1162                 dst_hold(&rt->dst);
1163         } else {
1164                 rt = NULL;
1165         }
1166         *prt = rt;
1167         return false;
1168 }
1169
1170 /* called with rcu_lock held */
1171 static struct rt6_info *ip6_create_rt_rcu(const struct fib6_result *res)
1172 {
1173         struct net_device *dev = res->nh->fib_nh_dev;
1174         struct fib6_info *f6i = res->f6i;
1175         unsigned short flags;
1176         struct rt6_info *nrt;
1177
1178         if (!fib6_info_hold_safe(f6i))
1179                 goto fallback;
1180
1181         flags = fib6_info_dst_flags(f6i);
1182         nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
1183         if (!nrt) {
1184                 fib6_info_release(f6i);
1185                 goto fallback;
1186         }
1187
1188         ip6_rt_copy_init(nrt, res);
1189         return nrt;
1190
1191 fallback:
1192         nrt = dev_net(dev)->ipv6.ip6_null_entry;
1193         dst_hold(&nrt->dst);
1194         return nrt;
1195 }
1196
1197 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_lookup(struct net *net,
1198                                              struct fib6_table *table,
1199                                              struct flowi6 *fl6,
1200                                              const struct sk_buff *skb,
1201                                              int flags)
1202 {
1203         struct fib6_result res = {};
1204         struct fib6_node *fn;
1205         struct rt6_info *rt;
1206
1207         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
1208                 flags &= ~RT6_LOOKUP_F_IFACE;
1209
1210         rcu_read_lock();
1211         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
1212 restart:
1213         res.f6i = rcu_dereference(fn->leaf);
1214         if (!res.f6i)
1215                 res.f6i = net->ipv6.fib6_null_entry;
1216         else
1217                 rt6_device_match(net, &res, &fl6->saddr, fl6->flowi6_oif,
1218                                  flags);
1219
1220         if (res.f6i == net->ipv6.fib6_null_entry) {
1221                 fn = fib6_backtrack(fn, &fl6->saddr);
1222                 if (fn)
1223                         goto restart;
1224
1225                 rt = net->ipv6.ip6_null_entry;
1226                 dst_hold(&rt->dst);
1227                 goto out;
1228         } else if (res.fib6_flags & RTF_REJECT) {
1229                 goto do_create;
1230         }
1231
1232         fib6_select_path(net, &res, fl6, fl6->flowi6_oif,
1233                          fl6->flowi6_oif != 0, skb, flags);
1234
1235         /* Search through exception table */
1236         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
1237         if (rt) {
1238                 if (ip6_hold_safe(net, &rt))
1239                         dst_use_noref(&rt->dst, jiffies);
1240         } else {
1241 do_create:
1242                 rt = ip6_create_rt_rcu(&res);
1243         }
1244
1245 out:
1246         trace_fib6_table_lookup(net, &res, table, fl6);
1247
1248         rcu_read_unlock();
1249
1250         return rt;
1251 }
1252
1253 struct dst_entry *ip6_route_lookup(struct net *net, struct flowi6 *fl6,
1254                                    const struct sk_buff *skb, int flags)
1255 {
1256         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_lookup);
1257 }
1258 EXPORT_SYMBOL_GPL(ip6_route_lookup);
1259
1260 struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr,
1261                             const struct in6_addr *saddr, int oif,
1262                             const struct sk_buff *skb, int strict)
1263 {
1264         struct flowi6 fl6 = {
1265                 .flowi6_oif = oif,
1266                 .daddr = *daddr,
1267         };
1268         struct dst_entry *dst;
1269         int flags = strict ? RT6_LOOKUP_F_IFACE : 0;
1270
1271         if (saddr) {
1272                 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
1273                 flags |= RT6_LOOKUP_F_HAS_SADDR;
1274         }
1275
1276         dst = fib6_rule_lookup(net, &fl6, skb, flags, ip6_pol_route_lookup);
1277         if (dst->error == 0)
1278                 return (struct rt6_info *) dst;
1279
1280         dst_release(dst);
1281
1282         return NULL;
1283 }
1284 EXPORT_SYMBOL(rt6_lookup);
1285
1286 /* ip6_ins_rt is called with FREE table->tb6_lock.
1287  * It takes new route entry, the addition fails by any reason the
1288  * route is released.
1289  * Caller must hold dst before calling it.
1290  */
1291
1292 static int __ip6_ins_rt(struct fib6_info *rt, struct nl_info *info,
1293                         struct netlink_ext_ack *extack)
1294 {
1295         int err;
1296         struct fib6_table *table;
1297
1298         table = rt->fib6_table;
1299         spin_lock_bh(&table->tb6_lock);
1300         err = fib6_add(&table->tb6_root, rt, info, extack);
1301         spin_unlock_bh(&table->tb6_lock);
1302
1303         return err;
1304 }
1305
1306 int ip6_ins_rt(struct net *net, struct fib6_info *rt)
1307 {
1308         struct nl_info info = { .nl_net = net, };
1309
1310         return __ip6_ins_rt(rt, &info, NULL);
1311 }
1312
1313 static struct rt6_info *ip6_rt_cache_alloc(const struct fib6_result *res,
1314                                            const struct in6_addr *daddr,
1315                                            const struct in6_addr *saddr)
1316 {
1317         struct fib6_info *f6i = res->f6i;
1318         struct net_device *dev;
1319         struct rt6_info *rt;
1320
1321         /*
1322          *      Clone the route.
1323          */
1324
1325         if (!fib6_info_hold_safe(f6i))
1326                 return NULL;
1327
1328         dev = ip6_rt_get_dev_rcu(res);
1329         rt = ip6_dst_alloc(dev_net(dev), dev, 0);
1330         if (!rt) {
1331                 fib6_info_release(f6i);
1332                 return NULL;
1333         }
1334
1335         ip6_rt_copy_init(rt, res);
1336         rt->rt6i_flags |= RTF_CACHE;
1337         rt->rt6i_dst.addr = *daddr;
1338         rt->rt6i_dst.plen = 128;
1339
1340         if (!rt6_is_gw_or_nonexthop(res)) {
1341                 if (f6i->fib6_dst.plen != 128 &&
1342                     ipv6_addr_equal(&f6i->fib6_dst.addr, daddr))
1343                         rt->rt6i_flags |= RTF_ANYCAST;
1344 #ifdef CONFIG_IPV6_SUBTREES
1345                 if (rt->rt6i_src.plen && saddr) {
1346                         rt->rt6i_src.addr = *saddr;
1347                         rt->rt6i_src.plen = 128;
1348                 }
1349 #endif
1350         }
1351
1352         return rt;
1353 }
1354
1355 static struct rt6_info *ip6_rt_pcpu_alloc(const struct fib6_result *res)
1356 {
1357         struct fib6_info *f6i = res->f6i;
1358         unsigned short flags = fib6_info_dst_flags(f6i);
1359         struct net_device *dev;
1360         struct rt6_info *pcpu_rt;
1361
1362         if (!fib6_info_hold_safe(f6i))
1363                 return NULL;
1364
1365         rcu_read_lock();
1366         dev = ip6_rt_get_dev_rcu(res);
1367         pcpu_rt = ip6_dst_alloc(dev_net(dev), dev, flags | DST_NOCOUNT);
1368         rcu_read_unlock();
1369         if (!pcpu_rt) {
1370                 fib6_info_release(f6i);
1371                 return NULL;
1372         }
1373         ip6_rt_copy_init(pcpu_rt, res);
1374         pcpu_rt->rt6i_flags |= RTF_PCPU;
1375
1376         if (f6i->nh)
1377                 pcpu_rt->sernum = rt_genid_ipv6(dev_net(dev));
1378
1379         return pcpu_rt;
1380 }
1381
1382 static bool rt6_is_valid(const struct rt6_info *rt6)
1383 {
1384         return rt6->sernum == rt_genid_ipv6(dev_net(rt6->dst.dev));
1385 }
1386
1387 /* It should be called with rcu_read_lock() acquired */
1388 static struct rt6_info *rt6_get_pcpu_route(const struct fib6_result *res)
1389 {
1390         struct rt6_info *pcpu_rt;
1391
1392         pcpu_rt = this_cpu_read(*res->nh->rt6i_pcpu);
1393
1394         if (pcpu_rt && pcpu_rt->sernum && !rt6_is_valid(pcpu_rt)) {
1395                 struct rt6_info *prev, **p;
1396
1397                 p = this_cpu_ptr(res->nh->rt6i_pcpu);
1398                 prev = xchg(p, NULL);
1399                 if (prev) {
1400                         dst_dev_put(&prev->dst);
1401                         dst_release(&prev->dst);
1402                 }
1403
1404                 pcpu_rt = NULL;
1405         }
1406
1407         return pcpu_rt;
1408 }
1409
1410 static struct rt6_info *rt6_make_pcpu_route(struct net *net,
1411                                             const struct fib6_result *res)
1412 {
1413         struct rt6_info *pcpu_rt, *prev, **p;
1414
1415         pcpu_rt = ip6_rt_pcpu_alloc(res);
1416         if (!pcpu_rt)
1417                 return NULL;
1418
1419         p = this_cpu_ptr(res->nh->rt6i_pcpu);
1420         prev = cmpxchg(p, NULL, pcpu_rt);
1421         BUG_ON(prev);
1422
1423         if (res->f6i->fib6_destroying) {
1424                 struct fib6_info *from;
1425
1426                 from = xchg((__force struct fib6_info **)&pcpu_rt->from, NULL);
1427                 fib6_info_release(from);
1428         }
1429
1430         return pcpu_rt;
1431 }
1432
1433 /* exception hash table implementation
1434  */
1435 static DEFINE_SPINLOCK(rt6_exception_lock);
1436
1437 /* Remove rt6_ex from hash table and free the memory
1438  * Caller must hold rt6_exception_lock
1439  */
1440 static void rt6_remove_exception(struct rt6_exception_bucket *bucket,
1441                                  struct rt6_exception *rt6_ex)
1442 {
1443         struct fib6_info *from;
1444         struct net *net;
1445
1446         if (!bucket || !rt6_ex)
1447                 return;
1448
1449         net = dev_net(rt6_ex->rt6i->dst.dev);
1450         net->ipv6.rt6_stats->fib_rt_cache--;
1451
1452         /* purge completely the exception to allow releasing the held resources:
1453          * some [sk] cache may keep the dst around for unlimited time
1454          */
1455         from = xchg((__force struct fib6_info **)&rt6_ex->rt6i->from, NULL);
1456         fib6_info_release(from);
1457         dst_dev_put(&rt6_ex->rt6i->dst);
1458
1459         hlist_del_rcu(&rt6_ex->hlist);
1460         dst_release(&rt6_ex->rt6i->dst);
1461         kfree_rcu(rt6_ex, rcu);
1462         WARN_ON_ONCE(!bucket->depth);
1463         bucket->depth--;
1464 }
1465
1466 /* Remove oldest rt6_ex in bucket and free the memory
1467  * Caller must hold rt6_exception_lock
1468  */
1469 static void rt6_exception_remove_oldest(struct rt6_exception_bucket *bucket)
1470 {
1471         struct rt6_exception *rt6_ex, *oldest = NULL;
1472
1473         if (!bucket)
1474                 return;
1475
1476         hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
1477                 if (!oldest || time_before(rt6_ex->stamp, oldest->stamp))
1478                         oldest = rt6_ex;
1479         }
1480         rt6_remove_exception(bucket, oldest);
1481 }
1482
1483 static u32 rt6_exception_hash(const struct in6_addr *dst,
1484                               const struct in6_addr *src)
1485 {
1486         static siphash_key_t rt6_exception_key __read_mostly;
1487         struct {
1488                 struct in6_addr dst;
1489                 struct in6_addr src;
1490         } __aligned(SIPHASH_ALIGNMENT) combined = {
1491                 .dst = *dst,
1492         };
1493         u64 val;
1494
1495         net_get_random_once(&rt6_exception_key, sizeof(rt6_exception_key));
1496
1497 #ifdef CONFIG_IPV6_SUBTREES
1498         if (src)
1499                 combined.src = *src;
1500 #endif
1501         val = siphash(&combined, sizeof(combined), &rt6_exception_key);
1502
1503         return hash_64(val, FIB6_EXCEPTION_BUCKET_SIZE_SHIFT);
1504 }
1505
1506 /* Helper function to find the cached rt in the hash table
1507  * and update bucket pointer to point to the bucket for this
1508  * (daddr, saddr) pair
1509  * Caller must hold rt6_exception_lock
1510  */
1511 static struct rt6_exception *
1512 __rt6_find_exception_spinlock(struct rt6_exception_bucket **bucket,
1513                               const struct in6_addr *daddr,
1514                               const struct in6_addr *saddr)
1515 {
1516         struct rt6_exception *rt6_ex;
1517         u32 hval;
1518
1519         if (!(*bucket) || !daddr)
1520                 return NULL;
1521
1522         hval = rt6_exception_hash(daddr, saddr);
1523         *bucket += hval;
1524
1525         hlist_for_each_entry(rt6_ex, &(*bucket)->chain, hlist) {
1526                 struct rt6_info *rt6 = rt6_ex->rt6i;
1527                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1528
1529 #ifdef CONFIG_IPV6_SUBTREES
1530                 if (matched && saddr)
1531                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1532 #endif
1533                 if (matched)
1534                         return rt6_ex;
1535         }
1536         return NULL;
1537 }
1538
1539 /* Helper function to find the cached rt in the hash table
1540  * and update bucket pointer to point to the bucket for this
1541  * (daddr, saddr) pair
1542  * Caller must hold rcu_read_lock()
1543  */
1544 static struct rt6_exception *
1545 __rt6_find_exception_rcu(struct rt6_exception_bucket **bucket,
1546                          const struct in6_addr *daddr,
1547                          const struct in6_addr *saddr)
1548 {
1549         struct rt6_exception *rt6_ex;
1550         u32 hval;
1551
1552         WARN_ON_ONCE(!rcu_read_lock_held());
1553
1554         if (!(*bucket) || !daddr)
1555                 return NULL;
1556
1557         hval = rt6_exception_hash(daddr, saddr);
1558         *bucket += hval;
1559
1560         hlist_for_each_entry_rcu(rt6_ex, &(*bucket)->chain, hlist) {
1561                 struct rt6_info *rt6 = rt6_ex->rt6i;
1562                 bool matched = ipv6_addr_equal(daddr, &rt6->rt6i_dst.addr);
1563
1564 #ifdef CONFIG_IPV6_SUBTREES
1565                 if (matched && saddr)
1566                         matched = ipv6_addr_equal(saddr, &rt6->rt6i_src.addr);
1567 #endif
1568                 if (matched)
1569                         return rt6_ex;
1570         }
1571         return NULL;
1572 }
1573
1574 static unsigned int fib6_mtu(const struct fib6_result *res)
1575 {
1576         const struct fib6_nh *nh = res->nh;
1577         unsigned int mtu;
1578
1579         if (res->f6i->fib6_pmtu) {
1580                 mtu = res->f6i->fib6_pmtu;
1581         } else {
1582                 struct net_device *dev = nh->fib_nh_dev;
1583                 struct inet6_dev *idev;
1584
1585                 rcu_read_lock();
1586                 idev = __in6_dev_get(dev);
1587                 mtu = idev->cnf.mtu6;
1588                 rcu_read_unlock();
1589         }
1590
1591         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
1592
1593         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
1594 }
1595
1596 #define FIB6_EXCEPTION_BUCKET_FLUSHED  0x1UL
1597
1598 /* used when the flushed bit is not relevant, only access to the bucket
1599  * (ie., all bucket users except rt6_insert_exception);
1600  *
1601  * called under rcu lock; sometimes called with rt6_exception_lock held
1602  */
1603 static
1604 struct rt6_exception_bucket *fib6_nh_get_excptn_bucket(const struct fib6_nh *nh,
1605                                                        spinlock_t *lock)
1606 {
1607         struct rt6_exception_bucket *bucket;
1608
1609         if (lock)
1610                 bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1611                                                    lockdep_is_held(lock));
1612         else
1613                 bucket = rcu_dereference(nh->rt6i_exception_bucket);
1614
1615         /* remove bucket flushed bit if set */
1616         if (bucket) {
1617                 unsigned long p = (unsigned long)bucket;
1618
1619                 p &= ~FIB6_EXCEPTION_BUCKET_FLUSHED;
1620                 bucket = (struct rt6_exception_bucket *)p;
1621         }
1622
1623         return bucket;
1624 }
1625
1626 static bool fib6_nh_excptn_bucket_flushed(struct rt6_exception_bucket *bucket)
1627 {
1628         unsigned long p = (unsigned long)bucket;
1629
1630         return !!(p & FIB6_EXCEPTION_BUCKET_FLUSHED);
1631 }
1632
1633 /* called with rt6_exception_lock held */
1634 static void fib6_nh_excptn_bucket_set_flushed(struct fib6_nh *nh,
1635                                               spinlock_t *lock)
1636 {
1637         struct rt6_exception_bucket *bucket;
1638         unsigned long p;
1639
1640         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1641                                            lockdep_is_held(lock));
1642
1643         p = (unsigned long)bucket;
1644         p |= FIB6_EXCEPTION_BUCKET_FLUSHED;
1645         bucket = (struct rt6_exception_bucket *)p;
1646         rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1647 }
1648
1649 static int rt6_insert_exception(struct rt6_info *nrt,
1650                                 const struct fib6_result *res)
1651 {
1652         struct net *net = dev_net(nrt->dst.dev);
1653         struct rt6_exception_bucket *bucket;
1654         struct fib6_info *f6i = res->f6i;
1655         struct in6_addr *src_key = NULL;
1656         struct rt6_exception *rt6_ex;
1657         struct fib6_nh *nh = res->nh;
1658         int max_depth;
1659         int err = 0;
1660
1661         spin_lock_bh(&rt6_exception_lock);
1662
1663         bucket = rcu_dereference_protected(nh->rt6i_exception_bucket,
1664                                           lockdep_is_held(&rt6_exception_lock));
1665         if (!bucket) {
1666                 bucket = kcalloc(FIB6_EXCEPTION_BUCKET_SIZE, sizeof(*bucket),
1667                                  GFP_ATOMIC);
1668                 if (!bucket) {
1669                         err = -ENOMEM;
1670                         goto out;
1671                 }
1672                 rcu_assign_pointer(nh->rt6i_exception_bucket, bucket);
1673         } else if (fib6_nh_excptn_bucket_flushed(bucket)) {
1674                 err = -EINVAL;
1675                 goto out;
1676         }
1677
1678 #ifdef CONFIG_IPV6_SUBTREES
1679         /* fib6_src.plen != 0 indicates f6i is in subtree
1680          * and exception table is indexed by a hash of
1681          * both fib6_dst and fib6_src.
1682          * Otherwise, the exception table is indexed by
1683          * a hash of only fib6_dst.
1684          */
1685         if (f6i->fib6_src.plen)
1686                 src_key = &nrt->rt6i_src.addr;
1687 #endif
1688         /* rt6_mtu_change() might lower mtu on f6i.
1689          * Only insert this exception route if its mtu
1690          * is less than f6i's mtu value.
1691          */
1692         if (dst_metric_raw(&nrt->dst, RTAX_MTU) >= fib6_mtu(res)) {
1693                 err = -EINVAL;
1694                 goto out;
1695         }
1696
1697         rt6_ex = __rt6_find_exception_spinlock(&bucket, &nrt->rt6i_dst.addr,
1698                                                src_key);
1699         if (rt6_ex)
1700                 rt6_remove_exception(bucket, rt6_ex);
1701
1702         rt6_ex = kzalloc(sizeof(*rt6_ex), GFP_ATOMIC);
1703         if (!rt6_ex) {
1704                 err = -ENOMEM;
1705                 goto out;
1706         }
1707         rt6_ex->rt6i = nrt;
1708         rt6_ex->stamp = jiffies;
1709         hlist_add_head_rcu(&rt6_ex->hlist, &bucket->chain);
1710         bucket->depth++;
1711         net->ipv6.rt6_stats->fib_rt_cache++;
1712
1713         /* Randomize max depth to avoid some side channels attacks. */
1714         max_depth = FIB6_MAX_DEPTH + prandom_u32_max(FIB6_MAX_DEPTH);
1715         while (bucket->depth > max_depth)
1716                 rt6_exception_remove_oldest(bucket);
1717
1718 out:
1719         spin_unlock_bh(&rt6_exception_lock);
1720
1721         /* Update fn->fn_sernum to invalidate all cached dst */
1722         if (!err) {
1723                 spin_lock_bh(&f6i->fib6_table->tb6_lock);
1724                 fib6_update_sernum(net, f6i);
1725                 spin_unlock_bh(&f6i->fib6_table->tb6_lock);
1726                 fib6_force_start_gc(net);
1727         }
1728
1729         return err;
1730 }
1731
1732 static void fib6_nh_flush_exceptions(struct fib6_nh *nh, struct fib6_info *from)
1733 {
1734         struct rt6_exception_bucket *bucket;
1735         struct rt6_exception *rt6_ex;
1736         struct hlist_node *tmp;
1737         int i;
1738
1739         spin_lock_bh(&rt6_exception_lock);
1740
1741         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1742         if (!bucket)
1743                 goto out;
1744
1745         /* Prevent rt6_insert_exception() to recreate the bucket list */
1746         if (!from)
1747                 fib6_nh_excptn_bucket_set_flushed(nh, &rt6_exception_lock);
1748
1749         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
1750                 hlist_for_each_entry_safe(rt6_ex, tmp, &bucket->chain, hlist) {
1751                         if (!from ||
1752                             rcu_access_pointer(rt6_ex->rt6i->from) == from)
1753                                 rt6_remove_exception(bucket, rt6_ex);
1754                 }
1755                 WARN_ON_ONCE(!from && bucket->depth);
1756                 bucket++;
1757         }
1758 out:
1759         spin_unlock_bh(&rt6_exception_lock);
1760 }
1761
1762 static int rt6_nh_flush_exceptions(struct fib6_nh *nh, void *arg)
1763 {
1764         struct fib6_info *f6i = arg;
1765
1766         fib6_nh_flush_exceptions(nh, f6i);
1767
1768         return 0;
1769 }
1770
1771 void rt6_flush_exceptions(struct fib6_info *f6i)
1772 {
1773         if (f6i->nh)
1774                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_flush_exceptions,
1775                                          f6i);
1776         else
1777                 fib6_nh_flush_exceptions(f6i->fib6_nh, f6i);
1778 }
1779
1780 /* Find cached rt in the hash table inside passed in rt
1781  * Caller has to hold rcu_read_lock()
1782  */
1783 static struct rt6_info *rt6_find_cached_rt(const struct fib6_result *res,
1784                                            const struct in6_addr *daddr,
1785                                            const struct in6_addr *saddr)
1786 {
1787         const struct in6_addr *src_key = NULL;
1788         struct rt6_exception_bucket *bucket;
1789         struct rt6_exception *rt6_ex;
1790         struct rt6_info *ret = NULL;
1791
1792 #ifdef CONFIG_IPV6_SUBTREES
1793         /* fib6i_src.plen != 0 indicates f6i is in subtree
1794          * and exception table is indexed by a hash of
1795          * both fib6_dst and fib6_src.
1796          * However, the src addr used to create the hash
1797          * might not be exactly the passed in saddr which
1798          * is a /128 addr from the flow.
1799          * So we need to use f6i->fib6_src to redo lookup
1800          * if the passed in saddr does not find anything.
1801          * (See the logic in ip6_rt_cache_alloc() on how
1802          * rt->rt6i_src is updated.)
1803          */
1804         if (res->f6i->fib6_src.plen)
1805                 src_key = saddr;
1806 find_ex:
1807 #endif
1808         bucket = fib6_nh_get_excptn_bucket(res->nh, NULL);
1809         rt6_ex = __rt6_find_exception_rcu(&bucket, daddr, src_key);
1810
1811         if (rt6_ex && !rt6_check_expired(rt6_ex->rt6i))
1812                 ret = rt6_ex->rt6i;
1813
1814 #ifdef CONFIG_IPV6_SUBTREES
1815         /* Use fib6_src as src_key and redo lookup */
1816         if (!ret && src_key && src_key != &res->f6i->fib6_src.addr) {
1817                 src_key = &res->f6i->fib6_src.addr;
1818                 goto find_ex;
1819         }
1820 #endif
1821
1822         return ret;
1823 }
1824
1825 /* Remove the passed in cached rt from the hash table that contains it */
1826 static int fib6_nh_remove_exception(const struct fib6_nh *nh, int plen,
1827                                     const struct rt6_info *rt)
1828 {
1829         const struct in6_addr *src_key = NULL;
1830         struct rt6_exception_bucket *bucket;
1831         struct rt6_exception *rt6_ex;
1832         int err;
1833
1834         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
1835                 return -ENOENT;
1836
1837         spin_lock_bh(&rt6_exception_lock);
1838         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
1839
1840 #ifdef CONFIG_IPV6_SUBTREES
1841         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1842          * and exception table is indexed by a hash of
1843          * both rt6i_dst and rt6i_src.
1844          * Otherwise, the exception table is indexed by
1845          * a hash of only rt6i_dst.
1846          */
1847         if (plen)
1848                 src_key = &rt->rt6i_src.addr;
1849 #endif
1850         rt6_ex = __rt6_find_exception_spinlock(&bucket,
1851                                                &rt->rt6i_dst.addr,
1852                                                src_key);
1853         if (rt6_ex) {
1854                 rt6_remove_exception(bucket, rt6_ex);
1855                 err = 0;
1856         } else {
1857                 err = -ENOENT;
1858         }
1859
1860         spin_unlock_bh(&rt6_exception_lock);
1861         return err;
1862 }
1863
1864 struct fib6_nh_excptn_arg {
1865         struct rt6_info *rt;
1866         int             plen;
1867 };
1868
1869 static int rt6_nh_remove_exception_rt(struct fib6_nh *nh, void *_arg)
1870 {
1871         struct fib6_nh_excptn_arg *arg = _arg;
1872         int err;
1873
1874         err = fib6_nh_remove_exception(nh, arg->plen, arg->rt);
1875         if (err == 0)
1876                 return 1;
1877
1878         return 0;
1879 }
1880
1881 static int rt6_remove_exception_rt(struct rt6_info *rt)
1882 {
1883         struct fib6_info *from;
1884
1885         from = rcu_dereference(rt->from);
1886         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1887                 return -EINVAL;
1888
1889         if (from->nh) {
1890                 struct fib6_nh_excptn_arg arg = {
1891                         .rt = rt,
1892                         .plen = from->fib6_src.plen
1893                 };
1894                 int rc;
1895
1896                 /* rc = 1 means an entry was found */
1897                 rc = nexthop_for_each_fib6_nh(from->nh,
1898                                               rt6_nh_remove_exception_rt,
1899                                               &arg);
1900                 return rc ? 0 : -ENOENT;
1901         }
1902
1903         return fib6_nh_remove_exception(from->fib6_nh,
1904                                         from->fib6_src.plen, rt);
1905 }
1906
1907 /* Find rt6_ex which contains the passed in rt cache and
1908  * refresh its stamp
1909  */
1910 static void fib6_nh_update_exception(const struct fib6_nh *nh, int plen,
1911                                      const struct rt6_info *rt)
1912 {
1913         const struct in6_addr *src_key = NULL;
1914         struct rt6_exception_bucket *bucket;
1915         struct rt6_exception *rt6_ex;
1916
1917         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
1918 #ifdef CONFIG_IPV6_SUBTREES
1919         /* rt6i_src.plen != 0 indicates 'from' is in subtree
1920          * and exception table is indexed by a hash of
1921          * both rt6i_dst and rt6i_src.
1922          * Otherwise, the exception table is indexed by
1923          * a hash of only rt6i_dst.
1924          */
1925         if (plen)
1926                 src_key = &rt->rt6i_src.addr;
1927 #endif
1928         rt6_ex = __rt6_find_exception_rcu(&bucket, &rt->rt6i_dst.addr, src_key);
1929         if (rt6_ex)
1930                 rt6_ex->stamp = jiffies;
1931 }
1932
1933 struct fib6_nh_match_arg {
1934         const struct net_device *dev;
1935         const struct in6_addr   *gw;
1936         struct fib6_nh          *match;
1937 };
1938
1939 /* determine if fib6_nh has given device and gateway */
1940 static int fib6_nh_find_match(struct fib6_nh *nh, void *_arg)
1941 {
1942         struct fib6_nh_match_arg *arg = _arg;
1943
1944         if (arg->dev != nh->fib_nh_dev ||
1945             (arg->gw && !nh->fib_nh_gw_family) ||
1946             (!arg->gw && nh->fib_nh_gw_family) ||
1947             (arg->gw && !ipv6_addr_equal(arg->gw, &nh->fib_nh_gw6)))
1948                 return 0;
1949
1950         arg->match = nh;
1951
1952         /* found a match, break the loop */
1953         return 1;
1954 }
1955
1956 static void rt6_update_exception_stamp_rt(struct rt6_info *rt)
1957 {
1958         struct fib6_info *from;
1959         struct fib6_nh *fib6_nh;
1960
1961         rcu_read_lock();
1962
1963         from = rcu_dereference(rt->from);
1964         if (!from || !(rt->rt6i_flags & RTF_CACHE))
1965                 goto unlock;
1966
1967         if (from->nh) {
1968                 struct fib6_nh_match_arg arg = {
1969                         .dev = rt->dst.dev,
1970                         .gw = &rt->rt6i_gateway,
1971                 };
1972
1973                 nexthop_for_each_fib6_nh(from->nh, fib6_nh_find_match, &arg);
1974
1975                 if (!arg.match)
1976                         goto unlock;
1977                 fib6_nh = arg.match;
1978         } else {
1979                 fib6_nh = from->fib6_nh;
1980         }
1981         fib6_nh_update_exception(fib6_nh, from->fib6_src.plen, rt);
1982 unlock:
1983         rcu_read_unlock();
1984 }
1985
1986 static bool rt6_mtu_change_route_allowed(struct inet6_dev *idev,
1987                                          struct rt6_info *rt, int mtu)
1988 {
1989         /* If the new MTU is lower than the route PMTU, this new MTU will be the
1990          * lowest MTU in the path: always allow updating the route PMTU to
1991          * reflect PMTU decreases.
1992          *
1993          * If the new MTU is higher, and the route PMTU is equal to the local
1994          * MTU, this means the old MTU is the lowest in the path, so allow
1995          * updating it: if other nodes now have lower MTUs, PMTU discovery will
1996          * handle this.
1997          */
1998
1999         if (dst_mtu(&rt->dst) >= mtu)
2000                 return true;
2001
2002         if (dst_mtu(&rt->dst) == idev->cnf.mtu6)
2003                 return true;
2004
2005         return false;
2006 }
2007
2008 static void rt6_exceptions_update_pmtu(struct inet6_dev *idev,
2009                                        const struct fib6_nh *nh, int mtu)
2010 {
2011         struct rt6_exception_bucket *bucket;
2012         struct rt6_exception *rt6_ex;
2013         int i;
2014
2015         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2016         if (!bucket)
2017                 return;
2018
2019         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2020                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
2021                         struct rt6_info *entry = rt6_ex->rt6i;
2022
2023                         /* For RTF_CACHE with rt6i_pmtu == 0 (i.e. a redirected
2024                          * route), the metrics of its rt->from have already
2025                          * been updated.
2026                          */
2027                         if (dst_metric_raw(&entry->dst, RTAX_MTU) &&
2028                             rt6_mtu_change_route_allowed(idev, entry, mtu))
2029                                 dst_metric_set(&entry->dst, RTAX_MTU, mtu);
2030                 }
2031                 bucket++;
2032         }
2033 }
2034
2035 #define RTF_CACHE_GATEWAY       (RTF_GATEWAY | RTF_CACHE)
2036
2037 static void fib6_nh_exceptions_clean_tohost(const struct fib6_nh *nh,
2038                                             const struct in6_addr *gateway)
2039 {
2040         struct rt6_exception_bucket *bucket;
2041         struct rt6_exception *rt6_ex;
2042         struct hlist_node *tmp;
2043         int i;
2044
2045         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2046                 return;
2047
2048         spin_lock_bh(&rt6_exception_lock);
2049         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2050         if (bucket) {
2051                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2052                         hlist_for_each_entry_safe(rt6_ex, tmp,
2053                                                   &bucket->chain, hlist) {
2054                                 struct rt6_info *entry = rt6_ex->rt6i;
2055
2056                                 if ((entry->rt6i_flags & RTF_CACHE_GATEWAY) ==
2057                                     RTF_CACHE_GATEWAY &&
2058                                     ipv6_addr_equal(gateway,
2059                                                     &entry->rt6i_gateway)) {
2060                                         rt6_remove_exception(bucket, rt6_ex);
2061                                 }
2062                         }
2063                         bucket++;
2064                 }
2065         }
2066
2067         spin_unlock_bh(&rt6_exception_lock);
2068 }
2069
2070 static void rt6_age_examine_exception(struct rt6_exception_bucket *bucket,
2071                                       struct rt6_exception *rt6_ex,
2072                                       struct fib6_gc_args *gc_args,
2073                                       unsigned long now)
2074 {
2075         struct rt6_info *rt = rt6_ex->rt6i;
2076
2077         /* we are pruning and obsoleting aged-out and non gateway exceptions
2078          * even if others have still references to them, so that on next
2079          * dst_check() such references can be dropped.
2080          * EXPIRES exceptions - e.g. pmtu-generated ones are pruned when
2081          * expired, independently from their aging, as per RFC 8201 section 4
2082          */
2083         if (!(rt->rt6i_flags & RTF_EXPIRES)) {
2084                 if (time_after_eq(now, rt->dst.lastuse + gc_args->timeout)) {
2085                         RT6_TRACE("aging clone %p\n", rt);
2086                         rt6_remove_exception(bucket, rt6_ex);
2087                         return;
2088                 }
2089         } else if (time_after(jiffies, rt->dst.expires)) {
2090                 RT6_TRACE("purging expired route %p\n", rt);
2091                 rt6_remove_exception(bucket, rt6_ex);
2092                 return;
2093         }
2094
2095         if (rt->rt6i_flags & RTF_GATEWAY) {
2096                 struct neighbour *neigh;
2097                 __u8 neigh_flags = 0;
2098
2099                 neigh = __ipv6_neigh_lookup_noref(rt->dst.dev, &rt->rt6i_gateway);
2100                 if (neigh)
2101                         neigh_flags = neigh->flags;
2102
2103                 if (!(neigh_flags & NTF_ROUTER)) {
2104                         RT6_TRACE("purging route %p via non-router but gateway\n",
2105                                   rt);
2106                         rt6_remove_exception(bucket, rt6_ex);
2107                         return;
2108                 }
2109         }
2110
2111         gc_args->more++;
2112 }
2113
2114 static void fib6_nh_age_exceptions(const struct fib6_nh *nh,
2115                                    struct fib6_gc_args *gc_args,
2116                                    unsigned long now)
2117 {
2118         struct rt6_exception_bucket *bucket;
2119         struct rt6_exception *rt6_ex;
2120         struct hlist_node *tmp;
2121         int i;
2122
2123         if (!rcu_access_pointer(nh->rt6i_exception_bucket))
2124                 return;
2125
2126         rcu_read_lock_bh();
2127         spin_lock(&rt6_exception_lock);
2128         bucket = fib6_nh_get_excptn_bucket(nh, &rt6_exception_lock);
2129         if (bucket) {
2130                 for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
2131                         hlist_for_each_entry_safe(rt6_ex, tmp,
2132                                                   &bucket->chain, hlist) {
2133                                 rt6_age_examine_exception(bucket, rt6_ex,
2134                                                           gc_args, now);
2135                         }
2136                         bucket++;
2137                 }
2138         }
2139         spin_unlock(&rt6_exception_lock);
2140         rcu_read_unlock_bh();
2141 }
2142
2143 struct fib6_nh_age_excptn_arg {
2144         struct fib6_gc_args     *gc_args;
2145         unsigned long           now;
2146 };
2147
2148 static int rt6_nh_age_exceptions(struct fib6_nh *nh, void *_arg)
2149 {
2150         struct fib6_nh_age_excptn_arg *arg = _arg;
2151
2152         fib6_nh_age_exceptions(nh, arg->gc_args, arg->now);
2153         return 0;
2154 }
2155
2156 void rt6_age_exceptions(struct fib6_info *f6i,
2157                         struct fib6_gc_args *gc_args,
2158                         unsigned long now)
2159 {
2160         if (f6i->nh) {
2161                 struct fib6_nh_age_excptn_arg arg = {
2162                         .gc_args = gc_args,
2163                         .now = now
2164                 };
2165
2166                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_age_exceptions,
2167                                          &arg);
2168         } else {
2169                 fib6_nh_age_exceptions(f6i->fib6_nh, gc_args, now);
2170         }
2171 }
2172
2173 /* must be called with rcu lock held */
2174 int fib6_table_lookup(struct net *net, struct fib6_table *table, int oif,
2175                       struct flowi6 *fl6, struct fib6_result *res, int strict)
2176 {
2177         struct fib6_node *fn, *saved_fn;
2178
2179         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2180         saved_fn = fn;
2181
2182         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2183                 oif = 0;
2184
2185 redo_rt6_select:
2186         rt6_select(net, fn, oif, res, strict);
2187         if (res->f6i == net->ipv6.fib6_null_entry) {
2188                 fn = fib6_backtrack(fn, &fl6->saddr);
2189                 if (fn)
2190                         goto redo_rt6_select;
2191                 else if (strict & RT6_LOOKUP_F_REACHABLE) {
2192                         /* also consider unreachable route */
2193                         strict &= ~RT6_LOOKUP_F_REACHABLE;
2194                         fn = saved_fn;
2195                         goto redo_rt6_select;
2196                 }
2197         }
2198
2199         trace_fib6_table_lookup(net, res, table, fl6);
2200
2201         return 0;
2202 }
2203
2204 struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table,
2205                                int oif, struct flowi6 *fl6,
2206                                const struct sk_buff *skb, int flags)
2207 {
2208         struct fib6_result res = {};
2209         struct rt6_info *rt = NULL;
2210         int strict = 0;
2211
2212         WARN_ON_ONCE((flags & RT6_LOOKUP_F_DST_NOREF) &&
2213                      !rcu_read_lock_held());
2214
2215         strict |= flags & RT6_LOOKUP_F_IFACE;
2216         strict |= flags & RT6_LOOKUP_F_IGNORE_LINKSTATE;
2217         if (net->ipv6.devconf_all->forwarding == 0)
2218                 strict |= RT6_LOOKUP_F_REACHABLE;
2219
2220         rcu_read_lock();
2221
2222         fib6_table_lookup(net, table, oif, fl6, &res, strict);
2223         if (res.f6i == net->ipv6.fib6_null_entry)
2224                 goto out;
2225
2226         fib6_select_path(net, &res, fl6, oif, false, skb, strict);
2227
2228         /*Search through exception table */
2229         rt = rt6_find_cached_rt(&res, &fl6->daddr, &fl6->saddr);
2230         if (rt) {
2231                 goto out;
2232         } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
2233                             !res.nh->fib_nh_gw_family)) {
2234                 /* Create a RTF_CACHE clone which will not be
2235                  * owned by the fib6 tree.  It is for the special case where
2236                  * the daddr in the skb during the neighbor look-up is different
2237                  * from the fl6->daddr used to look-up route here.
2238                  */
2239                 rt = ip6_rt_cache_alloc(&res, &fl6->daddr, NULL);
2240
2241                 if (rt) {
2242                         /* 1 refcnt is taken during ip6_rt_cache_alloc().
2243                          * As rt6_uncached_list_add() does not consume refcnt,
2244                          * this refcnt is always returned to the caller even
2245                          * if caller sets RT6_LOOKUP_F_DST_NOREF flag.
2246                          */
2247                         rt6_uncached_list_add(rt);
2248                         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
2249                         rcu_read_unlock();
2250
2251                         return rt;
2252                 }
2253         } else {
2254                 /* Get a percpu copy */
2255                 local_bh_disable();
2256                 rt = rt6_get_pcpu_route(&res);
2257
2258                 if (!rt)
2259                         rt = rt6_make_pcpu_route(net, &res);
2260
2261                 local_bh_enable();
2262         }
2263 out:
2264         if (!rt)
2265                 rt = net->ipv6.ip6_null_entry;
2266         if (!(flags & RT6_LOOKUP_F_DST_NOREF))
2267                 ip6_hold_safe(net, &rt);
2268         rcu_read_unlock();
2269
2270         return rt;
2271 }
2272 EXPORT_SYMBOL_GPL(ip6_pol_route);
2273
2274 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_input(struct net *net,
2275                                             struct fib6_table *table,
2276                                             struct flowi6 *fl6,
2277                                             const struct sk_buff *skb,
2278                                             int flags)
2279 {
2280         return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, skb, flags);
2281 }
2282
2283 struct dst_entry *ip6_route_input_lookup(struct net *net,
2284                                          struct net_device *dev,
2285                                          struct flowi6 *fl6,
2286                                          const struct sk_buff *skb,
2287                                          int flags)
2288 {
2289         if (rt6_need_strict(&fl6->daddr) && dev->type != ARPHRD_PIMREG)
2290                 flags |= RT6_LOOKUP_F_IFACE;
2291
2292         return fib6_rule_lookup(net, fl6, skb, flags, ip6_pol_route_input);
2293 }
2294 EXPORT_SYMBOL_GPL(ip6_route_input_lookup);
2295
2296 static void ip6_multipath_l3_keys(const struct sk_buff *skb,
2297                                   struct flow_keys *keys,
2298                                   struct flow_keys *flkeys)
2299 {
2300         const struct ipv6hdr *outer_iph = ipv6_hdr(skb);
2301         const struct ipv6hdr *key_iph = outer_iph;
2302         struct flow_keys *_flkeys = flkeys;
2303         const struct ipv6hdr *inner_iph;
2304         const struct icmp6hdr *icmph;
2305         struct ipv6hdr _inner_iph;
2306         struct icmp6hdr _icmph;
2307
2308         if (likely(outer_iph->nexthdr != IPPROTO_ICMPV6))
2309                 goto out;
2310
2311         icmph = skb_header_pointer(skb, skb_transport_offset(skb),
2312                                    sizeof(_icmph), &_icmph);
2313         if (!icmph)
2314                 goto out;
2315
2316         if (!icmpv6_is_err(icmph->icmp6_type))
2317                 goto out;
2318
2319         inner_iph = skb_header_pointer(skb,
2320                                        skb_transport_offset(skb) + sizeof(*icmph),
2321                                        sizeof(_inner_iph), &_inner_iph);
2322         if (!inner_iph)
2323                 goto out;
2324
2325         key_iph = inner_iph;
2326         _flkeys = NULL;
2327 out:
2328         if (_flkeys) {
2329                 keys->addrs.v6addrs.src = _flkeys->addrs.v6addrs.src;
2330                 keys->addrs.v6addrs.dst = _flkeys->addrs.v6addrs.dst;
2331                 keys->tags.flow_label = _flkeys->tags.flow_label;
2332                 keys->basic.ip_proto = _flkeys->basic.ip_proto;
2333         } else {
2334                 keys->addrs.v6addrs.src = key_iph->saddr;
2335                 keys->addrs.v6addrs.dst = key_iph->daddr;
2336                 keys->tags.flow_label = ip6_flowlabel(key_iph);
2337                 keys->basic.ip_proto = key_iph->nexthdr;
2338         }
2339 }
2340
2341 /* if skb is set it will be used and fl6 can be NULL */
2342 u32 rt6_multipath_hash(const struct net *net, const struct flowi6 *fl6,
2343                        const struct sk_buff *skb, struct flow_keys *flkeys)
2344 {
2345         struct flow_keys hash_keys;
2346         u32 mhash;
2347
2348         switch (ip6_multipath_hash_policy(net)) {
2349         case 0:
2350                 memset(&hash_keys, 0, sizeof(hash_keys));
2351                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2352                 if (skb) {
2353                         ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2354                 } else {
2355                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2356                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2357                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2358                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2359                 }
2360                 break;
2361         case 1:
2362                 if (skb) {
2363                         unsigned int flag = FLOW_DISSECTOR_F_STOP_AT_ENCAP;
2364                         struct flow_keys keys;
2365
2366                         /* short-circuit if we already have L4 hash present */
2367                         if (skb->l4_hash)
2368                                 return skb_get_hash_raw(skb) >> 1;
2369
2370                         memset(&hash_keys, 0, sizeof(hash_keys));
2371
2372                         if (!flkeys) {
2373                                 skb_flow_dissect_flow_keys(skb, &keys, flag);
2374                                 flkeys = &keys;
2375                         }
2376                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2377                         hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2378                         hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2379                         hash_keys.ports.src = flkeys->ports.src;
2380                         hash_keys.ports.dst = flkeys->ports.dst;
2381                         hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2382                 } else {
2383                         memset(&hash_keys, 0, sizeof(hash_keys));
2384                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2385                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2386                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2387                         hash_keys.ports.src = fl6->fl6_sport;
2388                         hash_keys.ports.dst = fl6->fl6_dport;
2389                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2390                 }
2391                 break;
2392         case 2:
2393                 memset(&hash_keys, 0, sizeof(hash_keys));
2394                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2395                 if (skb) {
2396                         struct flow_keys keys;
2397
2398                         if (!flkeys) {
2399                                 skb_flow_dissect_flow_keys(skb, &keys, 0);
2400                                 flkeys = &keys;
2401                         }
2402
2403                         /* Inner can be v4 or v6 */
2404                         if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
2405                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
2406                                 hash_keys.addrs.v4addrs.src = flkeys->addrs.v4addrs.src;
2407                                 hash_keys.addrs.v4addrs.dst = flkeys->addrs.v4addrs.dst;
2408                         } else if (flkeys->control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
2409                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2410                                 hash_keys.addrs.v6addrs.src = flkeys->addrs.v6addrs.src;
2411                                 hash_keys.addrs.v6addrs.dst = flkeys->addrs.v6addrs.dst;
2412                                 hash_keys.tags.flow_label = flkeys->tags.flow_label;
2413                                 hash_keys.basic.ip_proto = flkeys->basic.ip_proto;
2414                         } else {
2415                                 /* Same as case 0 */
2416                                 hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2417                                 ip6_multipath_l3_keys(skb, &hash_keys, flkeys);
2418                         }
2419                 } else {
2420                         /* Same as case 0 */
2421                         hash_keys.control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
2422                         hash_keys.addrs.v6addrs.src = fl6->saddr;
2423                         hash_keys.addrs.v6addrs.dst = fl6->daddr;
2424                         hash_keys.tags.flow_label = (__force u32)flowi6_get_flowlabel(fl6);
2425                         hash_keys.basic.ip_proto = fl6->flowi6_proto;
2426                 }
2427                 break;
2428         }
2429         mhash = flow_hash_from_keys(&hash_keys);
2430
2431         return mhash >> 1;
2432 }
2433
2434 /* Called with rcu held */
2435 void ip6_route_input(struct sk_buff *skb)
2436 {
2437         const struct ipv6hdr *iph = ipv6_hdr(skb);
2438         struct net *net = dev_net(skb->dev);
2439         int flags = RT6_LOOKUP_F_HAS_SADDR | RT6_LOOKUP_F_DST_NOREF;
2440         struct ip_tunnel_info *tun_info;
2441         struct flowi6 fl6 = {
2442                 .flowi6_iif = skb->dev->ifindex,
2443                 .daddr = iph->daddr,
2444                 .saddr = iph->saddr,
2445                 .flowlabel = ip6_flowinfo(iph),
2446                 .flowi6_mark = skb->mark,
2447                 .flowi6_proto = iph->nexthdr,
2448         };
2449         struct flow_keys *flkeys = NULL, _flkeys;
2450
2451         tun_info = skb_tunnel_info(skb);
2452         if (tun_info && !(tun_info->mode & IP_TUNNEL_INFO_TX))
2453                 fl6.flowi6_tun_key.tun_id = tun_info->key.tun_id;
2454
2455         if (fib6_rules_early_flow_dissect(net, skb, &fl6, &_flkeys))
2456                 flkeys = &_flkeys;
2457
2458         if (unlikely(fl6.flowi6_proto == IPPROTO_ICMPV6))
2459                 fl6.mp_hash = rt6_multipath_hash(net, &fl6, skb, flkeys);
2460         skb_dst_drop(skb);
2461         skb_dst_set_noref(skb, ip6_route_input_lookup(net, skb->dev,
2462                                                       &fl6, skb, flags));
2463 }
2464
2465 INDIRECT_CALLABLE_SCOPE struct rt6_info *ip6_pol_route_output(struct net *net,
2466                                              struct fib6_table *table,
2467                                              struct flowi6 *fl6,
2468                                              const struct sk_buff *skb,
2469                                              int flags)
2470 {
2471         return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, skb, flags);
2472 }
2473
2474 struct dst_entry *ip6_route_output_flags_noref(struct net *net,
2475                                                const struct sock *sk,
2476                                                struct flowi6 *fl6, int flags)
2477 {
2478         bool any_src;
2479
2480         if (ipv6_addr_type(&fl6->daddr) &
2481             (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL)) {
2482                 struct dst_entry *dst;
2483
2484                 /* This function does not take refcnt on the dst */
2485                 dst = l3mdev_link_scope_lookup(net, fl6);
2486                 if (dst)
2487                         return dst;
2488         }
2489
2490         fl6->flowi6_iif = LOOPBACK_IFINDEX;
2491
2492         flags |= RT6_LOOKUP_F_DST_NOREF;
2493         any_src = ipv6_addr_any(&fl6->saddr);
2494         if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr) ||
2495             (fl6->flowi6_oif && any_src))
2496                 flags |= RT6_LOOKUP_F_IFACE;
2497
2498         if (!any_src)
2499                 flags |= RT6_LOOKUP_F_HAS_SADDR;
2500         else if (sk)
2501                 flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs);
2502
2503         return fib6_rule_lookup(net, fl6, NULL, flags, ip6_pol_route_output);
2504 }
2505 EXPORT_SYMBOL_GPL(ip6_route_output_flags_noref);
2506
2507 struct dst_entry *ip6_route_output_flags(struct net *net,
2508                                          const struct sock *sk,
2509                                          struct flowi6 *fl6,
2510                                          int flags)
2511 {
2512         struct dst_entry *dst;
2513         struct rt6_info *rt6;
2514
2515         rcu_read_lock();
2516         dst = ip6_route_output_flags_noref(net, sk, fl6, flags);
2517         rt6 = (struct rt6_info *)dst;
2518         /* For dst cached in uncached_list, refcnt is already taken. */
2519         if (list_empty(&rt6->rt6i_uncached) && !dst_hold_safe(dst)) {
2520                 dst = &net->ipv6.ip6_null_entry->dst;
2521                 dst_hold(dst);
2522         }
2523         rcu_read_unlock();
2524
2525         return dst;
2526 }
2527 EXPORT_SYMBOL_GPL(ip6_route_output_flags);
2528
2529 struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig)
2530 {
2531         struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig;
2532         struct net_device *loopback_dev = net->loopback_dev;
2533         struct dst_entry *new = NULL;
2534
2535         rt = dst_alloc(&ip6_dst_blackhole_ops, loopback_dev, 1,
2536                        DST_OBSOLETE_DEAD, 0);
2537         if (rt) {
2538                 rt6_info_init(rt);
2539                 atomic_inc(&net->ipv6.rt6_stats->fib_rt_alloc);
2540
2541                 new = &rt->dst;
2542                 new->__use = 1;
2543                 new->input = dst_discard;
2544                 new->output = dst_discard_out;
2545
2546                 dst_copy_metrics(new, &ort->dst);
2547
2548                 rt->rt6i_idev = in6_dev_get(loopback_dev);
2549                 rt->rt6i_gateway = ort->rt6i_gateway;
2550                 rt->rt6i_flags = ort->rt6i_flags & ~RTF_PCPU;
2551
2552                 memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key));
2553 #ifdef CONFIG_IPV6_SUBTREES
2554                 memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key));
2555 #endif
2556         }
2557
2558         dst_release(dst_orig);
2559         return new ? new : ERR_PTR(-ENOMEM);
2560 }
2561
2562 /*
2563  *      Destination cache support functions
2564  */
2565
2566 static bool fib6_check(struct fib6_info *f6i, u32 cookie)
2567 {
2568         u32 rt_cookie = 0;
2569
2570         if (!fib6_get_cookie_safe(f6i, &rt_cookie) || rt_cookie != cookie)
2571                 return false;
2572
2573         if (fib6_check_expired(f6i))
2574                 return false;
2575
2576         return true;
2577 }
2578
2579 static struct dst_entry *rt6_check(struct rt6_info *rt,
2580                                    struct fib6_info *from,
2581                                    u32 cookie)
2582 {
2583         u32 rt_cookie = 0;
2584
2585         if (!from || !fib6_get_cookie_safe(from, &rt_cookie) ||
2586             rt_cookie != cookie)
2587                 return NULL;
2588
2589         if (rt6_check_expired(rt))
2590                 return NULL;
2591
2592         return &rt->dst;
2593 }
2594
2595 static struct dst_entry *rt6_dst_from_check(struct rt6_info *rt,
2596                                             struct fib6_info *from,
2597                                             u32 cookie)
2598 {
2599         if (!__rt6_check_expired(rt) &&
2600             rt->dst.obsolete == DST_OBSOLETE_FORCE_CHK &&
2601             fib6_check(from, cookie))
2602                 return &rt->dst;
2603         else
2604                 return NULL;
2605 }
2606
2607 static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
2608 {
2609         struct dst_entry *dst_ret;
2610         struct fib6_info *from;
2611         struct rt6_info *rt;
2612
2613         rt = container_of(dst, struct rt6_info, dst);
2614
2615         if (rt->sernum)
2616                 return rt6_is_valid(rt) ? dst : NULL;
2617
2618         rcu_read_lock();
2619
2620         /* All IPV6 dsts are created with ->obsolete set to the value
2621          * DST_OBSOLETE_FORCE_CHK which forces validation calls down
2622          * into this function always.
2623          */
2624
2625         from = rcu_dereference(rt->from);
2626
2627         if (from && (rt->rt6i_flags & RTF_PCPU ||
2628             unlikely(!list_empty(&rt->rt6i_uncached))))
2629                 dst_ret = rt6_dst_from_check(rt, from, cookie);
2630         else
2631                 dst_ret = rt6_check(rt, from, cookie);
2632
2633         rcu_read_unlock();
2634
2635         return dst_ret;
2636 }
2637
2638 static struct dst_entry *ip6_negative_advice(struct dst_entry *dst)
2639 {
2640         struct rt6_info *rt = (struct rt6_info *) dst;
2641
2642         if (rt) {
2643                 if (rt->rt6i_flags & RTF_CACHE) {
2644                         rcu_read_lock();
2645                         if (rt6_check_expired(rt)) {
2646                                 rt6_remove_exception_rt(rt);
2647                                 dst = NULL;
2648                         }
2649                         rcu_read_unlock();
2650                 } else {
2651                         dst_release(dst);
2652                         dst = NULL;
2653                 }
2654         }
2655         return dst;
2656 }
2657
2658 static void ip6_link_failure(struct sk_buff *skb)
2659 {
2660         struct rt6_info *rt;
2661
2662         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0);
2663
2664         rt = (struct rt6_info *) skb_dst(skb);
2665         if (rt) {
2666                 rcu_read_lock();
2667                 if (rt->rt6i_flags & RTF_CACHE) {
2668                         rt6_remove_exception_rt(rt);
2669                 } else {
2670                         struct fib6_info *from;
2671                         struct fib6_node *fn;
2672
2673                         from = rcu_dereference(rt->from);
2674                         if (from) {
2675                                 fn = rcu_dereference(from->fib6_node);
2676                                 if (fn && (rt->rt6i_flags & RTF_DEFAULT))
2677                                         WRITE_ONCE(fn->fn_sernum, -1);
2678                         }
2679                 }
2680                 rcu_read_unlock();
2681         }
2682 }
2683
2684 static void rt6_update_expires(struct rt6_info *rt0, int timeout)
2685 {
2686         if (!(rt0->rt6i_flags & RTF_EXPIRES)) {
2687                 struct fib6_info *from;
2688
2689                 rcu_read_lock();
2690                 from = rcu_dereference(rt0->from);
2691                 if (from)
2692                         rt0->dst.expires = from->expires;
2693                 rcu_read_unlock();
2694         }
2695
2696         dst_set_expires(&rt0->dst, timeout);
2697         rt0->rt6i_flags |= RTF_EXPIRES;
2698 }
2699
2700 static void rt6_do_update_pmtu(struct rt6_info *rt, u32 mtu)
2701 {
2702         struct net *net = dev_net(rt->dst.dev);
2703
2704         dst_metric_set(&rt->dst, RTAX_MTU, mtu);
2705         rt->rt6i_flags |= RTF_MODIFIED;
2706         rt6_update_expires(rt, net->ipv6.sysctl.ip6_rt_mtu_expires);
2707 }
2708
2709 static bool rt6_cache_allowed_for_pmtu(const struct rt6_info *rt)
2710 {
2711         return !(rt->rt6i_flags & RTF_CACHE) &&
2712                 (rt->rt6i_flags & RTF_PCPU || rcu_access_pointer(rt->from));
2713 }
2714
2715 static void __ip6_rt_update_pmtu(struct dst_entry *dst, const struct sock *sk,
2716                                  const struct ipv6hdr *iph, u32 mtu,
2717                                  bool confirm_neigh)
2718 {
2719         const struct in6_addr *daddr, *saddr;
2720         struct rt6_info *rt6 = (struct rt6_info *)dst;
2721
2722         /* Note: do *NOT* check dst_metric_locked(dst, RTAX_MTU)
2723          * IPv6 pmtu discovery isn't optional, so 'mtu lock' cannot disable it.
2724          * [see also comment in rt6_mtu_change_route()]
2725          */
2726
2727         if (iph) {
2728                 daddr = &iph->daddr;
2729                 saddr = &iph->saddr;
2730         } else if (sk) {
2731                 daddr = &sk->sk_v6_daddr;
2732                 saddr = &inet6_sk(sk)->saddr;
2733         } else {
2734                 daddr = NULL;
2735                 saddr = NULL;
2736         }
2737
2738         if (confirm_neigh)
2739                 dst_confirm_neigh(dst, daddr);
2740
2741         if (mtu < IPV6_MIN_MTU)
2742                 return;
2743         if (mtu >= dst_mtu(dst))
2744                 return;
2745
2746         if (!rt6_cache_allowed_for_pmtu(rt6)) {
2747                 rt6_do_update_pmtu(rt6, mtu);
2748                 /* update rt6_ex->stamp for cache */
2749                 if (rt6->rt6i_flags & RTF_CACHE)
2750                         rt6_update_exception_stamp_rt(rt6);
2751         } else if (daddr) {
2752                 struct fib6_result res = {};
2753                 struct rt6_info *nrt6;
2754
2755                 rcu_read_lock();
2756                 res.f6i = rcu_dereference(rt6->from);
2757                 if (!res.f6i)
2758                         goto out_unlock;
2759
2760                 res.fib6_flags = res.f6i->fib6_flags;
2761                 res.fib6_type = res.f6i->fib6_type;
2762
2763                 if (res.f6i->nh) {
2764                         struct fib6_nh_match_arg arg = {
2765                                 .dev = dst->dev,
2766                                 .gw = &rt6->rt6i_gateway,
2767                         };
2768
2769                         nexthop_for_each_fib6_nh(res.f6i->nh,
2770                                                  fib6_nh_find_match, &arg);
2771
2772                         /* fib6_info uses a nexthop that does not have fib6_nh
2773                          * using the dst->dev + gw. Should be impossible.
2774                          */
2775                         if (!arg.match)
2776                                 goto out_unlock;
2777
2778                         res.nh = arg.match;
2779                 } else {
2780                         res.nh = res.f6i->fib6_nh;
2781                 }
2782
2783                 nrt6 = ip6_rt_cache_alloc(&res, daddr, saddr);
2784                 if (nrt6) {
2785                         rt6_do_update_pmtu(nrt6, mtu);
2786                         if (rt6_insert_exception(nrt6, &res))
2787                                 dst_release_immediate(&nrt6->dst);
2788                 }
2789 out_unlock:
2790                 rcu_read_unlock();
2791         }
2792 }
2793
2794 static void ip6_rt_update_pmtu(struct dst_entry *dst, struct sock *sk,
2795                                struct sk_buff *skb, u32 mtu,
2796                                bool confirm_neigh)
2797 {
2798         __ip6_rt_update_pmtu(dst, sk, skb ? ipv6_hdr(skb) : NULL, mtu,
2799                              confirm_neigh);
2800 }
2801
2802 void ip6_update_pmtu(struct sk_buff *skb, struct net *net, __be32 mtu,
2803                      int oif, u32 mark, kuid_t uid)
2804 {
2805         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
2806         struct dst_entry *dst;
2807         struct flowi6 fl6 = {
2808                 .flowi6_oif = oif,
2809                 .flowi6_mark = mark ? mark : IP6_REPLY_MARK(net, skb->mark),
2810                 .daddr = iph->daddr,
2811                 .saddr = iph->saddr,
2812                 .flowlabel = ip6_flowinfo(iph),
2813                 .flowi6_uid = uid,
2814         };
2815
2816         dst = ip6_route_output(net, NULL, &fl6);
2817         if (!dst->error)
2818                 __ip6_rt_update_pmtu(dst, NULL, iph, ntohl(mtu), true);
2819         dst_release(dst);
2820 }
2821 EXPORT_SYMBOL_GPL(ip6_update_pmtu);
2822
2823 void ip6_sk_update_pmtu(struct sk_buff *skb, struct sock *sk, __be32 mtu)
2824 {
2825         int oif = sk->sk_bound_dev_if;
2826         struct dst_entry *dst;
2827
2828         if (!oif && skb->dev)
2829                 oif = l3mdev_master_ifindex(skb->dev);
2830
2831         ip6_update_pmtu(skb, sock_net(sk), mtu, oif, sk->sk_mark, sk->sk_uid);
2832
2833         dst = __sk_dst_get(sk);
2834         if (!dst || !dst->obsolete ||
2835             dst->ops->check(dst, inet6_sk(sk)->dst_cookie))
2836                 return;
2837
2838         bh_lock_sock(sk);
2839         if (!sock_owned_by_user(sk) && !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
2840                 ip6_datagram_dst_update(sk, false);
2841         bh_unlock_sock(sk);
2842 }
2843 EXPORT_SYMBOL_GPL(ip6_sk_update_pmtu);
2844
2845 void ip6_sk_dst_store_flow(struct sock *sk, struct dst_entry *dst,
2846                            const struct flowi6 *fl6)
2847 {
2848 #ifdef CONFIG_IPV6_SUBTREES
2849         struct ipv6_pinfo *np = inet6_sk(sk);
2850 #endif
2851
2852         ip6_dst_store(sk, dst,
2853                       ipv6_addr_equal(&fl6->daddr, &sk->sk_v6_daddr) ?
2854                       &sk->sk_v6_daddr : NULL,
2855 #ifdef CONFIG_IPV6_SUBTREES
2856                       ipv6_addr_equal(&fl6->saddr, &np->saddr) ?
2857                       &np->saddr :
2858 #endif
2859                       NULL);
2860 }
2861
2862 static bool ip6_redirect_nh_match(const struct fib6_result *res,
2863                                   struct flowi6 *fl6,
2864                                   const struct in6_addr *gw,
2865                                   struct rt6_info **ret)
2866 {
2867         const struct fib6_nh *nh = res->nh;
2868
2869         if (nh->fib_nh_flags & RTNH_F_DEAD || !nh->fib_nh_gw_family ||
2870             fl6->flowi6_oif != nh->fib_nh_dev->ifindex)
2871                 return false;
2872
2873         /* rt_cache's gateway might be different from its 'parent'
2874          * in the case of an ip redirect.
2875          * So we keep searching in the exception table if the gateway
2876          * is different.
2877          */
2878         if (!ipv6_addr_equal(gw, &nh->fib_nh_gw6)) {
2879                 struct rt6_info *rt_cache;
2880
2881                 rt_cache = rt6_find_cached_rt(res, &fl6->daddr, &fl6->saddr);
2882                 if (rt_cache &&
2883                     ipv6_addr_equal(gw, &rt_cache->rt6i_gateway)) {
2884                         *ret = rt_cache;
2885                         return true;
2886                 }
2887                 return false;
2888         }
2889         return true;
2890 }
2891
2892 struct fib6_nh_rd_arg {
2893         struct fib6_result      *res;
2894         struct flowi6           *fl6;
2895         const struct in6_addr   *gw;
2896         struct rt6_info         **ret;
2897 };
2898
2899 static int fib6_nh_redirect_match(struct fib6_nh *nh, void *_arg)
2900 {
2901         struct fib6_nh_rd_arg *arg = _arg;
2902
2903         arg->res->nh = nh;
2904         return ip6_redirect_nh_match(arg->res, arg->fl6, arg->gw, arg->ret);
2905 }
2906
2907 /* Handle redirects */
2908 struct ip6rd_flowi {
2909         struct flowi6 fl6;
2910         struct in6_addr gateway;
2911 };
2912
2913 INDIRECT_CALLABLE_SCOPE struct rt6_info *__ip6_route_redirect(struct net *net,
2914                                              struct fib6_table *table,
2915                                              struct flowi6 *fl6,
2916                                              const struct sk_buff *skb,
2917                                              int flags)
2918 {
2919         struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6;
2920         struct rt6_info *ret = NULL;
2921         struct fib6_result res = {};
2922         struct fib6_nh_rd_arg arg = {
2923                 .res = &res,
2924                 .fl6 = fl6,
2925                 .gw  = &rdfl->gateway,
2926                 .ret = &ret
2927         };
2928         struct fib6_info *rt;
2929         struct fib6_node *fn;
2930
2931         /* l3mdev_update_flow overrides oif if the device is enslaved; in
2932          * this case we must match on the real ingress device, so reset it
2933          */
2934         if (fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF)
2935                 fl6->flowi6_oif = skb->dev->ifindex;
2936
2937         /* Get the "current" route for this destination and
2938          * check if the redirect has come from appropriate router.
2939          *
2940          * RFC 4861 specifies that redirects should only be
2941          * accepted if they come from the nexthop to the target.
2942          * Due to the way the routes are chosen, this notion
2943          * is a bit fuzzy and one might need to check all possible
2944          * routes.
2945          */
2946
2947         rcu_read_lock();
2948         fn = fib6_node_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
2949 restart:
2950         for_each_fib6_node_rt_rcu(fn) {
2951                 res.f6i = rt;
2952                 if (fib6_check_expired(rt))
2953                         continue;
2954                 if (rt->fib6_flags & RTF_REJECT)
2955                         break;
2956                 if (unlikely(rt->nh)) {
2957                         if (nexthop_is_blackhole(rt->nh))
2958                                 continue;
2959                         /* on match, res->nh is filled in and potentially ret */
2960                         if (nexthop_for_each_fib6_nh(rt->nh,
2961                                                      fib6_nh_redirect_match,
2962                                                      &arg))
2963                                 goto out;
2964                 } else {
2965                         res.nh = rt->fib6_nh;
2966                         if (ip6_redirect_nh_match(&res, fl6, &rdfl->gateway,
2967                                                   &ret))
2968                                 goto out;
2969                 }
2970         }
2971
2972         if (!rt)
2973                 rt = net->ipv6.fib6_null_entry;
2974         else if (rt->fib6_flags & RTF_REJECT) {
2975                 ret = net->ipv6.ip6_null_entry;
2976                 goto out;
2977         }
2978
2979         if (rt == net->ipv6.fib6_null_entry) {
2980                 fn = fib6_backtrack(fn, &fl6->saddr);
2981                 if (fn)
2982                         goto restart;
2983         }
2984
2985         res.f6i = rt;
2986         res.nh = rt->fib6_nh;
2987 out:
2988         if (ret) {
2989                 ip6_hold_safe(net, &ret);
2990         } else {
2991                 res.fib6_flags = res.f6i->fib6_flags;
2992                 res.fib6_type = res.f6i->fib6_type;
2993                 ret = ip6_create_rt_rcu(&res);
2994         }
2995
2996         rcu_read_unlock();
2997
2998         trace_fib6_table_lookup(net, &res, table, fl6);
2999         return ret;
3000 };
3001
3002 static struct dst_entry *ip6_route_redirect(struct net *net,
3003                                             const struct flowi6 *fl6,
3004                                             const struct sk_buff *skb,
3005                                             const struct in6_addr *gateway)
3006 {
3007         int flags = RT6_LOOKUP_F_HAS_SADDR;
3008         struct ip6rd_flowi rdfl;
3009
3010         rdfl.fl6 = *fl6;
3011         rdfl.gateway = *gateway;
3012
3013         return fib6_rule_lookup(net, &rdfl.fl6, skb,
3014                                 flags, __ip6_route_redirect);
3015 }
3016
3017 void ip6_redirect(struct sk_buff *skb, struct net *net, int oif, u32 mark,
3018                   kuid_t uid)
3019 {
3020         const struct ipv6hdr *iph = (struct ipv6hdr *) skb->data;
3021         struct dst_entry *dst;
3022         struct flowi6 fl6 = {
3023                 .flowi6_iif = LOOPBACK_IFINDEX,
3024                 .flowi6_oif = oif,
3025                 .flowi6_mark = mark,
3026                 .daddr = iph->daddr,
3027                 .saddr = iph->saddr,
3028                 .flowlabel = ip6_flowinfo(iph),
3029                 .flowi6_uid = uid,
3030         };
3031
3032         dst = ip6_route_redirect(net, &fl6, skb, &ipv6_hdr(skb)->saddr);
3033         rt6_do_redirect(dst, NULL, skb);
3034         dst_release(dst);
3035 }
3036 EXPORT_SYMBOL_GPL(ip6_redirect);
3037
3038 void ip6_redirect_no_header(struct sk_buff *skb, struct net *net, int oif)
3039 {
3040         const struct ipv6hdr *iph = ipv6_hdr(skb);
3041         const struct rd_msg *msg = (struct rd_msg *)icmp6_hdr(skb);
3042         struct dst_entry *dst;
3043         struct flowi6 fl6 = {
3044                 .flowi6_iif = LOOPBACK_IFINDEX,
3045                 .flowi6_oif = oif,
3046                 .daddr = msg->dest,
3047                 .saddr = iph->daddr,
3048                 .flowi6_uid = sock_net_uid(net, NULL),
3049         };
3050
3051         dst = ip6_route_redirect(net, &fl6, skb, &iph->saddr);
3052         rt6_do_redirect(dst, NULL, skb);
3053         dst_release(dst);
3054 }
3055
3056 void ip6_sk_redirect(struct sk_buff *skb, struct sock *sk)
3057 {
3058         ip6_redirect(skb, sock_net(sk), sk->sk_bound_dev_if, sk->sk_mark,
3059                      sk->sk_uid);
3060 }
3061 EXPORT_SYMBOL_GPL(ip6_sk_redirect);
3062
3063 static unsigned int ip6_default_advmss(const struct dst_entry *dst)
3064 {
3065         struct net_device *dev = dst->dev;
3066         unsigned int mtu = dst_mtu(dst);
3067         struct net *net = dev_net(dev);
3068
3069         mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
3070
3071         if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss)
3072                 mtu = net->ipv6.sysctl.ip6_rt_min_advmss;
3073
3074         /*
3075          * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and
3076          * corresponding MSS is IPV6_MAXPLEN - tcp_header_size.
3077          * IPV6_MAXPLEN is also valid and means: "any MSS,
3078          * rely only on pmtu discovery"
3079          */
3080         if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr))
3081                 mtu = IPV6_MAXPLEN;
3082         return mtu;
3083 }
3084
3085 static unsigned int ip6_mtu(const struct dst_entry *dst)
3086 {
3087         struct inet6_dev *idev;
3088         unsigned int mtu;
3089
3090         mtu = dst_metric_raw(dst, RTAX_MTU);
3091         if (mtu)
3092                 goto out;
3093
3094         mtu = IPV6_MIN_MTU;
3095
3096         rcu_read_lock();
3097         idev = __in6_dev_get(dst->dev);
3098         if (idev)
3099                 mtu = idev->cnf.mtu6;
3100         rcu_read_unlock();
3101
3102 out:
3103         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3104
3105         return mtu - lwtunnel_headroom(dst->lwtstate, mtu);
3106 }
3107
3108 /* MTU selection:
3109  * 1. mtu on route is locked - use it
3110  * 2. mtu from nexthop exception
3111  * 3. mtu from egress device
3112  *
3113  * based on ip6_dst_mtu_forward and exception logic of
3114  * rt6_find_cached_rt; called with rcu_read_lock
3115  */
3116 u32 ip6_mtu_from_fib6(const struct fib6_result *res,
3117                       const struct in6_addr *daddr,
3118                       const struct in6_addr *saddr)
3119 {
3120         const struct fib6_nh *nh = res->nh;
3121         struct fib6_info *f6i = res->f6i;
3122         struct inet6_dev *idev;
3123         struct rt6_info *rt;
3124         u32 mtu = 0;
3125
3126         if (unlikely(fib6_metric_locked(f6i, RTAX_MTU))) {
3127                 mtu = f6i->fib6_pmtu;
3128                 if (mtu)
3129                         goto out;
3130         }
3131
3132         rt = rt6_find_cached_rt(res, daddr, saddr);
3133         if (unlikely(rt)) {
3134                 mtu = dst_metric_raw(&rt->dst, RTAX_MTU);
3135         } else {
3136                 struct net_device *dev = nh->fib_nh_dev;
3137
3138                 mtu = IPV6_MIN_MTU;
3139                 idev = __in6_dev_get(dev);
3140                 if (idev && idev->cnf.mtu6 > mtu)
3141                         mtu = idev->cnf.mtu6;
3142         }
3143
3144         mtu = min_t(unsigned int, mtu, IP6_MAX_MTU);
3145 out:
3146         return mtu - lwtunnel_headroom(nh->fib_nh_lws, mtu);
3147 }
3148
3149 struct dst_entry *icmp6_dst_alloc(struct net_device *dev,
3150                                   struct flowi6 *fl6)
3151 {
3152         struct dst_entry *dst;
3153         struct rt6_info *rt;
3154         struct inet6_dev *idev = in6_dev_get(dev);
3155         struct net *net = dev_net(dev);
3156
3157         if (unlikely(!idev))
3158                 return ERR_PTR(-ENODEV);
3159
3160         rt = ip6_dst_alloc(net, dev, 0);
3161         if (unlikely(!rt)) {
3162                 in6_dev_put(idev);
3163                 dst = ERR_PTR(-ENOMEM);
3164                 goto out;
3165         }
3166
3167         rt->dst.input = ip6_input;
3168         rt->dst.output  = ip6_output;
3169         rt->rt6i_gateway  = fl6->daddr;
3170         rt->rt6i_dst.addr = fl6->daddr;
3171         rt->rt6i_dst.plen = 128;
3172         rt->rt6i_idev     = idev;
3173         dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 0);
3174
3175         /* Add this dst into uncached_list so that rt6_disable_ip() can
3176          * do proper release of the net_device
3177          */
3178         rt6_uncached_list_add(rt);
3179         atomic_inc(&net->ipv6.rt6_stats->fib_rt_uncache);
3180
3181         dst = xfrm_lookup(net, &rt->dst, flowi6_to_flowi(fl6), NULL, 0);
3182
3183 out:
3184         return dst;
3185 }
3186
3187 static int ip6_dst_gc(struct dst_ops *ops)
3188 {
3189         struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
3190         int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
3191         int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;
3192         int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
3193         int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
3194         unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;
3195         unsigned int val;
3196         int entries;
3197
3198         entries = dst_entries_get_fast(ops);
3199         if (entries > rt_max_size)
3200                 entries = dst_entries_get_slow(ops);
3201
3202         if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
3203             entries <= rt_max_size)
3204                 goto out;
3205
3206         fib6_run_gc(atomic_inc_return(&net->ipv6.ip6_rt_gc_expire), net, true);
3207         entries = dst_entries_get_slow(ops);
3208         if (entries < ops->gc_thresh)
3209                 atomic_set(&net->ipv6.ip6_rt_gc_expire, rt_gc_timeout >> 1);
3210 out:
3211         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
3212         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));
3213         return entries > rt_max_size;
3214 }
3215
3216 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,
3217                                const struct in6_addr *gw_addr, u32 tbid,
3218                                int flags, struct fib6_result *res)
3219 {
3220         struct flowi6 fl6 = {
3221                 .flowi6_oif = cfg->fc_ifindex,
3222                 .daddr = *gw_addr,
3223                 .saddr = cfg->fc_prefsrc,
3224         };
3225         struct fib6_table *table;
3226         int err;
3227
3228         table = fib6_get_table(net, tbid);
3229         if (!table)
3230                 return -EINVAL;
3231
3232         if (!ipv6_addr_any(&cfg->fc_prefsrc))
3233                 flags |= RT6_LOOKUP_F_HAS_SADDR;
3234
3235         flags |= RT6_LOOKUP_F_IGNORE_LINKSTATE;
3236
3237         err = fib6_table_lookup(net, table, cfg->fc_ifindex, &fl6, res, flags);
3238         if (!err && res->f6i != net->ipv6.fib6_null_entry)
3239                 fib6_select_path(net, res, &fl6, cfg->fc_ifindex,
3240                                  cfg->fc_ifindex != 0, NULL, flags);
3241
3242         return err;
3243 }
3244
3245 static int ip6_route_check_nh_onlink(struct net *net,
3246                                      struct fib6_config *cfg,
3247                                      const struct net_device *dev,
3248                                      struct netlink_ext_ack *extack)
3249 {
3250         u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
3251         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3252         struct fib6_result res = {};
3253         int err;
3254
3255         err = ip6_nh_lookup_table(net, cfg, gw_addr, tbid, 0, &res);
3256         if (!err && !(res.fib6_flags & RTF_REJECT) &&
3257             /* ignore match if it is the default route */
3258             !ipv6_addr_any(&res.f6i->fib6_dst.addr) &&
3259             (res.fib6_type != RTN_UNICAST || dev != res.nh->fib_nh_dev)) {
3260                 NL_SET_ERR_MSG(extack,
3261                                "Nexthop has invalid gateway or device mismatch");
3262                 err = -EINVAL;
3263         }
3264
3265         return err;
3266 }
3267
3268 static int ip6_route_check_nh(struct net *net,
3269                               struct fib6_config *cfg,
3270                               struct net_device **_dev,
3271                               struct inet6_dev **idev)
3272 {
3273         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3274         struct net_device *dev = _dev ? *_dev : NULL;
3275         int flags = RT6_LOOKUP_F_IFACE;
3276         struct fib6_result res = {};
3277         int err = -EHOSTUNREACH;
3278
3279         if (cfg->fc_table) {
3280                 err = ip6_nh_lookup_table(net, cfg, gw_addr,
3281                                           cfg->fc_table, flags, &res);
3282                 /* gw_addr can not require a gateway or resolve to a reject
3283                  * route. If a device is given, it must match the result.
3284                  */
3285                 if (err || res.fib6_flags & RTF_REJECT ||
3286                     res.nh->fib_nh_gw_family ||
3287                     (dev && dev != res.nh->fib_nh_dev))
3288                         err = -EHOSTUNREACH;
3289         }
3290
3291         if (err < 0) {
3292                 struct flowi6 fl6 = {
3293                         .flowi6_oif = cfg->fc_ifindex,
3294                         .daddr = *gw_addr,
3295                 };
3296
3297                 err = fib6_lookup(net, cfg->fc_ifindex, &fl6, &res, flags);
3298                 if (err || res.fib6_flags & RTF_REJECT ||
3299                     res.nh->fib_nh_gw_family)
3300                         err = -EHOSTUNREACH;
3301
3302                 if (err)
3303                         return err;
3304
3305                 fib6_select_path(net, &res, &fl6, cfg->fc_ifindex,
3306                                  cfg->fc_ifindex != 0, NULL, flags);
3307         }
3308
3309         err = 0;
3310         if (dev) {
3311                 if (dev != res.nh->fib_nh_dev)
3312                         err = -EHOSTUNREACH;
3313         } else {
3314                 *_dev = dev = res.nh->fib_nh_dev;
3315                 dev_hold(dev);
3316                 *idev = in6_dev_get(dev);
3317         }
3318
3319         return err;
3320 }
3321
3322 static int ip6_validate_gw(struct net *net, struct fib6_config *cfg,
3323                            struct net_device **_dev, struct inet6_dev **idev,
3324                            struct netlink_ext_ack *extack)
3325 {
3326         const struct in6_addr *gw_addr = &cfg->fc_gateway;
3327         int gwa_type = ipv6_addr_type(gw_addr);
3328         bool skip_dev = gwa_type & IPV6_ADDR_LINKLOCAL ? false : true;
3329         const struct net_device *dev = *_dev;
3330         bool need_addr_check = !dev;
3331         int err = -EINVAL;
3332
3333         /* if gw_addr is local we will fail to detect this in case
3334          * address is still TENTATIVE (DAD in progress). rt6_lookup()
3335          * will return already-added prefix route via interface that
3336          * prefix route was assigned to, which might be non-loopback.
3337          */
3338         if (dev &&
3339             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3340                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3341                 goto out;
3342         }
3343
3344         if (gwa_type != (IPV6_ADDR_LINKLOCAL | IPV6_ADDR_UNICAST)) {
3345                 /* IPv6 strictly inhibits using not link-local
3346                  * addresses as nexthop address.
3347                  * Otherwise, router will not able to send redirects.
3348                  * It is very good, but in some (rare!) circumstances
3349                  * (SIT, PtP, NBMA NOARP links) it is handy to allow
3350                  * some exceptions. --ANK
3351                  * We allow IPv4-mapped nexthops to support RFC4798-type
3352                  * addressing
3353                  */
3354                 if (!(gwa_type & (IPV6_ADDR_UNICAST | IPV6_ADDR_MAPPED))) {
3355                         NL_SET_ERR_MSG(extack, "Invalid gateway address");
3356                         goto out;
3357                 }
3358
3359                 rcu_read_lock();
3360
3361                 if (cfg->fc_flags & RTNH_F_ONLINK)
3362                         err = ip6_route_check_nh_onlink(net, cfg, dev, extack);
3363                 else
3364                         err = ip6_route_check_nh(net, cfg, _dev, idev);
3365
3366                 rcu_read_unlock();
3367
3368                 if (err)
3369                         goto out;
3370         }
3371
3372         /* reload in case device was changed */
3373         dev = *_dev;
3374
3375         err = -EINVAL;
3376         if (!dev) {
3377                 NL_SET_ERR_MSG(extack, "Egress device not specified");
3378                 goto out;
3379         } else if (dev->flags & IFF_LOOPBACK) {
3380                 NL_SET_ERR_MSG(extack,
3381                                "Egress device can not be loopback device for this route");
3382                 goto out;
3383         }
3384
3385         /* if we did not check gw_addr above, do so now that the
3386          * egress device has been resolved.
3387          */
3388         if (need_addr_check &&
3389             ipv6_chk_addr_and_flags(net, gw_addr, dev, skip_dev, 0, 0)) {
3390                 NL_SET_ERR_MSG(extack, "Gateway can not be a local address");
3391                 goto out;
3392         }
3393
3394         err = 0;
3395 out:
3396         return err;
3397 }
3398
3399 static bool fib6_is_reject(u32 flags, struct net_device *dev, int addr_type)
3400 {
3401         if ((flags & RTF_REJECT) ||
3402             (dev && (dev->flags & IFF_LOOPBACK) &&
3403              !(addr_type & IPV6_ADDR_LOOPBACK) &&
3404              !(flags & (RTF_ANYCAST | RTF_LOCAL))))
3405                 return true;
3406
3407         return false;
3408 }
3409
3410 int fib6_nh_init(struct net *net, struct fib6_nh *fib6_nh,
3411                  struct fib6_config *cfg, gfp_t gfp_flags,
3412                  struct netlink_ext_ack *extack)
3413 {
3414         struct net_device *dev = NULL;
3415         struct inet6_dev *idev = NULL;
3416         int addr_type;
3417         int err;
3418
3419         fib6_nh->fib_nh_family = AF_INET6;
3420 #ifdef CONFIG_IPV6_ROUTER_PREF
3421         fib6_nh->last_probe = jiffies;
3422 #endif
3423         if (cfg->fc_is_fdb) {
3424                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3425                 fib6_nh->fib_nh_gw_family = AF_INET6;
3426                 return 0;
3427         }
3428
3429         err = -ENODEV;
3430         if (cfg->fc_ifindex) {
3431                 dev = dev_get_by_index(net, cfg->fc_ifindex);
3432                 if (!dev)
3433                         goto out;
3434                 idev = in6_dev_get(dev);
3435                 if (!idev)
3436                         goto out;
3437         }
3438
3439         if (cfg->fc_flags & RTNH_F_ONLINK) {
3440                 if (!dev) {
3441                         NL_SET_ERR_MSG(extack,
3442                                        "Nexthop device required for onlink");
3443                         goto out;
3444                 }
3445
3446                 if (!(dev->flags & IFF_UP)) {
3447                         NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3448                         err = -ENETDOWN;
3449                         goto out;
3450                 }
3451
3452                 fib6_nh->fib_nh_flags |= RTNH_F_ONLINK;
3453         }
3454
3455         fib6_nh->fib_nh_weight = 1;
3456
3457         /* We cannot add true routes via loopback here,
3458          * they would result in kernel looping; promote them to reject routes
3459          */
3460         addr_type = ipv6_addr_type(&cfg->fc_dst);
3461         if (fib6_is_reject(cfg->fc_flags, dev, addr_type)) {
3462                 /* hold loopback dev/idev if we haven't done so. */
3463                 if (dev != net->loopback_dev) {
3464                         if (dev) {
3465                                 dev_put(dev);
3466                                 in6_dev_put(idev);
3467                         }
3468                         dev = net->loopback_dev;
3469                         dev_hold(dev);
3470                         idev = in6_dev_get(dev);
3471                         if (!idev) {
3472                                 err = -ENODEV;
3473                                 goto out;
3474                         }
3475                 }
3476                 goto pcpu_alloc;
3477         }
3478
3479         if (cfg->fc_flags & RTF_GATEWAY) {
3480                 err = ip6_validate_gw(net, cfg, &dev, &idev, extack);
3481                 if (err)
3482                         goto out;
3483
3484                 fib6_nh->fib_nh_gw6 = cfg->fc_gateway;
3485                 fib6_nh->fib_nh_gw_family = AF_INET6;
3486         }
3487
3488         err = -ENODEV;
3489         if (!dev)
3490                 goto out;
3491
3492         if (idev->cnf.disable_ipv6) {
3493                 NL_SET_ERR_MSG(extack, "IPv6 is disabled on nexthop device");
3494                 err = -EACCES;
3495                 goto out;
3496         }
3497
3498         if (!(dev->flags & IFF_UP) && !cfg->fc_ignore_dev_down) {
3499                 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
3500                 err = -ENETDOWN;
3501                 goto out;
3502         }
3503
3504         if (!(cfg->fc_flags & (RTF_LOCAL | RTF_ANYCAST)) &&
3505             !netif_carrier_ok(dev))
3506                 fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
3507
3508         err = fib_nh_common_init(net, &fib6_nh->nh_common, cfg->fc_encap,
3509                                  cfg->fc_encap_type, cfg, gfp_flags, extack);
3510         if (err)
3511                 goto out;
3512
3513 pcpu_alloc:
3514         fib6_nh->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, gfp_flags);
3515         if (!fib6_nh->rt6i_pcpu) {
3516                 err = -ENOMEM;
3517                 goto out;
3518         }
3519
3520         fib6_nh->fib_nh_dev = dev;
3521         fib6_nh->fib_nh_oif = dev->ifindex;
3522         err = 0;
3523 out:
3524         if (idev)
3525                 in6_dev_put(idev);
3526
3527         if (err) {
3528                 lwtstate_put(fib6_nh->fib_nh_lws);
3529                 fib6_nh->fib_nh_lws = NULL;
3530                 if (dev)
3531                         dev_put(dev);
3532         }
3533
3534         return err;
3535 }
3536
3537 void fib6_nh_release(struct fib6_nh *fib6_nh)
3538 {
3539         struct rt6_exception_bucket *bucket;
3540
3541         rcu_read_lock();
3542
3543         fib6_nh_flush_exceptions(fib6_nh, NULL);
3544         bucket = fib6_nh_get_excptn_bucket(fib6_nh, NULL);
3545         if (bucket) {
3546                 rcu_assign_pointer(fib6_nh->rt6i_exception_bucket, NULL);
3547                 kfree(bucket);
3548         }
3549
3550         rcu_read_unlock();
3551
3552         if (fib6_nh->rt6i_pcpu) {
3553                 int cpu;
3554
3555                 for_each_possible_cpu(cpu) {
3556                         struct rt6_info **ppcpu_rt;
3557                         struct rt6_info *pcpu_rt;
3558
3559                         ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3560                         pcpu_rt = *ppcpu_rt;
3561                         if (pcpu_rt) {
3562                                 dst_dev_put(&pcpu_rt->dst);
3563                                 dst_release(&pcpu_rt->dst);
3564                                 *ppcpu_rt = NULL;
3565                         }
3566                 }
3567
3568                 free_percpu(fib6_nh->rt6i_pcpu);
3569         }
3570
3571         fib_nh_common_release(&fib6_nh->nh_common);
3572 }
3573
3574 void fib6_nh_release_dsts(struct fib6_nh *fib6_nh)
3575 {
3576         int cpu;
3577
3578         if (!fib6_nh->rt6i_pcpu)
3579                 return;
3580
3581         for_each_possible_cpu(cpu) {
3582                 struct rt6_info *pcpu_rt, **ppcpu_rt;
3583
3584                 ppcpu_rt = per_cpu_ptr(fib6_nh->rt6i_pcpu, cpu);
3585                 pcpu_rt = xchg(ppcpu_rt, NULL);
3586                 if (pcpu_rt) {
3587                         dst_dev_put(&pcpu_rt->dst);
3588                         dst_release(&pcpu_rt->dst);
3589                 }
3590         }
3591 }
3592
3593 static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
3594                                               gfp_t gfp_flags,
3595                                               struct netlink_ext_ack *extack)
3596 {
3597         struct net *net = cfg->fc_nlinfo.nl_net;
3598         struct fib6_info *rt = NULL;
3599         struct nexthop *nh = NULL;
3600         struct fib6_table *table;
3601         struct fib6_nh *fib6_nh;
3602         int err = -EINVAL;
3603         int addr_type;
3604
3605         /* RTF_PCPU is an internal flag; can not be set by userspace */
3606         if (cfg->fc_flags & RTF_PCPU) {
3607                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_PCPU");
3608                 goto out;
3609         }
3610
3611         /* RTF_CACHE is an internal flag; can not be set by userspace */
3612         if (cfg->fc_flags & RTF_CACHE) {
3613                 NL_SET_ERR_MSG(extack, "Userspace can not set RTF_CACHE");
3614                 goto out;
3615         }
3616
3617         if (cfg->fc_type > RTN_MAX) {
3618                 NL_SET_ERR_MSG(extack, "Invalid route type");
3619                 goto out;
3620         }
3621
3622         if (cfg->fc_dst_len > 128) {
3623                 NL_SET_ERR_MSG(extack, "Invalid prefix length");
3624                 goto out;
3625         }
3626         if (cfg->fc_src_len > 128) {
3627                 NL_SET_ERR_MSG(extack, "Invalid source address length");
3628                 goto out;
3629         }
3630 #ifndef CONFIG_IPV6_SUBTREES
3631         if (cfg->fc_src_len) {
3632                 NL_SET_ERR_MSG(extack,
3633                                "Specifying source address requires IPV6_SUBTREES to be enabled");
3634                 goto out;
3635         }
3636 #endif
3637         if (cfg->fc_nh_id) {
3638                 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
3639                 if (!nh) {
3640                         NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
3641                         goto out;
3642                 }
3643                 err = fib6_check_nexthop(nh, cfg, extack);
3644                 if (err)
3645                         goto out;
3646         }
3647
3648         err = -ENOBUFS;
3649         if (cfg->fc_nlinfo.nlh &&
3650             !(cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_CREATE)) {
3651                 table = fib6_get_table(net, cfg->fc_table);
3652                 if (!table) {
3653                         pr_warn("NLM_F_CREATE should be specified when creating new route\n");
3654                         table = fib6_new_table(net, cfg->fc_table);
3655                 }
3656         } else {
3657                 table = fib6_new_table(net, cfg->fc_table);
3658         }
3659
3660         if (!table)
3661                 goto out;
3662
3663         err = -ENOMEM;
3664         rt = fib6_info_alloc(gfp_flags, !nh);
3665         if (!rt)
3666                 goto out;
3667
3668         rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len,
3669                                                extack);
3670         if (IS_ERR(rt->fib6_metrics)) {
3671                 err = PTR_ERR(rt->fib6_metrics);
3672                 /* Do not leave garbage there. */
3673                 rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
3674                 goto out_free;
3675         }
3676
3677         if (cfg->fc_flags & RTF_ADDRCONF)
3678                 rt->dst_nocount = true;
3679
3680         if (cfg->fc_flags & RTF_EXPIRES)
3681                 fib6_set_expires(rt, jiffies +
3682                                 clock_t_to_jiffies(cfg->fc_expires));
3683         else
3684                 fib6_clean_expires(rt);
3685
3686         if (cfg->fc_protocol == RTPROT_UNSPEC)
3687                 cfg->fc_protocol = RTPROT_BOOT;
3688         rt->fib6_protocol = cfg->fc_protocol;
3689
3690         rt->fib6_table = table;
3691         rt->fib6_metric = cfg->fc_metric;
3692         rt->fib6_type = cfg->fc_type ? : RTN_UNICAST;
3693         rt->fib6_flags = cfg->fc_flags & ~RTF_GATEWAY;
3694
3695         ipv6_addr_prefix(&rt->fib6_dst.addr, &cfg->fc_dst, cfg->fc_dst_len);
3696         rt->fib6_dst.plen = cfg->fc_dst_len;
3697
3698 #ifdef CONFIG_IPV6_SUBTREES
3699         ipv6_addr_prefix(&rt->fib6_src.addr, &cfg->fc_src, cfg->fc_src_len);
3700         rt->fib6_src.plen = cfg->fc_src_len;
3701 #endif
3702         if (nh) {
3703                 if (rt->fib6_src.plen) {
3704                         NL_SET_ERR_MSG(extack, "Nexthops can not be used with source routing");
3705                         goto out_free;
3706                 }
3707                 if (!nexthop_get(nh)) {
3708                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
3709                         goto out_free;
3710                 }
3711                 rt->nh = nh;
3712                 fib6_nh = nexthop_fib6_nh(rt->nh);
3713         } else {
3714                 err = fib6_nh_init(net, rt->fib6_nh, cfg, gfp_flags, extack);
3715                 if (err)
3716                         goto out;
3717
3718                 fib6_nh = rt->fib6_nh;
3719
3720                 /* We cannot add true routes via loopback here, they would
3721                  * result in kernel looping; promote them to reject routes
3722                  */
3723                 addr_type = ipv6_addr_type(&cfg->fc_dst);
3724                 if (fib6_is_reject(cfg->fc_flags, rt->fib6_nh->fib_nh_dev,
3725                                    addr_type))
3726                         rt->fib6_flags = RTF_REJECT | RTF_NONEXTHOP;
3727         }
3728
3729         if (!ipv6_addr_any(&cfg->fc_prefsrc)) {
3730                 struct net_device *dev = fib6_nh->fib_nh_dev;
3731
3732                 if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) {
3733                         NL_SET_ERR_MSG(extack, "Invalid source address");
3734                         err = -EINVAL;
3735                         goto out;
3736                 }
3737                 rt->fib6_prefsrc.addr = cfg->fc_prefsrc;
3738                 rt->fib6_prefsrc.plen = 128;
3739         } else
3740                 rt->fib6_prefsrc.plen = 0;
3741
3742         return rt;
3743 out:
3744         fib6_info_release(rt);
3745         return ERR_PTR(err);
3746 out_free:
3747         ip_fib_metrics_put(rt->fib6_metrics);
3748         kfree(rt);
3749         return ERR_PTR(err);
3750 }
3751
3752 int ip6_route_add(struct fib6_config *cfg, gfp_t gfp_flags,
3753                   struct netlink_ext_ack *extack)
3754 {
3755         struct fib6_info *rt;
3756         int err;
3757
3758         rt = ip6_route_info_create(cfg, gfp_flags, extack);
3759         if (IS_ERR(rt))
3760                 return PTR_ERR(rt);
3761
3762         err = __ip6_ins_rt(rt, &cfg->fc_nlinfo, extack);
3763         fib6_info_release(rt);
3764
3765         return err;
3766 }
3767
3768 static int __ip6_del_rt(struct fib6_info *rt, struct nl_info *info)
3769 {
3770         struct net *net = info->nl_net;
3771         struct fib6_table *table;
3772         int err;
3773
3774         if (rt == net->ipv6.fib6_null_entry) {
3775                 err = -ENOENT;
3776                 goto out;
3777         }
3778
3779         table = rt->fib6_table;
3780         spin_lock_bh(&table->tb6_lock);
3781         err = fib6_del(rt, info);
3782         spin_unlock_bh(&table->tb6_lock);
3783
3784 out:
3785         fib6_info_release(rt);
3786         return err;
3787 }
3788
3789 int ip6_del_rt(struct net *net, struct fib6_info *rt, bool skip_notify)
3790 {
3791         struct nl_info info = {
3792                 .nl_net = net,
3793                 .skip_notify = skip_notify
3794         };
3795
3796         return __ip6_del_rt(rt, &info);
3797 }
3798
3799 static int __ip6_del_rt_siblings(struct fib6_info *rt, struct fib6_config *cfg)
3800 {
3801         struct nl_info *info = &cfg->fc_nlinfo;
3802         struct net *net = info->nl_net;
3803         struct sk_buff *skb = NULL;
3804         struct fib6_table *table;
3805         int err = -ENOENT;
3806
3807         if (rt == net->ipv6.fib6_null_entry)
3808                 goto out_put;
3809         table = rt->fib6_table;
3810         spin_lock_bh(&table->tb6_lock);
3811
3812         if (rt->fib6_nsiblings && cfg->fc_delete_all_nh) {
3813                 struct fib6_info *sibling, *next_sibling;
3814                 struct fib6_node *fn;
3815
3816                 /* prefer to send a single notification with all hops */
3817                 skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
3818                 if (skb) {
3819                         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
3820
3821                         if (rt6_fill_node(net, skb, rt, NULL,
3822                                           NULL, NULL, 0, RTM_DELROUTE,
3823                                           info->portid, seq, 0) < 0) {
3824                                 kfree_skb(skb);
3825                                 skb = NULL;
3826                         } else
3827                                 info->skip_notify = 1;
3828                 }
3829
3830                 /* 'rt' points to the first sibling route. If it is not the
3831                  * leaf, then we do not need to send a notification. Otherwise,
3832                  * we need to check if the last sibling has a next route or not
3833                  * and emit a replace or delete notification, respectively.
3834                  */
3835                 info->skip_notify_kernel = 1;
3836                 fn = rcu_dereference_protected(rt->fib6_node,
3837                                             lockdep_is_held(&table->tb6_lock));
3838                 if (rcu_access_pointer(fn->leaf) == rt) {
3839                         struct fib6_info *last_sibling, *replace_rt;
3840
3841                         last_sibling = list_last_entry(&rt->fib6_siblings,
3842                                                        struct fib6_info,
3843                                                        fib6_siblings);
3844                         replace_rt = rcu_dereference_protected(
3845                                             last_sibling->fib6_next,
3846                                             lockdep_is_held(&table->tb6_lock));
3847                         if (replace_rt)
3848                                 call_fib6_entry_notifiers_replace(net,
3849                                                                   replace_rt);
3850                         else
3851                                 call_fib6_multipath_entry_notifiers(net,
3852                                                        FIB_EVENT_ENTRY_DEL,
3853                                                        rt, rt->fib6_nsiblings,
3854                                                        NULL);
3855                 }
3856                 list_for_each_entry_safe(sibling, next_sibling,
3857                                          &rt->fib6_siblings,
3858                                          fib6_siblings) {
3859                         err = fib6_del(sibling, info);
3860                         if (err)
3861                                 goto out_unlock;
3862                 }
3863         }
3864
3865         err = fib6_del(rt, info);
3866 out_unlock:
3867         spin_unlock_bh(&table->tb6_lock);
3868 out_put:
3869         fib6_info_release(rt);
3870
3871         if (skb) {
3872                 rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
3873                             info->nlh, gfp_any());
3874         }
3875         return err;
3876 }
3877
3878 static int __ip6_del_cached_rt(struct rt6_info *rt, struct fib6_config *cfg)
3879 {
3880         int rc = -ESRCH;
3881
3882         if (cfg->fc_ifindex && rt->dst.dev->ifindex != cfg->fc_ifindex)
3883                 goto out;
3884
3885         if (cfg->fc_flags & RTF_GATEWAY &&
3886             !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway))
3887                 goto out;
3888
3889         rc = rt6_remove_exception_rt(rt);
3890 out:
3891         return rc;
3892 }
3893
3894 static int ip6_del_cached_rt(struct fib6_config *cfg, struct fib6_info *rt,
3895                              struct fib6_nh *nh)
3896 {
3897         struct fib6_result res = {
3898                 .f6i = rt,
3899                 .nh = nh,
3900         };
3901         struct rt6_info *rt_cache;
3902
3903         rt_cache = rt6_find_cached_rt(&res, &cfg->fc_dst, &cfg->fc_src);
3904         if (rt_cache)
3905                 return __ip6_del_cached_rt(rt_cache, cfg);
3906
3907         return 0;
3908 }
3909
3910 struct fib6_nh_del_cached_rt_arg {
3911         struct fib6_config *cfg;
3912         struct fib6_info *f6i;
3913 };
3914
3915 static int fib6_nh_del_cached_rt(struct fib6_nh *nh, void *_arg)
3916 {
3917         struct fib6_nh_del_cached_rt_arg *arg = _arg;
3918         int rc;
3919
3920         rc = ip6_del_cached_rt(arg->cfg, arg->f6i, nh);
3921         return rc != -ESRCH ? rc : 0;
3922 }
3923
3924 static int ip6_del_cached_rt_nh(struct fib6_config *cfg, struct fib6_info *f6i)
3925 {
3926         struct fib6_nh_del_cached_rt_arg arg = {
3927                 .cfg = cfg,
3928                 .f6i = f6i
3929         };
3930
3931         return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_del_cached_rt, &arg);
3932 }
3933
3934 static int ip6_route_del(struct fib6_config *cfg,
3935                          struct netlink_ext_ack *extack)
3936 {
3937         struct fib6_table *table;
3938         struct fib6_info *rt;
3939         struct fib6_node *fn;
3940         int err = -ESRCH;
3941
3942         table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table);
3943         if (!table) {
3944                 NL_SET_ERR_MSG(extack, "FIB table does not exist");
3945                 return err;
3946         }
3947
3948         rcu_read_lock();
3949
3950         fn = fib6_locate(&table->tb6_root,
3951                          &cfg->fc_dst, cfg->fc_dst_len,
3952                          &cfg->fc_src, cfg->fc_src_len,
3953                          !(cfg->fc_flags & RTF_CACHE));
3954
3955         if (fn) {
3956                 for_each_fib6_node_rt_rcu(fn) {
3957                         struct fib6_nh *nh;
3958
3959                         if (rt->nh && cfg->fc_nh_id &&
3960                             rt->nh->id != cfg->fc_nh_id)
3961                                 continue;
3962
3963                         if (cfg->fc_flags & RTF_CACHE) {
3964                                 int rc = 0;
3965
3966                                 if (rt->nh) {
3967                                         rc = ip6_del_cached_rt_nh(cfg, rt);
3968                                 } else if (cfg->fc_nh_id) {
3969                                         continue;
3970                                 } else {
3971                                         nh = rt->fib6_nh;
3972                                         rc = ip6_del_cached_rt(cfg, rt, nh);
3973                                 }
3974                                 if (rc != -ESRCH) {
3975                                         rcu_read_unlock();
3976                                         return rc;
3977                                 }
3978                                 continue;
3979                         }
3980
3981                         if (cfg->fc_metric && cfg->fc_metric != rt->fib6_metric)
3982                                 continue;
3983                         if (cfg->fc_protocol &&
3984                             cfg->fc_protocol != rt->fib6_protocol)
3985                                 continue;
3986
3987                         if (rt->nh) {
3988                                 if (!fib6_info_hold_safe(rt))
3989                                         continue;
3990                                 rcu_read_unlock();
3991
3992                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
3993                         }
3994                         if (cfg->fc_nh_id)
3995                                 continue;
3996
3997                         nh = rt->fib6_nh;
3998                         if (cfg->fc_ifindex &&
3999                             (!nh->fib_nh_dev ||
4000                              nh->fib_nh_dev->ifindex != cfg->fc_ifindex))
4001                                 continue;
4002                         if (cfg->fc_flags & RTF_GATEWAY &&
4003                             !ipv6_addr_equal(&cfg->fc_gateway, &nh->fib_nh_gw6))
4004                                 continue;
4005                         if (!fib6_info_hold_safe(rt))
4006                                 continue;
4007                         rcu_read_unlock();
4008
4009                         /* if gateway was specified only delete the one hop */
4010                         if (cfg->fc_flags & RTF_GATEWAY)
4011                                 return __ip6_del_rt(rt, &cfg->fc_nlinfo);
4012
4013                         return __ip6_del_rt_siblings(rt, cfg);
4014                 }
4015         }
4016         rcu_read_unlock();
4017
4018         return err;
4019 }
4020
4021 static void rt6_do_redirect(struct dst_entry *dst, struct sock *sk, struct sk_buff *skb)
4022 {
4023         struct netevent_redirect netevent;
4024         struct rt6_info *rt, *nrt = NULL;
4025         struct fib6_result res = {};
4026         struct ndisc_options ndopts;
4027         struct inet6_dev *in6_dev;
4028         struct neighbour *neigh;
4029         struct rd_msg *msg;
4030         int optlen, on_link;
4031         u8 *lladdr;
4032
4033         optlen = skb_tail_pointer(skb) - skb_transport_header(skb);
4034         optlen -= sizeof(*msg);
4035
4036         if (optlen < 0) {
4037                 net_dbg_ratelimited("rt6_do_redirect: packet too short\n");
4038                 return;
4039         }
4040
4041         msg = (struct rd_msg *)icmp6_hdr(skb);
4042
4043         if (ipv6_addr_is_multicast(&msg->dest)) {
4044                 net_dbg_ratelimited("rt6_do_redirect: destination address is multicast\n");
4045                 return;
4046         }
4047
4048         on_link = 0;
4049         if (ipv6_addr_equal(&msg->dest, &msg->target)) {
4050                 on_link = 1;
4051         } else if (ipv6_addr_type(&msg->target) !=
4052                    (IPV6_ADDR_UNICAST|IPV6_ADDR_LINKLOCAL)) {
4053                 net_dbg_ratelimited("rt6_do_redirect: target address is not link-local unicast\n");
4054                 return;
4055         }
4056
4057         in6_dev = __in6_dev_get(skb->dev);
4058         if (!in6_dev)
4059                 return;
4060         if (in6_dev->cnf.forwarding || !in6_dev->cnf.accept_redirects)
4061                 return;
4062
4063         /* RFC2461 8.1:
4064          *      The IP source address of the Redirect MUST be the same as the current
4065          *      first-hop router for the specified ICMP Destination Address.
4066          */
4067
4068         if (!ndisc_parse_options(skb->dev, msg->opt, optlen, &ndopts)) {
4069                 net_dbg_ratelimited("rt6_redirect: invalid ND options\n");
4070                 return;
4071         }
4072
4073         lladdr = NULL;
4074         if (ndopts.nd_opts_tgt_lladdr) {
4075                 lladdr = ndisc_opt_addr_data(ndopts.nd_opts_tgt_lladdr,
4076                                              skb->dev);
4077                 if (!lladdr) {
4078                         net_dbg_ratelimited("rt6_redirect: invalid link-layer address length\n");
4079                         return;
4080                 }
4081         }
4082
4083         rt = (struct rt6_info *) dst;
4084         if (rt->rt6i_flags & RTF_REJECT) {
4085                 net_dbg_ratelimited("rt6_redirect: source isn't a valid nexthop for redirect target\n");
4086                 return;
4087         }
4088
4089         /* Redirect received -> path was valid.
4090          * Look, redirects are sent only in response to data packets,
4091          * so that this nexthop apparently is reachable. --ANK
4092          */
4093         dst_confirm_neigh(&rt->dst, &ipv6_hdr(skb)->saddr);
4094
4095         neigh = __neigh_lookup(&nd_tbl, &msg->target, skb->dev, 1);
4096         if (!neigh)
4097                 return;
4098
4099         /*
4100          *      We have finally decided to accept it.
4101          */
4102
4103         ndisc_update(skb->dev, neigh, lladdr, NUD_STALE,
4104                      NEIGH_UPDATE_F_WEAK_OVERRIDE|
4105                      NEIGH_UPDATE_F_OVERRIDE|
4106                      (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER|
4107                                      NEIGH_UPDATE_F_ISROUTER)),
4108                      NDISC_REDIRECT, &ndopts);
4109
4110         rcu_read_lock();
4111         res.f6i = rcu_dereference(rt->from);
4112         if (!res.f6i)
4113                 goto out;
4114
4115         if (res.f6i->nh) {
4116                 struct fib6_nh_match_arg arg = {
4117                         .dev = dst->dev,
4118                         .gw = &rt->rt6i_gateway,
4119                 };
4120
4121                 nexthop_for_each_fib6_nh(res.f6i->nh,
4122                                          fib6_nh_find_match, &arg);
4123
4124                 /* fib6_info uses a nexthop that does not have fib6_nh
4125                  * using the dst->dev. Should be impossible
4126                  */
4127                 if (!arg.match)
4128                         goto out;
4129                 res.nh = arg.match;
4130         } else {
4131                 res.nh = res.f6i->fib6_nh;
4132         }
4133
4134         res.fib6_flags = res.f6i->fib6_flags;
4135         res.fib6_type = res.f6i->fib6_type;
4136         nrt = ip6_rt_cache_alloc(&res, &msg->dest, NULL);
4137         if (!nrt)
4138                 goto out;
4139
4140         nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE;
4141         if (on_link)
4142                 nrt->rt6i_flags &= ~RTF_GATEWAY;
4143
4144         nrt->rt6i_gateway = *(struct in6_addr *)neigh->primary_key;
4145
4146         /* rt6_insert_exception() will take care of duplicated exceptions */
4147         if (rt6_insert_exception(nrt, &res)) {
4148                 dst_release_immediate(&nrt->dst);
4149                 goto out;
4150         }
4151
4152         netevent.old = &rt->dst;
4153         netevent.new = &nrt->dst;
4154         netevent.daddr = &msg->dest;
4155         netevent.neigh = neigh;
4156         call_netevent_notifiers(NETEVENT_REDIRECT, &netevent);
4157
4158 out:
4159         rcu_read_unlock();
4160         neigh_release(neigh);
4161 }
4162
4163 #ifdef CONFIG_IPV6_ROUTE_INFO
4164 static struct fib6_info *rt6_get_route_info(struct net *net,
4165                                            const struct in6_addr *prefix, int prefixlen,
4166                                            const struct in6_addr *gwaddr,
4167                                            struct net_device *dev)
4168 {
4169         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4170         int ifindex = dev->ifindex;
4171         struct fib6_node *fn;
4172         struct fib6_info *rt = NULL;
4173         struct fib6_table *table;
4174
4175         table = fib6_get_table(net, tb_id);
4176         if (!table)
4177                 return NULL;
4178
4179         rcu_read_lock();
4180         fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
4181         if (!fn)
4182                 goto out;
4183
4184         for_each_fib6_node_rt_rcu(fn) {
4185                 /* these routes do not use nexthops */
4186                 if (rt->nh)
4187                         continue;
4188                 if (rt->fib6_nh->fib_nh_dev->ifindex != ifindex)
4189                         continue;
4190                 if (!(rt->fib6_flags & RTF_ROUTEINFO) ||
4191                     !rt->fib6_nh->fib_nh_gw_family)
4192                         continue;
4193                 if (!ipv6_addr_equal(&rt->fib6_nh->fib_nh_gw6, gwaddr))
4194                         continue;
4195                 if (!fib6_info_hold_safe(rt))
4196                         continue;
4197                 break;
4198         }
4199 out:
4200         rcu_read_unlock();
4201         return rt;
4202 }
4203
4204 static struct fib6_info *rt6_add_route_info(struct net *net,
4205                                            const struct in6_addr *prefix, int prefixlen,
4206                                            const struct in6_addr *gwaddr,
4207                                            struct net_device *dev,
4208                                            unsigned int pref)
4209 {
4210         struct fib6_config cfg = {
4211                 .fc_metric      = IP6_RT_PRIO_USER,
4212                 .fc_ifindex     = dev->ifindex,
4213                 .fc_dst_len     = prefixlen,
4214                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO |
4215                                   RTF_UP | RTF_PREF(pref),
4216                 .fc_protocol = RTPROT_RA,
4217                 .fc_type = RTN_UNICAST,
4218                 .fc_nlinfo.portid = 0,
4219                 .fc_nlinfo.nlh = NULL,
4220                 .fc_nlinfo.nl_net = net,
4221         };
4222
4223         cfg.fc_table = l3mdev_fib_table(dev) ? : RT6_TABLE_INFO;
4224         cfg.fc_dst = *prefix;
4225         cfg.fc_gateway = *gwaddr;
4226
4227         /* We should treat it as a default route if prefix length is 0. */
4228         if (!prefixlen)
4229                 cfg.fc_flags |= RTF_DEFAULT;
4230
4231         ip6_route_add(&cfg, GFP_ATOMIC, NULL);
4232
4233         return rt6_get_route_info(net, prefix, prefixlen, gwaddr, dev);
4234 }
4235 #endif
4236
4237 struct fib6_info *rt6_get_dflt_router(struct net *net,
4238                                      const struct in6_addr *addr,
4239                                      struct net_device *dev)
4240 {
4241         u32 tb_id = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT;
4242         struct fib6_info *rt;
4243         struct fib6_table *table;
4244
4245         table = fib6_get_table(net, tb_id);
4246         if (!table)
4247                 return NULL;
4248
4249         rcu_read_lock();
4250         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4251                 struct fib6_nh *nh;
4252
4253                 /* RA routes do not use nexthops */
4254                 if (rt->nh)
4255                         continue;
4256
4257                 nh = rt->fib6_nh;
4258                 if (dev == nh->fib_nh_dev &&
4259                     ((rt->fib6_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
4260                     ipv6_addr_equal(&nh->fib_nh_gw6, addr))
4261                         break;
4262         }
4263         if (rt && !fib6_info_hold_safe(rt))
4264                 rt = NULL;
4265         rcu_read_unlock();
4266         return rt;
4267 }
4268
4269 struct fib6_info *rt6_add_dflt_router(struct net *net,
4270                                      const struct in6_addr *gwaddr,
4271                                      struct net_device *dev,
4272                                      unsigned int pref)
4273 {
4274         struct fib6_config cfg = {
4275                 .fc_table       = l3mdev_fib_table(dev) ? : RT6_TABLE_DFLT,
4276                 .fc_metric      = IP6_RT_PRIO_USER,
4277                 .fc_ifindex     = dev->ifindex,
4278                 .fc_flags       = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT |
4279                                   RTF_UP | RTF_EXPIRES | RTF_PREF(pref),
4280                 .fc_protocol = RTPROT_RA,
4281                 .fc_type = RTN_UNICAST,
4282                 .fc_nlinfo.portid = 0,
4283                 .fc_nlinfo.nlh = NULL,
4284                 .fc_nlinfo.nl_net = net,
4285         };
4286
4287         cfg.fc_gateway = *gwaddr;
4288
4289         if (!ip6_route_add(&cfg, GFP_ATOMIC, NULL)) {
4290                 struct fib6_table *table;
4291
4292                 table = fib6_get_table(dev_net(dev), cfg.fc_table);
4293                 if (table)
4294                         table->flags |= RT6_TABLE_HAS_DFLT_ROUTER;
4295         }
4296
4297         return rt6_get_dflt_router(net, gwaddr, dev);
4298 }
4299
4300 static void __rt6_purge_dflt_routers(struct net *net,
4301                                      struct fib6_table *table)
4302 {
4303         struct fib6_info *rt;
4304
4305 restart:
4306         rcu_read_lock();
4307         for_each_fib6_node_rt_rcu(&table->tb6_root) {
4308                 struct net_device *dev = fib6_info_nh_dev(rt);
4309                 struct inet6_dev *idev = dev ? __in6_dev_get(dev) : NULL;
4310
4311                 if (rt->fib6_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
4312                     (!idev || idev->cnf.accept_ra != 2) &&
4313                     fib6_info_hold_safe(rt)) {
4314                         rcu_read_unlock();
4315                         ip6_del_rt(net, rt, false);
4316                         goto restart;
4317                 }
4318         }
4319         rcu_read_unlock();
4320
4321         table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
4322 }
4323
4324 void rt6_purge_dflt_routers(struct net *net)
4325 {
4326         struct fib6_table *table;
4327         struct hlist_head *head;
4328         unsigned int h;
4329
4330         rcu_read_lock();
4331
4332         for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
4333                 head = &net->ipv6.fib_table_hash[h];
4334                 hlist_for_each_entry_rcu(table, head, tb6_hlist) {
4335                         if (table->flags & RT6_TABLE_HAS_DFLT_ROUTER)
4336                                 __rt6_purge_dflt_routers(net, table);
4337                 }
4338         }
4339
4340         rcu_read_unlock();
4341 }
4342
4343 static void rtmsg_to_fib6_config(struct net *net,
4344                                  struct in6_rtmsg *rtmsg,
4345                                  struct fib6_config *cfg)
4346 {
4347         *cfg = (struct fib6_config){
4348                 .fc_table = l3mdev_fib_table_by_index(net, rtmsg->rtmsg_ifindex) ?
4349                          : RT6_TABLE_MAIN,
4350                 .fc_ifindex = rtmsg->rtmsg_ifindex,
4351                 .fc_metric = rtmsg->rtmsg_metric ? : IP6_RT_PRIO_USER,
4352                 .fc_expires = rtmsg->rtmsg_info,
4353                 .fc_dst_len = rtmsg->rtmsg_dst_len,
4354                 .fc_src_len = rtmsg->rtmsg_src_len,
4355                 .fc_flags = rtmsg->rtmsg_flags,
4356                 .fc_type = rtmsg->rtmsg_type,
4357
4358                 .fc_nlinfo.nl_net = net,
4359
4360                 .fc_dst = rtmsg->rtmsg_dst,
4361                 .fc_src = rtmsg->rtmsg_src,
4362                 .fc_gateway = rtmsg->rtmsg_gateway,
4363         };
4364 }
4365
4366 int ipv6_route_ioctl(struct net *net, unsigned int cmd, struct in6_rtmsg *rtmsg)
4367 {
4368         struct fib6_config cfg;
4369         int err;
4370
4371         if (cmd != SIOCADDRT && cmd != SIOCDELRT)
4372                 return -EINVAL;
4373         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4374                 return -EPERM;
4375
4376         rtmsg_to_fib6_config(net, rtmsg, &cfg);
4377
4378         rtnl_lock();
4379         switch (cmd) {
4380         case SIOCADDRT:
4381                 err = ip6_route_add(&cfg, GFP_KERNEL, NULL);
4382                 break;
4383         case SIOCDELRT:
4384                 err = ip6_route_del(&cfg, NULL);
4385                 break;
4386         }
4387         rtnl_unlock();
4388         return err;
4389 }
4390
4391 /*
4392  *      Drop the packet on the floor
4393  */
4394
4395 static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes)
4396 {
4397         struct dst_entry *dst = skb_dst(skb);
4398         struct net *net = dev_net(dst->dev);
4399         struct inet6_dev *idev;
4400         int type;
4401
4402         if (netif_is_l3_master(skb->dev) ||
4403             dst->dev == net->loopback_dev)
4404                 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
4405         else
4406                 idev = ip6_dst_idev(dst);
4407
4408         switch (ipstats_mib_noroutes) {
4409         case IPSTATS_MIB_INNOROUTES:
4410                 type = ipv6_addr_type(&ipv6_hdr(skb)->daddr);
4411                 if (type == IPV6_ADDR_ANY) {
4412                         IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
4413                         break;
4414                 }
4415                 fallthrough;
4416         case IPSTATS_MIB_OUTNOROUTES:
4417                 IP6_INC_STATS(net, idev, ipstats_mib_noroutes);
4418                 break;
4419         }
4420
4421         /* Start over by dropping the dst for l3mdev case */
4422         if (netif_is_l3_master(skb->dev))
4423                 skb_dst_drop(skb);
4424
4425         icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0);
4426         kfree_skb(skb);
4427         return 0;
4428 }
4429
4430 static int ip6_pkt_discard(struct sk_buff *skb)
4431 {
4432         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES);
4433 }
4434
4435 static int ip6_pkt_discard_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4436 {
4437         skb->dev = skb_dst(skb)->dev;
4438         return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES);
4439 }
4440
4441 static int ip6_pkt_prohibit(struct sk_buff *skb)
4442 {
4443         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES);
4444 }
4445
4446 static int ip6_pkt_prohibit_out(struct net *net, struct sock *sk, struct sk_buff *skb)
4447 {
4448         skb->dev = skb_dst(skb)->dev;
4449         return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES);
4450 }
4451
4452 /*
4453  *      Allocate a dst for local (unicast / anycast) address.
4454  */
4455
4456 struct fib6_info *addrconf_f6i_alloc(struct net *net,
4457                                      struct inet6_dev *idev,
4458                                      const struct in6_addr *addr,
4459                                      bool anycast, gfp_t gfp_flags)
4460 {
4461         struct fib6_config cfg = {
4462                 .fc_table = l3mdev_fib_table(idev->dev) ? : RT6_TABLE_LOCAL,
4463                 .fc_ifindex = idev->dev->ifindex,
4464                 .fc_flags = RTF_UP | RTF_NONEXTHOP,
4465                 .fc_dst = *addr,
4466                 .fc_dst_len = 128,
4467                 .fc_protocol = RTPROT_KERNEL,
4468                 .fc_nlinfo.nl_net = net,
4469                 .fc_ignore_dev_down = true,
4470         };
4471         struct fib6_info *f6i;
4472
4473         if (anycast) {
4474                 cfg.fc_type = RTN_ANYCAST;
4475                 cfg.fc_flags |= RTF_ANYCAST;
4476         } else {
4477                 cfg.fc_type = RTN_LOCAL;
4478                 cfg.fc_flags |= RTF_LOCAL;
4479         }
4480
4481         f6i = ip6_route_info_create(&cfg, gfp_flags, NULL);
4482         if (!IS_ERR(f6i)) {
4483                 f6i->dst_nocount = true;
4484
4485                 if (!anycast &&
4486                     (net->ipv6.devconf_all->disable_policy ||
4487                      idev->cnf.disable_policy))
4488                         f6i->dst_nopolicy = true;
4489         }
4490
4491         return f6i;
4492 }
4493
4494 /* remove deleted ip from prefsrc entries */
4495 struct arg_dev_net_ip {
4496         struct net_device *dev;
4497         struct net *net;
4498         struct in6_addr *addr;
4499 };
4500
4501 static int fib6_remove_prefsrc(struct fib6_info *rt, void *arg)
4502 {
4503         struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev;
4504         struct net *net = ((struct arg_dev_net_ip *)arg)->net;
4505         struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr;
4506
4507         if (!rt->nh &&
4508             ((void *)rt->fib6_nh->fib_nh_dev == dev || !dev) &&
4509             rt != net->ipv6.fib6_null_entry &&
4510             ipv6_addr_equal(addr, &rt->fib6_prefsrc.addr)) {
4511                 spin_lock_bh(&rt6_exception_lock);
4512                 /* remove prefsrc entry */
4513                 rt->fib6_prefsrc.plen = 0;
4514                 spin_unlock_bh(&rt6_exception_lock);
4515         }
4516         return 0;
4517 }
4518
4519 void rt6_remove_prefsrc(struct inet6_ifaddr *ifp)
4520 {
4521         struct net *net = dev_net(ifp->idev->dev);
4522         struct arg_dev_net_ip adni = {
4523                 .dev = ifp->idev->dev,
4524                 .net = net,
4525                 .addr = &ifp->addr,
4526         };
4527         fib6_clean_all(net, fib6_remove_prefsrc, &adni);
4528 }
4529
4530 #define RTF_RA_ROUTER           (RTF_ADDRCONF | RTF_DEFAULT)
4531
4532 /* Remove routers and update dst entries when gateway turn into host. */
4533 static int fib6_clean_tohost(struct fib6_info *rt, void *arg)
4534 {
4535         struct in6_addr *gateway = (struct in6_addr *)arg;
4536         struct fib6_nh *nh;
4537
4538         /* RA routes do not use nexthops */
4539         if (rt->nh)
4540                 return 0;
4541
4542         nh = rt->fib6_nh;
4543         if (((rt->fib6_flags & RTF_RA_ROUTER) == RTF_RA_ROUTER) &&
4544             nh->fib_nh_gw_family && ipv6_addr_equal(gateway, &nh->fib_nh_gw6))
4545                 return -1;
4546
4547         /* Further clean up cached routes in exception table.
4548          * This is needed because cached route may have a different
4549          * gateway than its 'parent' in the case of an ip redirect.
4550          */
4551         fib6_nh_exceptions_clean_tohost(nh, gateway);
4552
4553         return 0;
4554 }
4555
4556 void rt6_clean_tohost(struct net *net, struct in6_addr *gateway)
4557 {
4558         fib6_clean_all(net, fib6_clean_tohost, gateway);
4559 }
4560
4561 struct arg_netdev_event {
4562         const struct net_device *dev;
4563         union {
4564                 unsigned char nh_flags;
4565                 unsigned long event;
4566         };
4567 };
4568
4569 static struct fib6_info *rt6_multipath_first_sibling(const struct fib6_info *rt)
4570 {
4571         struct fib6_info *iter;
4572         struct fib6_node *fn;
4573
4574         fn = rcu_dereference_protected(rt->fib6_node,
4575                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4576         iter = rcu_dereference_protected(fn->leaf,
4577                         lockdep_is_held(&rt->fib6_table->tb6_lock));
4578         while (iter) {
4579                 if (iter->fib6_metric == rt->fib6_metric &&
4580                     rt6_qualify_for_ecmp(iter))
4581                         return iter;
4582                 iter = rcu_dereference_protected(iter->fib6_next,
4583                                 lockdep_is_held(&rt->fib6_table->tb6_lock));
4584         }
4585
4586         return NULL;
4587 }
4588
4589 /* only called for fib entries with builtin fib6_nh */
4590 static bool rt6_is_dead(const struct fib6_info *rt)
4591 {
4592         if (rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD ||
4593             (rt->fib6_nh->fib_nh_flags & RTNH_F_LINKDOWN &&
4594              ip6_ignore_linkdown(rt->fib6_nh->fib_nh_dev)))
4595                 return true;
4596
4597         return false;
4598 }
4599
4600 static int rt6_multipath_total_weight(const struct fib6_info *rt)
4601 {
4602         struct fib6_info *iter;
4603         int total = 0;
4604
4605         if (!rt6_is_dead(rt))
4606                 total += rt->fib6_nh->fib_nh_weight;
4607
4608         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings) {
4609                 if (!rt6_is_dead(iter))
4610                         total += iter->fib6_nh->fib_nh_weight;
4611         }
4612
4613         return total;
4614 }
4615
4616 static void rt6_upper_bound_set(struct fib6_info *rt, int *weight, int total)
4617 {
4618         int upper_bound = -1;
4619
4620         if (!rt6_is_dead(rt)) {
4621                 *weight += rt->fib6_nh->fib_nh_weight;
4622                 upper_bound = DIV_ROUND_CLOSEST_ULL((u64) (*weight) << 31,
4623                                                     total) - 1;
4624         }
4625         atomic_set(&rt->fib6_nh->fib_nh_upper_bound, upper_bound);
4626 }
4627
4628 static void rt6_multipath_upper_bound_set(struct fib6_info *rt, int total)
4629 {
4630         struct fib6_info *iter;
4631         int weight = 0;
4632
4633         rt6_upper_bound_set(rt, &weight, total);
4634
4635         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4636                 rt6_upper_bound_set(iter, &weight, total);
4637 }
4638
4639 void rt6_multipath_rebalance(struct fib6_info *rt)
4640 {
4641         struct fib6_info *first;
4642         int total;
4643
4644         /* In case the entire multipath route was marked for flushing,
4645          * then there is no need to rebalance upon the removal of every
4646          * sibling route.
4647          */
4648         if (!rt->fib6_nsiblings || rt->should_flush)
4649                 return;
4650
4651         /* During lookup routes are evaluated in order, so we need to
4652          * make sure upper bounds are assigned from the first sibling
4653          * onwards.
4654          */
4655         first = rt6_multipath_first_sibling(rt);
4656         if (WARN_ON_ONCE(!first))
4657                 return;
4658
4659         total = rt6_multipath_total_weight(first);
4660         rt6_multipath_upper_bound_set(first, total);
4661 }
4662
4663 static int fib6_ifup(struct fib6_info *rt, void *p_arg)
4664 {
4665         const struct arg_netdev_event *arg = p_arg;
4666         struct net *net = dev_net(arg->dev);
4667
4668         if (rt != net->ipv6.fib6_null_entry && !rt->nh &&
4669             rt->fib6_nh->fib_nh_dev == arg->dev) {
4670                 rt->fib6_nh->fib_nh_flags &= ~arg->nh_flags;
4671                 fib6_update_sernum_upto_root(net, rt);
4672                 rt6_multipath_rebalance(rt);
4673         }
4674
4675         return 0;
4676 }
4677
4678 void rt6_sync_up(struct net_device *dev, unsigned char nh_flags)
4679 {
4680         struct arg_netdev_event arg = {
4681                 .dev = dev,
4682                 {
4683                         .nh_flags = nh_flags,
4684                 },
4685         };
4686
4687         if (nh_flags & RTNH_F_DEAD && netif_carrier_ok(dev))
4688                 arg.nh_flags |= RTNH_F_LINKDOWN;
4689
4690         fib6_clean_all(dev_net(dev), fib6_ifup, &arg);
4691 }
4692
4693 /* only called for fib entries with inline fib6_nh */
4694 static bool rt6_multipath_uses_dev(const struct fib6_info *rt,
4695                                    const struct net_device *dev)
4696 {
4697         struct fib6_info *iter;
4698
4699         if (rt->fib6_nh->fib_nh_dev == dev)
4700                 return true;
4701         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4702                 if (iter->fib6_nh->fib_nh_dev == dev)
4703                         return true;
4704
4705         return false;
4706 }
4707
4708 static void rt6_multipath_flush(struct fib6_info *rt)
4709 {
4710         struct fib6_info *iter;
4711
4712         rt->should_flush = 1;
4713         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4714                 iter->should_flush = 1;
4715 }
4716
4717 static unsigned int rt6_multipath_dead_count(const struct fib6_info *rt,
4718                                              const struct net_device *down_dev)
4719 {
4720         struct fib6_info *iter;
4721         unsigned int dead = 0;
4722
4723         if (rt->fib6_nh->fib_nh_dev == down_dev ||
4724             rt->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4725                 dead++;
4726         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4727                 if (iter->fib6_nh->fib_nh_dev == down_dev ||
4728                     iter->fib6_nh->fib_nh_flags & RTNH_F_DEAD)
4729                         dead++;
4730
4731         return dead;
4732 }
4733
4734 static void rt6_multipath_nh_flags_set(struct fib6_info *rt,
4735                                        const struct net_device *dev,
4736                                        unsigned char nh_flags)
4737 {
4738         struct fib6_info *iter;
4739
4740         if (rt->fib6_nh->fib_nh_dev == dev)
4741                 rt->fib6_nh->fib_nh_flags |= nh_flags;
4742         list_for_each_entry(iter, &rt->fib6_siblings, fib6_siblings)
4743                 if (iter->fib6_nh->fib_nh_dev == dev)
4744                         iter->fib6_nh->fib_nh_flags |= nh_flags;
4745 }
4746
4747 /* called with write lock held for table with rt */
4748 static int fib6_ifdown(struct fib6_info *rt, void *p_arg)
4749 {
4750         const struct arg_netdev_event *arg = p_arg;
4751         const struct net_device *dev = arg->dev;
4752         struct net *net = dev_net(dev);
4753
4754         if (rt == net->ipv6.fib6_null_entry || rt->nh)
4755                 return 0;
4756
4757         switch (arg->event) {
4758         case NETDEV_UNREGISTER:
4759                 return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4760         case NETDEV_DOWN:
4761                 if (rt->should_flush)
4762                         return -1;
4763                 if (!rt->fib6_nsiblings)
4764                         return rt->fib6_nh->fib_nh_dev == dev ? -1 : 0;
4765                 if (rt6_multipath_uses_dev(rt, dev)) {
4766                         unsigned int count;
4767
4768                         count = rt6_multipath_dead_count(rt, dev);
4769                         if (rt->fib6_nsiblings + 1 == count) {
4770                                 rt6_multipath_flush(rt);
4771                                 return -1;
4772                         }
4773                         rt6_multipath_nh_flags_set(rt, dev, RTNH_F_DEAD |
4774                                                    RTNH_F_LINKDOWN);
4775                         fib6_update_sernum(net, rt);
4776                         rt6_multipath_rebalance(rt);
4777                 }
4778                 return -2;
4779         case NETDEV_CHANGE:
4780                 if (rt->fib6_nh->fib_nh_dev != dev ||
4781                     rt->fib6_flags & (RTF_LOCAL | RTF_ANYCAST))
4782                         break;
4783                 rt->fib6_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
4784                 rt6_multipath_rebalance(rt);
4785                 break;
4786         }
4787
4788         return 0;
4789 }
4790
4791 void rt6_sync_down_dev(struct net_device *dev, unsigned long event)
4792 {
4793         struct arg_netdev_event arg = {
4794                 .dev = dev,
4795                 {
4796                         .event = event,
4797                 },
4798         };
4799         struct net *net = dev_net(dev);
4800
4801         if (net->ipv6.sysctl.skip_notify_on_dev_down)
4802                 fib6_clean_all_skip_notify(net, fib6_ifdown, &arg);
4803         else
4804                 fib6_clean_all(net, fib6_ifdown, &arg);
4805 }
4806
4807 void rt6_disable_ip(struct net_device *dev, unsigned long event)
4808 {
4809         rt6_sync_down_dev(dev, event);
4810         rt6_uncached_list_flush_dev(dev_net(dev), dev);
4811         neigh_ifdown(&nd_tbl, dev);
4812 }
4813
4814 struct rt6_mtu_change_arg {
4815         struct net_device *dev;
4816         unsigned int mtu;
4817         struct fib6_info *f6i;
4818 };
4819
4820 static int fib6_nh_mtu_change(struct fib6_nh *nh, void *_arg)
4821 {
4822         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *)_arg;
4823         struct fib6_info *f6i = arg->f6i;
4824
4825         /* For administrative MTU increase, there is no way to discover
4826          * IPv6 PMTU increase, so PMTU increase should be updated here.
4827          * Since RFC 1981 doesn't include administrative MTU increase
4828          * update PMTU increase is a MUST. (i.e. jumbo frame)
4829          */
4830         if (nh->fib_nh_dev == arg->dev) {
4831                 struct inet6_dev *idev = __in6_dev_get(arg->dev);
4832                 u32 mtu = f6i->fib6_pmtu;
4833
4834                 if (mtu >= arg->mtu ||
4835                     (mtu < arg->mtu && mtu == idev->cnf.mtu6))
4836                         fib6_metric_set(f6i, RTAX_MTU, arg->mtu);
4837
4838                 spin_lock_bh(&rt6_exception_lock);
4839                 rt6_exceptions_update_pmtu(idev, nh, arg->mtu);
4840                 spin_unlock_bh(&rt6_exception_lock);
4841         }
4842
4843         return 0;
4844 }
4845
4846 static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg)
4847 {
4848         struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg;
4849         struct inet6_dev *idev;
4850
4851         /* In IPv6 pmtu discovery is not optional,
4852            so that RTAX_MTU lock cannot disable it.
4853            We still use this lock to block changes
4854            caused by addrconf/ndisc.
4855         */
4856
4857         idev = __in6_dev_get(arg->dev);
4858         if (!idev)
4859                 return 0;
4860
4861         if (fib6_metric_locked(f6i, RTAX_MTU))
4862                 return 0;
4863
4864         arg->f6i = f6i;
4865         if (f6i->nh) {
4866                 /* fib6_nh_mtu_change only returns 0, so this is safe */
4867                 return nexthop_for_each_fib6_nh(f6i->nh, fib6_nh_mtu_change,
4868                                                 arg);
4869         }
4870
4871         return fib6_nh_mtu_change(f6i->fib6_nh, arg);
4872 }
4873
4874 void rt6_mtu_change(struct net_device *dev, unsigned int mtu)
4875 {
4876         struct rt6_mtu_change_arg arg = {
4877                 .dev = dev,
4878                 .mtu = mtu,
4879         };
4880
4881         fib6_clean_all(dev_net(dev), rt6_mtu_change_route, &arg);
4882 }
4883
4884 static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = {
4885         [RTA_UNSPEC]            = { .strict_start_type = RTA_DPORT + 1 },
4886         [RTA_GATEWAY]           = { .len = sizeof(struct in6_addr) },
4887         [RTA_PREFSRC]           = { .len = sizeof(struct in6_addr) },
4888         [RTA_OIF]               = { .type = NLA_U32 },
4889         [RTA_IIF]               = { .type = NLA_U32 },
4890         [RTA_PRIORITY]          = { .type = NLA_U32 },
4891         [RTA_METRICS]           = { .type = NLA_NESTED },
4892         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
4893         [RTA_PREF]              = { .type = NLA_U8 },
4894         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
4895         [RTA_ENCAP]             = { .type = NLA_NESTED },
4896         [RTA_EXPIRES]           = { .type = NLA_U32 },
4897         [RTA_UID]               = { .type = NLA_U32 },
4898         [RTA_MARK]              = { .type = NLA_U32 },
4899         [RTA_TABLE]             = { .type = NLA_U32 },
4900         [RTA_IP_PROTO]          = { .type = NLA_U8 },
4901         [RTA_SPORT]             = { .type = NLA_U16 },
4902         [RTA_DPORT]             = { .type = NLA_U16 },
4903         [RTA_NH_ID]             = { .type = NLA_U32 },
4904 };
4905
4906 static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh,
4907                               struct fib6_config *cfg,
4908                               struct netlink_ext_ack *extack)
4909 {
4910         struct rtmsg *rtm;
4911         struct nlattr *tb[RTA_MAX+1];
4912         unsigned int pref;
4913         int err;
4914
4915         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
4916                                      rtm_ipv6_policy, extack);
4917         if (err < 0)
4918                 goto errout;
4919
4920         err = -EINVAL;
4921         rtm = nlmsg_data(nlh);
4922
4923         *cfg = (struct fib6_config){
4924                 .fc_table = rtm->rtm_table,
4925                 .fc_dst_len = rtm->rtm_dst_len,
4926                 .fc_src_len = rtm->rtm_src_len,
4927                 .fc_flags = RTF_UP,
4928                 .fc_protocol = rtm->rtm_protocol,
4929                 .fc_type = rtm->rtm_type,
4930
4931                 .fc_nlinfo.portid = NETLINK_CB(skb).portid,
4932                 .fc_nlinfo.nlh = nlh,
4933                 .fc_nlinfo.nl_net = sock_net(skb->sk),
4934         };
4935
4936         if (rtm->rtm_type == RTN_UNREACHABLE ||
4937             rtm->rtm_type == RTN_BLACKHOLE ||
4938             rtm->rtm_type == RTN_PROHIBIT ||
4939             rtm->rtm_type == RTN_THROW)
4940                 cfg->fc_flags |= RTF_REJECT;
4941
4942         if (rtm->rtm_type == RTN_LOCAL)
4943                 cfg->fc_flags |= RTF_LOCAL;
4944
4945         if (rtm->rtm_flags & RTM_F_CLONED)
4946                 cfg->fc_flags |= RTF_CACHE;
4947
4948         cfg->fc_flags |= (rtm->rtm_flags & RTNH_F_ONLINK);
4949
4950         if (tb[RTA_NH_ID]) {
4951                 if (tb[RTA_GATEWAY]   || tb[RTA_OIF] ||
4952                     tb[RTA_MULTIPATH] || tb[RTA_ENCAP]) {
4953                         NL_SET_ERR_MSG(extack,
4954                                        "Nexthop specification and nexthop id are mutually exclusive");
4955                         goto errout;
4956                 }
4957                 cfg->fc_nh_id = nla_get_u32(tb[RTA_NH_ID]);
4958         }
4959
4960         if (tb[RTA_GATEWAY]) {
4961                 cfg->fc_gateway = nla_get_in6_addr(tb[RTA_GATEWAY]);
4962                 cfg->fc_flags |= RTF_GATEWAY;
4963         }
4964         if (tb[RTA_VIA]) {
4965                 NL_SET_ERR_MSG(extack, "IPv6 does not support RTA_VIA attribute");
4966                 goto errout;
4967         }
4968
4969         if (tb[RTA_DST]) {
4970                 int plen = (rtm->rtm_dst_len + 7) >> 3;
4971
4972                 if (nla_len(tb[RTA_DST]) < plen)
4973                         goto errout;
4974
4975                 nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen);
4976         }
4977
4978         if (tb[RTA_SRC]) {
4979                 int plen = (rtm->rtm_src_len + 7) >> 3;
4980
4981                 if (nla_len(tb[RTA_SRC]) < plen)
4982                         goto errout;
4983
4984                 nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen);
4985         }
4986
4987         if (tb[RTA_PREFSRC])
4988                 cfg->fc_prefsrc = nla_get_in6_addr(tb[RTA_PREFSRC]);
4989
4990         if (tb[RTA_OIF])
4991                 cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]);
4992
4993         if (tb[RTA_PRIORITY])
4994                 cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]);
4995
4996         if (tb[RTA_METRICS]) {
4997                 cfg->fc_mx = nla_data(tb[RTA_METRICS]);
4998                 cfg->fc_mx_len = nla_len(tb[RTA_METRICS]);
4999         }
5000
5001         if (tb[RTA_TABLE])
5002                 cfg->fc_table = nla_get_u32(tb[RTA_TABLE]);
5003
5004         if (tb[RTA_MULTIPATH]) {
5005                 cfg->fc_mp = nla_data(tb[RTA_MULTIPATH]);
5006                 cfg->fc_mp_len = nla_len(tb[RTA_MULTIPATH]);
5007
5008                 err = lwtunnel_valid_encap_type_attr(cfg->fc_mp,
5009                                                      cfg->fc_mp_len, extack);
5010                 if (err < 0)
5011                         goto errout;
5012         }
5013
5014         if (tb[RTA_PREF]) {
5015                 pref = nla_get_u8(tb[RTA_PREF]);
5016                 if (pref != ICMPV6_ROUTER_PREF_LOW &&
5017                     pref != ICMPV6_ROUTER_PREF_HIGH)
5018                         pref = ICMPV6_ROUTER_PREF_MEDIUM;
5019                 cfg->fc_flags |= RTF_PREF(pref);
5020         }
5021
5022         if (tb[RTA_ENCAP])
5023                 cfg->fc_encap = tb[RTA_ENCAP];
5024
5025         if (tb[RTA_ENCAP_TYPE]) {
5026                 cfg->fc_encap_type = nla_get_u16(tb[RTA_ENCAP_TYPE]);
5027
5028                 err = lwtunnel_valid_encap_type(cfg->fc_encap_type, extack);
5029                 if (err < 0)
5030                         goto errout;
5031         }
5032
5033         if (tb[RTA_EXPIRES]) {
5034                 unsigned long timeout = addrconf_timeout_fixup(nla_get_u32(tb[RTA_EXPIRES]), HZ);
5035
5036                 if (addrconf_finite_timeout(timeout)) {
5037                         cfg->fc_expires = jiffies_to_clock_t(timeout * HZ);
5038                         cfg->fc_flags |= RTF_EXPIRES;
5039                 }
5040         }
5041
5042         err = 0;
5043 errout:
5044         return err;
5045 }
5046
5047 struct rt6_nh {
5048         struct fib6_info *fib6_info;
5049         struct fib6_config r_cfg;
5050         struct list_head next;
5051 };
5052
5053 static int ip6_route_info_append(struct net *net,
5054                                  struct list_head *rt6_nh_list,
5055                                  struct fib6_info *rt,
5056                                  struct fib6_config *r_cfg)
5057 {
5058         struct rt6_nh *nh;
5059         int err = -EEXIST;
5060
5061         list_for_each_entry(nh, rt6_nh_list, next) {
5062                 /* check if fib6_info already exists */
5063                 if (rt6_duplicate_nexthop(nh->fib6_info, rt))
5064                         return err;
5065         }
5066
5067         nh = kzalloc(sizeof(*nh), GFP_KERNEL);
5068         if (!nh)
5069                 return -ENOMEM;
5070         nh->fib6_info = rt;
5071         memcpy(&nh->r_cfg, r_cfg, sizeof(*r_cfg));
5072         list_add_tail(&nh->next, rt6_nh_list);
5073
5074         return 0;
5075 }
5076
5077 static void ip6_route_mpath_notify(struct fib6_info *rt,
5078                                    struct fib6_info *rt_last,
5079                                    struct nl_info *info,
5080                                    __u16 nlflags)
5081 {
5082         /* if this is an APPEND route, then rt points to the first route
5083          * inserted and rt_last points to last route inserted. Userspace
5084          * wants a consistent dump of the route which starts at the first
5085          * nexthop. Since sibling routes are always added at the end of
5086          * the list, find the first sibling of the last route appended
5087          */
5088         if ((nlflags & NLM_F_APPEND) && rt_last && rt_last->fib6_nsiblings) {
5089                 rt = list_first_entry(&rt_last->fib6_siblings,
5090                                       struct fib6_info,
5091                                       fib6_siblings);
5092         }
5093
5094         if (rt)
5095                 inet6_rt_notify(RTM_NEWROUTE, rt, info, nlflags);
5096 }
5097
5098 static bool ip6_route_mpath_should_notify(const struct fib6_info *rt)
5099 {
5100         bool rt_can_ecmp = rt6_qualify_for_ecmp(rt);
5101         bool should_notify = false;
5102         struct fib6_info *leaf;
5103         struct fib6_node *fn;
5104
5105         rcu_read_lock();
5106         fn = rcu_dereference(rt->fib6_node);
5107         if (!fn)
5108                 goto out;
5109
5110         leaf = rcu_dereference(fn->leaf);
5111         if (!leaf)
5112                 goto out;
5113
5114         if (rt == leaf ||
5115             (rt_can_ecmp && rt->fib6_metric == leaf->fib6_metric &&
5116              rt6_qualify_for_ecmp(leaf)))
5117                 should_notify = true;
5118 out:
5119         rcu_read_unlock();
5120
5121         return should_notify;
5122 }
5123
5124 static int fib6_gw_from_attr(struct in6_addr *gw, struct nlattr *nla,
5125                              struct netlink_ext_ack *extack)
5126 {
5127         if (nla_len(nla) < sizeof(*gw)) {
5128                 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_GATEWAY");
5129                 return -EINVAL;
5130         }
5131
5132         *gw = nla_get_in6_addr(nla);
5133
5134         return 0;
5135 }
5136
5137 static int ip6_route_multipath_add(struct fib6_config *cfg,
5138                                    struct netlink_ext_ack *extack)
5139 {
5140         struct fib6_info *rt_notif = NULL, *rt_last = NULL;
5141         struct nl_info *info = &cfg->fc_nlinfo;
5142         struct fib6_config r_cfg;
5143         struct rtnexthop *rtnh;
5144         struct fib6_info *rt;
5145         struct rt6_nh *err_nh;
5146         struct rt6_nh *nh, *nh_safe;
5147         __u16 nlflags;
5148         int remaining;
5149         int attrlen;
5150         int err = 1;
5151         int nhn = 0;
5152         int replace = (cfg->fc_nlinfo.nlh &&
5153                        (cfg->fc_nlinfo.nlh->nlmsg_flags & NLM_F_REPLACE));
5154         LIST_HEAD(rt6_nh_list);
5155
5156         nlflags = replace ? NLM_F_REPLACE : NLM_F_CREATE;
5157         if (info->nlh && info->nlh->nlmsg_flags & NLM_F_APPEND)
5158                 nlflags |= NLM_F_APPEND;
5159
5160         remaining = cfg->fc_mp_len;
5161         rtnh = (struct rtnexthop *)cfg->fc_mp;
5162
5163         /* Parse a Multipath Entry and build a list (rt6_nh_list) of
5164          * fib6_info structs per nexthop
5165          */
5166         while (rtnh_ok(rtnh, remaining)) {
5167                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5168                 if (rtnh->rtnh_ifindex)
5169                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5170
5171                 attrlen = rtnh_attrlen(rtnh);
5172                 if (attrlen > 0) {
5173                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5174
5175                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5176                         if (nla) {
5177                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5178                                                         extack);
5179                                 if (err)
5180                                         goto cleanup;
5181
5182                                 r_cfg.fc_flags |= RTF_GATEWAY;
5183                         }
5184                         r_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
5185
5186                         /* RTA_ENCAP_TYPE length checked in
5187                          * lwtunnel_valid_encap_type_attr
5188                          */
5189                         nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
5190                         if (nla)
5191                                 r_cfg.fc_encap_type = nla_get_u16(nla);
5192                 }
5193
5194                 r_cfg.fc_flags |= (rtnh->rtnh_flags & RTNH_F_ONLINK);
5195                 rt = ip6_route_info_create(&r_cfg, GFP_KERNEL, extack);
5196                 if (IS_ERR(rt)) {
5197                         err = PTR_ERR(rt);
5198                         rt = NULL;
5199                         goto cleanup;
5200                 }
5201                 if (!rt6_qualify_for_ecmp(rt)) {
5202                         err = -EINVAL;
5203                         NL_SET_ERR_MSG(extack,
5204                                        "Device only routes can not be added for IPv6 using the multipath API.");
5205                         fib6_info_release(rt);
5206                         goto cleanup;
5207                 }
5208
5209                 rt->fib6_nh->fib_nh_weight = rtnh->rtnh_hops + 1;
5210
5211                 err = ip6_route_info_append(info->nl_net, &rt6_nh_list,
5212                                             rt, &r_cfg);
5213                 if (err) {
5214                         fib6_info_release(rt);
5215                         goto cleanup;
5216                 }
5217
5218                 rtnh = rtnh_next(rtnh, &remaining);
5219         }
5220
5221         if (list_empty(&rt6_nh_list)) {
5222                 NL_SET_ERR_MSG(extack,
5223                                "Invalid nexthop configuration - no valid nexthops");
5224                 return -EINVAL;
5225         }
5226
5227         /* for add and replace send one notification with all nexthops.
5228          * Skip the notification in fib6_add_rt2node and send one with
5229          * the full route when done
5230          */
5231         info->skip_notify = 1;
5232
5233         /* For add and replace, send one notification with all nexthops. For
5234          * append, send one notification with all appended nexthops.
5235          */
5236         info->skip_notify_kernel = 1;
5237
5238         err_nh = NULL;
5239         list_for_each_entry(nh, &rt6_nh_list, next) {
5240                 err = __ip6_ins_rt(nh->fib6_info, info, extack);
5241                 fib6_info_release(nh->fib6_info);
5242
5243                 if (!err) {
5244                         /* save reference to last route successfully inserted */
5245                         rt_last = nh->fib6_info;
5246
5247                         /* save reference to first route for notification */
5248                         if (!rt_notif)
5249                                 rt_notif = nh->fib6_info;
5250                 }
5251
5252                 /* nh->fib6_info is used or freed at this point, reset to NULL*/
5253                 nh->fib6_info = NULL;
5254                 if (err) {
5255                         if (replace && nhn)
5256                                 NL_SET_ERR_MSG_MOD(extack,
5257                                                    "multipath route replace failed (check consistency of installed routes)");
5258                         err_nh = nh;
5259                         goto add_errout;
5260                 }
5261
5262                 /* Because each route is added like a single route we remove
5263                  * these flags after the first nexthop: if there is a collision,
5264                  * we have already failed to add the first nexthop:
5265                  * fib6_add_rt2node() has rejected it; when replacing, old
5266                  * nexthops have been replaced by first new, the rest should
5267                  * be added to it.
5268                  */
5269                 if (cfg->fc_nlinfo.nlh) {
5270                         cfg->fc_nlinfo.nlh->nlmsg_flags &= ~(NLM_F_EXCL |
5271                                                              NLM_F_REPLACE);
5272                         cfg->fc_nlinfo.nlh->nlmsg_flags |= NLM_F_CREATE;
5273                 }
5274                 nhn++;
5275         }
5276
5277         /* An in-kernel notification should only be sent in case the new
5278          * multipath route is added as the first route in the node, or if
5279          * it was appended to it. We pass 'rt_notif' since it is the first
5280          * sibling and might allow us to skip some checks in the replace case.
5281          */
5282         if (ip6_route_mpath_should_notify(rt_notif)) {
5283                 enum fib_event_type fib_event;
5284
5285                 if (rt_notif->fib6_nsiblings != nhn - 1)
5286                         fib_event = FIB_EVENT_ENTRY_APPEND;
5287                 else
5288                         fib_event = FIB_EVENT_ENTRY_REPLACE;
5289
5290                 err = call_fib6_multipath_entry_notifiers(info->nl_net,
5291                                                           fib_event, rt_notif,
5292                                                           nhn - 1, extack);
5293                 if (err) {
5294                         /* Delete all the siblings that were just added */
5295                         err_nh = NULL;
5296                         goto add_errout;
5297                 }
5298         }
5299
5300         /* success ... tell user about new route */
5301         ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5302         goto cleanup;
5303
5304 add_errout:
5305         /* send notification for routes that were added so that
5306          * the delete notifications sent by ip6_route_del are
5307          * coherent
5308          */
5309         if (rt_notif)
5310                 ip6_route_mpath_notify(rt_notif, rt_last, info, nlflags);
5311
5312         /* Delete routes that were already added */
5313         list_for_each_entry(nh, &rt6_nh_list, next) {
5314                 if (err_nh == nh)
5315                         break;
5316                 ip6_route_del(&nh->r_cfg, extack);
5317         }
5318
5319 cleanup:
5320         list_for_each_entry_safe(nh, nh_safe, &rt6_nh_list, next) {
5321                 if (nh->fib6_info)
5322                         fib6_info_release(nh->fib6_info);
5323                 list_del(&nh->next);
5324                 kfree(nh);
5325         }
5326
5327         return err;
5328 }
5329
5330 static int ip6_route_multipath_del(struct fib6_config *cfg,
5331                                    struct netlink_ext_ack *extack)
5332 {
5333         struct fib6_config r_cfg;
5334         struct rtnexthop *rtnh;
5335         int last_err = 0;
5336         int remaining;
5337         int attrlen;
5338         int err;
5339
5340         remaining = cfg->fc_mp_len;
5341         rtnh = (struct rtnexthop *)cfg->fc_mp;
5342
5343         /* Parse a Multipath Entry */
5344         while (rtnh_ok(rtnh, remaining)) {
5345                 memcpy(&r_cfg, cfg, sizeof(*cfg));
5346                 if (rtnh->rtnh_ifindex)
5347                         r_cfg.fc_ifindex = rtnh->rtnh_ifindex;
5348
5349                 attrlen = rtnh_attrlen(rtnh);
5350                 if (attrlen > 0) {
5351                         struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
5352
5353                         nla = nla_find(attrs, attrlen, RTA_GATEWAY);
5354                         if (nla) {
5355                                 err = fib6_gw_from_attr(&r_cfg.fc_gateway, nla,
5356                                                         extack);
5357                                 if (err) {
5358                                         last_err = err;
5359                                         goto next_rtnh;
5360                                 }
5361
5362                                 r_cfg.fc_flags |= RTF_GATEWAY;
5363                         }
5364                 }
5365                 err = ip6_route_del(&r_cfg, extack);
5366                 if (err)
5367                         last_err = err;
5368
5369 next_rtnh:
5370                 rtnh = rtnh_next(rtnh, &remaining);
5371         }
5372
5373         return last_err;
5374 }
5375
5376 static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5377                               struct netlink_ext_ack *extack)
5378 {
5379         struct fib6_config cfg;
5380         int err;
5381
5382         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5383         if (err < 0)
5384                 return err;
5385
5386         if (cfg.fc_nh_id &&
5387             !nexthop_find_by_id(sock_net(skb->sk), cfg.fc_nh_id)) {
5388                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
5389                 return -EINVAL;
5390         }
5391
5392         if (cfg.fc_mp)
5393                 return ip6_route_multipath_del(&cfg, extack);
5394         else {
5395                 cfg.fc_delete_all_nh = 1;
5396                 return ip6_route_del(&cfg, extack);
5397         }
5398 }
5399
5400 static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
5401                               struct netlink_ext_ack *extack)
5402 {
5403         struct fib6_config cfg;
5404         int err;
5405
5406         err = rtm_to_fib6_config(skb, nlh, &cfg, extack);
5407         if (err < 0)
5408                 return err;
5409
5410         if (cfg.fc_metric == 0)
5411                 cfg.fc_metric = IP6_RT_PRIO_USER;
5412
5413         if (cfg.fc_mp)
5414                 return ip6_route_multipath_add(&cfg, extack);
5415         else
5416                 return ip6_route_add(&cfg, GFP_KERNEL, extack);
5417 }
5418
5419 /* add the overhead of this fib6_nh to nexthop_len */
5420 static int rt6_nh_nlmsg_size(struct fib6_nh *nh, void *arg)
5421 {
5422         int *nexthop_len = arg;
5423
5424         *nexthop_len += nla_total_size(0)        /* RTA_MULTIPATH */
5425                      + NLA_ALIGN(sizeof(struct rtnexthop))
5426                      + nla_total_size(16); /* RTA_GATEWAY */
5427
5428         if (nh->fib_nh_lws) {
5429                 /* RTA_ENCAP_TYPE */
5430                 *nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5431                 /* RTA_ENCAP */
5432                 *nexthop_len += nla_total_size(2);
5433         }
5434
5435         return 0;
5436 }
5437
5438 static size_t rt6_nlmsg_size(struct fib6_info *f6i)
5439 {
5440         int nexthop_len;
5441
5442         if (f6i->nh) {
5443                 nexthop_len = nla_total_size(4); /* RTA_NH_ID */
5444                 nexthop_for_each_fib6_nh(f6i->nh, rt6_nh_nlmsg_size,
5445                                          &nexthop_len);
5446         } else {
5447                 struct fib6_nh *nh = f6i->fib6_nh;
5448
5449                 nexthop_len = 0;
5450                 if (f6i->fib6_nsiblings) {
5451                         nexthop_len = nla_total_size(0)  /* RTA_MULTIPATH */
5452                                     + NLA_ALIGN(sizeof(struct rtnexthop))
5453                                     + nla_total_size(16) /* RTA_GATEWAY */
5454                                     + lwtunnel_get_encap_size(nh->fib_nh_lws);
5455
5456                         nexthop_len *= f6i->fib6_nsiblings;
5457                 }
5458                 nexthop_len += lwtunnel_get_encap_size(nh->fib_nh_lws);
5459         }
5460
5461         return NLMSG_ALIGN(sizeof(struct rtmsg))
5462                + nla_total_size(16) /* RTA_SRC */
5463                + nla_total_size(16) /* RTA_DST */
5464                + nla_total_size(16) /* RTA_GATEWAY */
5465                + nla_total_size(16) /* RTA_PREFSRC */
5466                + nla_total_size(4) /* RTA_TABLE */
5467                + nla_total_size(4) /* RTA_IIF */
5468                + nla_total_size(4) /* RTA_OIF */
5469                + nla_total_size(4) /* RTA_PRIORITY */
5470                + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */
5471                + nla_total_size(sizeof(struct rta_cacheinfo))
5472                + nla_total_size(TCP_CA_NAME_MAX) /* RTAX_CC_ALGO */
5473                + nla_total_size(1) /* RTA_PREF */
5474                + nexthop_len;
5475 }
5476
5477 static int rt6_fill_node_nexthop(struct sk_buff *skb, struct nexthop *nh,
5478                                  unsigned char *flags)
5479 {
5480         if (nexthop_is_multipath(nh)) {
5481                 struct nlattr *mp;
5482
5483                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5484                 if (!mp)
5485                         goto nla_put_failure;
5486
5487                 if (nexthop_mpath_fill_node(skb, nh, AF_INET6))
5488                         goto nla_put_failure;
5489
5490                 nla_nest_end(skb, mp);
5491         } else {
5492                 struct fib6_nh *fib6_nh;
5493
5494                 fib6_nh = nexthop_fib6_nh(nh);
5495                 if (fib_nexthop_info(skb, &fib6_nh->nh_common, AF_INET6,
5496                                      flags, false) < 0)
5497                         goto nla_put_failure;
5498         }
5499
5500         return 0;
5501
5502 nla_put_failure:
5503         return -EMSGSIZE;
5504 }
5505
5506 static int rt6_fill_node(struct net *net, struct sk_buff *skb,
5507                          struct fib6_info *rt, struct dst_entry *dst,
5508                          struct in6_addr *dest, struct in6_addr *src,
5509                          int iif, int type, u32 portid, u32 seq,
5510                          unsigned int flags)
5511 {
5512         struct rt6_info *rt6 = (struct rt6_info *)dst;
5513         struct rt6key *rt6_dst, *rt6_src;
5514         u32 *pmetrics, table, rt6_flags;
5515         unsigned char nh_flags = 0;
5516         struct nlmsghdr *nlh;
5517         struct rtmsg *rtm;
5518         long expires = 0;
5519
5520         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*rtm), flags);
5521         if (!nlh)
5522                 return -EMSGSIZE;
5523
5524         if (rt6) {
5525                 rt6_dst = &rt6->rt6i_dst;
5526                 rt6_src = &rt6->rt6i_src;
5527                 rt6_flags = rt6->rt6i_flags;
5528         } else {
5529                 rt6_dst = &rt->fib6_dst;
5530                 rt6_src = &rt->fib6_src;
5531                 rt6_flags = rt->fib6_flags;
5532         }
5533
5534         rtm = nlmsg_data(nlh);
5535         rtm->rtm_family = AF_INET6;
5536         rtm->rtm_dst_len = rt6_dst->plen;
5537         rtm->rtm_src_len = rt6_src->plen;
5538         rtm->rtm_tos = 0;
5539         if (rt->fib6_table)
5540                 table = rt->fib6_table->tb6_id;
5541         else
5542                 table = RT6_TABLE_UNSPEC;
5543         rtm->rtm_table = table < 256 ? table : RT_TABLE_COMPAT;
5544         if (nla_put_u32(skb, RTA_TABLE, table))
5545                 goto nla_put_failure;
5546
5547         rtm->rtm_type = rt->fib6_type;
5548         rtm->rtm_flags = 0;
5549         rtm->rtm_scope = RT_SCOPE_UNIVERSE;
5550         rtm->rtm_protocol = rt->fib6_protocol;
5551
5552         if (rt6_flags & RTF_CACHE)
5553                 rtm->rtm_flags |= RTM_F_CLONED;
5554
5555         if (dest) {
5556                 if (nla_put_in6_addr(skb, RTA_DST, dest))
5557                         goto nla_put_failure;
5558                 rtm->rtm_dst_len = 128;
5559         } else if (rtm->rtm_dst_len)
5560                 if (nla_put_in6_addr(skb, RTA_DST, &rt6_dst->addr))
5561                         goto nla_put_failure;
5562 #ifdef CONFIG_IPV6_SUBTREES
5563         if (src) {
5564                 if (nla_put_in6_addr(skb, RTA_SRC, src))
5565                         goto nla_put_failure;
5566                 rtm->rtm_src_len = 128;
5567         } else if (rtm->rtm_src_len &&
5568                    nla_put_in6_addr(skb, RTA_SRC, &rt6_src->addr))
5569                 goto nla_put_failure;
5570 #endif
5571         if (iif) {
5572 #ifdef CONFIG_IPV6_MROUTE
5573                 if (ipv6_addr_is_multicast(&rt6_dst->addr)) {
5574                         int err = ip6mr_get_route(net, skb, rtm, portid);
5575
5576                         if (err == 0)
5577                                 return 0;
5578                         if (err < 0)
5579                                 goto nla_put_failure;
5580                 } else
5581 #endif
5582                         if (nla_put_u32(skb, RTA_IIF, iif))
5583                                 goto nla_put_failure;
5584         } else if (dest) {
5585                 struct in6_addr saddr_buf;
5586                 if (ip6_route_get_saddr(net, rt, dest, 0, &saddr_buf) == 0 &&
5587                     nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5588                         goto nla_put_failure;
5589         }
5590
5591         if (rt->fib6_prefsrc.plen) {
5592                 struct in6_addr saddr_buf;
5593                 saddr_buf = rt->fib6_prefsrc.addr;
5594                 if (nla_put_in6_addr(skb, RTA_PREFSRC, &saddr_buf))
5595                         goto nla_put_failure;
5596         }
5597
5598         pmetrics = dst ? dst_metrics_ptr(dst) : rt->fib6_metrics->metrics;
5599         if (rtnetlink_put_metrics(skb, pmetrics) < 0)
5600                 goto nla_put_failure;
5601
5602         if (nla_put_u32(skb, RTA_PRIORITY, rt->fib6_metric))
5603                 goto nla_put_failure;
5604
5605         /* For multipath routes, walk the siblings list and add
5606          * each as a nexthop within RTA_MULTIPATH.
5607          */
5608         if (rt6) {
5609                 if (rt6_flags & RTF_GATEWAY &&
5610                     nla_put_in6_addr(skb, RTA_GATEWAY, &rt6->rt6i_gateway))
5611                         goto nla_put_failure;
5612
5613                 if (dst->dev && nla_put_u32(skb, RTA_OIF, dst->dev->ifindex))
5614                         goto nla_put_failure;
5615         } else if (rt->fib6_nsiblings) {
5616                 struct fib6_info *sibling, *next_sibling;
5617                 struct nlattr *mp;
5618
5619                 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
5620                 if (!mp)
5621                         goto nla_put_failure;
5622
5623                 if (fib_add_nexthop(skb, &rt->fib6_nh->nh_common,
5624                                     rt->fib6_nh->fib_nh_weight, AF_INET6,
5625                                     0) < 0)
5626                         goto nla_put_failure;
5627
5628                 list_for_each_entry_safe(sibling, next_sibling,
5629                                          &rt->fib6_siblings, fib6_siblings) {
5630                         if (fib_add_nexthop(skb, &sibling->fib6_nh->nh_common,
5631                                             sibling->fib6_nh->fib_nh_weight,
5632                                             AF_INET6, 0) < 0)
5633                                 goto nla_put_failure;
5634                 }
5635
5636                 nla_nest_end(skb, mp);
5637         } else if (rt->nh) {
5638                 if (nla_put_u32(skb, RTA_NH_ID, rt->nh->id))
5639                         goto nla_put_failure;
5640
5641                 if (nexthop_is_blackhole(rt->nh))
5642                         rtm->rtm_type = RTN_BLACKHOLE;
5643
5644                 if (READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode) &&
5645                     rt6_fill_node_nexthop(skb, rt->nh, &nh_flags) < 0)
5646                         goto nla_put_failure;
5647
5648                 rtm->rtm_flags |= nh_flags;
5649         } else {
5650                 if (fib_nexthop_info(skb, &rt->fib6_nh->nh_common, AF_INET6,
5651                                      &nh_flags, false) < 0)
5652                         goto nla_put_failure;
5653
5654                 rtm->rtm_flags |= nh_flags;
5655         }
5656
5657         if (rt6_flags & RTF_EXPIRES) {
5658                 expires = dst ? dst->expires : rt->expires;
5659                 expires -= jiffies;
5660         }
5661
5662         if (!dst) {
5663                 if (rt->offload)
5664                         rtm->rtm_flags |= RTM_F_OFFLOAD;
5665                 if (rt->trap)
5666                         rtm->rtm_flags |= RTM_F_TRAP;
5667         }
5668
5669         if (rtnl_put_cacheinfo(skb, dst, 0, expires, dst ? dst->error : 0) < 0)
5670                 goto nla_put_failure;
5671
5672         if (nla_put_u8(skb, RTA_PREF, IPV6_EXTRACT_PREF(rt6_flags)))
5673                 goto nla_put_failure;
5674
5675
5676         nlmsg_end(skb, nlh);
5677         return 0;
5678
5679 nla_put_failure:
5680         nlmsg_cancel(skb, nlh);
5681         return -EMSGSIZE;
5682 }
5683
5684 static int fib6_info_nh_uses_dev(struct fib6_nh *nh, void *arg)
5685 {
5686         const struct net_device *dev = arg;
5687
5688         if (nh->fib_nh_dev == dev)
5689                 return 1;
5690
5691         return 0;
5692 }
5693
5694 static bool fib6_info_uses_dev(const struct fib6_info *f6i,
5695                                const struct net_device *dev)
5696 {
5697         if (f6i->nh) {
5698                 struct net_device *_dev = (struct net_device *)dev;
5699
5700                 return !!nexthop_for_each_fib6_nh(f6i->nh,
5701                                                   fib6_info_nh_uses_dev,
5702                                                   _dev);
5703         }
5704
5705         if (f6i->fib6_nh->fib_nh_dev == dev)
5706                 return true;
5707
5708         if (f6i->fib6_nsiblings) {
5709                 struct fib6_info *sibling, *next_sibling;
5710
5711                 list_for_each_entry_safe(sibling, next_sibling,
5712                                          &f6i->fib6_siblings, fib6_siblings) {
5713                         if (sibling->fib6_nh->fib_nh_dev == dev)
5714                                 return true;
5715                 }
5716         }
5717
5718         return false;
5719 }
5720
5721 struct fib6_nh_exception_dump_walker {
5722         struct rt6_rtnl_dump_arg *dump;
5723         struct fib6_info *rt;
5724         unsigned int flags;
5725         unsigned int skip;
5726         unsigned int count;
5727 };
5728
5729 static int rt6_nh_dump_exceptions(struct fib6_nh *nh, void *arg)
5730 {
5731         struct fib6_nh_exception_dump_walker *w = arg;
5732         struct rt6_rtnl_dump_arg *dump = w->dump;
5733         struct rt6_exception_bucket *bucket;
5734         struct rt6_exception *rt6_ex;
5735         int i, err;
5736
5737         bucket = fib6_nh_get_excptn_bucket(nh, NULL);
5738         if (!bucket)
5739                 return 0;
5740
5741         for (i = 0; i < FIB6_EXCEPTION_BUCKET_SIZE; i++) {
5742                 hlist_for_each_entry(rt6_ex, &bucket->chain, hlist) {
5743                         if (w->skip) {
5744                                 w->skip--;
5745                                 continue;
5746                         }
5747
5748                         /* Expiration of entries doesn't bump sernum, insertion
5749                          * does. Removal is triggered by insertion, so we can
5750                          * rely on the fact that if entries change between two
5751                          * partial dumps, this node is scanned again completely,
5752                          * see rt6_insert_exception() and fib6_dump_table().
5753                          *
5754                          * Count expired entries we go through as handled
5755                          * entries that we'll skip next time, in case of partial
5756                          * node dump. Otherwise, if entries expire meanwhile,
5757                          * we'll skip the wrong amount.
5758                          */
5759                         if (rt6_check_expired(rt6_ex->rt6i)) {
5760                                 w->count++;
5761                                 continue;
5762                         }
5763
5764                         err = rt6_fill_node(dump->net, dump->skb, w->rt,
5765                                             &rt6_ex->rt6i->dst, NULL, NULL, 0,
5766                                             RTM_NEWROUTE,
5767                                             NETLINK_CB(dump->cb->skb).portid,
5768                                             dump->cb->nlh->nlmsg_seq, w->flags);
5769                         if (err)
5770                                 return err;
5771
5772                         w->count++;
5773                 }
5774                 bucket++;
5775         }
5776
5777         return 0;
5778 }
5779
5780 /* Return -1 if done with node, number of handled routes on partial dump */
5781 int rt6_dump_route(struct fib6_info *rt, void *p_arg, unsigned int skip)
5782 {
5783         struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg;
5784         struct fib_dump_filter *filter = &arg->filter;
5785         unsigned int flags = NLM_F_MULTI;
5786         struct net *net = arg->net;
5787         int count = 0;
5788
5789         if (rt == net->ipv6.fib6_null_entry)
5790                 return -1;
5791
5792         if ((filter->flags & RTM_F_PREFIX) &&
5793             !(rt->fib6_flags & RTF_PREFIX_RT)) {
5794                 /* success since this is not a prefix route */
5795                 return -1;
5796         }
5797         if (filter->filter_set &&
5798             ((filter->rt_type  && rt->fib6_type != filter->rt_type) ||
5799              (filter->dev      && !fib6_info_uses_dev(rt, filter->dev)) ||
5800              (filter->protocol && rt->fib6_protocol != filter->protocol))) {
5801                 return -1;
5802         }
5803
5804         if (filter->filter_set ||
5805             !filter->dump_routes || !filter->dump_exceptions) {
5806                 flags |= NLM_F_DUMP_FILTERED;
5807         }
5808
5809         if (filter->dump_routes) {
5810                 if (skip) {
5811                         skip--;
5812                 } else {
5813                         if (rt6_fill_node(net, arg->skb, rt, NULL, NULL, NULL,
5814                                           0, RTM_NEWROUTE,
5815                                           NETLINK_CB(arg->cb->skb).portid,
5816                                           arg->cb->nlh->nlmsg_seq, flags)) {
5817                                 return 0;
5818                         }
5819                         count++;
5820                 }
5821         }
5822
5823         if (filter->dump_exceptions) {
5824                 struct fib6_nh_exception_dump_walker w = { .dump = arg,
5825                                                            .rt = rt,
5826                                                            .flags = flags,
5827                                                            .skip = skip,
5828                                                            .count = 0 };
5829                 int err;
5830
5831                 rcu_read_lock();
5832                 if (rt->nh) {
5833                         err = nexthop_for_each_fib6_nh(rt->nh,
5834                                                        rt6_nh_dump_exceptions,
5835                                                        &w);
5836                 } else {
5837                         err = rt6_nh_dump_exceptions(rt->fib6_nh, &w);
5838                 }
5839                 rcu_read_unlock();
5840
5841                 if (err)
5842                         return count += w.count;
5843         }
5844
5845         return -1;
5846 }
5847
5848 static int inet6_rtm_valid_getroute_req(struct sk_buff *skb,
5849                                         const struct nlmsghdr *nlh,
5850                                         struct nlattr **tb,
5851                                         struct netlink_ext_ack *extack)
5852 {
5853         struct rtmsg *rtm;
5854         int i, err;
5855
5856         if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
5857                 NL_SET_ERR_MSG_MOD(extack,
5858                                    "Invalid header for get route request");
5859                 return -EINVAL;
5860         }
5861
5862         if (!netlink_strict_get_check(skb))
5863                 return nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
5864                                               rtm_ipv6_policy, extack);
5865
5866         rtm = nlmsg_data(nlh);
5867         if ((rtm->rtm_src_len && rtm->rtm_src_len != 128) ||
5868             (rtm->rtm_dst_len && rtm->rtm_dst_len != 128) ||
5869             rtm->rtm_table || rtm->rtm_protocol || rtm->rtm_scope ||
5870             rtm->rtm_type) {
5871                 NL_SET_ERR_MSG_MOD(extack, "Invalid values in header for get route request");
5872                 return -EINVAL;
5873         }
5874         if (rtm->rtm_flags & ~RTM_F_FIB_MATCH) {
5875                 NL_SET_ERR_MSG_MOD(extack,
5876                                    "Invalid flags for get route request");
5877                 return -EINVAL;
5878         }
5879
5880         err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
5881                                             rtm_ipv6_policy, extack);
5882         if (err)
5883                 return err;
5884
5885         if ((tb[RTA_SRC] && !rtm->rtm_src_len) ||
5886             (tb[RTA_DST] && !rtm->rtm_dst_len)) {
5887                 NL_SET_ERR_MSG_MOD(extack, "rtm_src_len and rtm_dst_len must be 128 for IPv6");
5888                 return -EINVAL;
5889         }
5890
5891         for (i = 0; i <= RTA_MAX; i++) {
5892                 if (!tb[i])
5893                         continue;
5894
5895                 switch (i) {
5896                 case RTA_SRC:
5897                 case RTA_DST:
5898                 case RTA_IIF:
5899                 case RTA_OIF:
5900                 case RTA_MARK:
5901                 case RTA_UID:
5902                 case RTA_SPORT:
5903                 case RTA_DPORT:
5904                 case RTA_IP_PROTO:
5905                         break;
5906                 default:
5907                         NL_SET_ERR_MSG_MOD(extack, "Unsupported attribute in get route request");
5908                         return -EINVAL;
5909                 }
5910         }
5911
5912         return 0;
5913 }
5914
5915 static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
5916                               struct netlink_ext_ack *extack)
5917 {
5918         struct net *net = sock_net(in_skb->sk);
5919         struct nlattr *tb[RTA_MAX+1];
5920         int err, iif = 0, oif = 0;
5921         struct fib6_info *from;
5922         struct dst_entry *dst;
5923         struct rt6_info *rt;
5924         struct sk_buff *skb;
5925         struct rtmsg *rtm;
5926         struct flowi6 fl6 = {};
5927         bool fibmatch;
5928
5929         err = inet6_rtm_valid_getroute_req(in_skb, nlh, tb, extack);
5930         if (err < 0)
5931                 goto errout;
5932
5933         err = -EINVAL;
5934         rtm = nlmsg_data(nlh);
5935         fl6.flowlabel = ip6_make_flowinfo(rtm->rtm_tos, 0);
5936         fibmatch = !!(rtm->rtm_flags & RTM_F_FIB_MATCH);
5937
5938         if (tb[RTA_SRC]) {
5939                 if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr))
5940                         goto errout;
5941
5942                 fl6.saddr = *(struct in6_addr *)nla_data(tb[RTA_SRC]);
5943         }
5944
5945         if (tb[RTA_DST]) {
5946                 if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr))
5947                         goto errout;
5948
5949                 fl6.daddr = *(struct in6_addr *)nla_data(tb[RTA_DST]);
5950         }
5951
5952         if (tb[RTA_IIF])
5953                 iif = nla_get_u32(tb[RTA_IIF]);
5954
5955         if (tb[RTA_OIF])
5956                 oif = nla_get_u32(tb[RTA_OIF]);
5957
5958         if (tb[RTA_MARK])
5959                 fl6.flowi6_mark = nla_get_u32(tb[RTA_MARK]);
5960
5961         if (tb[RTA_UID])
5962                 fl6.flowi6_uid = make_kuid(current_user_ns(),
5963                                            nla_get_u32(tb[RTA_UID]));
5964         else
5965                 fl6.flowi6_uid = iif ? INVALID_UID : current_uid();
5966
5967         if (tb[RTA_SPORT])
5968                 fl6.fl6_sport = nla_get_be16(tb[RTA_SPORT]);
5969
5970         if (tb[RTA_DPORT])
5971                 fl6.fl6_dport = nla_get_be16(tb[RTA_DPORT]);
5972
5973         if (tb[RTA_IP_PROTO]) {
5974                 err = rtm_getroute_parse_ip_proto(tb[RTA_IP_PROTO],
5975                                                   &fl6.flowi6_proto, AF_INET6,
5976                                                   extack);
5977                 if (err)
5978                         goto errout;
5979         }
5980
5981         if (iif) {
5982                 struct net_device *dev;
5983                 int flags = 0;
5984
5985                 rcu_read_lock();
5986
5987                 dev = dev_get_by_index_rcu(net, iif);
5988                 if (!dev) {
5989                         rcu_read_unlock();
5990                         err = -ENODEV;
5991                         goto errout;
5992                 }
5993
5994                 fl6.flowi6_iif = iif;
5995
5996                 if (!ipv6_addr_any(&fl6.saddr))
5997                         flags |= RT6_LOOKUP_F_HAS_SADDR;
5998
5999                 dst = ip6_route_input_lookup(net, dev, &fl6, NULL, flags);
6000
6001                 rcu_read_unlock();
6002         } else {
6003                 fl6.flowi6_oif = oif;
6004
6005                 dst = ip6_route_output(net, NULL, &fl6);
6006         }
6007
6008
6009         rt = container_of(dst, struct rt6_info, dst);
6010         if (rt->dst.error) {
6011                 err = rt->dst.error;
6012                 ip6_rt_put(rt);
6013                 goto errout;
6014         }
6015
6016         if (rt == net->ipv6.ip6_null_entry) {
6017                 err = rt->dst.error;
6018                 ip6_rt_put(rt);
6019                 goto errout;
6020         }
6021
6022         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
6023         if (!skb) {
6024                 ip6_rt_put(rt);
6025                 err = -ENOBUFS;
6026                 goto errout;
6027         }
6028
6029         skb_dst_set(skb, &rt->dst);
6030
6031         rcu_read_lock();
6032         from = rcu_dereference(rt->from);
6033         if (from) {
6034                 if (fibmatch)
6035                         err = rt6_fill_node(net, skb, from, NULL, NULL, NULL,
6036                                             iif, RTM_NEWROUTE,
6037                                             NETLINK_CB(in_skb).portid,
6038                                             nlh->nlmsg_seq, 0);
6039                 else
6040                         err = rt6_fill_node(net, skb, from, dst, &fl6.daddr,
6041                                             &fl6.saddr, iif, RTM_NEWROUTE,
6042                                             NETLINK_CB(in_skb).portid,
6043                                             nlh->nlmsg_seq, 0);
6044         } else {
6045                 err = -ENETUNREACH;
6046         }
6047         rcu_read_unlock();
6048
6049         if (err < 0) {
6050                 kfree_skb(skb);
6051                 goto errout;
6052         }
6053
6054         err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
6055 errout:
6056         return err;
6057 }
6058
6059 void inet6_rt_notify(int event, struct fib6_info *rt, struct nl_info *info,
6060                      unsigned int nlm_flags)
6061 {
6062         struct sk_buff *skb;
6063         struct net *net = info->nl_net;
6064         u32 seq;
6065         int err;
6066
6067         err = -ENOBUFS;
6068         seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6069
6070         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6071         if (!skb)
6072                 goto errout;
6073
6074         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6075                             event, info->portid, seq, nlm_flags);
6076         if (err < 0) {
6077                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6078                 WARN_ON(err == -EMSGSIZE);
6079                 kfree_skb(skb);
6080                 goto errout;
6081         }
6082         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6083                     info->nlh, gfp_any());
6084         return;
6085 errout:
6086         if (err < 0)
6087                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6088 }
6089
6090 void fib6_rt_update(struct net *net, struct fib6_info *rt,
6091                     struct nl_info *info)
6092 {
6093         u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
6094         struct sk_buff *skb;
6095         int err = -ENOBUFS;
6096
6097         /* call_fib6_entry_notifiers will be removed when in-kernel notifier
6098          * is implemented and supported for nexthop objects
6099          */
6100         call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_REPLACE, rt, NULL);
6101
6102         skb = nlmsg_new(rt6_nlmsg_size(rt), gfp_any());
6103         if (!skb)
6104                 goto errout;
6105
6106         err = rt6_fill_node(net, skb, rt, NULL, NULL, NULL, 0,
6107                             RTM_NEWROUTE, info->portid, seq, NLM_F_REPLACE);
6108         if (err < 0) {
6109                 /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */
6110                 WARN_ON(err == -EMSGSIZE);
6111                 kfree_skb(skb);
6112                 goto errout;
6113         }
6114         rtnl_notify(skb, net, info->portid, RTNLGRP_IPV6_ROUTE,
6115                     info->nlh, gfp_any());
6116         return;
6117 errout:
6118         if (err < 0)
6119                 rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err);
6120 }
6121
6122 static int ip6_route_dev_notify(struct notifier_block *this,
6123                                 unsigned long event, void *ptr)
6124 {
6125         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
6126         struct net *net = dev_net(dev);
6127
6128         if (!(dev->flags & IFF_LOOPBACK))
6129                 return NOTIFY_OK;
6130
6131         if (event == NETDEV_REGISTER) {
6132                 net->ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = dev;
6133                 net->ipv6.ip6_null_entry->dst.dev = dev;
6134                 net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev);
6135 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6136                 net->ipv6.ip6_prohibit_entry->dst.dev = dev;
6137                 net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev);
6138                 net->ipv6.ip6_blk_hole_entry->dst.dev = dev;
6139                 net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev);
6140 #endif
6141          } else if (event == NETDEV_UNREGISTER &&
6142                     dev->reg_state != NETREG_UNREGISTERED) {
6143                 /* NETDEV_UNREGISTER could be fired for multiple times by
6144                  * netdev_wait_allrefs(). Make sure we only call this once.
6145                  */
6146                 in6_dev_put_clear(&net->ipv6.ip6_null_entry->rt6i_idev);
6147 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6148                 in6_dev_put_clear(&net->ipv6.ip6_prohibit_entry->rt6i_idev);
6149                 in6_dev_put_clear(&net->ipv6.ip6_blk_hole_entry->rt6i_idev);
6150 #endif
6151         }
6152
6153         return NOTIFY_OK;
6154 }
6155
6156 /*
6157  *      /proc
6158  */
6159
6160 #ifdef CONFIG_PROC_FS
6161 static int rt6_stats_seq_show(struct seq_file *seq, void *v)
6162 {
6163         struct net *net = (struct net *)seq->private;
6164         seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n",
6165                    net->ipv6.rt6_stats->fib_nodes,
6166                    net->ipv6.rt6_stats->fib_route_nodes,
6167                    atomic_read(&net->ipv6.rt6_stats->fib_rt_alloc),
6168                    net->ipv6.rt6_stats->fib_rt_entries,
6169                    net->ipv6.rt6_stats->fib_rt_cache,
6170                    dst_entries_get_slow(&net->ipv6.ip6_dst_ops),
6171                    net->ipv6.rt6_stats->fib_discarded_routes);
6172
6173         return 0;
6174 }
6175 #endif  /* CONFIG_PROC_FS */
6176
6177 #ifdef CONFIG_SYSCTL
6178
6179 static int ipv6_sysctl_rtcache_flush(struct ctl_table *ctl, int write,
6180                               void *buffer, size_t *lenp, loff_t *ppos)
6181 {
6182         struct net *net;
6183         int delay;
6184         int ret;
6185         if (!write)
6186                 return -EINVAL;
6187
6188         net = (struct net *)ctl->extra1;
6189         delay = net->ipv6.sysctl.flush_delay;
6190         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
6191         if (ret)
6192                 return ret;
6193
6194         fib6_run_gc(delay <= 0 ? 0 : (unsigned long)delay, net, delay > 0);
6195         return 0;
6196 }
6197
6198 static struct ctl_table ipv6_route_table_template[] = {
6199         {
6200                 .procname       =       "flush",
6201                 .data           =       &init_net.ipv6.sysctl.flush_delay,
6202                 .maxlen         =       sizeof(int),
6203                 .mode           =       0200,
6204                 .proc_handler   =       ipv6_sysctl_rtcache_flush
6205         },
6206         {
6207                 .procname       =       "gc_thresh",
6208                 .data           =       &ip6_dst_ops_template.gc_thresh,
6209                 .maxlen         =       sizeof(int),
6210                 .mode           =       0644,
6211                 .proc_handler   =       proc_dointvec,
6212         },
6213         {
6214                 .procname       =       "max_size",
6215                 .data           =       &init_net.ipv6.sysctl.ip6_rt_max_size,
6216                 .maxlen         =       sizeof(int),
6217                 .mode           =       0644,
6218                 .proc_handler   =       proc_dointvec,
6219         },
6220         {
6221                 .procname       =       "gc_min_interval",
6222                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6223                 .maxlen         =       sizeof(int),
6224                 .mode           =       0644,
6225                 .proc_handler   =       proc_dointvec_jiffies,
6226         },
6227         {
6228                 .procname       =       "gc_timeout",
6229                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_timeout,
6230                 .maxlen         =       sizeof(int),
6231                 .mode           =       0644,
6232                 .proc_handler   =       proc_dointvec_jiffies,
6233         },
6234         {
6235                 .procname       =       "gc_interval",
6236                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_interval,
6237                 .maxlen         =       sizeof(int),
6238                 .mode           =       0644,
6239                 .proc_handler   =       proc_dointvec_jiffies,
6240         },
6241         {
6242                 .procname       =       "gc_elasticity",
6243                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_elasticity,
6244                 .maxlen         =       sizeof(int),
6245                 .mode           =       0644,
6246                 .proc_handler   =       proc_dointvec,
6247         },
6248         {
6249                 .procname       =       "mtu_expires",
6250                 .data           =       &init_net.ipv6.sysctl.ip6_rt_mtu_expires,
6251                 .maxlen         =       sizeof(int),
6252                 .mode           =       0644,
6253                 .proc_handler   =       proc_dointvec_jiffies,
6254         },
6255         {
6256                 .procname       =       "min_adv_mss",
6257                 .data           =       &init_net.ipv6.sysctl.ip6_rt_min_advmss,
6258                 .maxlen         =       sizeof(int),
6259                 .mode           =       0644,
6260                 .proc_handler   =       proc_dointvec,
6261         },
6262         {
6263                 .procname       =       "gc_min_interval_ms",
6264                 .data           =       &init_net.ipv6.sysctl.ip6_rt_gc_min_interval,
6265                 .maxlen         =       sizeof(int),
6266                 .mode           =       0644,
6267                 .proc_handler   =       proc_dointvec_ms_jiffies,
6268         },
6269         {
6270                 .procname       =       "skip_notify_on_dev_down",
6271                 .data           =       &init_net.ipv6.sysctl.skip_notify_on_dev_down,
6272                 .maxlen         =       sizeof(int),
6273                 .mode           =       0644,
6274                 .proc_handler   =       proc_dointvec_minmax,
6275                 .extra1         =       SYSCTL_ZERO,
6276                 .extra2         =       SYSCTL_ONE,
6277         },
6278         { }
6279 };
6280
6281 struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net)
6282 {
6283         struct ctl_table *table;
6284
6285         table = kmemdup(ipv6_route_table_template,
6286                         sizeof(ipv6_route_table_template),
6287                         GFP_KERNEL);
6288
6289         if (table) {
6290                 table[0].data = &net->ipv6.sysctl.flush_delay;
6291                 table[0].extra1 = net;
6292                 table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh;
6293                 table[2].data = &net->ipv6.sysctl.ip6_rt_max_size;
6294                 table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6295                 table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout;
6296                 table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval;
6297                 table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity;
6298                 table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires;
6299                 table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss;
6300                 table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval;
6301                 table[10].data = &net->ipv6.sysctl.skip_notify_on_dev_down;
6302
6303                 /* Don't export sysctls to unprivileged users */
6304                 if (net->user_ns != &init_user_ns)
6305                         table[0].procname = NULL;
6306         }
6307
6308         return table;
6309 }
6310 #endif
6311
6312 static int __net_init ip6_route_net_init(struct net *net)
6313 {
6314         int ret = -ENOMEM;
6315
6316         memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template,
6317                sizeof(net->ipv6.ip6_dst_ops));
6318
6319         if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0)
6320                 goto out_ip6_dst_ops;
6321
6322         net->ipv6.fib6_null_entry = fib6_info_alloc(GFP_KERNEL, true);
6323         if (!net->ipv6.fib6_null_entry)
6324                 goto out_ip6_dst_entries;
6325         memcpy(net->ipv6.fib6_null_entry, &fib6_null_entry_template,
6326                sizeof(*net->ipv6.fib6_null_entry));
6327
6328         net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template,
6329                                            sizeof(*net->ipv6.ip6_null_entry),
6330                                            GFP_KERNEL);
6331         if (!net->ipv6.ip6_null_entry)
6332                 goto out_fib6_null_entry;
6333         net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6334         dst_init_metrics(&net->ipv6.ip6_null_entry->dst,
6335                          ip6_template_metrics, true);
6336         INIT_LIST_HEAD(&net->ipv6.ip6_null_entry->rt6i_uncached);
6337
6338 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6339         net->ipv6.fib6_has_custom_rules = false;
6340         net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
6341                                                sizeof(*net->ipv6.ip6_prohibit_entry),
6342                                                GFP_KERNEL);
6343         if (!net->ipv6.ip6_prohibit_entry)
6344                 goto out_ip6_null_entry;
6345         net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6346         dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst,
6347                          ip6_template_metrics, true);
6348         INIT_LIST_HEAD(&net->ipv6.ip6_prohibit_entry->rt6i_uncached);
6349
6350         net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template,
6351                                                sizeof(*net->ipv6.ip6_blk_hole_entry),
6352                                                GFP_KERNEL);
6353         if (!net->ipv6.ip6_blk_hole_entry)
6354                 goto out_ip6_prohibit_entry;
6355         net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops;
6356         dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst,
6357                          ip6_template_metrics, true);
6358         INIT_LIST_HEAD(&net->ipv6.ip6_blk_hole_entry->rt6i_uncached);
6359 #ifdef CONFIG_IPV6_SUBTREES
6360         net->ipv6.fib6_routes_require_src = 0;
6361 #endif
6362 #endif
6363
6364         net->ipv6.sysctl.flush_delay = 0;
6365         net->ipv6.sysctl.ip6_rt_max_size = 4096;
6366         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
6367         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
6368         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
6369         net->ipv6.sysctl.ip6_rt_gc_elasticity = 9;
6370         net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ;
6371         net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40;
6372         net->ipv6.sysctl.skip_notify_on_dev_down = 0;
6373
6374         atomic_set(&net->ipv6.ip6_rt_gc_expire, 30*HZ);
6375
6376         ret = 0;
6377 out:
6378         return ret;
6379
6380 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6381 out_ip6_prohibit_entry:
6382         kfree(net->ipv6.ip6_prohibit_entry);
6383 out_ip6_null_entry:
6384         kfree(net->ipv6.ip6_null_entry);
6385 #endif
6386 out_fib6_null_entry:
6387         kfree(net->ipv6.fib6_null_entry);
6388 out_ip6_dst_entries:
6389         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6390 out_ip6_dst_ops:
6391         goto out;
6392 }
6393
6394 static void __net_exit ip6_route_net_exit(struct net *net)
6395 {
6396         kfree(net->ipv6.fib6_null_entry);
6397         kfree(net->ipv6.ip6_null_entry);
6398 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6399         kfree(net->ipv6.ip6_prohibit_entry);
6400         kfree(net->ipv6.ip6_blk_hole_entry);
6401 #endif
6402         dst_entries_destroy(&net->ipv6.ip6_dst_ops);
6403 }
6404
6405 static int __net_init ip6_route_net_init_late(struct net *net)
6406 {
6407 #ifdef CONFIG_PROC_FS
6408         proc_create_net("ipv6_route", 0, net->proc_net, &ipv6_route_seq_ops,
6409                         sizeof(struct ipv6_route_iter));
6410         proc_create_net_single("rt6_stats", 0444, net->proc_net,
6411                         rt6_stats_seq_show, NULL);
6412 #endif
6413         return 0;
6414 }
6415
6416 static void __net_exit ip6_route_net_exit_late(struct net *net)
6417 {
6418 #ifdef CONFIG_PROC_FS
6419         remove_proc_entry("ipv6_route", net->proc_net);
6420         remove_proc_entry("rt6_stats", net->proc_net);
6421 #endif
6422 }
6423
6424 static struct pernet_operations ip6_route_net_ops = {
6425         .init = ip6_route_net_init,
6426         .exit = ip6_route_net_exit,
6427 };
6428
6429 static int __net_init ipv6_inetpeer_init(struct net *net)
6430 {
6431         struct inet_peer_base *bp = kmalloc(sizeof(*bp), GFP_KERNEL);
6432
6433         if (!bp)
6434                 return -ENOMEM;
6435         inet_peer_base_init(bp);
6436         net->ipv6.peers = bp;
6437         return 0;
6438 }
6439
6440 static void __net_exit ipv6_inetpeer_exit(struct net *net)
6441 {
6442         struct inet_peer_base *bp = net->ipv6.peers;
6443
6444         net->ipv6.peers = NULL;
6445         inetpeer_invalidate_tree(bp);
6446         kfree(bp);
6447 }
6448
6449 static struct pernet_operations ipv6_inetpeer_ops = {
6450         .init   =       ipv6_inetpeer_init,
6451         .exit   =       ipv6_inetpeer_exit,
6452 };
6453
6454 static struct pernet_operations ip6_route_net_late_ops = {
6455         .init = ip6_route_net_init_late,
6456         .exit = ip6_route_net_exit_late,
6457 };
6458
6459 static struct notifier_block ip6_route_dev_notifier = {
6460         .notifier_call = ip6_route_dev_notify,
6461         .priority = ADDRCONF_NOTIFY_PRIORITY - 10,
6462 };
6463
6464 void __init ip6_route_init_special_entries(void)
6465 {
6466         /* Registering of the loopback is done before this portion of code,
6467          * the loopback reference in rt6_info will not be taken, do it
6468          * manually for init_net */
6469         init_net.ipv6.fib6_null_entry->fib6_nh->fib_nh_dev = init_net.loopback_dev;
6470         init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev;
6471         init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6472   #ifdef CONFIG_IPV6_MULTIPLE_TABLES
6473         init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev;
6474         init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6475         init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev;
6476         init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev);
6477   #endif
6478 }
6479
6480 #if IS_BUILTIN(CONFIG_IPV6)
6481 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6482 DEFINE_BPF_ITER_FUNC(ipv6_route, struct bpf_iter_meta *meta, struct fib6_info *rt)
6483
6484 BTF_ID_LIST(btf_fib6_info_id)
6485 BTF_ID(struct, fib6_info)
6486
6487 static const struct bpf_iter_seq_info ipv6_route_seq_info = {
6488         .seq_ops                = &ipv6_route_seq_ops,
6489         .init_seq_private       = bpf_iter_init_seq_net,
6490         .fini_seq_private       = bpf_iter_fini_seq_net,
6491         .seq_priv_size          = sizeof(struct ipv6_route_iter),
6492 };
6493
6494 static struct bpf_iter_reg ipv6_route_reg_info = {
6495         .target                 = "ipv6_route",
6496         .ctx_arg_info_size      = 1,
6497         .ctx_arg_info           = {
6498                 { offsetof(struct bpf_iter__ipv6_route, rt),
6499                   PTR_TO_BTF_ID_OR_NULL },
6500         },
6501         .seq_info               = &ipv6_route_seq_info,
6502 };
6503
6504 static int __init bpf_iter_register(void)
6505 {
6506         ipv6_route_reg_info.ctx_arg_info[0].btf_id = *btf_fib6_info_id;
6507         return bpf_iter_reg_target(&ipv6_route_reg_info);
6508 }
6509
6510 static void bpf_iter_unregister(void)
6511 {
6512         bpf_iter_unreg_target(&ipv6_route_reg_info);
6513 }
6514 #endif
6515 #endif
6516
6517 int __init ip6_route_init(void)
6518 {
6519         int ret;
6520         int cpu;
6521
6522         ret = -ENOMEM;
6523         ip6_dst_ops_template.kmem_cachep =
6524                 kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0,
6525                                   SLAB_HWCACHE_ALIGN, NULL);
6526         if (!ip6_dst_ops_template.kmem_cachep)
6527                 goto out;
6528
6529         ret = dst_entries_init(&ip6_dst_blackhole_ops);
6530         if (ret)
6531                 goto out_kmem_cache;
6532
6533         ret = register_pernet_subsys(&ipv6_inetpeer_ops);
6534         if (ret)
6535                 goto out_dst_entries;
6536
6537         ret = register_pernet_subsys(&ip6_route_net_ops);
6538         if (ret)
6539                 goto out_register_inetpeer;
6540
6541         ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep;
6542
6543         ret = fib6_init();
6544         if (ret)
6545                 goto out_register_subsys;
6546
6547         ret = xfrm6_init();
6548         if (ret)
6549                 goto out_fib6_init;
6550
6551         ret = fib6_rules_init();
6552         if (ret)
6553                 goto xfrm6_init;
6554
6555         ret = register_pernet_subsys(&ip6_route_net_late_ops);
6556         if (ret)
6557                 goto fib6_rules_init;
6558
6559         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_NEWROUTE,
6560                                    inet6_rtm_newroute, NULL, 0);
6561         if (ret < 0)
6562                 goto out_register_late_subsys;
6563
6564         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_DELROUTE,
6565                                    inet6_rtm_delroute, NULL, 0);
6566         if (ret < 0)
6567                 goto out_register_late_subsys;
6568
6569         ret = rtnl_register_module(THIS_MODULE, PF_INET6, RTM_GETROUTE,
6570                                    inet6_rtm_getroute, NULL,
6571                                    RTNL_FLAG_DOIT_UNLOCKED);
6572         if (ret < 0)
6573                 goto out_register_late_subsys;
6574
6575         ret = register_netdevice_notifier(&ip6_route_dev_notifier);
6576         if (ret)
6577                 goto out_register_late_subsys;
6578
6579 #if IS_BUILTIN(CONFIG_IPV6)
6580 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6581         ret = bpf_iter_register();
6582         if (ret)
6583                 goto out_register_late_subsys;
6584 #endif
6585 #endif
6586
6587         for_each_possible_cpu(cpu) {
6588                 struct uncached_list *ul = per_cpu_ptr(&rt6_uncached_list, cpu);
6589
6590                 INIT_LIST_HEAD(&ul->head);
6591                 spin_lock_init(&ul->lock);
6592         }
6593
6594 out:
6595         return ret;
6596
6597 out_register_late_subsys:
6598         rtnl_unregister_all(PF_INET6);
6599         unregister_pernet_subsys(&ip6_route_net_late_ops);
6600 fib6_rules_init:
6601         fib6_rules_cleanup();
6602 xfrm6_init:
6603         xfrm6_fini();
6604 out_fib6_init:
6605         fib6_gc_cleanup();
6606 out_register_subsys:
6607         unregister_pernet_subsys(&ip6_route_net_ops);
6608 out_register_inetpeer:
6609         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6610 out_dst_entries:
6611         dst_entries_destroy(&ip6_dst_blackhole_ops);
6612 out_kmem_cache:
6613         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6614         goto out;
6615 }
6616
6617 void ip6_route_cleanup(void)
6618 {
6619 #if IS_BUILTIN(CONFIG_IPV6)
6620 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
6621         bpf_iter_unregister();
6622 #endif
6623 #endif
6624         unregister_netdevice_notifier(&ip6_route_dev_notifier);
6625         unregister_pernet_subsys(&ip6_route_net_late_ops);
6626         fib6_rules_cleanup();
6627         xfrm6_fini();
6628         fib6_gc_cleanup();
6629         unregister_pernet_subsys(&ipv6_inetpeer_ops);
6630         unregister_pernet_subsys(&ip6_route_net_ops);
6631         dst_entries_destroy(&ip6_dst_blackhole_ops);
6632         kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep);
6633 }